No title

WATER RESEARCH A Journal of the International Water Association

Editor-in-Chief Mark van Loosdrecht Delft University of Technology Department of Biochemical Engineering Julianalaan 67 2628 BC Delft The Netherlands Tel: +31 15 27 81618 E-mail:[email protected]

Editors J. Block Université H. Poincaré, Nancy I France David Dixon University of Melbourne Australia Hiroaki Furumai The University of Tokyo Japan Xiaodi Hao Beijing University of Civil Engineering and Architecture China Gregory Korshin University of Washington USA Anna Ledin Formas Sweden Eberhard Morgenroth Swiss Federal Institute of Aquatic Science and Technology (EAWAG) Switzerland W. Rauch University Innsbruck Austria Maria Reis Universidade Nova de Lisboa/FCT Portugal Hang-Shik Shin Korea Advanced Institute of Science and Technology Korea Thomas Ternes Bundesanstalt für Gewässerkunde Germany Stefan Wuertz Univ. of California, Davis USA

Associate Editors Andrew Baker University of New South Wales Australia

Damien Batstone The University of Queensland Australia G-H. Chen The Hong Kong University of Science & Technology Hong Kong China Tom Curtis Univ. of Newcastle upon Tyne UK Ana Deletic Monash University USA Francis de los Reyes III North Carolina State University USA Rob Eldridge The University of Melbourne Australia Rosina Girones University of Barcelona Spain Stephen Gray Victoria University Australia Kate Grudpan Chiang Mai University Thailand E.E. Herricks University of Illinois - Urbana USA Peter Hillis United Utilities Plc UK H-Y. Hu Tsinghua University China P.M. Huck University of Waterloo Canada Bruce Jefferson Cranfield University UK Ulf Jeppsson Lund University Sweden Sergey Kalyuzhnyi Moscow State University Russian Federation Jaehong Kim Georgia Institute of Technology USA Jes La Cour Jansen Lund Institute of Technology Sweden G. Langergraber BOKU/Univ. of Natural Res. and Applied Life Scs. Austria S-L. Lo National Taiwan University Taiwan Dionisis Mantzavinos Technical University of Crete Greece

Y. Matsui Hokkaido University Japan A. Maul Université Paul Verlaine-Metz France Max Maurer EAWAG Switzerland How Yong Ng National University of Singapore Singapore Satoshi Okabe Hokkaido University Japan S.L. Ong National University of Singapore Singapore Jong M. Park Pohang University of Science & Technology Korea Susan Richardson U.S. Environmental Protection Agency USA Miguel Salgot University of Barcelona Spain David Sedlak University of California, Berkeley USA Jean-Philippe Steyer LBE-INRA France M. Takahashi Hokkaido University Japan Kai Udert EAWAG Switzerland V.P. Venugopalan BARC Facilities India E. von Sperling Federal University of Minas Gerais Brazil Hanqing Yu University of Science & Technology of China China J. Zilles University of Illinois Urbana USA A.I. Zouboulis Aristotle University Greece

Editorial Office Mailing address Water Research Radarweg 29 1043 NX Amsterdam The Netherlands E-mail: [email protected]

Publication information: Water Research (ISSN 0043-1354). For 2011, volume 45 is scheduled for publication. Subscription prices are available upon ­request from the publisher or from the Elsevier Customer Service Department nearest you or from this journal’s website (http://www.elsevier.com/locate/watres). Further information is available on this journal and other Elsevier products through Elsevier’s website (http://www.elsevier.com). Subscriptions are accepted on a prepaid basis only and are entered on a calendar year basis. Issues are sent by standard mail (surface within Europe, air delivery outside Europe). Priority rates are available upon request. Claims for missing issues should be made within six months of the date of dispatch. Author Enquiries: For enquiries relating to the submission of articles (including electronic submission) please visit this journal’s homepage at http://www.elsevier.com/ locate/watres. Contact details for questions arising after acceptance of an article, especially those relating to proofs, will be provided by the publisher. You can track accepted articles at http://www.elsevier.com/trackarticle. You can also check our Author FAQs at http://www.elsevier.com/authorFAQ and/or contact Customer Support via http://support.elsevier.com. Orders, claims, and journal enquiries: Please contact the Elsevier Customer Service Department nearest you: St. Louis: Elsevier Customer Service Department, 3251 Riverport Lane, Maryland Heights, MO 63043, USA; phone: (877) 8397126 [toll free within the USA]; (+1) (314) 4478878 [outside the USA]; fax: (+1) (314) 4478077; e-mail: [email protected] Oxford: Elsevier Customer Service Department, The Boulevard, Langford Lane, Kidlington OX5 1GB, UK; phone: (+44) (1865) 843434; fax: (+44) (1865) 843970; e-mail: [email protected] Tokyo: Elsevier Customer Service Department, 4F Higashi-Azabu, 1-Chome Bldg, 1-9-15 Higashi-Azabu, Minato-ku, Tokyo 106-0044, Japan; phone: (+81) (3) 5561 5037; fax: (+81) (3) 5561 5047; e-mail: [email protected] Singapore: Elsevier Customer Service Department, 3 Killiney Road, #08-01 Winsland House I, Singapore 239519; phone: (+65) 63490222; fax: (+65) 67331510; e-mail: [email protected] Application for membership of International Water Association should be made to: Executive Director, IWA, Alliance House, 12 Caxton Street, London SW1H 0QS, U.K. (Tel.: +44 207 654 5500; Fax: +44 207 654 5555; e-mail: [email protected]; website: http://www.IWAhq.org.uk). Registered Charity (England) No. 289269. Individual membership is available from £30 upwards. For details contact IWA.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Review

Occurrence and fate of bulk organic matter and pharmaceutically active compounds in managed aquifer recharge: A review Sung Kyu Maeng a,*, Saroj K. Sharma b, Karin Lekkerkerker-Teunissen c,d, Gary L. Amy e a

Water Environment Center, Korea Institute of Science and Technology, P.O. Box. 131, Cheongryang, Seoul 130-650, South Korea UNESCO-IHE Institute for Water Education, P.O. Box 3015, 2601 DA Delft, The Netherlands c Technical University of Delft, Stevinweg 1, 2628 CN Delft, The Netherlands d Dunea, P.O. Box 34, 2270 AA, Voorburg, The Netherlands e King Abdullah University of Science and Technology, Thuwal 23955-6000, Saudi Arabia b

article info

abstract

Article history:

Managed aquifer recharge (MAR) is a natural water treatment process that induces surface

Received 23 August 2010

water to flow in response to a hydraulic gradient through soil/sediment and into a vertical or

Received in revised form

horizontal well. It is a relatively cost-effective, robust and sustainable technology. Detailed

7 February 2011

characteristics of bulk organic matter and the occurrence and fate of pharmaceutically active

Accepted 12 February 2011

compounds (PhACs) during MAR processes such as bank filtration (BF) and artificial recharge

Available online 19 February 2011

(AR) were reviewed. Understanding the fate of bulk organic matter during BF and AR is an essential step in determining pre- and/or post-treatment requirements. Analysis of organic

Keywords:

matter characteristics using a suite of analytical tools suggests that there is a preferential

Artificial recharge

removal of non-humic substances during MAR. Different classes of PhACs were found to

Bank filtration

behave differently during BF and AR. Antibiotics, non-steroidal anti-inflammatory drugs

Bulk organic matter

(NSAIDs), beta blockers, and steroid hormones generally exhibited good removal efficiencies,

Pharmaceutically active compounds

especially for compounds having hydrophobic-neutral characteristics. However, anticonvulsants showed a persistent behavior during soil passage. There were also some redoxdependent PhACs. For example, X-ray contrast agents measured, as adsorbable organic iodine (AOI), and sulfamethoxazole (an antibiotic) degraded more favorably under anoxic conditions compared to oxic conditions. Phenazone-type pharmaceuticals (NSAIDs) exhibited better removal under oxic conditions. The redox transition from oxic to anoxic conditions during soil passage can enhance the removal of PhACs that are sensitive to redox conditions. In general, BF and AR can be included in a multi-barrier treatment system for the removal of PhACs. ª 2011 Published by Elsevier Ltd.

Contents 1. 2. 3.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3016 Managed aquifer recharge systems: bank filtration and artificial recharge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3018 Factors influencing removal of PhACs during MAR system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3019

* Corresponding author. Tel.: þ82 2 958 6769; fax: þ82 2 958 6854. E-mail address: [email protected] (S.K. Maeng). 0043-1354/$ e see front matter ª 2011 Published by Elsevier Ltd. doi:10.1016/j.watres.2011.02.017

3016

4.

5.

6. 7.

1.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

3.1. Effects of the properties of PhACs on biodegradation in managed aquifer recharge . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Effect of the properties of PhACs on sorption in managed aquifer recharge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Occurrence, fate, and removal of bulk organic matter in managed aquifer recharge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. Total organic carbon and dissolved organic carbon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2. Specific UV absorbance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3. XAD resin fractionation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Size exclusion chromatography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5. Fluorescence excitationeemission matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6. Polarity rapid assessment method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Occurrence, fate, and removal of organic micropollutants in managed aquifer recharge . . . . . . . . . . . . . . . . . . . . . . . . . 5.1. Antibiotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2. Non-steroidal anti-inflammatory drug (NSAIDs) and analgesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3. Anticonvulsants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4. Antidepressants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5. Beta blockers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6. Lipid regulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.7. X-ray contrast media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.8. Steroid hormones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Practical implications of MAR and further research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Introduction

Bulk organic matter in natural waters mainly consists of natural organic matter (NOM), derived from allochthonous and autochthonous sources. NOM does not pose a direct health threat to humans with respect to drinking water quality, but it is a precursor of organic disinfection by-products (DBP) (Siddiqui et al., 2000; Lu et al., 2009). In wastewater effluenteimpacted (or wastewater effluentedominated) surface waters, bulk organic matter is a mixture of NOM and effluent organic matter (EfOM), which originates from wastewater treatment plants. While EfOM has not been studied as extensively as NOM has, EfOM is composed of different types of organics: refractory compounds, residual degradable substrates, intermediates, complex organic compounds, and soluble microbial products (SMPs) (Barker and Stuckey, 1999). SMPs are biodegradable organic matter produced from substrate metabolism and biomass decay, and they are known as major foulants for reverse osmosis (RO), nanofiltration (NF), and ultrafiltration (UF) membranes (Jarusutthirak and Amy, 2006). Moreover, SMPs are precursor materials of nitrogenous disinfection by-products (N-DBPs) and can lead to bacterial regrowth in drinking water distribution systems (Amy and Drewes, 2007). EfOM also consists of emerging contaminants such as pharmaceutically active compounds (PhACs), endocrine disrupting compounds (EDCs), and personal care products (PCPs). PhACs and transformation products enter surface water primarily through discharged effluent from wastewater treatment plants resulting from patient excretion in both urine and feces (Cunningham et al., 2006; Zhou et al., 2009). Previous studies conducted by Kasprzyk-Hordern et al. (2009) and Zhou et al. (2009) demonstrated the impact of wastewater effluent containing organic micropollutants (e.g.,

3019 3019 3019 3019 3020 3020 3020 3022 3022 3022 3022 3023 3026 3026 3027 3027 3027 3028 3028 3029 3029 3029

PhACs) on the quality of receiving waters. Non-point sources, such as overland flow (i.e., runoff) during rainfall or land drainage in agricultural areas, also deliver PhACs (e.g., veterinary medicines) to surface water or groundwater (Boxall et al., 2004). These situations create the possibility for the occurrence of organic micropollutants such as PhACs, PCPs, and EDCs in drinking water sources. Currently, the total consumption of PhACs and PCPs in the world is not known because many of these compounds significantly vary with respect to application and consumption from one country to another (Cunningham, 2004). Many of them are slightly transformed or unchanged during municipal wastewater treatment (Chefetz et al., 2008). Moreover, the growing use of PhACs, EDCs, and PCPs for human and veterinary purposes has contributed to their frequent detection in the aquatic environment and in wastewater (Heberer, 2002; Tixier et al., 2003; Jjemba, 2006). Growing concern over the safety of drinking water containing PhACs, EDCs and PCPs has resulted in increased research worldwide (Mechlinski and Heberer, 2005; Kim et al., 2007; Ku¨mmerer, 2009; Madden et al., 2009; Mompelat et al., 2009). Many water utilities in developed countries are adopting advanced water treatment processes to provide a reliable supply of safe drinking water. However, little is known about the fate of transformation products formed in drinking water treatment processes such as advanced oxidation processes (AOPs) and biodegradation (Mompelat et al., 2009). Snyder et al. (2007) evaluated the removal of EDCs and PhACs in 13 full-scale water treatment facilities. Conventional coagulation, flocculation, and sedimentation processes were ineffective at removing most of the target EDCs and PCPs. Slow sand filtration and flocculation by iron (III) chloride were also ineffective for selected pharmaceuticals (bezafibrate, clofibric acid, carbamazepine, and diclofenac) (Ternes et al.,

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

2002). However, AOPs (e.g., UV/H2O2 and ozone/H2O2) were highly effective at removing many organic micropollutants (Gebhardt and Schro¨er, 2007; Snyder et al., 2007; Klavarioti et al., 2009). NF and RO membrane systems have been known to be effective processes for removing pharmaceutical residues (Yoon et al., 2006; Yangali-Quintanilla et al., 2009). Radjenovic et al. (2008) demonstrated that NF and RO membranes appear to be very effective for most of the pharmaceutical residues in a full-scale drinking water treatment plant. However, construction and maintenance costs for both membrane and AOP systems are factors that limit implementation in a large treatment facility. Retrofitting conventional water treatment facilities with membrane or AOP systems to remove pharmaceutical residues is also relatively costly; this leads to a high unit treatment cost of water, which makes water unaffordable. In contrast, managed aquifer recharge (MAR) processes are robust and cost-effective systems for obtaining a safe water supply, and they include a wide variety of systems for different applications (e.g., aquifer storage and recovery, aquifer storage and transfer and recovery, BF, dune infiltration, infiltration ponds, percolation tanks, soil aquifer treatment, sand dams, underground dams and recharge releases) (Dillon, 2005). MAR systems such as BF and AR (artificial recharge) systems are generally adopted by water utilities if their raw water quality (e.g., surface water) is inadequate or if the amount of raw water (e.g., groundwater) is not sufficient. Often, water utilities using BF and AR are located downstream from municipal wastewater treatment plants, especially those water utilities that use a river that flows through many cities or countries (e.g., Rhine River, Germany). In this case, it is nearly impossible to withdraw raw water that is not affected by wastewater effluent, and the fraction of wastewater effluent in the river can be varied. For example, BF facilities located in Berlin, Germany are managed by Berlin Water Works (Berliner Wasserbetriebe); their source water (Lake Tegel) is influenced by well-treated domestic wastewater effluent between 15 and 30% (Jekel and Gru¨nheid,   et al. (1996) investigated a BF 2005; Pekdeger, 2006). Cosovi c site (Zagreb, Croatia) under extreme conditions in which the chemical oxygen demand (COD) of the surface water was several thousand mg O2/L, most of which is biodegradable organic matter from the local yeast industry and from the pharmaceutical industry, which produces antibiotics and synthetic organic compounds. If source water for BF and AR systems contains EfOM that originates from a wastewater treatment plant, it is necessary to assess the characteristics of the bulk organic matter and the fate of pharmaceuticals and transformation products in order to determine the pre- and/or post-treatment requirements of BF and AR. Recalcitrant wastewater-derived contaminants (PhACs) such as gemfibrozil and sulfamethoxazole were more difficult to remove in biodegradable dissolved organic carbon (BDOC) derived from wastewater effluentedominated surface water compared to BDOC derived from aquatic plants, suggesting that the source (i.e., the characteristics) of the BDOC also needs to be considered (Lim and Snyder, 2008). BDOC derived from wastewater effluent may have different effects on the removal of PhACs. Therefore, the amount of BDOC, which is a limiting factor for co-metabolism, is not the only important factor for enhancing the removal (i.e., transformation) of PhACs; the

3017

type of BDOC is also significant. Moreover, Rauch-Williams et al. (2010) explored the role of organic matter in the removal of emerging contaminants during soil passage and found that the concentration and characteristics of bulk organic carbon (colloidal organic carbon, hydrophobic acids, and hydrophilic carbon) affect the removal of contaminants, suggesting that organic carbon characteristics affect the performance of MAR in regards to contaminant removal. However, further study is required to determine whether the organic matter characteristics control the removal of contaminants by increasing the diversity of microorganisms or their activity (biomass). It is also important to determine if microorganisms in the aquatic environment are able to use the very low contaminant concentrations (ng/L) remaining in the environment for their growth (i.e., as carbon sources). Some microorganisms degrade PhACs more efficiently than others. For example, the removal of ten selected PhACs using an enriched nitrifier culture was higher than the removal achieved by microorganisms from conventional activated sludge processes; this was confirmed through the by addition of inhibitors (Tran et al., 2009). It is important to note that these selected PhACs are removed by biotransformation through co-metabolism or are used as a carbon source for a growth substrate. Kagle et al. (2009) summarized bacterial strains and consortia that grow on PCPs, EDCs, and PhACs; these strains consist mostly of bacteria that survive under oxic conditions only (aerobic bacteria). In the case of biotransformation (i.e., co-metabolism), the pathways of the PhACs are important to know in relation to residuals (i.e., the biotransformation products) remaining in the aquatic environment. However, pathway information for the biodegradation of PhACs is limited. In addition to studying bacterial isolates, mixed cultures, and consortia (microbial community structures) capable of growth on PhACs (a growth substrate), it is also important to investigate microbial activity or biomass associated with aquifer materials. There are various methods available, and Rauch-Williams and Drewes (2005) summarized them and grouped them into five categories of biomass detection methods available for soil: direct total counts by microscopy, extraction of cell constituents, molecular biological methods, enzyme activities and cultural methods. More detailed methods of soil biomass detection are described in the previous study (Rauch-Williams and Drewes, 2005), and various have been practiced in MAR systems (La˚ngmark et al., 2004; Maeng et al., 2008; Rauch-Williams and Drewes, 2006). As surface water infiltrates through the riverbed sediments and aquifer materials, most suspended solids, biodegradable organic compounds and other contaminants are removed (Ray, 2008). The main mechanisms for the removal of PhACs in MAR processes are biodegradation and sorption. Biodegradation and sorption are considered to be important mechanisms during soil passage (Schoenheinz et al., 2005). The most important desirable removal mechanism is biodegradation, because it is a sustainable process and can result in endproducts consisting of inorganic compounds (i.e., the process of mineralization) (Howard, 2000). Biodegradation is also an important mechanism in the removal of PhACs during wastewater treatment processes, but some PhACs are discharged into the aquatic environment as a result of incomplete removal by wastewater treatment plants. Another

3018

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

important mechanism for the removal of PhACs during MAR is sorption, but this is a non-sustainable process because the sorption site becomes exhausted or desorption occurs. Moreover, retardation of an organic micropollutant is important to consider because there is a reduction in transport of the pollutant through an aquifer due to sorption and desorption of the aquifer matrices. For example, the retardation value of carbamazepine is 1.84 (Scheytt et al., 2006), this compound will never be detected in a production well at the travel time of bulk water. Carbamazepine will reach a production well after a duration that is about 1.84 times the bulk water travel time. Thus, the retardation value of a target compound during soil passage must be known in order to allow for proper investigation of the performance of MAR systems related to the removal of PhACs. However, some polar and refractory PhACs are often detected in infiltrated water, necessitating monitoring of their fate and mobility, and their levels are often highly dependent on the physicochemical properties of PhACs. For proper understanding of the removal of PhACs during soil passage, the role of the physicochemical properties of PhACs with respect to biodegradation and sorption must be well understood. In this review, the occurrence and removal of bulk organic matter and PhACs are evaluated in reference to MAR systems (i.e., BF and AR). Understanding the behavior of bulk organic matter during soil passage provides an insight into the factors that affect the removal of PhACs. Physicochemical properties of bulk organic matter and PhACs in relation to BF and AR systems are also addressed. This review collects data from full-scale studies carried out at BF and AR field sites around the world; if no data are available from field sites, data from laboratory-scale and pilot studies are used. Finally, suggestions are made regarding the practical implications of MAR and the further research required with respect to understanding the fate of PhACs during soil passage.

2. Managed aquifer recharge systems: bank filtration and artificial recharge BF is a robust and cost-effective treatment technology that has been proven to be an excellent option for the attenuation of organic micropollutants often found in surface waters (Schmidt et al., 2007). This technology offers several advantages over (direct) surface water intake resulting from its ability to remove suspended solids, biodegradable organic matter, bacteria, viruses, and parasites and to achieve partial elimination of adsorbable compounds through mixing, biodegradation, and sorption (Hiscock and Grischek, 2002). Water utilities in the Netherlands and some in Germany that use BF or AR as a water treatment process have been supplying drinking water without a disinfection process (e.g., chlorination). This is not possible at all BF sites on account of the site-specific differences in source water qualities, hydrogeological conditions, residence times, and travel distances. Recently, BF and AR have become attractive as processes that are as part of protective multi-barrier treatment for emerging contaminants (e.g., PhACs, PCPs, and EDCs) and that reduce organic/biological fouling of membrane systems (Nederlof et al., 2000; Speth et al., 2002). BF has been shown to be a robust system against chemical spills or accidents; it is

capable of reducing the pollutants through sorption, biodegradation, and mixing (Ray et al., 2002c; Schmidt et al., 2007). All of these benefits make MAR systems such as BF and AR more attractive than other treatment processes. The hydraulic gradient between the river/lake/infiltration basin and a production well is created by pumping from the production well, and the gradient induces infiltrated water from the river/lake/infiltration basin to move toward one or more production wells resulting in a mixture of groundwater originally in the aquifer and the infiltrated water from the river/lake/infiltration basin (Schmidt et al., 2003). Naturally, during flooding conditions, the elevated level of the river water causes it to move toward the aquifer under nonpumping conditions (Ray et al., 2002c). The degree of hydraulic connection between surface waters and the aquifer is an important factor that determines the feasibility of the process and the location of pumping well. MAR processes have been studied and practiced in (1) Europe (Bourg and Bertin, 1993; Hiscock and Grischek, 2002; Irmscher and Teermann, 2002; Hiemstra et al., 2003; Eckert and Irmscher, 2006; Kedziorek et al., 2008; Maeng et al., 2010); (2) North America (Ray et al., 2002b; Gupta et al., 2009); (3) Africa (Shamrukh and Abdel-Wahab, 2008); and (4) Asia (Lee et al., 2009; Wu et al., 2007). BF has been applied for over a century in some parts of Central Europe, especially Germany (Irmscher and Teermann, 2002; Ray et al., 2002a; Tufenkji et al., 2002; Jekel and Gru¨nheid, 2003; Eckert and Irmscher, 2006; Ray, 2008). It is a popular treatment process in Europe (primarily in Germany, the Netherlands, France, and Hungary), and the majority of MAR sites (BF and AR) are located in Europe (70%), followed by North America (23%), Asia (4%), Australia (2%), and Africa (1%) (Gru¨tzmacher et al., 2009). In Europe, BF and AR are considered to be major components in drinking water treatment, but in North America, BF and AR have been applied mainly as pre-treatments intended to reduce treatment costs by lowering chemical and energy requirements for the removal of contaminants like pathogens, particles, suspended matter and some dissolved organic carbon (Gru¨nheid et al., 2005). Weiss et al. (2005) demonstrated the potential for bank filtration as a significant barrier to the transport of Cryptosporidium and Giardia from surface water sources to the production wells. BF or AR sites in North America and Europe generally have favorable conditions, such as a high-quality water source (e.g., rivers and lakes) and a site with well-defined hydrogeological conditions. However, this is not the case in many developing countries, where rivers are often contaminated by wastes from many different sources (e.g., agricultural runoff and municipal wastewaters) and where hydrogeological information is lacking. Recently, many cities from developing or recently developed countries have begun investigating the feasibility of MAR for their part of the water treatment processes. German scientists have carried out field investigations of three potential sites for bank filtration in Delhi, India through the EU co-funded project TECHNEAU (Gru¨tzmacher et al., 2009). Gru¨tzmacher et al. (2009) reviewed MAR sites (AR and BF) used for water treatment or pre-treatment around the world. Since 2006, the city of Changwon, along the Nakdong River in South Korea, has been extracting 60,000 m3/day for its drinking water supply through BF

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

systems using vertical and horizontal collector wells. The city of Gimhae, South Korea, is installing a BF system that will produce 180,000 m3/day. Moreover, K-Water, the public water utility company in South Korea, is currently investigating potential BF sites that could supply 280,000 m3/day to cities located near the Nakdong River. More information on potential MAR sites all over the world is available from Ray (2008) and Gru¨tzmacher et al. (2009), who summarized the worldwide potential sites for riverbank filtration; Ray’s work focuses on developing countries and the work of Gru¨tzmacher et al. focuses on current worldwide MAR (BF and AR) sites.

3. Factors influencing removal of PhACs during MAR system 3.1. Effects of the properties of PhACs on biodegradation in managed aquifer recharge The removal of PhACs by biodegradation is often considered an important mechanism during soil passage, contributing to compound attenuation. Biodegradation depends on the characteristics of the PhACs and on the specific biomass, microbial activity, and microbes present in the soil. The complete conversion of organic compounds by biodegradation into inorganic products (i.e., mineralization) in waters and soils is mostly attributable to microbial activity (Alexander, 1981). The potential for PhACs to be biotransformed or mineralized (i.e., ultimate biodegradation) is an important aspect of their fate during soil passage. Both chemical structures and physicochemical properties significantly influence the degree of biodegradation during soil passage. The relative biodegradability of an organic compound based on its structural properties was summarized by Howard (2000). This study summarized seven different structural properties (namely, branching, aliphatic functional groups, aromatic functional groups, aliphatic amines, halophenols, polycyclic aromatics, and triazines) that determine the biodegradability of a compound. For example, electrone withdrawing aromatic substituents, such as nitro groups and halogens, decrease biodegradability and make it more difficult for enzymes to degrade a compound. In contrast, the aromatic substituents with donating functionalities (e.g., carboxylic acids and amines) increase biodegradability (Howard, 2000). Next to PhACs properties, also the local biomass conditions influence the biodegradation potential. Key factors to be considered evolve specific microbes present, analyzed for example by DNA identification; microbial activity often measured as ATP; available carbon source, meausured as BDOC and redox condiotions.

3.2. Effect of the properties of PhACs on sorption in managed aquifer recharge The removal of PhACs by sorption is also an important mechanism during soil passage because it contributes to compound attenuation. The organic carbon partition coefficient, Koc, is often used to assess the sorption and distribution behavior of non-polar organic micropollutants in soils and sediments. The organic carbon content of soils and sediments greatly contributes to the sorption of non-polar PhACs (e.g., pesticides and

3019

industrial chemicals). However, Koc may not properly describe the distribution behavior between soil and water for some PhACs that contain charged sites and that exist as ionic compounds in the aquatic environment (Cunningham, 2004). The estimated log Kow of bezafibrate is 4.25 (US EPA, 2009). Values of log Kow greater than 3 indicate that a compound is more likely to partition to soil or sediment, but this is not always true for ionic compounds. The estimated log Dow of bezafibrate at pH 8 is 0.69 (ADME/Tox Web Software). According to Cunningham (2004), a chemical with log Dow value less than 1 is unlikely to sorb or bioconcentrate on organic matter. The degree of ionization becomes a key factor in sorption mechanisms for acidic pharmaceuticals such as non-steroidal anti-inflammatory drugs and lipid regulators. Therefore, it is important to know the acid dissociation constants (pKa) of PhACs which control the degree of ionization; the pH must also be considered when estimating parameters that influence the fate of acidic PhACs during soil passage.

4. Occurrence, fate, and removal of bulk organic matter in managed aquifer recharge There are many factors that affect the fate of organic micropollutants during soil passage. Hydrogeology (e.g., sediment porosity, permeability and groundwater flow) geochemical and nutrient conditions, temperature, and redox conditions are important factors identified by previous studies (Massmann et al., 2006, 2008; Dı´az-Cruz and Barcelo´, 2008). Our investigation indicates that insufficient attention has been given to the effects of bulk organic matter characteristics on the removal of PhACs. Recently, Rauch-Williams et al. (2010) revealed that organic matter characteristics affect the removal of organic micropollutants during managed aquifer recharge. Consequently, this study collected a suite of both routine and innovative analytical tools used for investigating the characteristics of bulk organic matter from MAR sites around the world and explained its behavior during soil passage.

4.1.

Total organic carbon and dissolved organic carbon

Chemical oxygen demand (CODCr) and five-day biochemical oxygen demand (BOD5) are still widely used to determine total organic matter and biodegradable organic matter, respectively. However, there are very limited CODCr or BOD5 data available from previous studies conducted at MAR sites. Currently, total organic carbon (TOC) and dissolved organic carbon (DOC) (i.e., elemental analysis of organic matter) are more often used to measure total organic matter in water for many applications, especially water research. Subtracting DOC values of BF and AR filtrates from source water can be used to determine the removal of biodegradable organic carbon during soil passage. However, DOC reduction was not always utilized by microorganisms for energy or microbial growth; rather the DOC was merely adsorbed onto aquifer materials such as sand. It is necessary to determine adenosine triphosphate (ATP) concentration or other indicators that measure the microbial activity in a sample. Trulleyova´ and Rulı´k (2004) showed that there will be a certain overestimation in biodegradable organic carbon results if some

3020

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

part of the adsorbed DOC is resistant to biodegradation. During soil passage from source water to a production well, bulk organic matter is attenuated through physical, chemical, and biological processes (i.e., filtration, sorption and biodegradation). The fate of bulk organic matter during soil passage can be characterized using a suite of both routine and innovative analytical tools. Generally, TOC and DOC removal efficiencies range from 30 to 88% and 33 to 88%, respectively (Table 1). The removal of TOC and DOC dominantly occurs within the first few meters of infiltration and is mainly due to biodegradation (i.e., a biologically active colmation layer) (Hiscock and Grischek, 2002; Quanrud et al., 2003). Biocides such as sodium azide can be used to isolate biodegradation and determine the role of biodegradation in the removal of organic matter during soil passage, but this process cannot be used at MAR field sites. Many factors control the biodegradation of organic matter during soil passage (e.g., soil biomass and redox conditions); thus, the reduction of organic matter resulting from biodegradation varies from site to site.

4.2.

Specific UV absorbance

Specific UV absorbance (SUVA) is the ratio of DOC to UV absorbance at 254 nm; it represents the relative aromaticity of organic matter (Amy and Drewes, 2007). SUVA was observed to increase from 2.1 L/m-mg to 2.4 L/m-mg during infiltration under oxic conditions at the Lake Tegel BF site (Berlin, Germany). This increase suggested a preferential removal of aliphatic organic matter during soil passage (Gru¨nheid et al., 2005). At a BF site in Parkville (Missouri, USA), the average SUVA from the production wells was higher than that of the surface water (Missouri river) during one month of monitoring (Weiss et al., 2004). The Sweetwater recharge facilities located in Tucson (Arizona, USA) monitored SUVA from 1996 to 2001; the SUVA increased by 30e80% during infiltration through the first meter of infiltration basin sediments (Quanrud et al., 2003). Cha et al. (2004) and Maeng et al. (2008) used river water and canal water, respectively, to investigate the fate of bulk organic matter during BF using columns. They also reported that SUVA values increased with travel distances and residence times. In contrast, SUVA values from BF sites in Jeffersonville and Terre Haute (USA) were highly variable, and exhibited no consistent trends (Weiss et al., 2004). However, other NOM characterization tools described below have demonstrated the preferential removal of non-humic substances (i.e., aliphatic organic matter) during soil passage. Thus, it can be concluded that aliphatic organic matter is preferentially removed during BF and AR.

4.3.

XAD resin fractionation

XAD resin fractionation is a method to distinguish different fractions of NOM based on chromatography. The XAD-8 resin column separates NOM into hydrophobic and hydrophilic fractions. The XAD-8 fractionation is often used to characterize NOM fractions in many applications. Weiss et al. (2004) investigated the change in NOM fractions during RBF using XAD-8 columns, and they suggested that there was no clear removal relationship from different NOM fractions. Removal efficiencies of the hydrophilic fraction from three BF sites in

the USA (Jeffersonville, Terre Haute, and Parkville) ranged from 40 to 70%, and removal efficiencies of the hydrophobic fraction ranged from 35 to 60% (Weiss et al., 2004). Column studies conducted by Rauch-Williams and Drewes (2004) found that hydrophilic organic matter in secondary effluent isolated using chromatographic resin (XAD-8) was removed more effectively than was hydrophobic organic matter. Xue et al. (2009) used XAD-8/XAD-4 resin chromatography to fractionate five classes of NOM, and they showed that 80% of the hydrophilic organic matter was removed by aerobic biodegradation. They found that hydrophilic organic matter exhibits the highest biodegradability compared to the other four NOM fractions (i.e., hydrophobic acid, hydrophobic neutral, transphilic acid and transphilic neutral). During a five year program conducted at the Sweetwater recharge facilities (Tucson, Arizona, USA), hydrophilic organic matter was also preferentially removed during soil passage (Quanrud et al., 2003). Based on the results obtained using XAD-8 and XAD-4 resin chromatography, it can be concluded that hydrophilic organic matter is preferentially removed during soil passage. Hydrophilic organic matter contains a relatively high amount of aliphatic organic matter (Rauch-Williams and Drewes, 2004), as indicated by the SUVA results discussed in the previous section; both approaches suggest that aliphatic organic matter is preferentially removed.

4.4.

Size exclusion chromatography

An LC-OCD (Liquid ChromatographyeOrganic Carbon Detector) system(i.e., size exclusion chromatography) using a liquid chromatography method describes the molecular weight (MW) distribution and classification of organic matter according to biopolymers, humic substances, building blocks, neutrals and low MW acids. These fractions are quantified according to organic carbon concentration using an organic carbon detector and characterized by a UV detector. Additional details regarding this system are provided in Huber and Frimmel (1992). The major NOM fraction change during soil passage (Lake Tegel site, Berlin, Germany) was the biopolymer fraction, which is the largest MW fraction (MW > 20,000 Da) in an LC-OCD chromatogram and which is comprised of nonhumic substances (Gru¨nheid and Jekel, 2005; Jekel and Gru¨nheid, 2005). Samples from the production well at the Lake Tegel BF site showed almost complete removal of the biopolymer fraction, even at the monitoring well closest to the lake. An AR site located at Lake Tegel, Berlin (Germany) treats the same lake water as does the BF site, but it uses infiltration basins instead of extraction from the lake caused by a hydraulic gradient (Gru¨nheid et al., 2005). Again, a significant amount of biopolymer fraction was removed, and other fractions were partially removed. Kolehmainen et al. (2007) suggested that, during AR, large MW fractions in the river water were removed more efficiently than were the smaller fractions. Thus, significant removal of the biopolymer fraction in infiltrated water was observed. The biopolymer fraction is comprised of easily biodegradable (i.e., non-humic) organic matter such as proteins and polysaccharides. The preferential removal of the biopolymer (non-humic) fraction corresponds to the increase of the SUVA, which was due to the removal of aliphatic organic matter during soil passage.

Table 1 e Overview of TOC and DOC removal efficiencies in managed aquifer recharge systems worldwide. Co, (TOC/DOC) mg/L

TOC removal (%)

Well type

14/ 9/ 6/ 2.9/ 3.0/2.7 3.0/2.7 4.7/4.1 4.7/4.1 4.5/3.6 4.5/3.6 /1.5e7 /7.5 /7.2e7.5 /7.2e7.5 /4.3 /14.1

88 77 73 30 60 75 67 88 41 40

11

ARc ARc ARc RBFa RBFa RBFa RBFa RBFa RBFa RBFa RBFa LBFb LBFb ARc RBFa ARc

33 58 74 64 88 35 36 71 42 34e40 34e40 40 66

Vd Vd Vd He Vd Vd He Vd Vd Vd Vd Vd Vd Vd He Vd

35

35

ARc

/14.1

93

Vd

(Amy and Drewes, 2007)

388

6e18 (month)

ARc

/6.10

76

Vd

(Amy and Drewes, 2007)

655

6e19 (month)

ARc

/6.10

71

Vd

(Amy and Drewes, 2007)

885

6e20 (month)

ARc

/6.10

75

Vd

(Amy and Drewes, 2007)

1950

12e96 (month)

ARc

/6.10

81

Vd

(Amy and Drewes, 2007)

1950

12e96 (month)

ARc

/6.10

88

Vd

(Amy and Drewes, 2007)

2700

12e96 (month)

ARc

/6.10

82

Vd

(Amy and Drewes, 2007)

Ha¨meenlinna, Finland Jyva¨skyla¨, Finland Tuusula, Finland Louisville, Kentucky, USA Jeffersonville, Indiana, USA Jeffersonville, Indiana, USA Terre Haute, Indiana, USA Terre Haute, Indiana, USA Parkville, Missouri, USA Parkville, Missouri, USA Pembroke, New Hampshire, USA Berlin (Lake Tegel), Germany Berlin (Lake Tegel), Germany Berlin (Lake Tegel), Germany Du¨sseldorf, Germany Monitoring well-MW5, Tuscon, Arizona, USA Monitoring well WR199, Tuscon, Arizona, USA Monitoring well-NW4, Mesa, Arizona, USA Monitoring well-NW3, Mesa, Arizona, USA Monitoring well-NW2, Mesa, Arizona, USA Monitoring well-10U, Mesa, Arizona, USA Monitoring well-26U, Mesa, Arizona, USA Monitoring well-44U, Mesa, Arizona, USA

1000e1300 200e550 500e700 30.5 61 177 27 122 37 24 55 100 77 32 50 6

90 15e30 30e60 120 3e5 13e19 13e19 e e e 5 135 117 50

a b c d e

Residence time (day)

Capacity (m3/s)

References

DOC removal (%)

Distance (meter)

0.88 0.23 0.23 0.53 0.044 0.075 0.075

(Kolehmainen et al., 2007) (Kolehmainen et al., 2007) (Kolehmainen et al., 2007) (Wang et al., 2002) (Weiss et al., 2004) (Weiss et al., 2004) (Weiss et al., 2004) (Weiss et al., 2004) (Weiss et al., 2004) (Weiss et al., 2004) (Partinoudi and Collins, 2007) (Grunheid et al., 2005) (Grunheid et al., 2005) (Grunheid et al., 2005) (Schubert, 2002) (Amy and Drewes, 2007)

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Types

Site

RBF: riverbank filtration. LBF: lake bank filtration. AR: artificial recharge. V: vertical well. H: horizontal well (radial collector well).

3021

3022

4.5.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Fluorescence excitationeemission matrices

Fluorescence excitation (ex) e emission (em) matrices (F-EEM) provide fluorescence intensity (FI) peaks at known wavelengths of humic-like peaks P1 (ex/em ¼ 250e260/380e480 nm) and P2 (ex/em ¼ 330e350/420e480 nm) and protein-like peak, P3 (ex/ em ¼ 270e280/320e350 nm) (Leenheer and Croue, 2003). Several studies have demonstrated that P1, P2, and P3 are associated with humic-like and protein-like substances (Coble, 1996; Chen et al., 2003; Henderson et al., 2009). Column studies simulating BF showed that there was a preferential removal of protein-like substances compared to the removal of humic-like substances during infiltration (Maeng et al., 2008). Previous studies have also reported that significant amounts of proteinlike substances were attenuated during soil passage using reclaimed wastewater (Amy and Drewes, 2007; Xue et al., 2009). These results correspond to LC-OCD results from Lake Tegel, which indicated that the biopolymer fraction (e.g., protein-like substances) was effectively removed by BF. On the basis of the results of SUVA, F-EEM, and LC-OCD, it can be concluded that there is a preferential removal of non-humic substances during soil passage, and this removal is attributed to microbial activity.

4.6.

Polarity rapid assessment method

The polarity rapid assessment method (PRAM) characterizes the polarity of bulk organic matter by measuring the fraction of substances adsorbed onto different solid-phase extraction (SPE) sorbents (Rosario-Ortiz et al., 2004, 2007a, 2009). Nonpolar sorbents (C18, C8 and C2), polar sorbents (CN, silica and diol) and anionic sorbents (NH2 and SAX) are commonly used to characterize the polarity of bulk organic matter under ambient conditions. Non-polar sorbents and polar sorbents extract hydrophobic organic matter and hydrophilic organic matter, respectively. Anionic sorbents extract negatively charged organic matter. PRAM uses the normalized UV absorbance breakthrough curve to determine the amount of total materials adsorbed, which is defined as a retention coefficient (RC). RC is expressed as 1  Cmax/C0 (Cmax is the maximum absorbance after breakthrough and C0 is the absorbance of the original sample) (Rosario-Ortiz et al., 2007b; Philibert et al., 2008). Using different properties of the sorbents, PRAM analysis depicts the different degrees of polarity in bulk organic matter originating from different sources or water treatment steps. Since the recent introduction of PRAM, there have been only a few published data on its use in studies related to soil passage. PRAM was carried out in a study using soil columns (Maeng et al., 2008). Three different solvents (C18, silica, and amino) were used for samples derived from column studies simulating BF. PRAM demonstrated that the non-polar character of organic matter (i.e., the hydrophobic component) slightly decreased during infiltration using column studies simulating BF, with fractions in influent and effluent samples corresponding to 20 and 15%, respectively. The uncharged polar character (hydrophilic organic matter) significantly decreased during infiltration, which corresponds to a reduction in the biopolymer fraction (i.e., hydrophilic-neutral) determined by LC-OCD analysis, as explained above. The

anionic fraction of NOM slightly increased after infiltration as a result of a reduction of neutral organic matter (i.e., aliphatic organic matter). PRAM can be used as an analytical tool that provides insightful information about variability in bulk organic matter characteristics during soil passage.

5. Occurrence, fate, and removal of organic micropollutants in managed aquifer recharge A summarized literature review regarding the removal efficiencies of PhACs grouped according to therapeutic uses into eight categories during BF or AR is now presented. Each BF or AR site was analyzed with respect to well types, travel distances, residence times, and redox conditions. Detailed information related to site characteristics and well information is presented in Tables 2 and 3, respectively.

5.1.

Antibiotics

Antibiotics comprise the most important pharmaceutical group. Antibiotics detected in the environment come from human medicine, veterinary medicine, plant agriculture and aquaculture (Ku¨mmerer, 2008a). Antibiotics are one of the most widely used groups of pharmaceutical compounds preventing or treating animal diseases (Boxall et al., 2003). The occurrence and fate of antibiotics in the aquatic environment is of growing interest on account of the possible existence and proliferation of antibiotic microorganisms, which may be related to the presence of antibiotics in the environment (Ku¨mmerer, 2008b). These unwanted effects of antibiotics require further investigation so that the impact of antibiotics on the aquatic environment can be better understood. In general, the concentration usuallydetected in the environment is not high enough to inhibit biological processes such as nitrification; however, it has been shown that a prolonged period of exposure to antibiotics can significantly reduce these processes (Halling-Sørensen, 2001). Manure and slurry containing veterinary medicines, including antibiotics, can partition into soil and leach into groundwater, especially in regards to compounds with high mobility (i.e., low log Kow). Many MAR systems, including BF, are influenced by groundwater (i.e., the mixing effect); thus, filtered water from MAR systems usually contains some portion of groundwater, even when a production well is located very close to surface water (e.g., rivers or lakes); therefore, groundwater contaminated with antibiotics can affect the water quality in a production well. The degree of mixing effects in groundwater varies with hydrogeological conditions and well locations. Only a few studies have been carried out to investigate the fate of antibiotics during BF and AR. Heberer et al. (2008) investigated 19 targeted antibiotics at a lake BF site located in Berlin (Germany) for 2.5 years. This site is well characterized and instrumented in terms of both production wells and transects of monitoring wells, where travel distances, residence times, and redox conditions are defined. They detected 7 out of 19 target antibiotics in Lake Wannsee water, which is used for BF: sulfamethoxazole, acetyl-sulfamethoxazole, anhydroerythromycin, clarithromycin, roxithromycin, trimethoprim, and clindamycin. However, all antibiotics were completely

3023

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Table 2 e Summary of managed aquifer recharge field sites investigated for occurrence of pharmaceutically active compounds. Location

Aquifer thickness (m)

Hydraulic permeability, k,(m/s)

Type

Mineralogy

References

Gru¨nheid et al. (2005), Pekdeger (2006) Gru¨nheid et al. (2005), Massmann et al. (2006)

Schmidt et al. (2007)

BFa

<40

2e8104, clogged sand 5  106 2e8104

<40

1  104 to 1  106

BFa

12e15

12  103

BFa

0.16e1.3% carbonate, 0.02e0.08% organic carbon, 1-2 g/kg iron 0.0e2.3% carbonate, 0.0e2.1% organic carbon, 1.8 g/kg iron, 0.1 g/kg manganese 0e1% carbonate, 0.1e2% organic carbon, 0.2-1 g/kg iron e

10e12

3  103 to 6  103

BFa

e

Schmidt et al. (2007)

40e55

0.6  103

BFa

e

Schmidt et al. (2007)

5

101 to 103

BFa

e

Schmidt et al. (2007)

37e45 30 (UAUc)

9  104 to 7  103 7  105 to 1  103

ARb ARb

e

Fox et al., (2001) Mansell and Drewes (2004)

Lake Tegel, Germany Lake Tegel, Germany

<40

Lake Wannsee, Germany RhineA, Germany RhineB, Germany Elbe, Germany Ruhr, Germany Tucson Mesa

ARb

Pekdeger (2006)

a BF: bank filtration. b AR: artificial recharge. c UAU: upper alluvial unit.

removed after 2e4 months of travel time except for sulfamethoxazole. Sulfamethoxazole was detected at the production well, but it was present at a concentration (2 ng/L) that was much lower than to the concentration found in Lake Wannsee (155 ng/L). Based on the results of several monitoring wells installed along the transect, Heberer et al. (2008) suggested that sulfamethoxazole was removed more efficiently under anoxic conditions (w99%) than under oxic conditions (w52%). Gru¨nheid et al. (2005) compared the bulk organic matter and organic micropollutant removal of BF and AR sites located in Berlin (Germany). The BF site exhibited oxic conditions followed by prolonged anoxic conditions, whereas the AR site mainly showed oxic conditions. The removal of sulfamethoxazole at the BF site (75%) was higher than that at the AR site (50%). However, Heberer et al. (2008) suggested that it is not possible to determine if redox conditions controlled the degradation of sulfamethoxazole in the study conducted by Gru¨nheid et al. (2005) on account of the different residence times at the BF site and the AR site. Longer residence times at the BF site could have contributed to a better removal of sulfamethoxazole. Schmidt et al. (2007) investigated sulfamethoxazole, clarithromycin, trimethoprim and clindamycin at four BF sites located along the Rhine river (Rhine A and Rhine B), the Elbe river and the Ruhr river. They found that selected antibiotics were removed at amounts greater than 70% except sulfamethoxazole, which experienced relatively low removal (0e25%) at Rhine A and Rhine B (where mainly oxic conditions were present). However, removal efficiencies of sulfamethoxazole greater than 90% were achieved at the Ruhr BF site, which has residence times that are short compared with those of Rhine A and Rhine B; however, the Ruhr BF site mainly exhibits anaerobic conditions. This study supports the results of Gru¨nheid et al. (2005), which indicated that sulfamethoxazole is a redox-dependent compound that degrades more effectively

under anoxic/anaerobic conditions. Fig. 1a and Fig. 1b show the performance of the MAR systems listed in Table 2 regarding the removal of antibiotics with respect to residence time and travel distance, respectively. The removal of antibiotics gradually increases as residence time and travel distance increase. From the results described above, it can be concluded that MAR is an effective treatment step in a multi-barrier system for removing antibiotics from drinking water supplies.

5.2. Non-steroidal anti-inflammatory drug (NSAIDs) and analgesics NSAIDs, also known as pain killers, are commonly used through the world for symptoms of arthritis, bursitis, gout, swelling, stiffness, and joint pain (Metcalfe et al., 2004). Large amounts of NSAIDs are sold by prescription or without prescription (i.e., over the counter drugs) worldwide (Heberer, 2002). High concentrations of NSAIDs have been detected in aquatic environments and wastewater as a result of their high consumption in human medical care and, to some degree, as a result of their persistent characteristics (Zhou et al., 2009). Many NSAIDs mentioned in this paper have been removed at rates greater than 50% during BF and AR. A number of field and laboratory-scale studies have shown significant removals of diclofenac, ibuprofen, naproxen, and phenazone during soil passage (Heberer and Adam, 2004; Massmann et al., 2006, 2008; Schmidt et al., 2007; Snyder et al., 2007). Biodegradation or biotransformation (biotic) and sorption (abiotic) are possible mechanisms for the removal of NSAIDs. Diclofenac, ibuprofen, indomethacin, and naproxen have moderately high octanole water partition coefficients (log Kow > 2.5), and sorption would likely be the main mechanism of their removal during soil passage. However, several studies demonstrated the

3024

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Table 3 e Summary of production and monitoring wells investigated for occurrence of pharmaceutically active compounds. Location (Type)

Well number

Production/ Monitoring well

Travel distance. (meter)

Residence time (day)

Redox conditions

References

Lake Tegel (LBFa) Lake Tegel (LBFa)

12

Production well

90

135

Anoxic

Scheytt et al. (2004) Verstraeten et al. (2002) Gru¨nheid et al. (2005) Scheytt et al. (2004) Schittko et al. (2004) Heberer et al. (2004) Jekel and Gru¨nheid (2005) Mechlinski and Heberer (2005) Verstraeten et al. (2002) Scheytt et al. (2004) Verstraeten et al. (2002) Zuehlke et al. (2004) Heberer et al. (2004) Schittko et al. (2004) Jekel and Gru¨nheid (2005) Mechlinski and Heberer (2005) Gru¨nheid et al. (2005) Verstraeten et al. (2002) Jekel and Gru¨nheid (2005) Mechlinski and Heberer (2005) Gru¨nheid et al. (2005) Verstraeten et al. (2002) Jekel and Gru¨nheid (2005) Mechlinski and Heberer (2005) Gru¨nheid et al. (2005) Verstraeten et al. (2002) Mechlinski and Heberer (2005) Gru¨nheid et al. (2005) Schittko et al. (2004) Mechlinski and Heberer (2005) Heberer et al. (2004) Gru¨nheid et al. (2005)

13

Production well

90

135

Anoxic

Lake Tegel (LBFa) Lake Tegel (LBFa)

14

Production well

90

135

Anoxic

3301

Monitoring well

25e40

90

Anoxic

Lake Tegel (LBFa)

3302

Monitoring well

55e70

90

Anoxic

Lake Tegel (LBFa)

3303

Monitoring well

77e95

117

Anoxic

Lake Tegel (LBFa) Lake Tegel (LBFa)

3310

Monitoring well

2

<1

Oxic

3311

Monitoring well

0

<1

Oxic

Lake Tegel (LBFa) Lake Tegel (ARb)

337UP

Monitoring well

30

84

Anoxic

20

Production well

50

50

Oxic

Lake Tegel (ARb) Lake Tegel (ARb) Lake Tegel (ARb) Lake Wannsee (LBFa)

365

Monitoring well

<10

<4

Oxic

368UP

Monitoring well

10

25

Oxic

Gru¨nheid et al. (2005) Zuehlke et al. (2004) Heberer and Adam (2004) Massmann et al. (2006) Gru¨nheid et al. (2005) Gru¨nheid et al. (2005)

369UP

Monitoring well

32

50

Oxic

Gru¨nheid et al. (2005)

Well#3

Production well

75

>120

Anoxic

Lake Wannsee (LBFa) Lake Wannsee (LBFa) Lake Wannsee (LBFa) Lake Wannsee (LBFa) Lake Wannsee (LBFa)

Well#4

Production well

100

>120

Anoxic

Well#5

Production well

>120

Anoxic

BE202OP

Monitoring well

60e120

Oxic

BE203

Monitoring well

60e120

Oxic

BE205

Monitoring well

20

<30

Anoxic

Lake Wannsee (LBFa)

BE206

Monitoring well

1.5

<30

Oxic

Lake Wannsee (LBFa)

3339

Monitoring well

40

65

Oxic

Heberer et al. (2008) Heberer et al. (2004) Gru¨nheid and Jekel (2005) Mechlinski and Heberer (2005) Heberer et al. (2003a) Heberer et al. (2004) Heberer et al. (2003a) Heberer et al. (2004) Heberer et al. (2003a) Gru¨nheid and Jekel (2005) Heberer et al. (2008) Gru¨nheid and Jekel (2005) Heberer et al. (2008) Massmann et al. (2008) Gru¨nheid and Jekel (2005) Heberer et al. (2008) Massmann et al. (2008) Mechlinski and Heberer (2005) Gru¨nheid and Jekel (2005) Heberer et al. (2008) Heberer et al. (2003b) and Schmidt et al. (2007)

3025

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Table 3 (continued) Location (Type) Lake Wannsee (LBFa) Lake Wannsee (LBFa) Lake Wannsee (LBFa) Lake Wannsee (LBFa) Rhine A (RBFc) Rhine B (RBFc) Elbe (RBFc) Ruhr (RBFc) Tuscon-WR205 (ARb) Mesa-NW2 (ARb) Mesa-NW4 (ARb) Mesa-NW2 (ARb) Mesa-NW2 (ARb) Mesa-2U (ARb) Mesa-6U (ARb)

Well number

Production/ Monitoring well

Travel distance. (meter)

Residence time (day)

Redox conditions

References

3338

Monitoring well

10

30

Oxic

3337

Monitoring well

5

20

Oxic

3335

Monitoring well

16

<30

Oxic

BR#4

Production well

33

45

Anoxic

Production well

160

7e20

Oxic

Heberer et al. (2003b) Schmidt et al. (2007) Heberer et al. (2003b) Schmidt et al. (2007) Heberer et al. (2003b) Schmidt et al. (2007) Heberer et al. (2003b) Schmidt et al. (2007) Schmidt et al. (2007)

Production well

70

12e60

Anoxic

Schmidt et al. (2007)

270

45e300

Anoxic

Schmidt et al. (2007)

125

5e15

Anaerobic

Schmidt et al. (2007)

Secondary Effluent Tertiary Effluent Tertiary Effluent Tertiary Effluent Tertiary Effluent Tertiary Effluent Tertiary Effluent

2190

Drewes et al. (2002)

Monitoring well

360

Mansell and Drewes (2004)

Monitoring well

540

Mansell and Drewes (2004)

Monitoring well

724

Mansell and Drewes (2004)

Monitoring well

2920

Mansell and Drewes (2004)

Monitoring well

360

Mansell and Drewes (2004)

Monitoring well

360

Mansell and Drewes (2004)

a LBF: lake bank filtration. b AR: artificial recharge. c RBF: riverbank filtration.

biodegradation potential of some NSAIDs in laboratory-scale experiments (Kagle et al., 2009). For example, ibuprofen was biologically transformed in microcosms prepared with sediment and fortified lake water (Buser et al., 1999; Lin et al., 2006). Transformation products (e.g., carboxyibuprofen and hydroxyibuprofen) were detected in a cultivated river water reactor, confirming the possibility of the partial biodegradation of ibuprofen (Winkler et al., 2001). The majority of NSAIDs are acidic compounds that exist as ionic species during soil passage. The pKa values of many NSAIDs are lower than the pH of aquatic environments (e.g., diclofenac: pKa 4.2; ibuprofen: pKa 4.9; naproxen: pKa 4.2). KOWWIN was used to estimate log Kow (US EPA, 2009); naproxen exhibits 3.18 for log Kow, which indicates moderate hydrophobicity. However, naproxen predominantly exists in ionized form in the aqueous phase on account of its low pKa (4.2). The log D (distribution coefficient at pH 8) of naproxen is 0.53, suggesting that ionic interactions may have more influence on the removal of naproxen than sorption during soil passage does. Accordingly, pH is an important parameter to consider when investigating the fate of acidic compounds; in addition, log D (instead of log Kow) should be used to indicate the characteristics of acidic PhACs such as NSAIDs (Cunningham, 2004).

Phenazone-type pharmaceuticals, another group of NSAIDs, are redox-dependent compounds that are more thoroughly removed under oxic conditions than under anoxic conditions. According to Massmann et al. (2006), the removal of phenazone at an AR site varied by season and was higher in winter than it was in summer. This occurred because oxic conditions were present at the AR site during the low temperatures of winter. Moreover, the AR site exhibited higher removal efficiencies for phenazone-type pharmaceuticals, including phenazone (91%), propyphenazone (100%), and DMAA (Dimethylaminophenazone) metabolites such as formylaminoantipyrine (FAA) (89%) and acetoaminoantipyrine (AAA) (96%). However, AMDOPH, which is another DMAA metabolite, was almost persistent (removal <5%) under all conditions. Massmann et al. (2008) found similar results at a BF site located at Lake Wannsee (Berlin, Germany) using two monitoring wells installed between Lake Wannsee and the production well. The removal of phenazone in anoxic monitoring well was relatively high (66%), whereas the phenazone removal in an anoxic monitoring well was very low (10%). Their study confirms that there is a preferential removal of phenazone-type pharmaceuticals under oxic conditions. Massmann et al. (2006) suggested that biodegradation by aerobic bacteria was the main removal mechanism of phenazone-type pharmaceuticals during soil passage. Based on

3026

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

a

most wastewater treatment plants (Ternes, 1998; Stamatelatou et al., 2003; Zhang et al., 2008; Kasprzyk-Hordern et al., 2009). Moreover, Clara et al. (2004) proposed the use of carbamazepine as a marker for anthropogenic influences in the aquatic environment. In addition, a number of laboratory and field studies on BF and AR have revealed low removals of primidone and dilantin during infiltration (Heberer and Adam, 2004; Heberer et al., 2004). Drewes et al. (2002) found that there was no change in carbamazepine and primidone concentrations during soil aquifer treatment for estimated travel times up to six years. Based on the performance of selected anticonvulsants in these studies, it can be concluded that BF and AR are not effective for the removal of anticonvulsants.

80

60

40

Sulfamethoxazole Acetyl-sulfamethoxazole Clarithromycin Roxithromycin Trimethoprim Clindamycin

20

0 0

20

40

60

80

100

120

140

160

5.4.

Residence Time (day)

Removal efficiency (%)

b

Antidepressants

Snyder et al. (2007) investigated three antidepressants (fluoxetine and meprobamate) in relation to their fate during BF using a pilot plant. Among the studied antidepressants, fluoxetine was found to be removed significantly (>99%), and the average removal efficiencies observed in column studies

100

80

60

a

Sulfamethoxazole Acetyl-sulfamethoxazole Clarithromycin Roxithromycin Trimethoprim Clindamycin

40

20

0 0

50

100

150

200

250

300

Residence Distance (meter)

Fig. 1 e Antibiotic concentration at MAR systems: (a) residence time, (b) travel distance (Grunheid et al., 2005; Grunheid and Jekel, 2005; Heberer et al., 2008; Jekel and Grunheid, 2005; Schmidt et al., 2007).

Removal efficiency (%)

Removal efficiency (%)

100

100

80

60 Diclofenac Ibuprofen Indomethacin Naproxen Phenazone FAA (formylaminoantipyrine) AAA (acetoaminoantipyrine) Pentoxifyline

40

20

0

0

20

40

60

80

100

120

140

160

Residence Time (day)

5.3.

Anticonvulsants

Anticonvulsant pharmaceuticals are the most persistent type during BF and AR. Among the anticonvulsants, carbamazepine has been extensively studied because it is the most frequently detected anticonvulsant in the environment (Drewes et al., 2002; Heberer et al., 2002; Cordy et al., 2004; Heberer and Adam, 2004; Mechlinski and Heberer, 2005; Massmann et al., 2006; Schmidt et al., 2007). Carbamazepine has shown a persistent behavior in the aquatic environment, and the poor biodegradability of carbamazepine results in low removal (<10%) in

b Removal efficiency (%)

previous studies described above, it can be concluded that the pH should be monitored during soil passage because many NSAIDs remain as ionic species in the aquatic environment. Moreover, some NSAIDs (e.g., phenazone-type pharmaceuticals) are redox dependent; thus, it is equally important to monitor redox conditions during BF and AR. Fig. 2a and b show that many NSAIDs and analgesics were still effectively removed despite short residence times and short travel distances, respectively.

100

80

60 Diclofenac Ibuprofen Indomethacin Naproxen Phenazone FAA (formylaminoantipyrine) AAA (acetoaminoantipyrine) Pentoxifyline

40

20

0 0

50

100

150

200

250

300

Travel Distance (meter)

Fig. 2 e Non-steroidal anti-inflammatory drugs (NSAIDs) and analgesics concentrations at MAR systems: (a) residence time, (b) travel distance (Heberer and Adam., 2004; Heberer et al., 2003b; Heberer et al., 2003a; Massmann et al., 2008; Schmidt et al., 2007; Snyder et al., 2007).

3027

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

for meprobamate was 66%. The higher removal of fluoxetine can be attributed mainly to sorption on account of its high octanolewater partition coefficient (log Kow: 4.69). As a result of the limited information and research regarding antidepressants during subsurface infiltration, it is not possible to characterize the performance of BF and AR regarding the removal of antidepressants.

Beta blockers

5.6.

Lipid regulators

Like NSAIDs, many lipid regulators remain as ionic species in the aquatic environment (Mompelat et al., 2009). Thus, the pH during subsurface infiltration plays an important role in the removal mechanism of lipid regulators. It is important to monitor metabolites of lipid regulators (e.g., clofibric acid and fenofibric acid) during soil passage because they are derived from “prodrugs”, which are administered in an inactive form. A prodrug undergoes metabolic conversion of the parent compound to an active metabolite; it is this metabolite that produces the pharmaceutical effect, the parent inactive parent compound (e.g., clofibrate, and fenofibrate). Clofibric acid, fenofibric acid and salicylic acid are common metabolites originating from clofibrate, fenofibrate, and aspirin, respectively. Clofibric acid is one of the most common metabolites studied in relation to BF and AR, and it is often detected in the aquatic environment, wetlands and wastewater treatment plants (Winkler et al., 2001; Heberer, 2002; Matamoros et al., 2008; Dordio et al., 2009; Kagle et al., 2009). According to Heberer et al. (2004), clofibric acid concentrations increased at the production wells located at the Lake Tegel BF site in Berlin (Germany). This increase was due to the high consumption of fibrate-based lipid regulators during the 1990s. The use of fibrate-based lipid regulators has significantly decreased over the last ten years, but clofibric acid is still present in deeper layers of the aquifer (Heberer et al., 2004). Consequently, “old” bank filtrate still exhibits a high concentration of clofibric acid from the 1990’s. However, BF sites along the Rhine river and an AR site located at Lake Tegel

5.7.

X-ray contrast media

Gru¨nheid et al. (2005) conducted field studies at BF and AR sites (Lake Tegel, Berlin, Germany) to investigate the fate of X-ray contrast agents, which can be measured as adsorbable organic iodine (AOI). Redox conditions at the BF site gradually changed from oxic to prolonged anoxic conditions along the flow pathway, but at the AR site, mainly oxic conditions prevailed. Both the BF and AR sites used water from Lake Tegel for infiltration. AOI removal efficiencies at the BF site and the AR site were 60% and 30%, respectively. The removal efficiency of AOI at the BF site was higher than that

a Removal efficiency (%)

According to Schmidt et al. (2007), four beta blockers (atenolol, metoprolol, bisoprolol and sotalol) were removed greater than 70% at BF sites located along the rivers Rhine, Elbe, and Ruhr. Five beta blockerseatenolol (0.36 g/L), sotalol (1.3 g/L), celiprolol (0.28 g/L), propranolol (0.18 g/L), and metoprolol (1.7 g/ L)ein secondary effluent were found to be below the limit of quantification (0.025 g/L) after soil aquifer treatment using agricultural fields in Braunschweig, Germany (Ternes et al., 2007). This study suggests that selected beta blockers were removed by sorption and/or biodegradation during soil passage. In contrast, in some recent studies, low removals of beta blockers were observed during wastewater treatment processes (Castiglioni et al., 2005; Kasprzyk-Hordern et al., 2009; Wick et al., 2009). As a consequence, beta blockers are often detected in the aquatic environment in the ng/L to mg/L range (Wick et al., 2009). Further study is required to explain the removal of beta blockers observed during subsurface infiltration as compared to the removal resulting from wastewater treatment processes.

100

80

60

40

Bezafibrate Fenofibric acid Clofibric acid

20

0 0

20

40

60

80

100

120

140

160

Residence Time (day)

b Removal efficiency (%)

5.5.

removed clofibric acid at rates greater than 70% (Heberer and Adam, 2004; Heberer et al., 2004; Schmidt et al., 2007). Moreover, bezafibrate was found to be significantly removed during subsurface infiltration (Heberer, 2002). Thus, the performance of BF and AR systems in relation to the removal of lipid regulators varied from site to site, but most of the lipid regulators at MAR systems reviewed in this study were removed at rates greater than 50% (Fig. 3a and b).

100

80

60

40 Bezafibrate Fenofibric acid Clofibric acid

20

0 0

50

100

150

200

250

300

Travel Distance (meter)

Fig. 3 e Lipid regulators concentrations at MAR systems: (a) residence time, (b) travel distance (Heberer et al., 2004; Heberer and Adam., 2004; Heberer et al., 2003b; Scheytt et al., 2004; Schmidt et al., 2007; Snyder et al., 2007; Verstraeten et al., 2002).

3028

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Steroid hormones

Steroid hormones, such as synthetic estrogens (e.g., 17-ethinylestradiol) and natural estrogens (e.g., estrone and 17estradiol) are of special concern because they produce potentially adverse effects on human health and aquatic life, even at very low concentrations (ng/L) (Purdom et al., 1994; Tyler et al., 2005). A number of laboratory-scale and field studies on the fate of estrogens during soil passage have been carried out (Mansell and Drewes, 2004; Mansell et al., 2004; Snyder et al., 2004; Zuehlke et al., 2004). A field study carried out by Zuehlke et al. (2004) showed that 17-estradiol (E2) and 17-ethinylestradiol (EE2) were not detected in the surface water from Berlin (LOQ: 0.2 ng/L), and estrone (E1) was removed greater than 80% at a monitoring well located close to the lake shore. This study demonstrated that significant removal of E1 was possible during soil passage even within a short distance. Estrogen compounds are generally hydrophobic andare typically neutral; thus, sorption is likely to be the main removal mechanism. Mansell and Drewes (2004) performed field studies combined with laboratory-scale experiments and showed that estriol (E3) and testosterone were reduced to quantities below their detection limits (E3: <0.6 ng/L; testosterone: < 0.5 ng/L). Their study also suggested that biodegradation increased removal efficiencies of E1, E2, and EE2. Snyder et al. (2004) produced similar results using batch experiments and field studies; they demonstrated biodegradation and significant attenuation of E1, E2, EE2, and testosterone. Mansell et al. (2004) found that steroids were reduced to quantities below the detection limit by a combination of sorption and biodegradation; thus, the removal of estrogen compounds is not only dependent on sorption, but is also affected by biodegradation. Ying et al. (2003) conducted an experiment with E2 in aquifer materials using groundwater and found that E2 was rapidly degraded in aquifer materials under aerobic conditions but that it remained almost unchanged under anaerobic conditions. This study suggests that the removal of E2 can be enhanced by aerobic bacteria, but this removal was not observed under anaerobic conditions. Thus, aerobic bacteria also play an important role in the removal of E2. Based on the studies mentioned above, it can be

6. Practical implications of MAR and further research This review demonstrates that MAR is an effective barrier in the multi-barrier approach to removing organic micropollutants in drinking water supplies. The removal efficiencies of MAR for these contaminants can be maximized by proper well design that takes source water quality characteristics and local hydrogeological conditions into consideration. Regarding the practical implications of this review, many organic micropollutants can be reduced or transformed during soil passage, although to varying extents. Generally, oxic conditions followed by anoxic conditions, and varying redox conditions are effective for removing redoxesensitive organic micropollutants during MAR. MAR is the only drinking water treatment process that provides both oxic and anoxic conditions. One important conclusion of this review is that MAR (such as BF or AR) alone is not capable of removing all organic micropollutants; consequently, it is important to choose the post-treatment process

a Removal efficiency (%)

5.8.

concluded that MAR is an effective and reliable treatment for removing estrogenic compounds.

100

80

60

AOI (Absorbable organic iodine) Iopromide Iopamidol Iomeprol IIoxhexol Diatrizoate

40

20

0 0

20

40

60

80

100

120

140

160

Residence Time (day)

b 100

Removal efficiency (%)

at the AR site. Gru¨nheid et al. (2005) found that AOI removal efficiencies and ORP (oxidation-reduction potential) were inversely correlated and that dehalogenation of AOI was enhanced under anoxic conditions. They also showed that AOI concentrations in infiltrated water changed by according to the season as a result of the discharge of wastewater effluent into the lake; the changes were attributable to varying dilutions in the wastewater effluent discharges. Schittko et al. (2004) obtained similar results at the same BF site, where 63% of AOI was removed during soil passage. Besides AOI, four individual iodinated X-ray contrast agents (iopromide, iopamidol, iomeprol, and ioxhexol) were measured at various BF sites located in Germany (Schittko et al., 2004; Gru¨nheid et al., 2005; Schmidt et al., 2007). Iopromide, iopamidol and iomeprol were significantly removed (>80%) and were found to be compounds that BF can easily remove (Fig. 4a and b).

80

60 AOI (Absorbable organic iodine) Iopromide Iopamidol Iomeprol Ioxhexol Diatrizoate

40

20 0

50

100

150

200

250

300

Travel Distance (meter)

Fig. 4 e X-ray contrast media concentrations at MAR systems: (a) residence time, (b) travel distance (Gru ¨ nheid et al., 2005; Schmidt et al., 2007; Schittko et al., 2004; Snyder et al., 2007).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

that maximizes the removal of organic micropollutants. For example, BF followed by nanofiltration is an effective combination in which organic micropollutants will be removed first by biodegradation then by the membrane through the mechanisms of size exclusion and advanced oxidation processes as a pre-treatment process for AR. Although biodegradation, sorption, and mixing with natural groundwater achieve a reduction of OMPs during MAR, the best method of attenuating PhACs will involve avoiding or minimizing the use of organic micropollutants in the household and adopting an advanced wastewater treatment process, such as a membrane bioreactor, to control point sources. Further study is required to provide more definitions of the physicohemical properties of PhACs (e.g., log Kow and log D) that can used to predict their fate during soil passage. The guidelines for the groups of organic micropollutants associated with soil can be developed to increase their reliability by including more recent literature sources. Detailed studies on the possibility of direct uptake of PhACs as a carbon source by microorganisms will be necessary if PhACs are not always transformed by co-metabolism. Further research is necessary to investigate the fate of emerging organic micropollutants (i.e., transformation products) during soil passage. More research should be carried out on the combination of managed aquifer recharge with other advanced water treatment processes (e.g., advanced oxidation, membrane and ion exchange systems). This would help drinking water utilities to identify the best combination of water treatment processes for removing organic micropollutants. MAR systems can provide a robust and natural barrier in the multi-barrier approach and when combined with an advanced treatment process as a pre or post-treatment process.

7.

Conclusions

MAR systems, such as BF and AR, can be included in a multibarrier treatment system for the removal of PhACs in drinking water treatment as part of indirect potable reuse schemes. Based on the studies reviewed in this paper, the following conclusions can be drawn: - TOC and DOC reductions resulting from BF and AR varied from site to site; generally, TOC and DOC removal efficiencies ranged from 30 to 88% and from 33 to 88%, respectively. - The preferential removal of non-humic substances (i.e., aliphatic organic matter or biopolymers) was observed during soil passage as a result of biodegradation. - Ionic interactions may be a key factor in the sorption mechanism for acidic pharmaceuticals such as nonsteroidal anti-inflammatory drugs (NSAIDs) and lipid regulators. Thus, log D should be used to explain the fate of acidic PhACs during BF and AR. - Antibiotics, NSAIDs, beta blockers, and steroid hormones generally exhibited good removal efficiencies, especially for compounds having hydrophobic-neutral characteristics. - Anticonvulsant drugs were generally difficult to remove through BF and AR. - Phenazone-type pharmaceuticals (NSAIDs) exhibited better removal efficiencies under oxic conditions, except

3029

AMDOPH, which was persistent under all conditions during BF and AR. - Some PhACs, including X-ray contrast agents (AOI) and sulfamethoxazole, experienced no significant removal under oxic conditions, but were removed under anoxic conditions. - Biodegradation plays an important role in removing PhACs during soil passage. - BF is effective for the removal of redox-dependent PhACs because redox conditions during BF gradually change from oxic to anoxic during soil passage.

Acknowledgement This work was supported by EU SWITCH Project No. 018530-2 under the Sixth Framework Programme. The authors thank Veolia Water and the Berlin Water Company (BWB) for their support on data collected from their sites. We would like to express our gratitude to Dr. Gesche Gru¨tzmacher from Kompetenzzentrum Wasser Berlin.

references

Alexander, M., 1981. Biodegradation of chemicals of environmental concern. Science 211 (4478), 132e138. Amy, G., Drewes, J., 2007. Soil aquifer treatment (SAT) as natural and sustainable wastewater reclamation/reuse technology: fate of wastewater effluent organic matter (EfOM) and trace organic compounds. Environ. Monit. Assess. 129 (1e3), 19e26. Barker, D.J., Stuckey, D.C., 1999. A review of soluble microbial products (SMP) in wastewater treatment systems. Water Res. 33 (14), 3063e3082. Bourg, A.C.M., Bertin, C., 1993. Biogeochemical processes during the infiltration of river water into an alluvial aquifer. Environ. Sci. Technol. 27 (4), 661e666. Boxall, A.B.A., Kolpin, D.W., Halling-Sørensen, B., Tolls, J., 2003. Are veterinary medicines causing environmental Risks? Environ. Sci. Technol. 37 (15), 286Ae294A. Boxall, A.B.A., Kay, P., Blackwell, P.A., Fogg, L.A., 2004. Fate of veterinary medicines applied to soils. In: Ku¨mmerer, K. (Ed.), Pharmaceutical in the Environment, second ed. Springer, Berlin Heidelberg. Buser, H.-R., Poiger, T., Muller, M.D., 1999. Occurrence and environmental behavior of the chiral pharmaceutical drug ibuprofen in surface waters and in wastewater. Environ. Sci. Technol. 33 (15), 2529e2535. Castiglioni, S., Bagnati, R., Fanelli, R., Pomati, F., Calamari, D., Zuccato, E., 2005. Removal of pharmaceuticals in sewage treatment plants in Italy. Environ. Sci. Technol. 40 (1), 357e363. Cha, W., Choi, H., Kim, J., Kim, I.S., 2004. Evaluation of wastewater effluents for soil aquifer treatment in South Korea. Water Sci. Technol. 50 (2), 315e322. Chefetz, B., Mualem, T., Ben-Ari, J., 2008. Sorption and mobility of pharmaceutical compounds in soil irrigated with reclaimed wastewater. Chemosphere 73 (8), 1335e1343. Chen, W., Westerhoff, P., Leenheer, J.A., Booksh, K., 2003. Fluorescence excitation-emission matrix regional integration to quantify spectra for dissolved organic matter. Environ. Sci. Technol. 37 (24), 5701e5710.

3030

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Clara, M., Strenn, B., Kreuzinger, N., 2004. Carbamazepine as a possible anthropogenic marker in the aquatic environment: investigations on the behaviour of carbamazepine in wastewater treatment and during groundwater infiltration. Water Res. 38 (4), 947e954. Coble, P.G., 1996. Characterization of marine and terrestrial DOM in seawater using excitation-emission matrix spectroscopy. Mar. Chem. 51 (4), 325e346. Cordy, G.E., Duran, N.L., Bouwer, H., Rice, R.C., Furlong, E.T., Zaugg, S.D., Meyer, M.T., Barber, L.B., Kolpin, D.W., 2004. Do pharmaceuticals, pathogens, and other organic wast water compounds persist when waste water is used for recharge? Ground Water Monit. R 24 (2), 58e69.  , B., Hr , V., Krznaric , D., 1996. Cosovi c sak, D., Vojvodic Transformation of organic matter and bank filtration from a polluted stream. Water Res. 30 (12), 2921e2928. Cunningham, V.L., 2004. Special Characteristics of Pharmaceuticals Related to Environmental Fate. Springer, Berlin. Cunningham, V.L., Buzby, M., Hutchinson, T., Mastrocco, F., Parke, N., Roden, N., 2006. Effects of human pharmaceuticals on aquatic life: next steps. Environ. Sci. Technol. 40 (11), 3456e3462. Dı´az-Cruz, M.S., Barcelo´, D., 2008. Trace organic chemicals contamination in ground water recharge. Chemosphere 72 (3), 333e342. Dillon, P., 2005. Future management of aquifer recharge. Hydrogeol J. 13 (1), 313e316. Dordio, A.V., Estevao Candeias, A.J., Pinto, A.P., Costa, Teixeira da, Palace, C., Carvalho, A.J., 2009. Preliminary media screening for application in the removal of clofibric acid, carbamazepine and ibuprofen by SSF-constructed wetlands. Ecol. Eng. 35 (2), 290e302. Drewes, J.E., Heberer, T., Reddersen, K., 2002. Fate of pharmaceuticals during indirect potable reuse. Water Sci. Technol. 46 (3), 73e80. Eckert, P., Irmscher, R., 2006. Practical paper over 130 years of experience with riverbank filtration in Dusseldorf, Germany. J. Water Supply Res. T. 55 (4), 283e291. Fox, P., Houston, S., Westerhoff, P., Drewes, J.E., Nellor, M., Yanko, W., Baird, R., Rincon, M., Arnold, R., Lansey, K., Bassett, R., Gerba, C., Karpiscak, M., Amy, G., Reinhard, M., 2001. Soil Aquifer Treatment for Sustainable Water Reuse. Awwa Research Foundation, Denver, CO. Gebhardt, W., Schro¨er, H.F., 2007. Liquid chromatography(tandem) mass spectrometry for the follow-up of the elimination of persistent pharmaceuticals during wastewater treatment applying biological wastewater treatment and advanced oxidation. J. Chromatogr. A 1160 (1e2), 34e43. Gru¨nheid, S., Jekel, M., 2005. Fate of bulk organics during bank filtration of wastewater-impacted surface water. In: Proceedings of the 5th International Symposium on Management of Aquifer Recharge Berlin, 10e16 June 2005. Gru¨nheid, S., Amy, G., Jekel, M., 2005. Removal of bulk dissolved organic carbon (DOC) and trace organic compounds by bank filtration and artificial recharge. Water Res. 39 (14), 3219. Gru¨tzmacher, G., Hu¨lshoff, I., Wiese, B., Golvan, Y., Sprenger, C., Lorenzen, G., Pekdeger, A., 2009. Function and Relevance of Aquifer Recharge Techniques to Enable Sustainable Water Resources Management in Developing or Newly-industrialized Countries. IWA Publishing. Gupta, V., Johnson, W.P., Shafieian, P., Ryu, H., Alum, A., Abbaszadegan, M., Hubbs, S.A., Rauch-Williams, T., 2009. Riverbank filtration: comparison of pilot scale transport with theory. Environ. Sci. Technol. 43 (8), 2992. Halling-Sørensen, B., 2001. Inhibition of aerobic growth and nitrification of bacteria in sewage sludge by antibacterial agents. Arch. Environ. Contam. Toxicol. 40 (4), 451e460.

Heberer, T., 2002. Occurrence, fate, and removal of pharmaceutical residues in the aquatic environment: a review of recent research data. Toxicol. Lett. 131 (1e2), 5e17. Heberer, T., Adam, M., 2004. Transport and attenuation of pharmaceutical residues during artificial groundwater Replenishment. Environ. Chem. 1 (1), 22e25. Heberer, T., Reddersen, K., Mechlinski, A., 2002. From municipal sewage to drinking water: fate and removal of pharmaceutical residues in the aquatic environment in urban areas. Water Sci. Technol. 46 (3), 81e88. Heberer, T., Fanck, B., Mechlinski, A., Zu¨hlke, S., Adam, M., Voigt, M., Wicke, D., Du¨nnbier, U., 2003a. Occurrence and Fate of Drug Residues and Related Polar Contaminants during Bank Filtration. Berlin Centre for Water Competence, Berlin. Heberer, T., Mechlinski, A., Fanck, B., 2003b. NASRI e Occurrence and fate of pharmaceuticals during bank filtration. In: Conference Wasser Berlin 2003. Berlin Centre for Water Competence, Berlin. Heberer, T., Mechlinski, A., Fanck, B., Knappe, A., Massmann, G., Pekdeger, A., Fritz, B., 2004. Field studies on the fate and transport of pharmaceutical residues in bank filtration. Ground Water Monit. R 24 (2), 70e77. Heberer, T., Massmann, G., Fanck, B., Taute, T., Du¨nbier, U., 2008. Behaviour and redox sensitivity of antimicrobial residues during bank filtration. Chemosphere 73 (4), 451e460. Henderson, R.K., Baker, A., Murphy, K.R., Hambly, A., Stuetz, R.M., Khan, S.J., 2009. Fluorescence as a potential monitoring tool for recycled water systems: a review. Water Res. 43 (3), 863e881. Hiemstra, P., Kolpa, R.J., Eekhout, J.M.J.M.v., Kessel, T.A.L.v., Adamse, E.D., Paassen, J.A.M.v, 2003. Natural recharge of groundwater: bank infiltration in the Netherlands. J. Water Supply Res. T. 52 (1), 37e47. Hiscock, K.M., Grischek, T., 2002. Attenuation of groundwater pollution by bank filtration. J. Hydrol 266 (3e4), 139e144. Howard, P.H., 2000. Biodegradation. Handbook of Property Estimation Methods for Chemicals and Health Sciences. CRC Press LLC, Boca Raton. Huber, S., Frimmel, F.H., 1992. A liquid chromatographic system with multi-detection for the direct analysis of hydrophilic organic compounds in natural waters. Fresenius J. Anal. Chem. 342 (1e2), 198e200. Irmscher, R., Teermann, I., 2002. Riverbank filtration for drinking water supply e a proven method, perfect to face today’s challenge. Water Sci. Technol. 2 (5e6), 1e8. Jarusutthirak, C., Amy, G., 2006. Role of soluble microbial products (SMP) in membrane fouling and flux decline. Environ. Sci. Technol. 40 (3), 969e974. Jekel, M., Gru¨nheid, S., 2003. NASRIeRemoval of organic substances. In: Proceedings of the Conference Wasser Berlin 7e11 April 2003. Jekel, M., Gru¨nheid, S., 2005. Bank filtration and groundwater recharge for treatment of polluted surface waters. Water Sci. Technol. 5 (5), 57e66. Jjemba, P.K., 2006. Excretion and ecotoxicity of pharmaceutical and personal care products in the environment. Ecotox. Environ. Safe 63 (1), 113e130. Kagle, J., Porter, A.W., Murdoch, R.W., Rivera-Cancel, G., Hay, A.G., 2009. Biodegradation of pharmaceuticals and personal care products. Adv. Appl. Microbiol. 67, 65e108. Kasprzyk-Hordern, B., Dinsdale, R.M., Guwy, A.J., 2009. The removal of pharmaceuticals, personal care products, endocrine disruptors and illicit drugs during wastewater treatment and its impact on the quality of receiving waters. Water Res. 43 (2), 363e380. Kedziorek, M.A.M., Geoffriau, S., Bourg, A.C.M., 2008. Organic matter and modeling redox reactions during river bank filtration in an alluvial aquifer of the Lot River, France. Environ. Sci. Technol. 42 (8), 2793e2798.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Kim, S.D., Cho, J., Kim, I.S., Vanderford, B.J., Snyder, S.A., 2007. Occurrence and removal of pharmaceuticals and endocrine disruptors in South Korean surface, drinking, and waste waters. Water Res. 41 (5), 1013e1021. Klavarioti, M., Mantzavinos, D., Kassinos, D., 2009. Removal of residual pharmaceuticals from aqueous systems by advanced oxidation processes. Environ. Int. 35 (2), 402e417. Kolehmainen, R.E., Langwaldt, J.g.H., Puhakka, J.A., 2007. Natural organic matter (NOM) removal and structural changes in the bacterial community during artificial groundwater recharge with humic lake water. Water Res. 41 (12), 2715e2725. Ku¨mmerer, K., 2009. The presence of pharmaceuticals in the environment due to human use e present knowledge and future challenges. J. Environ. Manage. 90 (8), 2354e2366. Ku¨mmerer, K., 2008a. Antibiotics in the environment. In: Ku¨mmerer, K. (Ed.), Pharmaceutical in the Environment, third ed. Springer, Berlin Heidelberg. Ku¨mmerer, K., 2008b. Effect of antibiotics and virustatics in the environment. In: Ku¨mmerer, K. (Ed.), Pharmaceutical in the Environment, third ed. Springer, Berlin Heidelberg. La˚ngmark, J., Storey, M.V., Ashbolt, N.J., Stenstro¨m, T.A., 2004. Artificial groundwater treatment: biofilm activity and organic carbon removal performance. Water Res. 38 (3), 740e748. Lee, J.-H., Hamm, S.-Y., Cheong, J.-Y., Kim, H.-S., Ko, E.-J., Lee, K.S., Lee, S.-I., 2009. Characterizing riverbank-filtered water and river water qualities at a site in the lower Nakdong River basin, Republic of Korea. J. Hydrol 376 (1e2), 209e220. Leenheer, J.A., Croue, J.-P., 2003. Characterizing aquatic dissolved organic matter. Environ. Sci. Technol. 37 (1), 18Ae26A. Lin, A.Y.-C., Plumlee, M.H., Reinhard, M., 2006. Natural attenuation of pharmaceuticals and alkylphenol polyethoxylate metabolites during river transport: photochemical and biological transformation. Environ. Toxicol. Chem. 25 (6), 1458e1464. Lim, M.-H., Snyder, S.A., 2008. Use of biodegradable dissolved organic carbon (BDOC) to assess the potential for transformation of wastewater-derived contaminants in surface waters. Water Res. 42 (12), 2943e2952. Lu, J., Zhang, T., Ma, J., Chen, Z., 2009. Evaluation of disinfection by-products formation during chlorination and chloramination of dissolved natural organic matter fractions isolated from a filtered river water. J. Hazard. Mater. 162 (1), 140e145. Madden, J.C., Enoch, S.J., Hewitt, M., Cronin, M.T.D., 2009. Pharmaceuticals in the environment: good practice in predicting acute ecotoxicological effects. Toxicol. Lett. 185 (2), 85e101. Maeng, S.K., Sharma, S.K., Amy, G., Magic-Knezev, A., 2008. Fate of effluent organic matter (EfOM) and natural organic matter (NOM) through riverbank filtration. Water Sci. Technol. 57 (12), 1999e2007. Maeng, S.K., Ameda, E., Sharma, S.K., Gru¨tzmacher, G., Amy, G., 2010. Organic micropollutant removal from wastewater effluent-impacted drinking water sources during bank filtration and artificial recharge. Water Res. 44 (14), 4003e4014. Mansell, J., Drewes, J.E., 2004. Fate of steroidal hormones during soil-aquifer treatment. Ground Water Monit. R 24 (2), 94e101. Mansell, J., Drewes, J.E., Rauch-Williams, T., 2004. Removal mechanism of endocrine disrupting compounds (steroids) soil aquifer treatment. Water Sci. Technol. 50 (2), 229e237. Massmann, G., Greskowiak, J., Du¨nnbier, U., Zuehlke, S., 2006. The impact of variable temperatures on the redox conditions and the behavior of pharmaceutical residues during artificial recharge. J. Hydrol 328 (1e2), 141e156. Massmann, G., Dunnbier, U., Heberer, T., Taute, T., 2008. Behaviour and redox sensitivity of pharmaceutical residues during bank filtration e Investigation of residues of phenazone-type analgesics. Chemosphere 71 (8), 1476e1485.

3031

Matamoros, V.t., Caselles-Osorio, A., Garc, J., Bayona, J.M., 2008. Behaviour of pharmaceutical products and biodegradation intermediates in horizontal subsurface flow constructed wetland. A microcosm experiment. Sci. Total Environ. 394 (1), 171e176. Mechlinski, A., Heberer, T., 2005. Fate and transport of pharmaceutical residues during bank filtration. In: Proceedings of 5th International Symposium on Management of Aquifer Recharge, Berlin, 10e16 June 2005. Metcalfe, C., Miao, X.-S., Hua, W., Letcher, R., Servos, M., 2004. Pharmaceuticals in the Canadian Environment. Springer, Berlin. Mompelat, S., LeBot, B., Thomas, O., 2009. Occurrence and fate of pharmaceutical products and by-products, from resource to drinking water. Environ. Int. 35 (5), 803e814. Nederlof, M.M., Kruithof, J.C., Taylor, J.S., van der Kooij, D., Schippers, J.C., 2000. Comparison of NF/RO membrane performance in integrated membrane systems. Desalination 131 (1e3), 257e269. Partinoudi, V., Collins, M.R., 2007. Assessing RBF reduction/ removal mechanisms for microbial and organic DBP precursors. J. AWWA 99 (12), 61e71. Pekdeger, A., 2006. Hydrogeological-hydrogeochemical Processes during Bank Filtration and Groundwater Recharge Using a Multi-tracer Approach. NASRI Project, Berlin. Philibert, M., Bush, S., Rosario-Ortiz, F.L., Suffet, I.H., 2008. Advances in the characterization of the polarity of DOM under ambient water quality conditions using the polarity rapid assessment method. Water Sci. Technol. 8 (6), 725e733. Purdom, C.E., Hardiman, P.A., Bye, V.J., Eno, N.C., Tyler, C.R., Sumpter, J.P., 1994. Estrogenic effects of effluents from sewage treatment works. Chem. Ecol. 8, 275e285. Quanrud, D.M., Hafer, J., Karpiscak, M.M., Zhang, J., Lansey, K.E., Arnold, R.G., 2003. Fate of organics during soil-aquifer treatment: sustainability of removals in the field. Water Res. 37 (14), 3401e3411. Radjenovic, J., Petrovic, M., Ventura, F., Barcel, D., 2008. Rejection of pharmaceuticals in nanofiltration and reverse osmosis membrane drinking water treatment. Water Res. 42 (14), 3601e3610. Rauch-Williams, T., Drewes, J.E., 2004. Assessing the removal potential of soil-aquifer treatment systems for bulk organic matter. Water Sci. Technol. 50 (2), 245e253. Rauch-Williams, T., Drewes, J., 2005. Quantifying biological organic carbon removal in groundwater recharge systems. J. Environ. Eng-ASCE 131 (6), 909e923. Rauch-Williams, T., Drewes, J.E., 2006. Using soil biomass as an indicator for the biological removal of effluent-derived organic carbon during soil infiltration. Water Res. 40 (5), 961e968. Rauch-Williams, T., Hoppe-Jones, C., Drewes, J.E., 2010. The role of organic matter in the removal of emerging trace organic chemicals during managed aquifer recharge. Water Res. 44 (2), 449e460. Ray, C., 2008. Worldwide potential of riverbank filtration. Clean Technol. Envir 10 (3), 223e225. Ray, C., Grischek, T., Schubert, J., Wang, J., Speth, T., 2002a. A perspective of riverbank filtration. J. AWWA 94 (3), 149e160. Ray, C., Schubert, J., Linsky, R.B., Melin, G., 2002b. In: Ray, C., Melin, G., Linsky, R.B. (Eds.), Introduction. Riverbank Filtration: Improving Source-Water Quality. Kluwer Academic Publishers, Dordrecht. Ray, C., Soong, T.W., Lian, Y.Q., Roadcap, G.S., 2002c. Effect of flood-induced chemical load on filtrate quality at bank filtration sites. J. Hydrol 266 (3e4), 235e258. Rosario-Ortiz, F.L., Kozawa, K., Al-Samarrai, H.N., Gerringer, F.W., Gabelich, C.J., Suffet, I.H., 2004. Characterization of the changes in polarity of natural organic matter using solidphase extraction: introducing the NOM polarity rapid

3032

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

assessment method (NOM-PRAM). Water Sci. Technol. 40 (4), 11e18. Rosario-Ortiz, F.L., Snyder, S., Suffet, I.H., 2007a. Characterization of the polarity of natural organic matter under ambient conditions by the polarity rapid assessment method (PRAM). Environ. Sci. Technol. 41 (14), 4895e4900. Rosario-Ortiz, F.L., Snyder, S.A., Suffet, I.H., 2007b. Characterization of dissolved organic matter in drinking water sources impacted by multiple tributaries. Water Res. 41 (18), 4115e4128. Rosario-Ortiz, F.L., Gerringer, F.W., Suffet, I.H., 2009. Application of a novel polarity method for the characterization of natural organic matter during water treatment. J. Water Supply Res. T. 58 (3), 159e169. Scheytt, T., Mersmann, P., Leidig, M., Pekdeger, A., Heberer, T., 2004. Transport of pharmaceutically active compounds in saturated laboratory columns. Ground Water 42 (5), 767e773. Scheytt, T.J., Mermann, P., Heberer, T., 2006. Mobility of pharmaceuticals carbamazepine, diclofenac, ibuprofen, and propyphenazone in miscible-displacement experiments. J. Contam. Hydrol 83, 53e69. Schittko, S., Putschew, A., Jekel, M., 2004. Bank filtration: a suitable process for the removal of iodinated X-ray contrast media. Water Sci. Technol. 50 (5), 261e268. Schmidt, C., Lange, F., Brauch, H., Kuhn, W., 2003. Experiences with riverbank filtration and infiltration in Germany. In: International Symposium on Artificial Recharge of Groundwater. K-WATER, Daejon, Korea, pp. 117e131. Schmidt, C.K., Lange, F.T., Brauch, H.J., 2007. Characteristics and evaluation of natural attenuation processes for organic micropollutant removal during riverbank filtration. Water Sci. Technol. 7 (3), 1e7. Schoenheinz, D., Bornick, H., Worch, E., 2005. Temperature effects on organics removal during river bank filtration. In: Proceedings of the 5th International Symposium on Management of Aquifer Recharge Berlin, 10e16 June 2005. Schubert, J., 2002. Hydraulic aspects of riverbank filtration-field studies. J. Hydrol 266 (3e4), 145e161. Shamrukh, M., Abdel-Wahab, A., 2008. Riverbank filtration for sustainable water supply: application to a large-scale facility on the Nile River. Clean Technol. Envir 10 (4), 351e358. Siddiqui, M., Amy, G., Ryan, J., Odem, W., 2000. Membranes for the control of natural organic matter from surface waters. Wat. Res. 34 (13), 3355e3370. Snyder, S.A., Leising, J., Westerhoff, P., Yoon, Y., Mash, H., Vanderford, B., 2004. Biological and physical attenuation of endocrine disruptors and pharmaceuticals: implications for water reuse. Ground Water Monit. R 24 (2), 108e118. Snyder, S.A., Wert, E.C., Lei, H., Westerhoff, P., Yoon, Y., 2007. Removal of EDCs and Pharmaceuticals in Drinking and Reuse Treatment Processes. Awwa Research Foundation, Denver, CO. Speth, T.F., Merkel, T., Gusses, A.M., 2002. Riverbank filtration as a pretreatment for nanofiltration membranes. In: Ray, C., Melin, G., Linsky, R.B. (Eds.), Riverbank Filtration: Improving Source-water Quality. Kluwer Academic Publishers, Dordrecht. Stamatelatou, K., Frouda, C., Fountoulakis, M.S., Drillia, P., Kornaros, M., Lyberatos, G., 2003. Pharmaceuticals and health care products in wastewater effluents: the example of carbamazepine. Water Sci. Technol. Water Supply 3 (4), 131e137. Ternes, T.A., 1998. Occurence of drugs in German sewage treatment plants and rivers. Water Res. 32 (11), 3245e3260. Ternes, T.A., Meisenheimer, M., McDowell, D., Sacher, F., Brauch, H.J., Haist-Gulde, B., Preuss, G., Wilme, U., ZuleiSeibert, N., 2002. Removal of pharmaceuticals during

drinking water treatment. Environ. Sci. Technol. 36 (17), 3855e3863. Ternes, T.A., Bonerz, M., Herrmann, N., Teiser, B., Andersen, H.R., 2007. Irrigation of treated wastewater in Braunschweig, Germany: an option to remove pharmaceuticals and musk fragrances. Chemosphere 66 (5), 894e904. Tixier, C., Singer, H.P., Oellers, S., Muller, S.R., 2003. Occurrence and fate of carbamazepine, clofibric acid, diclofenac, ibuprofen, ketoprofen, and naproxen in surface waters. Environ. Sci. Technol. 37 (16), 1061e1068. Tran, N.H., Urase, T., Kusakabe, O., 2009. The characteristics of enriched nitrifier culture in the degradation of selected pharmaceutically active compounds. J. Hazard. Mater. 171, 1051e1057. Trulleyova´, S., Rulı´k, M., 2004. Determination of biodegradable dissolved organic carbon in waters: comparison of batch methods. Sci. Total Environ. 332 (1e3), 253e260. Tufenkji, N., Ryan, J.N., Elimelech, M., 2002. The promise of bank filtration. Environ. Sci. Technol. 36 (21), 422Ae428A. Tyler, C.R., Spary, C., Gibson, R., Santos, E.M., Shears, J., Hill, E.M., 2005. Accounting for differences in estrogenic responses in Rainbow Trout (Oncorhynchus mykiss: Salmonidae) and Roach (Rutilus rutilus: Cyprinidae) Exposed to effluents from wastewater treatment works. Environ. Sci. Technol. 39 (8), 2599e2607. US EPA, 2009. Estimation Programs Interface Suite for Microsoft Windows, V. 4.00. United States Environmental Protection Agency, Washington, DC, USA. Verstraeten, I.M., Heberer, T., Scheytt, T., 2002. Occurrence, characteristics, and transport and fate of pesticides, pharmaceutical active compounds, and industrial and personal care products at bank-filtration sites. In: Riverbank Filtration e Improving Source Water Quality. Kluwer Academic Publishers, Dordrecht. Wang, J., Hubbs, S.A., Song, R., 2002. Evaluation of Riverbank Filtration as a Drinking Water Treatment Process. AWWA Research Foundation and AWWA. Weiss, W.J., Bouwer, E.J., Aboytes, R., LeChevallier, M.W., O’Melia, C.R., Le, B.T., Schwab, K.J., 2005. Riverbank filtration for control of microorganisms: results from field monitoring. Water Res. 39 (10), 1990e2001. Weiss, W.J., Bouwer, E.J., Ball, W.P., O’Melia, C.R., Aboytes, R., Speth, T.F., 2004. Riverbank filtration: effect of ground passage on NOM character. J. Water Supply Res. T. 53 (2), 61e83. Wick, A., Fink, G., Joss, A., Siegrist, H., Ternes, T.A., 2009. Fate of beta blockers and psycho-active drugs in conventional wastewater treatment. Water Res. 43 (4), 1060e1074. Winkler, M., Lawrence, J.R., Neu, T.R., 2001. Selective degradation of ibuprofen and clofibric acid in two model river biofilm systems. Water Res. 35 (13), 3197e3205. Wu, Y., Hui, L., Wang, H., Li, Y., Zeng, R., 2007. Effectiveness of riverbank filtration for removal of nitrogen from heavily polluted rivers: a case study of Kuihe River, Xuzhou, Jiangsu. China Environ. Geol. 52 (1), 19e25. Xue, S., Zhao, Q.L., Wei, L.L., Ren, N.Q., 2009. Behavior and characteristics of dissolved organic matter during column studies of soil aquifer treatment. Water Res. 43 (2), 499e507. Yangali-Quintanilla, V., Sadmani, A., McConville, M., Kennedy, M., Amy, G., 2009. Rejection of pharmaceutically active compounds and endocrine disrupting compounds by clean and fouled nanofiltration membranes. Water Res. 43 (9), 2349e2362. Ying, G.-G., Kookana, R.S., Dillon, P., 2003. Sorption and degradation of selected five endocrine disrupting chemicals in aquifer material. Water Res. 37 (15), 3785e3791. Yoon, Y., Westerhoff, P., Snyder, S.A., Wert, E.C., 2006. Nanofiltration and ultrafiltration of endocrine disrupting compounds, pharmaceuticals and personal care products. J. Membr. Sci. 270 (1e2), 88e100.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 1 5 e3 0 3 3

Zhang, Y., Geissen, S.-U., Gal, C., 2008. Carbamazepine and diclofenac: removal in wastewater treatment plants and occurrence in water bodies. Chemosphere 73 (8), 1151e1161. Zhou, J.L., Zhang, Z.L., Banks, E., Grover, D., Jiang, J.Q., 2009. Pharmaceutical residues in wastewater treatment works

3033

effluents and their impact on receiving river water. J. Hazard. Mater. 166 (2e3), 655e661. Zuehlke, S., Duennbier, U., Heberer, T., Fritz, B., 2004. Analysis of endocrine disrupting steroids: investigation of their Release into the environment and their behavior during bank filtration. Ground Water Monit. R 24 (2), 78e85.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Biological Cr(VI) removal coupled with biomass growth, biomass decay, and multiple substrate limitation E.M. Contreras a,b,*, A.M. Ferro Orozco a, N.E. Zaritzky a,b a

Centro de Investigacio´n y Desarrollo en Criotecnologı´a de Alimentos (CIDCA), CCT La Plata CONICET, Fac. de Cs. Exactas, UNLP. 47 y 116, B1900AJJ La Plata, Argentina b Fac. de Ingenierı´a, UNLP. 47 y 1, B1900AJJ La Plata, Argentina

article info

abstract

Article history:

In this work, a mathematical model for the biological reduction of Cr(VI), carbon and

Received 26 October 2010

nitrogen sources consumption, and biomass growth under fully aerobic conditions was

Received in revised form

developed. The model comprises three types of aerobic heterotrophic cells (non-growing

2 February 2011

cells, growing cells with chromate reductase activity, and growing cells that have lost the

Accepted 8 March 2011

chromate reductase activity), and five soluble compounds (organic substrate, ammonia

Available online 16 March 2011

nitrogen, non-metabolizable soluble products, dissolved oxygen, and hexavalent chromium). Two processes are considered responsible for the reduction of Cr(VI). The first one

Keywords:

is the reduction of Cr(VI) coupled with growth, the second process is coupled with the

Cr(VI) reduction

endogenous decay of the biomass. The model was calibrated using the results obtained in

Chromium toxicity

batch cultures in the absence of carbon and nitrogen sources, using different initial Cr(VI)

Kinetic parameter

concentrations (0e100 mgCr L
1), two carbon sources (cheese whey and lactose), and

Activated sludge

different initial nitrogen to carbon ratio (0e50 mgN gCOD
1). The calibrated model was

Nitrogen to carbon ratio

used to calculate steady-state values of TSS, soluble COD, TAN and Cr(VI) in continuous

Mathematical modeling

systems, obtaining a good agreement with the experimental data. The model also accurately predicted the transient concentration of Cr(VI) as a function of time in response to step changes of the inlet Cr(VI) concentration in continuous systems. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Biotransformation of metals is a promising technique to convert more toxic heavy metals into less toxic forms, and therefore, can be potentially useful for bioremediation of industrial wastewaters. A good example of biotransformation is the case of chromium. The extensive use of chromium in several industries such as petroleum refining, metal finishing, leather tanning, iron and steel industries, inorganic chemical production, textile manufacturing and pulp production have largely contributed to its wide spread in the environment (Katz and Salem, 1994;

Guertin et al., 2005). Although chromium has several oxidation states, chromium compounds mainly occur as Cr(III) or Cr (VI). The former is the most stable under reduced conditions, it is relatively immobile because it has a strong affinity for negative charged ions and colloids in soils, and gives sparingly soluble compounds such as Cr(OH)3 that dominate at pH values from 4 to 8. Conversely, Cr(VI) is more soluble, mobile, and bio-available than Cr(III). Cr(VI) occurs in the form of oxoanions under most environmental conditions. Equilibria of hexavalent chromium compounds are pH and concentration dependent. Chromate ion (CrO42
) represents more than 98% of the total Cr(VI) species for pH higher than 8. Chromate

* Corresponding author. Centro de Investigacio´n y Desarrollo en Criotecnologı´a de Alimentos (CIDCA), CCT La Plata CONICET, Fac. de Cs. Exactas, UNLP. 47 y 116, B1900AJJ La Plata, Argentina. Tel./fax: þ54 221 4254853. E-mail address: [email protected] (E.M. Contreras). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.011

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

Nomenclature Model compounds Cr (mgCr L
1) soluble hexavalent chromium concentration S (mgCOD L
1) organic substrate concentration P (mgCOD L
1) soluble products concentration released during the biomass decay NH (mgN L
1) total ammonia nitrogen (TAN) concentration XNG (mgCOD L
1) non-growing biomass concentration XGA (mgCOD L
1) growing active (with chromate reductase activity) biomass concentration XGNA (mgCOD L
1) growing non-active (without chromate reductase activity) biomass concentration Stoichiometric coefficients Y (mgCOD mgCOD
1) true biomass yield iN,BM (mgN mgCOD
1) nitrogen content of the biomass iN,P (mgN mgCOD
1) nitrogen content of the products released during the endogenous decay fP (mgCOD mgCOD
1) fraction of the endogenous biomass converted into soluble inert products

monoacid (HCrO4
), and dichromate (Cr2O72
) ions are the main species when pH is lower than 5. In addition, as the total Cr(VI) concentration increases, the fraction of HCrO4
decreases due to the formation of dichromate. In all cases, the fraction of chromic acid (H2CrO4) is negligible. Within the normal pH range of most biological wastewater treatment systems (pH about 5e9) and for Cr(VI) concentrations usually found in wastewaters (usually less than 2 mM), Cr2O72
ion account for less than 3% of the total Cr(VI); thus, HCrO4
and CrO42
are the dominant species (Contreras et al., in press). The reduction of Cr(VI) to Cr(III) is of great environmental importance, because Cr(III) is less hazardous. Chromium is an essential micro-nutrient in the diet of animals and humans, as it is indispensable for the normal sugar, lipid and protein metabolism of mammals (USEPA, 1998a). Conversely, Cr(VI) is highly toxic to all forms of living organisms, mutagenic in bacteria, mutagenic and carcinogenic in humans and animals (USEPA, 1998b). For these reasons, reducing Cr(VI) to Cr(III) is beneficial in eliminating the toxicity of Cr(VI) from the environment. For many years, conventional Cr(VI) removal was achieved by chemical reduction, ion exchange or adsorption. Recently, researchers have focused attention on biodetoxification of hexavalent chromium. In contrast to the conventional methods, biodetoxification is cost-effective (Li et al., 2007). Trivalent and hexavalent forms of chromium can interconvert; within normal conditions, the reduction of Cr(VI) to Cr(III) due to the presence of organic compounds is favored. For example, the change of free energy at pH 7.0 and 25  C corresponding to the reaction between acetic acid (electron donor) and chromate ion (electron acceptor) is about
83.3 kJ per mole of electrons transferred. Thus, from a thermodynamic point of view, Cr(VI) is capable of oxidizing acetic acid and also most of the organic compounds commonly found in wastewaters (Contreras et al., in press). However, the

3035

RC (mgCr mgCOD
1) Cr(VI) removed per unit mass of XGA converted into XGNA iSS (mgTSS mgCOD
1) conversion factor from biomass COD to TSS units Kinetic coefficients kA (h
1) first order lag phase constant mm0 (h
1) maximum specific growth rate in the absence of Cr(VI) mmN (h
1) maximum specific growth rate at infinite Cr(VI) qSm0 (mgCOD mgCOD
1 h
1) maximum specific substrate consumption rate in the absence of Cr(VI) qSmN (mgCOD mgCOD
1 h
1) maximum specific substrate consumption rate at infinite Cr(VI) concentration K (L mgCr
1) inhibition constant due to Cr(VI) toxicity KS (mgCOD L
1) half-saturation coefficient for S KNH (mgN/L) half-saturation coefficient for NH b (h
1) specific endogenous decay rate qCrGm (mgCr mgCOD
1 h
1) maximum specific Cr(VI) consumption rate associated to the growth phase qCrDm (mgCr mgCOD
1 h
1) maximum specific Cr(VI) consumption rate associated to the decay phase

reduction of Cr(VI) to Cr(III) by organic compounds is frequently slow; thus, in the absence of a catalyst, such as the activated sludge biomass, redox systems are far from equilibrium (Stumm and Morgan, 1996). Therefore, the presence of cells that act as catalyst, and a suitable electron donor are necessary to achieve the reduction of Cr(VI). The ability of Cr(VI) reduction has been found in many bacterial genera including Pseudomonas, Micrococcus, Bacillus, Achromobacter, Microbacterium, Arthrobacter, and Corynebacterium (McLean et al., 2000; Pattanapipitpaisal et al., 2001; McLean and Beveridge, 2001; Megharaj et al., 2003). The mechanisms through which bacterial strains reduce Cr(VI) to Cr(III) are variable and species dependent (McLean et al., 2000). Anaerobic bacteria may use chromate as a terminal-electron acceptor or reduce chromate in periplasmatic space by hydrogenase or cytocrome c3 (Michel et al., 2001; Puzon et al., 2002). In aerobic bacteria, Cr(VI) reduction may be carried out by cellular reducing agents (the primary reductant is glutathione) and NADH-dependent chromate reductase (Shen and Wang, 1994; Garbisu et al., 1998). The mechanisms for Cr(VI) reduction might be a secondary utilization or cometabolism as suggested for Shewanella onoidensis MR-1 (Middleton et al., 2003). Although there are many studies concerning the removal of Cr(VI) by pure cultures of different microorganisms, its applicability to field removal processes is limited. The main disadvantage of using pure cultures to remove chromium compounds is related to the use of sterile conditions to prevent external microbial contamination, increasing the treatment costs. If sterile conditions are not employed (due to economic reasons, for example), indigenous bacteria may overcome the added Cr(VI)-reducing microorganisms. Besides, most countries have severe restrictions concerning the introduction of new species. For these reasons, results obtained in the study of Cr(VI) removal by mixed cultures,

3036

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

such as activated sludge, are more relevant to optimize design and operation of actual biological Cr(VI) reduction systems. Many factors affect the reduction of Cr(VI) by activated sludge. In previous works it was demonstrated that the presence of suitable carbon and nitrogen sources are necessary to enhance the Cr(VI) reduction capacity of activated sludge (Ferro Orozco et al., 2007, 2010a). Several studies (Shen and Wang, 1994; Wang and Shen, 1995, 1997) demonstrated that the rate and extent of Cr(VI) reduction in batch cultures depend on the initial biomass concentration, regardless of the subsequent growth. Recently, Ferro Orozco et al. (2010a) demonstrated that the net biological Cr(VI) removal is the consequence of two processes: a fast Cr(VI) removal process, associated to the biomass growth, and a slow removal process that is independent of the presence of both the carbonaceous substrate and the nitrogen source. While the former process is important when there is no limitation in carbon or nitrogen sources (within the exponential growth phase in batch systems, for example), the latter may be significant in continuous cultures in which substrate concentrations are normally low (Ferro Orozco et al., 2010a). The utilization of biological processes for the treatment of wastewaters containing toxic compounds, such as Cr(VI), emphasizes the practical requirement for developing adequate mathematical models to be used for the design and operation of these processes. The knowledge of microbial substrate utilization kinetics and the effect of toxic compounds on the growth rate are both important for the accurate prediction of the quality of the treatment process effluent. Accurate models and kinetic parameters also help engineers to optimize operational conditions in order to meet discharge requirements minimizing the operational costs. In this work, a mathematical model that describes the biological removal of Cr(VI) coupled with biomass growth, biomass decay, and multiple substrate limitation was developed. The model takes into account the effect of the initial Cr (VI) and substrates (carbon and nitrogen sources) concentrations, and Cr(VI) toxicity on the observed growth rate of activated sludge and on Cr(VI) removal. The model was calibrated using batch experiments under different conditions, and it was validated by comparing model results with those obtained in continuous systems.

2.

Mathematical model

2.1.

General description

Three types of aerobic heterotrophic cells were considered in the model: non-growing cells (XNG), growing cells with chromate reductase activity (XGA), and growing cells that have lost the chromate reductase activity (XGNA). Non-growing cells (XNG) are converted to growing cells with chromate reductase activity (XGA) following a first order kinetics; this process was introduced to simulate the lag phase observed in batch cultures (Ferro Orozco et al., 2010b). In addition, due to toxic effects of Cr(VI), XGA cells loose its chromate reductase activity to produce XGNA (Yamamoto et al., 1993; Wang and Shen, 1997; Ferro Orozco et al., 2008). Organic substrate (S ) and ammonia nitrogen (NH ) can limit the growth rate of XGA and XGNA; for

this reason, Monod saturation terms corresponding to each substrate were included. Under endogenous decay, the three types of cells release non-metabolizable soluble products (P), and ammonia nitrogen (NH ). Growth of XGA produces new XGA cells, however, during the growth of XGNA new cells with chromate reductase activity (XGA) are formed. In addition, to reduce model complexity, growth kinetics corresponding to XGA and XGNA were assumed identical. Two processes are considered responsible for the reduction of Cr(VI) to Cr(III) (Ferro Orozco et al., 2010a). The first one is the reduction of Cr(VI) coupled with the growth of XGA; during this process some electrons of the organic substrate are transferred to Cr(VI). Although this process is fast, it depends on the availability of S and NH. Because both S and NH can limit the growth of XGA, if one or more of those substrates are absent then this process stops. The second Cr (VI) reduction process is coupled with the endogenous decay of the biomass; in this case, the source of electrons to reduce Cr(VI) is the biomass itself. This process is slower than the first one but not depends on the availability of substrates (Ferro Orozco et al., 2010a).

2.2.

Definitions of model compounds

The following compounds are used in the model: S (mgCOD L
1) ¼ organic substrate concentration, carbon and energy source for biomass growth. P (mgCOD L
1) ¼ soluble products concentration, released during the biomass decay. NH (mgN L
1) ¼ total ammonia nitrogen (TAN) concentration, nitrogen source for biomass growth. XNG (mgCOD L
1) ¼ non-growing biomass concentration (biomass in Lag phase). XGA (mgCOD L
1) ¼ growing active (with chromate reductase activity) biomass concentration. XGNA (mgCOD L
1) ¼ growing non-active (without chromate reductase activity) biomass concentration. Cr (mgCr L
1) ¼ soluble hexavalent chromium concentration.

2.3.

Stoichiometry and kinetics

The stoichiometric matrix nj,i of the model is shown in Table 1; all empty elements of nj,i indicate zero values. The model has six stoichiometric coefficients: Y (mgCOD mgCOD
1) ¼ true biomass yield iN,BM (mgN mgCOD
1) ¼ nitrogen content of the biomass iN,P (mgN mgCOD
1) ¼ nitrogen content of the products released during the endogenous decay fP (mgCOD mgCOD
1) ¼ fraction of the endogenous biomass converted into soluble inert products RC (mgCr mgCOD
1) ¼ Cr(VI) removed per unit mass of XGA converted into XGNA iSS (mgTSS mgCOD
1): conversion factor from biomass COD to TSS units A first order kinetics with a constant kA was assumed for the conversion of non-growing cells (XNG) to growing cells

3037

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

Table 1 e Stoichiometric matrix (nj,i) of the proposed model and process rate expressions (rj). Process ( j )

Rate (rj)

Compound (i) 1 XNG

1 2

Activation of XNG to XGA Growth of XGA

3

Growth of XGNA

2 XGA


1

3 XGNA

4S

1 1

1 Y 1
Y


1

4 Decay of XNG 5 Decay of XGA 6 Decay of XGNA 7 Cr reduction by growth of XGA 8 Cr reduction by decay of XNG 9 Cr reduction by decay of XGA Observed conversion rates


1
1


ri ¼

P

1 RC

1 RC


1

5 NH
iN,BM

6 DO


7P

9 Cr kA XNG m XGA

1
Y


iN,BM

Y 1
Y
Y

(iN,BM
iN,P) (iN,BM
iN,P) (iN,BM
iN,P)


(1
fP)
(1
fP)
(1
fP)

m XGNA fP fP fP
1
1
1

b XNG b XGA b XGNA qCrG XGA qCrD XNG qCrD XGA

nj;i rj

j

where m, qS, qCrG, qCrD, and Y are defined by Eqs. (1)e(5), respectively.

with chromate reductase activity (XGA) to simulate the lag phase (Ferro Orozco et al., 2010b). In addition, the following kinetic expressions corresponding to the specific growth rate (m), specific substrate consumption rate (qS), specific Cr(VI) consumption rate associated to growth phase (qCrG) and to endogenous decay phase (qCrD) were proposed: m¼


   mm0 þ mmN KCr S NH 1 þ KCr KS þ S KNH þ NH

qS ¼

 
 qSm0 þ qSmN KCr S NH 1 þ KCr KS þ S KNH þ NH 

qCrG ¼ qCrGm

KCr 1 þ KCr

 qCrD ¼ qCrDm

  S NH KS þ S KNH þ NH

Y¼

m qS

(5)

(1)

From model outputs, soluble COD (CODS), total suspended solids (TSS), and oxygen uptake rate (OUR) can be computed as follows:

(2)

CODS ¼ S þ P

(6)

TSS ¼ iSS ðXNG þ XGA þ XGNA Þ

(7)



 KCr 1 þ KCr

both carbon (S ) and nitrogen sources (NH ) can limit the growth process, Monod-type saturation terms were included. The true biomass growth yield (Y ) can be calculated by combining Eqs. (1) and (2) as follows:

(3)

(4)

where: mm0 (h
1) ¼ maximum specific growth rate in the absence of Cr (VI) mmN (h
1) ¼ maximum specific growth rate at infinite Cr(VI) concentration qSm0 (mgCOD mgCOD
1 h
1) ¼ maximum specific substrate consumption rate in the absence of Cr(VI) qSmN (mgCOD mgCOD
1 h
1) ¼ maximum specific substrate consumption rate at infinite Cr(VI) concentration K (L mgCr
1) ¼ inhibition constant due to Cr(VI) toxicity KS (mgCOD L
1) ¼ half-saturation coefficient for S KNH (mgN L
1) ¼ half-saturation coefficient for NH b (h
1) ¼ specific endogenous decay rate qCrGm (mgCr mgCOD
1 h
1) ¼ maximum specific Cr(VI) consumption rate associated to growth qCrDm (mgCr mgCOD
1 h
1) ¼ maximum specific Cr(VI) consumption rate associated to decay Eqs. (1) and (2) take into account the toxic effect of Cr(VI) on the specific growth rate (m) and specific substrate consumption (qS) of activated sludge. The removal of Cr(VI) associated with the growth and the endogenous decay of activated sludge were represented by Eqs. (3) and (4), respectively. Because

OUR ¼

    1
Y mðXGA þXGNA Þþ 1
fP bðXNG þXGA þXGNA Þ Y

(8)

It must be pointed out that the present model was developed for fully aerobic conditions. Aeration conditions were assumed high enough as to maintain the dissolved oxygen (DO) close to the saturation level. However, if this is not the case, Eqs. (1)e(3) can be modified to include Monod saturation terms corresponding to DO. Although the model allowed predicting the OUR (Eq. 8), the effect of the DO concentration on the kinetics is out of the scope of the present work.

3.

Materials and methods

3.1.

Biological and chemical materials

All reagents used in the present work were commercial products of reagent grade from Anedra (San Fernando, Argentina). Activated sludge used in all the experiments were harvested from a continuous aerobic laboratory-scale (4.5 L) activated sludge reactor with partial biomass recycle. Aeration was provided by an air pump; air was pumped near the bottom of the reactor and it was enough to maintain the dissolved oxygen concentration above 4 mgO2 L
1. The reactor was fed with a synthetic wastewater with the following composition: dehydrated cheese whey (Food S.A. Villa Maipu´, Argentina)

3038

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

1.5 g, (NH4)2SO4 0.90 g, and NaHCO3 1.03 g dissolved in 1 L of tap water. Soluble chemical oxygen demand (CODS) of the synthetic wastewater was 1500 mg L
1. The hydraulic retention time was 2 d; the sludge age was maintained at 45 d by daily wasting of mixed liquor directly from the reactor. During the experiments the temperature of the reactor was 20  2  C. Under steady-state conditions dissolved oxygen concentration (DO) was above 4 mg L
1, pH was 7.9  0.5, CODS of the effluent ranged between 30 and 160 mg L
1, and total suspended solid (TSS) concentration ranged from 3700 to 5400 mgTSS L
1.

provides not only organic matter but also organic nitrogen, cheese whey was replaced by lactose in order to control the initial nitrogen to carbon ratio (N0/S0). Therefore, in these experiments the culture medium composition was the following: lactose (carbon and energy source) 5.6 gCOD L
1, ammonium sulfate (nitrogen source) 0 to 250 mgN L
1, and micro-nutrient solutions M1 and M2 (1 mL L
1). At predetermined time intervals, biomass (X ), soluble COD (CODS), total ammonia nitrogen (TAN), and Cr(VI) concentrations were measured.

3.2. Cr(VI) removal by activated sludge in different systems

3.2.4.

3.2.1.

Cr(VI) removal in the absence of substrates

The removal of Cr(VI) under endogenous decay conditions of activated sludge were studied in 250 mL aerated vessels at constant temperature (20  2  C) and initial pH ¼ 7.0  0.1. Small air pumps were employed to aerate and to agitate the vessels. Air was pumped near the bottom of each vessel via a flexible hose; aeration was enough to maintain the dissolved oxygen concentration above 5 mgO2 L
1. Activated sludge samples were obtained from the reactor described previously (Section 3.1). The samples were washed three times with phosphate buffer (KH2PO4 2 g L
1, K2HPO4 0.5 g L
1, pH ¼ 7) before performing the assays in order to remove all remnant of substrates. Activated sludge washed samples were re-suspended in phosphate buffer to obtain an initial biomass concentration of 3800  200 mgTSS L
1. A Cr(VI) stock solution (10 gCr(VI) L
1) was prepared using analytical grade K2Cr2O7; an appropriate volume of this solution was added to obtain 10e100 mgCr(VI) L
1. At predetermined time intervals, samples were withdrawn from the vessels to determine biomass (X ), soluble COD (CODS), total ammonia nitrogen (TAN), and Cr(VI) concentrations.

3.2.2. Effect of Cr(VI) on activated sludge growth and Cr(VI) removal These experiments were performed in 250 mL aerated vessels at room temperature (20  2  C) and initial pH ¼ 7.0  0.1. Activated sludge samples were washed before performing the assays. In these experiments the initial biomass concentration was 700  50 mgTSS L
1. The culture medium composition was the following: dehydrated cheese whey (carbon and energy source) 5 gCOD L
1, ammonium sulfate (nitrogen source) 212 mgN L
1, and micro-nutrient solutions M1 and M2 (1 mL L
1). The composition of M1 was (expressed as g 100 mL
1): FeSO4.7H2O 1.5, ZnSO4.7H2O 0.5, MnSO4.H2O 0.3, CuSO4.5H2O 0.075, CoCl2.6H2O 0.015, and citric acid 0.6. M2 solution contained the following salts (gg 100 mL
1): (NH4)6Mo7O24.4H2O 0.05, BO3H3 0.01, KI 0.01. Tested Cr(VI) concentrations were 0, 10, 25, 50, and 100 mg L
1. At predetermined time intervals samples were taken to determine biomass (X ), soluble COD (CODS), and Cr(VI) concentrations, and oxygen uptake rate (OUR).

3.2.3. Effect of the initial nitrogen to carbon ratio (N0/S0) on activated sludge growth and Cr(VI) removal Cr(VI) removal batch assays, similar to those described in the previous section, were performed. Because the cheese whey

Cr(VI) removal in continuous systems

The removal of Cr(VI) in continuous systems was also studied. These experiments were performed in the activated sludge reactors described in Section 2.1. Two sludge ages (qC) were tested 20 and 45 d; sludge age was maintained at the desired value by daily wasting of mixed liquor directly from the reactor. In both cases the hydraulic retention time was 2 d. The synthetic wastewater described in Section 2.1 was used to feed the reactors; an appropriate volume of a Cr(VI) stock solution (10 gCr(VI) L
1) was added to the synthetic wastewater to obtain the desired inlet Cr(VI) concentration. For qC ¼ 20 d, the tested inlet Cr(VI) concentrations were 0, 10, 25, and 50 mgCr L
1; for qC ¼ 45 d, 0 and 10 mgCr L
1 were tested. Reactors were considered to run under steady-state conditions after operating for one sludge age, at least. Then, samples were withdrawn from the reactor to determine biomass (X ), soluble COD (CODS), total ammonia nitrogen (TAN), and Cr(VI) concentrations.

3.3.

Analytical techniques

Total suspended solids (TSS, mg L
1) were used to measure the biomass concentration (X ). Known sample volumes (8 mL in this work) were poured into pre-weighted centrifuge tubes, centrifuged and washed twice with distilled water, and placed at 105  C for 24 h; TSS of each sample was calculated as the difference between final weight (dry sample þ tube) and initial weight (tube alone) divided by the sample volume. It must be pointed out that TSS is a lumped parameter that included the three types of aerobic heterotrophic cells: non-growing cells (XNG), growing cells with chromate reductase activity (XGA), and growing cells that have lost the chromate reductase activity (XGNA). In all batch growth experiments duplicate biomass measurements were performed to reduce experimental errors; mean and maximum errors for TSS were 4 and 13%, respectively. Soluble chemical oxygen demand (CODs) was determined as follows: 3 mL of culture samples were centrifuged for 5 min at 13 000 rpm (Eppendorf 5415C); then, the supernatant was filtered through 0.45 mm cellulosic membranes (Osmonics Inc.). Soluble COD of the filtrate was determined using commercial reagents (Hach Company, Loveland, CO). Total ammonia nitrogen (TAN) concentration of the filtrate was measured by the Nessler colorimetric method using commercial reagents (Hach Company, Loveland, CO). Cr(VI) of the filtrate was determined colorimetrically using a spectrophotometer (Hach DR 2000) at 540 nm by reaction with 1,5-diphenylcarbazide in acid solution (APHA, 2005).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

Oxygen uptake rate (OUR, mgO2 L
1) measurements were performed using a closed respirometer consisted in a 30 mL glass vessel maintained at 20  0.5  C by means of a water bath. The vessel was filled with the tested sample, air was supplied until oxygen saturation level was reached; then, the vessel was sealed with the insertion of a polarographic oxygen probe (YSI model 5739). The sample was continuously stirred with a magnetic stirring and the decay of the dissolved oxygen (DO) concentration as a function of time was recorded. Data were acquired by a personal computer interfaced to the DO monitor (YSI model 58). OUR was calculated as the slope of the straight line obtained by plotting the DO concentration as a function of time. All the results reported in the present work are average values of two experiments, at least.

3.4. Estimation of the model coefficients and dynamic simulations The estimation of the coefficients of the mathematical model proposed in this paper and the dynamic simulations were performed using the software package Gepasi 3 (Mendes, 1993). Gepasi integrates the systems of differential equations with the routine LSODA (Livermore Solver of Ordinary Differential Equations). LSODA algorithm measures the stiffness of the equations and switches the integration method dynamically according to this measure. For non-stiff regions, the Adams integration method with variable step size and variable order up to 12th order is used; for stiff regions the Gear (or BDF) method with variable step size and variable order up to 5th order is used. Among the optimization methods available in Gepasi 3, the Multistart Optimization algorithm (with LevenbergeMarquardt local optimization) was selected. Multistart is a hybrid stochastic-deterministic optimization method. Rather than run a simple local optimization (e.g. gradient descent methods), Multistart runs several of them, each time starting from a different initial guess. The first start takes for initial guess the parameter values entered by the user. The initial guesses for the subsequent starts are generated randomly within the boundaries for the adjustable parameters. The local optimizer used is the LevenbergeMarquardt method as this has proved the most efficient gradient optimizer used in Gepasi 3. In order to reduce fitting errors of the adjustable coefficients, initial concentrations were adjusted within 5% of their measured values. This procedure takes into account the degree of uncertainty in the initial conditions due to analytical errors (Mendes and Kell, 1998). For more details see the supplementary data.

4.

Results and discussion

4.1.

Cr(VI) removal in the absence of substrates

In order to study the effect of Cr(VI) on the endogenous decay of activated sludge, Cr(VI) removal experiments in the absence of substrates were performed. Fig. 1 shows that no trend can be observed with regard to the biomass decay, ammonia release, and soluble COD production as a function of the initial Cr(VI) concentration. These results indicate that Cr(VI) exerted

3039

a negligible effect on the decay rate of activated sludge; therefore, for modeling purposes a unique value corresponding to the endogenous decay rate constant (b) was assumed. This approach was also adopted by other authors (Elangovan and Philip, 2009). The proposed model was fitted to the data shown in Fig. 1 to obtain the following coefficients: b, iN,P, fP, and qCrDm (Table 2). Obtained coefficients are within the range reported by other authors. For example, the specific endogenous decay rate (b) obtained in the present work was (3.3  0.4)  10
3 h
1. Elangovan and Philip (2009) reported a value of b ¼ 0.002 h
1 corresponding to an activated sludge removing Cr(VI) in aerobic batch systems. The default value for the aerobic endogenous respiration rate recommended in the Activated Sludge Model #3 ranged between 0.004 and 0.008 h
1 (Gujer et al., 1999). The value corresponding to the specific Cr(VI) consumption rate in the absence of substrates obtained in this work, qCrDm ¼ (2.2  1.0)  10
5 mgCr mgCOD
1 h
1, is close to the value reported in a previous paper (qCrDm ¼ 2.3  10
5 mgCr mgCOD
1 h
1) (Ferro Orozco et al., 2010a).

4.2. Effect of Cr(VI) on activated sludge growth and Cr (VI) removal Fig. 2 shows the time course of biomass (X ), soluble COD, oxygen uptake rate (OUR), and Cr(VI) for different initial Cr(VI) concentrations. In all cases, the increase in Cr(VI) concentration produced a longer lag phase; this effect was also reported by other authors (Gikas and Romanos, 2006; Li et al., 2007; Elangovan and Philip, 2009; Gikas et al., 2009; Sengor et al., 2009). After the lag phase, biomass and OUR increased, and soluble COD decreased as a function of time. In all cases, when the soluble COD was depleted, biomass entered to endogenous decay phase and OUR dropped to endogenous levels. Using the values of b, iN,P, fP, and qCrDm obtained in the previous section (Table 2), the proposed model was fitted to the data shown in Fig. 2 to obtain the following coefficients: mm0, mmN, qSm0, qSmN, and K (Table 2). Fig. 2 shows that the proposed model predicts satisfactorily the values of soluble COD, X, OUR, and Cr(VI) as a function of time for all the tested conditions. Although the effect of Cr(VI) on the lag phase was not included explicitly in the proposed model, it was predicted quite well. Taking into account that Cr(VI) exerted a negligible effect on the decay rate of activated sludge (Fig. 1a), this phenomenon can be attributed to the toxic effect of Cr(VI) on mm. For this reason, the term between brackets in Eq. (2) was included to represent the inhibition of mm as a function of Cr (VI). Because it was assumed that the first order constant (kA) for the conversion of XNG to XGA was not dependent on the Cr (VI) concentration, the increase of the lag phase was only due to the inhibition of mm by Cr(VI). A different approach was used by Gikas et al. (2009) and Sengor et al. (2009) concerning the effect of metals on the lag phase. In the model developed by those authors, the transformation of resting cells to actively growing cells was represented by the metabolic potential function. This function starts at zero value and rises to unity through a convolution of the history of S. Those authors proposed that the lag time is a function of the local metal concentration (Sengor et al., 2009). Although the model of

3040

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

Fig. 1 e Time course of (a) total suspended solids (TSS), (b) soluble COD, (c) total ammonia nitrogen (TAN), and (d) hexavalent chromium (Cr(VI)) in batch cultures. Tested initial Cr(VI) concentrations were the following: (C) 0 mgCr LL1, (;) 10 mgCr LL1, (-) 25 mgCr LL1, (A) 50 mgCr LL1, and (:) 100 mgCr LL1. Lines indicate the proposed model.

Table 2 e Coefficients of the proposed model. Coefficient kA (cheese whey) kA (lactose) mm0 (cheese whey) mm0 (lactose) mmN (cheese whey) mmN (lactose) qSm0 (cheese whey) qSm0 (lactose) qSmN (cheese whey) qSmN (lactose) K KS KNH b qCrGm qCrDm iN,BM iN,P fP RC iSS

Units

Value

Reference/experiment

h h
1 h
1 h
1 h
1 h
1 mgCOD mgCOD
1 h
1 mgCOD mgCOD
1 h
1 mgCOD mgCOD
1 h
1 mgCOD mgCOD
1 h
1 L mgCr
1 mgCOD L
1 mgN L
1 h
1 mgCr mgCOD
1 h
1 mgCr mgCOD
1 h
1 mgN mgCOD
1 mgN mgCOD
1 mgCOD mgCOD
1 mgCr mgCOD
1 mgSS mgCOD
1

0.012 0.013  0.002 0.303  0.004 0.176  0.055 0.023  0.001 0.001  0.005 0.519  0.006 0.413  0.127 0.086  0.001 0.018  0.012 0.453  0.014 188 10 (3.3  0.4)  10
3 1.5  10
4 (2.2  1.0)  10
5 0.07 0.045  0.005 0.133  0.002 0.019 0.77

Ferro Orozco et al., 2010b Section 4.3 Section 4.2 Section 4.3 Section 4.2 Section 4.3 Section 4.2 Section 4.3 Section 4.2 Section 4.3 Section 4.2 Ferro Orozco et al., 2010b Contreras et al., 2008 Section 4.1 Ferro Orozco et al., 2008 Section 4.1 Gujer et al., 1999 Section 4.1 Section 4.1 Ferro Orozco et al., 2008 Gujer et al., 1999


1

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

3041

Fig. 2 e Time course of (a) total suspended solids (TSS), (b) soluble COD, (c) oxygen uptake rate (OUR), and (d) hexavalent chromium (Cr(VI)) in batch cultures with cheese whey as the carbon source. Cr(VI) initial concentrations were: (C) 0 mgCr LL1, (;) 10 mgCr LL1, (-) 25 mgCr LL1, (A) 50 mgCr LL1, and (:) 100 mgCr LL1. Lines indicate the proposed model.

those authors may provide good results regarding the effect of metals on the lag phase, a mathematically simpler approach such as the proposed in the present work was enough to represent our experimental results with an acceptable accuracy (Fig. 2). The inhibition term in Eq. (2) is similar to the one proposed by other authors (Yamamoto et al., 1993; Elangovan and Philip, 2009). Fig. 3 shows the effect of Cr(VI) on the dimensionless specific growth rate (mm/mm0) calculated by Eq. (1) using the values for mm0, mmN, and K shown in Table 2; for comparison purposes, literature data concerning the effect of Cr(VI) of the ratio mm/mm0 are also depicted. In agreement to the proposed expression (Eq. 1), in all cases a similar trend can be observed. Hexavalent chromium exerts a strong inhibition on mm within the concentration range 0e20 mgCr L
1; then, a much lesser effect of Cr(VI) on mm can be noticed for higher Cr(VI) concentrations (Fig. 3). The proposed model also predicts that the true biomass yield (Y ) decreases as a function of Cr(VI). This effect of Cr(VI) on Y is in accordance with results of Fig. 2a which shows that the maximum biomass concentration decreases as a function of the initial Cr(VI) concentration. In addition, several authors report this effect of Cr(VI) on Y (Srinath et al., 2002; JuveraEspinosa et al., 2006; Villegas et al., 2008). According to Rittmann and Saez (1993), this behavior is typical of inhibitors

Fig. 3 e Effect of Cr(VI) on the dimensionless specific growth rate (mm/mm0) calculated with Eq. (1) using the coefficients shown in Table 2 (dotted line) and those adapted from Elangovan and Philip (2009) (continuous line), Bae et al. (2000) (circles), and Stasinakis et al. (2002) triangles.

3042

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

that uncouple substrate oxidation and biomass synthesis. Uncouplers usually make the cytoplasmatic membrane permeable for protons, lowering the proton motive force across the membrane and therefore, the amount of ATP synthesized per unit of substrate oxidized. Because ATP and biomass synthesis are related, as a result, the presence of uncouplers lowers the biomass yield. The effect of Cr(VI) on lowering the biomass yield can be explained taking into account that the reduction of chromate to Cr(III) is a protonconsuming reaction (Katz and Salem, 1994); thus, this reaction could lower the proton motive force across the cytoplasmatic membrane, in a similar manner than an uncoupler.

4.3. Effect of the initial nitrogen to carbon ratio (N0/S0) on activated sludge growth and Cr(VI) removal Fig. 4 shows some examples of the time course of biomass (X ), soluble COD, total ammonia nitrogen (TAN), and Cr(VI) for different initial TAN concentrations. In these cases, cheese whey was replaced by lactose as the carbon source in order to control the initial nitrogen to carbon ratio (N0/S0). In general, the lag phases observed in these experiments (Fig. 4) were longer than the corresponding to the experiment with cheese whey as the carbon source in the presence of 25 mgCr L
1 (Fig. 2). While the lag phase with lactose was about 70 h, only 30 h was necessary to observe growth when cheese whey was tested. This difference in the lag phase occurred because the

model wastewater used to feed the parent activated sludge reactor had cheese whey as the carbon source, thus, microorganisms were adapted to this substrate but not to lactose. Fig. 2 shows that in the experiments with cheese whey, the increase in Cr(VI) concentration produced longer lag phases. Conversely, the lag phase observed in the experiments with lactose was constant (Fig. 4); taking into account that in these experiments the initial Cr(VI) concentration was constant, this result confirms that the lag phase depended on the Cr(VI) concentration but not on the ratio N0/S0. Fig. 4 shows that the removal of Cr(VI) in the absence of a nitrogen source was very low. When the initial TAN concentration was 250 mgN L
1, the soluble COD (lactose) was almost depleted but TAN values were about 70 mgN L
1 at t ¼ 140 h; therefore, this case corresponded to a carbonlimited assay (high N0/S0 ratio). Beyond this time, Cr(VI) removal rate drastically decreased and TAN concentration increased due to the biomass decay. When the initial TAN concentration was 60 mgN L
1, the nitrogen source was depleted at t ¼ 90 h; although the carbonaceous substrate concentration was around 3000 mgCOD L
1, the biomass growth stopped. Hence, in this case the biomass growth was limited by the nitrogen source (low N0/S0 ratio). Because lactose was the carbon but also the energy source, biomass concentration remained almost constant; however, Cr(VI) removal rate decreased due to the depletion of the nitrogen source.

Fig. 4 e Time course of (a) total suspended solids (TSS), (b) soluble COD, (c) total ammonia nitrogen (TAN), and (d) hexavalent chromium (Cr(VI)) in batch cultures with lactose as the carbon source. Initial TAN concentrations were: (C) 0 mgN LL1, (;) 60 mgN LL1, (-) 130 mgN LL1, and (A) 250 mgN LL1. Lines indicate the proposed model.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

The following coefficients of the proposed model were fitted to the Cr(VI) removal experiments with different initial TAN concentrations: KA, mm0, mmN, qSm0, qSmN; these coefficients were adjusted because they represent the growth kinetics of activated sludge using lactose as the carbon source. Table 2 shows that the kinetic coefficients corresponding to the growth of activated sludge using cheese whey were higher than those obtained with lactose. In contrast, KA values corresponding to both substrates were similar. The true biomass growth yield (Y ) values in the absence of Cr(VI) calculated by means of Eq. (5) for cheese whey and lactose were 0.58 and 0.43 gCOD gCOD
1, respectively. These results indicate that the growth of activated sludge is favored by a complex carbon source, such as cheese whey, in comparison with a defined medium with lactose as the carbon source. Fig. 4 shows that the proposed model predicts quite well the values for biomass, soluble COD, TAN, and Cr(VI) as a function of time for all the tested conditions. The effect of the ratio between initial concentration of the nitrogen source and the carbonaceous substrate (N0/S0) on the removal of Cr(VI) is shown in Fig. 5; in all cases, Cr(VI) removal was calculated at t ¼ 140 h Cr(VI) removal increased from 2 mgCr L
1 in the case of assays without nitrogen source addition (N0/S0 ¼ 0 mgN gCOD
1) to about 15 mgCr L
1 for N0/S0 ranging between 19 and 28 mgN gCOD
1. Further increments on the ratio N0/S0 did not enhance the removal of Cr(VI) because the carbonaceous substrate was completely consumed before the nitrogen source; in these cases, growth and Cr(VI) removal were limited by the carbonaceous source. The initial N0/S0 ratio determines the substrate that limits the biomass growth. The stoichiometric N0/S0 ratio, (N0/S0)St,

Fig. 5 e Effect of the initial nitrogen to carbon ratio (N0/S0) on the removal of Cr(VI). Initial conditions: Cr (VI) [ 25 mg LL1, X [ 700 ± 50 mgTSS LL1, Lactose [ 5 g LL1. In all cases Cr(VI) removal were calculated at t [ 140 h. Bars indicate the standard deviation. The shaded band indicates the range of stoichiometric N0/S0 values obtained by the proposed model.

3043

can be defined as the initial N0/S0 ratio at which both N and S are depleted at the same time; in this case both N and S are limiting substrates simultaneously. If N0/S0 is higher than (N0/ S0)St, N is in excess with respect to the stoichiometric requirements and S will be the growth limiting substrate. Conversely, when N0/S0 is lower than (N0/S0)St, the carbonaceous substrate is in excess with respect to the stoichiometric requirements and N will limit the biomass growth. If assimilation (incorporation of the nitrogen source into biomass) is the only nitrogen consumption process (e.g. if nitrification can be neglected), (N0/S0)St can be calculated as the product between the nitrogen content of the biomass (iN,BM), and the biomass growth yield (Y ). Although the value corresponding to iN,BM for activated sludge can be considered as a constant, Y depends on many factors, such as wastewater characteristics, culture conditions, and the presence of toxic compounds (e.g. Cr(VI)). Using the values of the coefficients mm0, mmN, qSm0, and qSmN shown in Table 2 corresponding to lactose, Y was calculated as a function of Cr(VI) by combining Eqs. (1), (2), and (5); then, considering iN,BM ¼ 0.07 mgN mgCOD
1, (N0/S0)St was calculated as a function of Cr(VI). Because Y varies with Cr(VI), calculated (N0/S0)St values also depended on Cr(VI); thus, two limiting conditions were considered. If the consumption of Cr (VI) is negligible then Cr(VI) ¼ 25 mgCr L
1 (the initial CrVI concentration in these experiments); in this case (N0/ S0)St ¼ 21 mgN gCOD
1. Conversely, if Cr(VI) is depleted then (N0/S0)St ¼ 30 mgN gCOD
1. Fig. 5 shows that this range was in accordance to the experimental values (19e28 mgN gCOD
1).

4.4. Cr(VI) removal in continuous systems and model validation The removal of Cr(VI) using activated sludge reactors operated at two sludge ages (20 and 45 d) and different inlet Cr(VI) concentrations were studied; in all cases cheese whey was the carbon source. Fig. 6a shows that steady-state values of TSS decreased as the inlet Cr(VI) concentration increased. This effect is in accordance with the decrease of the biomass as a function of the initial Cr(VI) concentration shown in Fig. 2a, confirming that the biomass yield (Y ) decreases as a function of Cr(VI). Besides, no trend can be observed with respect to steady-state values of the soluble COD and TAN (Fig. 6b,c). In addition, Fig. 6d shows that the Cr(VI) in the outlet stream increased as a function of the inlet Cr(VI) concentration. Steady-state values of TSS, soluble COD, TAN and Cr(VI) measured in continuous systems were compared to steadystate values calculated by the proposed model using the coefficients shown in Table 2 corresponding to cheese whey as the carbon source. Fig. 6 shows that, taking into account the experimental errors, in all cases model calculations were in agreement with the experimental results. Moreover, Fig. 7 shows that the model accurately predicts the transient concentrations of Cr(VI) as a function of time in response of step changes in the inlet Cr(VI) concentration. These results demonstrate that the proposed model could be a powerful tool to predict the effect of the operating conditions on the performance of an activated sludge reactor treating Cr(VI)containing wastewaters in the presence of a readily biodegradable organic matter (cheese whey in this work) under fully aerobic conditions. The presence of high organic matter

3044

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

Fig. 6 e Steady-state values of (a) total suspended solids (TSS), (b) soluble COD, (c) total ammonia nitrogen (TAN), and (d) hexavalent chromium (Cr(VI)) as a function of the inlet Cr(VI) concentration in continuous systems feed with cheese whey as the carbon source and operating at different sludge ages: (C, continuous line) 20 d; (-, dashed line) 45 d. In all cases the hydraulic retention time was 2 d. Dots indicate experimental results, bars show the standard deviation. Lines indicate the proposed model.

concentrations along with Cr(VI) can occur when wastewaters from more than one industry are mixed (Elangovan and Philip, 2009; Cokgor et al., 2008, 2009). In the case of Cr(VI)-containing wastewaters with low organic matter content, such as electroplating, pigmentation, and wood preservation (Katz and Salem, 1994), a carbon source has to be supplied externally; thus, the addition of cheese whey (a residue from dairy industries) could be a suitable alternative to enhance the Cr (VI) reduction due to its low cost.

5.

Fig. 7 e Time course of Cr(VI) in a continuous activated sludge reactor feed with cheese whey as the carbon source and operating at a sludge age of 20 d and hydraulic retention time of 2 d. Bars indicate the standard deviation. Dotted line indicates the inlet Cr(VI) concentration. Continuous line represents the proposed model.

Conclusions

The mathematical model developed in the present work adequately describes the biological reduction of Cr(VI), carbon and nitrogen sources consumption, and biomass growth under fully aerobic conditions. Model coefficients were obtained by adjusting the model to the results obtained in batch cultures under different conditions such as: in the absence of carbon and nitrogen sources, different initial Cr(VI) concentrations, two carbon sources, and different initial nitrogen to carbon ratio. Then, the calibrated model was used to calculate steady-state values of TSS, soluble COD, TAN and

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

Cr(VI) in continuous systems, obtaining a good accordance with the experimental data. The model also accurately predicted the transient concentration of Cr(VI) in continuous systems as a function of time in response to step changes of the inlet Cr(VI) concentration. The proposed model could be a powerful tool to predict the effect of the operating conditions on the performance of an activated sludge reactor treating Cr(VI)-containing wastewaters in the presence of a readily biodegradable organic matter under fully aerobic conditions.

Acknowledgments The authors gratefully acknowledge the financial support given by Universidad Nacional de La Plata (UNLP), Consejo Nacional de Investigaciones Cientı´ficas y Te´cnicas (CONICET), and Agencia Nacional de Promocio´n Cientı´fica y Tecnolo´gica Argentina (ANPCyT).

Appendix. Supplementary data Supplementary data related to this article can be found online at doi:10.1016/j.watres.2011.03.011.

references

APHA, 2005. Standard methods for the examination of water and wastewater. In: Eaton, A.D., Clesceri, L.S., Rice, E.W., Greenberg, A.E., Franson, M.A.H. (Eds.), 21st ed. American Public Health Association, American Water Works Association, Water Environment Federation, USA (centennial edition). Bae, W.C., Kang, T.G., Kang, I.K., Won, Y.J., Jeong, B.C., 2000. Reduction of hexavalent chromium by Escherichia coli ATCC 33456 in batch and continuous cultures. The Journal of Microbiology 38 (8), 36e39. Cokgor, E.U., Insel, G., Aydin, E., Orhon, D., 2009. Respirometric evaluation of a mixture of organic chemicals with different biodegradation kinetics. Journal of Hazardous Materials 161 (1), 35e41. Cokgor, E.U., Karahan, O., Orhon, D., 2008. The effect of mixing pharmaceutical and tannery wastewaters on the biodegradation characteristics of the effluents. Journal of Hazardous Materials 156 (1e3), 292e299. Contreras, E.M., Ferro Orozco, A.M., Zaritzky, N.E. Factors affecting the biological removal of hexavalent chromium using activated sludges. In: Balart M. (Ed.) Management of Hazardous Residues Containing Cr(VI). Nova Science Publishers, Inc. New York, in press. Contreras, E.M., Ruiz, F., Bertola, N.C., 2008. Kinetic modelling of inhibition of ammonia oxidation by nitrite under low dissolved oxygen conditions. Journal of Environmental Engineering 134 (3), 184e190. Elangovan, R., Philip, L., 2009. Performance evaluation of various bioreactors for the removal of Cr(VI) and organic matter from industrial effluent. Biochemical Engineering Journal 44 (2e3), 174e186. Ferro Orozco, A.M., Contreras, E.M., Bertola, N.C., Zaritzky, N.E., 2007. Hexavalent chromium removal using aerobic activated

3045

sludge batch systems added with powdered activated carbon. Water SA 33 (2), 239e244. Ferro Orozco, A.M., Contreras, E.M., Zaritzky, N.E., 2008. Modelling Cr(VI) removal by a combined carbon-activated sludge system. Journal of Hazardous Materials 150 (1), 46e52. Ferro Orozco, A.M., Contreras, E.M., Zaritzky, N.E., 2010a. Cr(VI) reduction capacity of activated sludge as affected by nitrogen and carbon sources, microbial acclimation and cell multiplication. Journal of Hazardous Materials 176 (1e3), 657e665. Ferro Orozco, A.M., Contreras, E.M., Zaritzky, N.E., 2010b. Dynamic response of combined activated sludge-powdered activated carbon batch systems. Chemical Engineering Journal 157 (2e3), 331e338. Garbisu, C., Alkorta, I., Llama, M.J., Serra, J.L., 1998. Aerobic chromate reduction by Bacillus subtilis. Biodegradation 9 (2), 133e141. Gikas, P., Romanos, P., 2006. Effects of tri-valent (Cr(III)) and hexavalent (Cr(VI)) chromium on the growth of activated sludge. Journal of Hazardous Materials 133 (1e3), 212e217. Gikas, P., Sengor, S.S., Ginn, T., Moberly, J., Peyton, B., 2009. The effects of heavy metals and temperature on microbial growth and lag. Global NEST Journal 11 (3), 325e332. Guertin, J., Jacobs, J.A., Avakian, C.P., 2005. Chromium (VI) Handbook. IETEG: Independent Environmental Technical Evaluation Group. CRC Press. Gujer, W., Henze, M., Mino, T., van Loosdrecht, M., 1999. Activated sludge model no. 3. Water Science and Technology 39 (1), 183e193. Juvera-Espinosa, J., Morales-Barrera, L., Cristiani-Urbina, E., 2006. Isolation and characterization of a yeast strain capable of removing Cr(VI). Enzyme and Microbial Technology 40 (1), 114e121. Katz, S.A., Salem, H., 1994. The Biological and Environmental Chemistry of Chromium. VCH Publishers Inc., New York, USA. Li, Y., Low, G.K.-C., Scott, J.A., Amal, R., 2007. Microbial reduction of hexavalent chromium by landfill leachate. Journal of Hazardous Materials 142 (1e2), 153e159. McLean, J.S., Beveridge, T.J., 2001. Chromate reduction by a pseudomonad isolated from a site contaminated with chromated copper arsenate. Applied and Environmental Microbiology 67 (3), 1076e1084. McLean, J.S., Beveridge, T.J., Phipps, D., 2000. Isolation and characterization of chromium-reducing bacterium from a chromated copper arsenate-contaminated site. Environmental Microbiollogy 2 (6), 611e619. Megharaj, M., Avudainayagam, S., Naidu, R., 2003. Toxicity of hexavalent chromium and its reduction by bacteria isolated from soil contaminated with tannery waste. Current Microbiology 47, 51e54. Mendes, P., 1993. GEPASI: a software package for modelling the dynamics, steady states and control of biochemical and other systems. Computer Applications in the Biosciences 9 (5), 563e571. Mendes, P., Kell, D.B., 1998. Non-linear optimization of biochemical pathways: applications to metabolic engineering and parameter estimation. Bioinformatics 14 (10), 869e883. Michel, C., Brugna, M., Aubert, C., Bernadac, A., Bruschi, M., 2001. Enzymatic reduction of chromate: comparative studies using sulfate-reducing bacteria. Key role of polyheme cytochromes c and hydrogenases. Applied Microbiology and Biotechnology 55 (1), 95e100. Middleton, S.S., Latmani, R.B., Mackey, M.R., Ellisman, M.H., Tebo, B.M., Criddle, C.S., 2003. Cometabolism of Cr(VI) by Shewanella oneidensis MR-1 produces cell-associated reduced chromium and inhibits growth. Biotechnology and Bioengineering 83 (6), 627e636. Pattanapipitpaisal, P., Brown, N.L., Macaskie, L.E., 2001. Chromate reduction and 16S rRNA identification of bacteria isolated

3046

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 3 4 e3 0 4 6

from a Cr(VI)-contaminated site. Applied Microbiology and Biotechnology 57 (1e2), 257e261. Puzon, G.J., Petersen, J.N., Roberts, A.G., Kramer, D.M., Xun, L., 2002. A bacterial flavin reductase system reduces chromate to a soluble chromium(III)-NAD(þ) complex. Biochemical and Biophysical Research Communications 294 (1), 76e81. Rittmann, B.E., Saez, P.B., 1993. Modeling biological processes involved in degradation of hazardous organic substrates. In: Levin, M., Gealt, M. (Eds.), Biotreatment of Industrial and Hazardous Wastes. McGraw Hill Book Co., New York, pp. 113e136. Sengor, S.S., Barua, S., Gikas, P., Ginn, T.R., Peyton, B., Sani, R.K., Spycher, N.F., 2009. Influence of heavy metals on microbial growth kinetics including lag time: mathematical modeling and experimental verification. Enironmental Toxicology and Chemistry 28 (10), 2020e2029. Shen, H., Wang, Y.T., 1994. Biological reduction of chromium by E. coli. Journal of Environmental Engineering 120 (3), 560e572. Srinath, T., Verma, T., Ramteke, P.W., Garg, S.K., 2002. Chrumium (VI) biosorption and bioaccumulation by chromate resistant bacteria. Chemosphere 48 (4), 427e435. Stasinakis, A.S., Mamais, D., Thomaidis, N.S., Lekkas, T.D., 2002. Effect of chromium (VI) on bacterial kinetics of heterotrophic biomass of activated sludge. Water Research 36 (13), 3341e3350.

Stumm, W., Morgan, J.J., 1996. Aquatic Chemistry. Chemical Equilibria and Rates in Natural Waters, third ed. Wiley, New York. 425e513. US Environmental Protection Agency, 1998a. Toxicological review of hexavalent trivalent. In: Support of Summary Information on the Integrated Risk Information System (IRIS) (CAS No. 16065-83-1). US. Environmental Protection Agency, Washington, DC. US Environmental Protection Agency, 1998b. Toxicological review of hexavalent chromium. In: Support of Summary Information on the Integrated Risk Information System (IRIS) (CAS No.18540-29-9). US. Environmental Protection Agency, Washington, DC. Villegas, L.B., Ferna´ndez, P.M., Amoroso, M.J., de Figueroa, L.I.C., 2008. Chromate removal by yeast isolated from sediments of a tanning and a mine site in Argentina. Biometals 21 (5), 591e600. Wang, Y.T., Shen, H., 1995. Bacterial reduction of hexavalent chromium. Journal of Industrial Microbiology and Biotechnology 14 (2), 159e163. Wang, Y.T., Shen, H., 1997. Modeling Cr(VI) reduction by pure bacterial cultures. Water Research 31 (4), 727e732. Yamamoto, K., Kato, J., Yano, T., Ohtake, H., 1993. Kinetics and modeling of hexavalent chromium reduction in Enterobacter cloacae. Biotechnology and Bioengineering 41 (1), 129e133.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 4 7 e3 0 5 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Comparative studies on the retardation and reduction of glyphosate during subsurface passage N.T. Litz a,*, A. Weigert c, B. Krause a, S. Heise a, G. Gru¨tzmacher b a

Federal Environmental Agency (UBA), Section Drinking Water Resource Protection and Water Treatment e Center for Aquatic Simulations, Marienfelde, Schichauweg 58, D-12307 Berlin, Germany b KompetenzZentrum Wasser Berlin, Cicerostraße 24, D-10709 Berlin, Germany c University of Dresden, Institute for Urban Water Management, George-Ba¨hr-Str. 1, 01062 Dresden, Germany

article info

abstract

Article history:

The herbicide Glyphosate was detected in River Havel (Berlin, Germany) in concentrations

Received 18 November 2010

between 0.1 and 2 mg/L (single maximum outlier: 5 mg/L). As the river indirectly acts as

Received in revised form

drinking water source for the city’s 3.4 Mio inhabitants potential risks for drinking water

10 February 2011

production needed to be assessed. For this reason laboratory (sorption and degradation

Accepted 12 February 2011

studies) and technical scale investigations (bank filtration and slow sand filter experi-

Available online 8 March 2011

ments) were carried out.

Keywords:

concentrations between 0.1 and 100 mg/L. Degradation experiments at 8  C with oxygen

Mitigation

limitation resulted in a decrease of Glyphosate concentrations in the liquid phase probably

Adsorption

due to slow adsorption (half life: 30 days).

Batch adsorption experiments with Glyphosate yielded a low KF of 1.89 (1/n ¼ 0.48) for

Degradation

During technical scale slow sand filter (SSF) experiments Glyphosate attenuation was

Glyphosate retardation

70e80% for constant inlet concentrations of 0.7, 3.5 and 11.6 mg/L, respectively. Relevant

Bank filtration

retardation of Glyphosate breakthrough was observed despite the low adsorption potential

Slow sand filter

of the sandy filter substrate and the relatively high flow velocity. The VisualCXTFit model

Modelling

was applied with data from typical Berlin bank filtration sites to extrapolate the results to a realistic field setting and yielded sufficient attenuation within a few days of travel time. Experiments on an SSF planted with Phragmites australis and an unplanted SSF with mainly vertical flow conditions to which Glyphosate was continuously dosed showed that in the planted SSF Glyphosate retardation exceeds 54% compared to 14% retardation in the unplanted SSF. The results show that saturated subsurface passage has the potential to efficiently attenuate glyphosate, favorably with aerobic conditions, long travel times and the presence of planted riparian boundary buffer strips. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

The herbicide Glyphosate [N-(phosphonomethyl)glycine] and its metabolite AMPA (aminomethylphosphonic acid) have been reported to occur in surface and ground water throughout the world (GEUS, 2004, Struger et al., 2008). Wide-

spread application of Glyphosate in agriculture, in silvicultural and urban areas requires special attention to their possible transport from terrestrial to aquatic environments (Borggaard and Gimsing, 2008; Soerensen et al., 2006). Based on their chemical properties and analyzed concentrations an increased risk for drinking water supply can not be ruled out

* Corresponding author. E-mail address: [email protected] (N.T. Litz). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.02.015

3048

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 4 7 e3 0 5 4

(EC, 2002, Vereecken, 2005). A possible breakthrough would mean additional expenses for purification techniques, like oxidation, which could bring along new risks e.g. by producing other metabolites. Subsurface passage represents an important natural barrier for many diffuse pollutants in the anthropogenically affected water cycle. Infiltration of water takes place during natural groundwater recharge, after surface runoff in rural riparian zones or (constructed) wetlands as well as during artificial groundwater recharge and bank filtration. In order to assess the risk associated with the occurrence of unwanted substances in the environment and to design possible mitigation strategies it is therefore necessary to quantify possible removal capacities in the subsurface for a wide range of scenarios. Within the frame of two research projects funded by the Berliner Wasserbetriebe (BWB) and Veolia Water investigations were carried out on Glyphosate attenuation by saturated subsurface passage which occurs during drinking water production via bank filtration or artificial groundwater recharge. In a pre-study the relevance of Glyphosate and AMPA was assessed for the Berlin water supply, which utilizes bank filtration and artificial recharge along the city’s abundant rivers and lakes for the production of about 60% of its drinking water (Zippel and Hannappel, 2008). This pre-study was initiated after Glyphosate was detected in the region’s surface waters by Schmidt and Boas (2006) with concentrations of up to 5 mg/L in the river Havel upstream and downstream of the city. Thus, a screening for Glyphosate and AMPA was carried out within the city’s limits on two occasions (March and June 2008), yielding detectable concentrations of Glyphosate and AMPA in 78% and 100% of the analyzed samples, respectively. In 4 of 22 sampling sites Glyphosate concentration exceeded the European threshold value for drinking water (EU, 1998) of 0.1 mg/L (maximum 0.5 mg/L). AMPA concentrations exceeded the permitted limit for relevant pesticide metabolites at 16 sampling points (maximum 3 mg/L). Maximum concentrations were found at places where surface water is influenced by urban runoff or waste water (for further detail see: Krause et al., 2009). The investigations described below aimed at identifying and quantifying the processes responsible for glyphosate attenuation during subsurface passage using so-called enclosures and semi-technical scale vertical slow sand filters (SSFs) at the UBA’s Center for Aquatic Simulations (CAS) in Berlin. In addition batch experiments and degradation studies were conducted under defined laboratory conditions to investigate the adsorption and degradation potential of Glyphosate for the specific conditions at the CAS. Finally the one-dimensional substance transport model VisualCXTFit was applied to obtain substance specific parameters for Glyphosate and to transfer these results to conditions met in the field.

2.

Materials and methods

In all experimental settings e laboratory batch, enclosure and SSF tests- the same filter material was used. The texture of the

applied sandy substrate can be characterized as follows: in average 2% fine sand (0.1e0.2 mm), 43% medium sand (>0.2e0.5 mm), 49% coarse sand (>0.5e2.0 mm) and 6% fine gravel (>2 mm), no clay or silt with only traces of organic matter and an effective porosity of 0.38e0.4% (Table 1). The pH value of the percolated water was w7.7. Solid Glyphosate produced by SigmaeAldrich with a purity degree of 98.7%, dissolved in a 0.01 M CaCl2-solution, was used for the experiments. Glyphosate concentrations were analyzed according to the German Standard DIN 38407-22 (2001). The quantitative determination of AMPA and Glyphosate was done using a Waters HPLC system with a fluorescence detector and two Knauer 64 as reagent pumps. The analytical column for Glyphosate was a Supelco SAX column (25  4 mm), for the quantification of AMPA a cation exchange column (Pickering) was applied (15  4 mm), because in field samples the AMPA peak was interfered by matrix peaks. The run conditions were: 0.4 ml/min, isocratic, phosphate buffer pH 2.05  0.1 at 50  C. Retention time for Glyphosate was 13.6 min on the anion exchange column and for AMPA 13.9 min on the cation exchange column. The detection limits were 0.02 mg/L and 0.005 mg/L, the quantification limit 0.07 mg/L and 0.02 mg/L for Glyphosate, for AMPA, respectively. The two analytes AMPA and Glyphosate were detected after a 2-step post-column derivatization. The first step was an oxidation with a phosphate buffer containing sodium hypochlorite (0.4 mL/min) in a 10 m reaction coil of PEEK tubing (i. d. 0.25 mm, volume 500 mL) at 50  C, the second a transformation into fluorescing compounds by reaction with phthaldialdehyde and 2-mercaptoethanol in an alkaline borate buffer (0.3 mL/min) in a 2 m reaction coil of PEEK tubing (i. d. 0.25 mm, volume 100 mL) at ambient temperature. The excitation wavelength of the resulting compounds was 390 nm and the emission wave length 450 nm. All solutions were degassed and filtered through 0.45 mm prior to use. Samples of the filter substrate were extracted according to Bo¨rjesson and Torstensson (2000): 10 g of the sample were brought into contact for 30 min with 25 mL of 1 M NaOH. Subsequently the mixture was centrifuged for 15 min at 3000 rpm. The supernatant was abstracted with a pipette and the extraction was repeated. 4.2 mL concentrated HCl was added to the combined supernatants. After dilution of the sample with deionised water to a volume of 200 mL the

Table 1 e Characterisation of the enclosure filling material (Massmann et al., 2004). Characteristics

Clogging layer

Filter substrate

Drainage stratum

Soil type Thickness [m] CUa/CGa Fe(ox) [mg/kg] Mn(ox) [mg/kg] Corg/Canorg [%]

n.a. 0.05b n.a. 605 68 0.343/1.4

mS, gS, fg 1 3.2/0.7 275 11 0.022/0.12

fG, mg 0.25 2.0/1.0 n.a. n.a. n.a.

a Parameters for classification of non-structured sediments (uniformity coefficient, coefficient of gradation). b Clogging layer is situated in the upper layer of the filter substrate, n.a. ¼ not analyzed.

3049

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 4 7 e3 0 5 4

analytes Glyphosate and AMPA were determined as described above. The cleanup of the water samples was also performed according to the abovementioned German standard method DIN 38407-22. Water samples obtained from laboratory-, and enclosure experiments (typically 100e500 mL) were filtrated through glass fiber filters and adjusted with hydrochloric acid to pH 2  0.1. The filtrate was applied to a column filled with a cation exchange resin which had been loaded with Fe3þ ions. Subsequently the column was rinsed with 20 mL water and 40 mL 0.02 M HCl. The analyteeiron complex was eluted with 10 mL 6 M HCl and 4 mL 32% HCl were added to the eluate. This solution was applied to an anion exchange column. By elution of the column with 6 M HCl the iron was retained on the column.

3.

Laboratory experiments

3.1.

Batch experiments

The batch experiments were conducted according to OECD 106 (OECD 2000) using the filter substrate and deionized water with Glyphosate concentrations of 0.1 mg/L, 1 mg/L, 10 mg/L and 100 mg/L and a soil/water-ratio of 1:2, shaking the mixture for 4 h to establish an equilibrium. The chosen concentrations were applied in three parallels. After centrifugation the supernatant was carefully extracted and prepared for measurement. The FREUNDLICH adsorption isothermal model was used to describe the nonlinear water/sediment distribution relations (KF) over the total concentration range. The equation’s first differentiation was used to describe also the linear distribution coefficient (KD) and to estimate retardation factors (RF).

3.2.

Degradation experiment

Degradation studies were carried out by taking a defined sediment sample of 450 g wet material and mixing it with 10 mg Glyphosate per kg filter substrate. The vessels were stored in the dark at a temperature of around 8  C for a period

water column

A - enclosure

of up to 73 days to allow for biological degradation processes to take place. The airtight stoppers of the vessels sealed the sample from the atmosphere. During the experiment the vessels were left undisturbed. The redox potential, oxygen content, pH value and the temperature in the supernatant were determined after the respective vessels were opened and sampled. At intervals (7, 14, 21, 28 and 73 days) two experiment vessels were opened at a time. This experimental arrangement was intended to simulate naturally deposited filter substrate under partly reducing conditions, as it would be expected in slowly flowing groundwater.

4.

Technical scale experiments

4.1.

Enclosure experiment

Water production pre-treatment via bank filtration or/and slow sand filtration is commonly used if drinking water is produced from surface water. In enclosure experiments the attenuation of compounds can be determined simulating conditions that occur during slow sand filtration or within the first meter of infiltration. The enclosures are three metal cylinders with an area of 1 m2 and a height of 1.85 m (filtration length 1.00 m) (see Fig. 1A). They are situated within an infiltration pond (area: 90m2) in order to be exposed to natural environmental conditions. Three different concentration levels of Glyphosate were continuously dosed to the supernatant of the enclosures over a time period of 14 d from 20th October to 6th November 2007, yielding average inlet concentrations of 0.7, 3.5 and 11.6 mg/L. Water samples for Glyphosate and AMPA analysis were taken for 34 days from the supernatant, from sampling points within the filter material and from the filter effluent. The flow rate was set at 50 cm/d and was controlled by adjustable pumps connected to the enclosure outlets. The depth of the supernatant was kept constant by siphoning the water out of the infiltration pond into the enclosure without additional pumping. The water in the infiltration pond originates from a large storage pond (volume of 7000 m3) with relatively high mineralization (average electrical conductivity: 1000 mS/m) but low nutrient status (nitrate < 1 mg/L, ortho-

B - SSL planted - unplanted Slow Sand Filter

sediment surface

filter substrate

0,2 m P1

inlet mixing cell

0,4 m P 2

1,0 m

pond

0,6 0,6 m mP3

Sand

0,8 m P 4

o gravel layer

effluent

o

drainage pipes o

0,8 m

outlet

1,0 m

Fig. 1 e Schematic cross section and location of sampling ports in enclosures (A) and slow sand filter e infiltration site with inlet and outlet devise (B) according to Gru ¨ tzmacher et al. (2006).

3050

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 4 7 e3 0 5 4

phosphate < 1 mg/L, DOC 3e4 mg/L) thus representing oligotrophic surface water.

4.2.

Slow sand filter (SSF) experiments

The SSF experiments were conducted at two vertical-flow experimental SSFs: (Fig. 1B) one without vegetation cover (average area 60 m2, filter depth 0.8 m, filter volume 48 m3) and the other with a 3 year old vegetation cover of Phragmites australis (average area 68 m2, filter depth 1.2 m, filter volume 81.6 m3) to simulate processes in grown planted bank filtration sites along rivers or surface water lakes. Due to the arrangement of inflow, water reservoir and drainage pipes, water flow through the SSFs was assumed to be predominantly vertical simulating conditions that occur during the first meter of bank filtration. The water fluxes of the unplanted and the planted SSF were regulated at the outlet and were regularly controlled by discharge measurements. Their yield amounted in average to approximately 0.41 and 0.45 m3/h, respectively (corresponding to a filtration velocity of 0.16 and 0.18 m/d). Physico-chemical parameters of the water (pH, redox potential, and temperature) as well as DOC, PO43 and NO3 concentrations were also monitored to gain insights into controlling processes. After an equilibration phase of 1 month during which nitrate and phosphate were dosed to target 10 mg/L N and 1 mg/L PO43 in the supernatant, Glyphosate was additionally applied for 22 days with a target concentration of 20 mg/L.

5.

Results and discussion

5.1.

Batch experiments

Glyphosate exhibits under different site conditions a complex adsorption behavior in the environment (Borggard and Gimsing, 2008) which is influenced by pH and by variation of soil constituents and the chemical Glyphosate species (Vereeken, 2005). In order to determine the distribution coefficient of Glyphosate degree of adsorption in the filter substrate batch experiments were conducted. The resulting linear regression with a FREUNDLICH sorption coefficient (KF) of 1.90 and a FREUNDLICH exponential of 0.48 confirms the poor adsorptive characteristics of the sandy material and indicates beginning saturation at higher concentrations (Table 2). With sorption data from different concentration ranges a calculation of the adsorption coefficients (KD-value) was carried out for different concentration ranges. Due to lower adsorption at high concentrations the KD-values decrease by 3 orders of magnitude when regarding the complete range of concentrations from 0.1 to 100 mg/L. This is in agreement with comparable experiments of Mentler et al. (2006) with sandy material, which is well comparable to the one used in this study, where a KD-range of 1.5e2.9 L/kg was determined. Compared to other studies on Glyphosate adsorption with soils showing KD values that range from 62 to 410 L/kg (Strange-Hansen et al., 2004; Bergstro¨m et al., 2010) these values are quite low. This is most probably due to the low content of clay, iron and aluminum oxide or organic matter content (Vereeken, 2005) in the filter material. Only

Table 2 e Estimated retardation of Glyphosate in the filter substrate on the basis of FREUNDLICH distribution equation. Concentration (caq) [mg/L] 100 10 1 0.1 0.02

Gradient from first differentiation (G)a [L/kg]

Retardation factor (Rf)b

0.08 0.28 0.9 3 7

1.4 2.2 5 14 31

a G ¼ 1/n  KF  (caq)1/n1 with KF: 1.9 mg11/n  L1/n kg1 and 1/n: 0.48. b Rf ¼ 1 þ (rb/ne)  G, with an effective porosity (ne): 0,37 and bulk density (rb): 1.59 kg/L.

some iron and organic matter content may have influenced the sorption in our filter material and should be responsible for slightly elevated adsorption coefficients (5.4 L/kg) at least with low Glyphosate concentrations (0.1e1 mg/L).

5.2.

Degradation experiment

It is well known that Glyphosate degrades more easily under aerobic conditions compared to anaerobic conditions (Borggard and Gimsing, 2008). Fig. 2 shows the residual Glyphosate concentrations, obtained from the analysis of the solvent samples in the batch degradation experiment under anaerobic conditions. As it is not clear, if the reduction of concentrations was due to degradation or adsorption, the term dissipation will be used in the following. The development of the redox potential and oxygen content during the degradation experiment showed that oxygen-free conditions were partially achieved. The oxygen in the supernatant was almost completely consumed (data not shown) whereas the pH value remained constant at around 7. 7. Dissipation of 50% (DT50) of the Glyphosate in the supernatant was calculated to be achieved after 30.5 days yielding a rate of dissipation of 0.0227 d1. A mass balance approach was carried out taking into account the initially applied amount of Glyphosate, the concentrations measured in solution and the adsorbed fraction. During the first 30 days the decrease in dissolved concentration is due to a continuous in adsorption in this time (data not shown). Degradation must therefore be negligible in the first time. Similar findings in anoxic substrate have been reported by So¨rensen et al. (2006). A single non reliable Glyphosate measurement of the solid and solvent sample after 72 days let assume that degradation could finally occur under this conditions (data not shown). The results of laboratory degradation studies were in contrast to differed from our findings in the outside door enclosure experiments, which were carried out under more aerobic and temperate conditions.

5.3.

Enclosure experiments

By simulating slow sand filter conditions enclosure experiments can help to verify the risk for groundwater pollution by

3051

y = 15e

16

R = 0.901

14 12 10 8 6 4

DT 50

2 0

0

10

20

30

40

50

60

70

80

Duration of the experiment [d]

AMPA Glyphosate quantification limit - AMPA (0,02 µg/L) quantification limit - Glyphosate (0,07 µg/L) Tracer

1170 1120 1070

0.4 1020

0.0

920 20

25

1707

2.0

1507

1.5 1307 1.0 1107

0.5

30

907 5

10

15

20

25

30

35

Fig. 4 e Glyphosate and AMPA concentrations in the outlet of enclosure III (with an average inlet Glyphosate concentration of 11.6 mg/L).

970

15

2.5

Duration of the experiment [d]

Electrical conductivity [µS/m]

Glyphosate- and AMPA- concentration [µg/L]

0.8

10

1907

day 6 in enclosure III and on day 12 in enclosure II. Our experiments for simulation of the mobility of Glyphosate in slow sand filter reflect the results of de Jonge et al. (2000) concerning the mobility behavior of Glyphosate in a comparable sandy soil. Whereas Strange-Hansen et al. (2004) found higher outlet concentrations, however for in shorter and smaller laboratory soil columns. The vertical concentration profile is illustrated exemplary for enclosure III in Fig. 5. This illustrates that retardation and degradation processes are distributed almost linearly along the filtration depth as this was also observed in experiments by Bergstro¨m et al. (2010). Tracer and Glyphosate concentrations at the outlets of enclosures II and III were modeled using the computer program VisualCXTFit. On the basis of the hydrodynamic properties of the filter substrate obtained from the tracer experiment (R2 ¼ 0.95 and 0.93 for enclosures II and III, respectively (data not shown), it was possible to assess the retardation and degradation capacity of the enclosures for Glyphosate. The modeled results of the Glyphosate concentrations in enclosures II and III corresponds well compared to

1270 1220

5

2107

3.0

0

contaminants entering from surface waters. Glyphosate and AMPA concentrations in enclosures II and III for the time of the experiment (34 days) are given in Figs. 3 and 4. Glyphosate was continuously dosed for 14 days to both enclosures reaching in average concentrations of 3.5 and 11.6 mg/L, respectively, with a standard deviation of 20%. The two concentrations reflect medium and maximum levels generally observed in surface water. In enclosure II the Glyphosate concentrations at the outlet reached a maximum value of 0.7 mg/L towards the end of the experiment (after 34 days). Since the experiment was terminated before the concentrations decreased again the point in time for the peak value could only been estimated. A breakthrough curve was observed in enclosure III, to which the highest Glyphosate concentration was applied. The maximum outlet concentration for Glyphosate of 2.7 mg/L occurred after 23 days. After 8 days (enclosure III) and after 17 days (enclosure II) nearly all observed Glyphosate concentrations exceeded the European limit for pesticides in drinking water of 0.1 mg/L. AMPA concentrations above 0.1 mg/L have been observed since

0

3.5

0.0

Fig. 2 e Glyphosate partitioning between solid and aqueous phase during degradation batch experiments (points represent samples from 2 replicates for each sampling date).

1.2

AMPA Glyphosate q. limit - AMPA (0,02 µg/L) q. limit - Glyphosate (0,07 µg/L) Tracer

4.0

-0.0227x

2

Electrical conductivity [µS/m]

18

Glyphosate- and AMPA- concentration [µg/L]

Glyphosate concentration in solution [mg/L]

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 4 7 e3 0 5 4

35

Duration of the experiment [d]

Fig. 3 e Glyphosate and AMPA concentrations in the outlet of enclosure II (with an average inlet Glyphosate concentration of 3.5 mg/L).

Supernatant - 20 cm - 40 cm - 60 cm - 80 cm Outlet 0

5

10

15

Glyphosate concentration [µg/L]

Fig. 5 e Vertical distribution of Glyphosate concentrations in enclosure III on 05.11.2007 (16 days after dosing commenced).

3052

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 4 7 e3 0 5 4

5.4.

Slow sand filter experiments

For simulating Glyphosate attenuation in a riparian zone studies with an adapted planted SSF and unplanted SSF were conducted. The hydro-chemical analyses (tracer tests, breakthrough curves of nitrate, see Jacinthe et al. (2008)) indicated that the planted SSF does not show a homogeneous vertical flow pattern. Thus the planted SSF is divided into two zones (right and left) with different hydraulic and subsequently hydro-chemical characteristics and an estimation of the hydraulically effective surface area was carried out. These estimations showed a reduction in average surface area of the planted SSF to around 67% of the unplanted SSF, confirming that the flux in the planted SSF seems to be partly inhibited. The lowering to around 67% of the average surface area could be explained by collimation due to high production of biomass which at constant hydraulic head results in a decrease of pore velocities or even blocking of pore volume. The concentrations of Glyphosate measured in the mixing cell, in the supernatant, in 40 cm depth and in the outlet of the planted SSF (left site) are given in Fig. 7. In the mixing cell of the planted SSF the average Glyphosate concentration of 21.2 mg/L was slightly higher than the targeted level of 20 mg/L. In the left zone of the planted SSF only little reduction was observed in the water reservoir above the SSF surface (19 mg L1 in average). In 40 cm depth the maximum concentration of Glyphosate was retarded by 11 days and reduced to approximately 7 mg/L (63% of the average concentration in the supernatant). In the right zone (data not shown) the concentrations decreased by more than 50% between mixing cell and surface water of the SSF. Glyphosate was completely removed from solution in 40 cm depth, which seems to be due to lower inlet concentrations, higher residence times and therefore higher efficiency of reduction. In the combined outlet (left and right zone) the fluxes of all sampling sites rejoined and resulted in a maximum concentration of 1.4 mg/L. The final measurements at the end of the experiment show a reduction for about 93% of the applied Glyphosate compared to the inlet concentration.

dosing

35

2.5

30

surface

25

outlet

Glyphosate conc. [µg/L]

3

5

70 80 Time [d]

07 09

.1

1.

07 04

.1

1.

07 30

.1

0.

07 0. .1 25

0.

07

0

.1

Fig. 6 e Modeled length of the filter substrate (from left to right: 1.25; 2.0; 3.0; 3.5 and 3.75 m) in order to ensure a reduction of the Glyphosate concentrations below the European threshold for drinking water of 0.1 mg/L (enclosure III).

07

60

20

50

0.

40

.1

30

07

20

0.

10

.1

0

10

15

0

15

07

0.5

0.

1

20

.1

1.5

mixing cell 40 cm depth

10

2

05

Glyphosate concentration [µg/L]

the observed breakthrough curves. Based on the recovered concentrations at the outlet the applied Glyphosate was reduced by 78e80%. Modeling yielded a retardation factor of 25 and 18 and a degradation rate of 0.0069 d1 and 0.092 d1 in enclosures II and III, respectively. The enclosure experiments confirm the results of the laboratory experiments that the adsorption behavior of Glyphosate is linked to concentration as we observe a decrease of retardation with increasing concentrations. However, the retardation factors observed in the enclosure experiments range far below those calculated on the basis of the laboratory experiments (31 and higher). This could be explained by the artificial conditions in the laboratory adsorption test (e.g. homogeneous mixing of sand and water). In addition the amount of binding sites in the enclosures is reduced by permanent contact of the filter substrate grains at their touch points (Fehse, 2004). The half-lives derived from the modeled degradation rates, amounted to 10 d (enclosure II) and 7.5 d (enclosure III) respectively, and correspond well to the values mentioned in literature with 2e14 d for aerobic conditions (EC, 2002). The slightly higher degradation in enclosure III could be related to the higher Glyphosate concentrations in the liquid phase and a resulting better access of microorganisms to Glyphosate. With the obtained parameters data it was attempted to predict the necessary depth of filter substrate to ensure an attenuation of Glyphosate to values below the European threshold for drinking water starting from source water concentrations of 3.5 mg/L (enclosure II) and 11.6 mg/L (enclosure III). The modeled filtration length for a sufficient attenuation in enclosure II and III would be about 2.75 m and 3.75 m, respectively (Fig. 6). Model calculations assuming conditions occurring at existing bank filtration well fields yielded in all cases no contamination risk for the water used in drinking water production. Similar findings have been published by Malaguerra et al. (2010) who used a more complex finiteeelement model to simulate Glyphosate and AMPA concentrations in the pumping wells and found only negligible concentrations as well.

Fig. 7 e Glyphosate distribution in the left zone of the vegetated SSF.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 4 7 e3 0 5 4

While the planted SSF had to be divided into two zones the unplanted SSF can be regarded as homogenous (Fig. 8). The inlet concentrations of the unplanted SSF did not reach the targeted level of 20 mg/L. In average it was lower and characterized by strong fluctuations probably due to degradation processes in the stock solution (17.6 mg L1 in average). The concentration gradient between the level of Glyphosate in the mixing cell corresponds well to the concentrations measured in the supernatant. In contrast to the planted SSF where an increase in 40 cm depth was found only after 10 days, low concentrations of Glyphosate were observed here from the very beginning in the unplanted SSF. This is clearly a result of enhanced attenuation could be interpreted as retardation by the biomass of the root zone. Maximum Glyphosate concentrations decreased to 9 mg/L after 40 cm of the filter passage (49% reduction of average supernatant concentration). The concentration in the outlet did not reach the climax of the breakthrough curve. The maximum concentration detected here was 4.5 mg/L. Comparing the concentrations in 40 cm depth and in the effluent of the unplanted SSF with those of the left zone as representative for the planted SSF there was slightly higher Glyphosate reduction in the planted SSF (63% in 40 cm depth, compared to 49% in the unplanted filter), although the inlet concentrations were slightly higher and the residence time was lower. The higher reduction rate of Glyphosate in the planted SSF could be due to the strong biological activity, which was concluded from the lower oxygen contents. The redox potential in 40 cm depth varied strongly in both SSFs and amounted to an average of 200 eV in the left zone as representative for the planted and þ235 eV in the unplanted SSF. The decisive factor seems to be the availability of organic carbon, due to vegetal growth. The influence of phragmites buffer strips along surface water on Glyphosate retardation has not been not studied by other experts before. Only Moore et al. (2008) investigated the effect of planted and unplanted artificial ditches used for runoff retardation of diazinon and permithrin. Their findings revealed that attenuation in the planted plot was more effective compared to the unplanted one, thus confirming our findings e though for other substances. Studies on Glyphosate attenuation 35

Glyphosate conc. [µg/L]

mixing cell

dosing

30

surface 40 cm depth

25

outlet

20 15

3053

during artificial recharge bank filtration were carried out by Skark et al. (2006) who found no contamination above the detection limit. Whereas Post et al. (2000) and Stuyzand et al. (2004) reported detectable traces of Glyphosate at river bank filtration sites. These contradictory findings are differentiated by our results, showing a high natural variability of subsurface mobility for Glyphosate depending on site characteristics.

6.

Summary and conclusions

Laboratory studies were conducted to characterize the substrate of the enclosures and the slow sand filters with regard to Glyphosate removal processes. Batch adsorption studies yielded a very low adsorption capacity for Glyphosate with a KF of 1.9 in the sandy material. This is presumably due to the low organic mater content compared to studies carried out with soils, especially with those of a higher iron and aluminum oxide content. Anaerobic dissipation studies under laboratory conditions at 10  C resulted in a half-life of 30.5 d with dissipation rate of 0.023 d1 in the solvent phase. However, it could not be proven, that degradation is the main removal process for short subsurface passage as complete recovery was achieved from the solid phase after 30 d. In the further course of the experiment, however, significant degradation was observed. In the enclosure experiments a rapid degradation was observed due to the aerobic conditions and higher temperatures with a halflife of 7.5e10.5 d1, with lower initial concentrations (3.5e12 mg/L) compared to the lab experiments. The enclosure experiments showed that between 78 and 80% of continuously applied Glyphosate (3.5 mg L1 or 11.6 mg/L1 in average) can be attenuated despite of low adsorption capacity of the filter substrate and high filtration velocity. The necessary length of the filter substrate in order to ensure a reduction of the Glyphosate concentrations below the European threshold for drinking water of 0.1 mg/L was modeled with VisualCXTfit and must exceed 2.75 or 3.75 m for an initial Glyphosate concentration of 3.5 mg/L (enclosure II) or 11.6 mg/L (enclosure III), respectively. In the SSF experiments the SSF covered with P. australis showed a 2e5 times higher removal capacity (57%) for Glyphosate than the one without reed cover (14%). Thus, the following conclusions can be drawn for the attenuation of Glyphosate during subsurface passage: At low concentrations adsorption may play an important role, however, degradation needs to be considered as the main process for Glyphosate attenuation. Favourable for Glyphosate removal at bank filtration sites are oxic conditions, planted sediment surfaces and travel times of more than 10 d.

10 5

Acknowledgements 7 1. 0 09 .1

07 .1 1. 04

.0 7 .1 0

7 30

10 .0 25 .

.0 7 .1 0

07 20

.1 0.

7 15

10 .0 10 .

05 .1

0. 0

7

0

Fig. 8 e Glyphosate distribution in the unplanted SSF.

We greatly appreciate the financial support from Veolia Water and Berliner Wasser Betriebe (BWB) and the coordination of the project by the Berlin Centre of Competence for Water (KompetenzZentrum Wasser Berlin, KWB).

3054

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 4 7 e3 0 5 4

references

Bergstro¨m, L., Bo¨rjesson, E., Stenstro¨m, J., 2011. Laboratory and lysimeter studies of glyphosate and aminomethylphosphonic acid in a sand and a clay soil. J. Environm. Qual. 40, 98e108. Borggard, O.K., Gimsing, A.L., 2008. Fate of glyphosate in soil and the possibility of leaching to ground and surface waters: a review. Pest Manag. Sci. 64, 441e456. Bo¨rjesson, E., Torstenson, L., 2000. New methods for determination of glyphosate and (aminomethyl)phosphonic acid in water and soil. J. Chromatogr. A 886, 207e216. DIN 38407-22, 2001. Determination of Glyphosate and Aminomethyl Phosphonic Acid (AMPA) in Water Using High Performance Liquid Chromatography (HPLC), Post Column Derivatization and Fluorescence Detection. Published by Deutsches Institut fu¨r Normung e.V., Beuth Verlag, Berlin (in German). EC, 2002. Review Report for the Active Substance Glyphosate Finalized in the Standing Committee on Plant Health at Its Meeting on 29 June 2001, European Commission. Health And Consumer Protection Directorate General, Brussels, Belgium. EU, 1998. European Community Council Directive concerning the quality of water intended for human consumption. Off. J. Eur. Commun. L330, 32e54. Fehse, K.J. (2004) Consideration of the solid:solution ratio, Dissertation. Published by Institut fu¨r Agrartechnik und Landeskultur der Landwirtschaftlichen Fakulta¨t, MartinLuther-Universita¨t Halle-Wittenberg, Halle/Saale. GEUS, 2004. Grundvandsovervagning 1998e2003. Geological Survey of Denmark and Greenland Ministry of Environment, Copenhagen, Denmark (In Danish with English Summary). Gru¨tzmacher, G., Bartel, H., Wiese, B., 2006. Simulating bank filtration and artificial recharge on a technical scale. In: Recharge Systems for Protecting and Enhancing Groundwater Resources. Proceedings of the 5th International Symposium on Management of Aquifer Recharge ISMAR 5, Berlin, Germany, 11th-16th June 2005, IHP-VI. Series on Groundwater, No. 13. UNESCO. Jacinthe, P., Vidon, P., Tedesco, L., Krause, B., Weigert, A., Litz, N., 2009. Effects of Vegetation and Glyphosate on Denitrification in Constructed Wetlands (Work Package 3). project acronym: AQUISAFE 1. e KWB report Aquisafe 1 D 3.1. Komptenzzentrum Wasser, Berlin. Krause, B., Weigert, A., Heise, S., Litz, N., 2009. Organic trace substances relevant for drinking water e assessing their elimination by bank filtration. KWB report. http://www. kompetenz-wasser.de/fileadmin/user_upload/pdf/forschung/ TRACE_final_report_09092009.pdf, 113 p. Malaguerra, F., Albrechtsen, H.-J., Binning, P.J., 2010. Contamination of drinking water supply wells by pesticides from surface water. In: Carrera, J. (Ed.), XVIII International Conference on Water Resources CMWR 2010. CIMNE, Barcelona 2010. Massmann, G., Taute, T., Bartels, H., Ohm, B., 2004. Characteristics of Sediments used for Batch-, Enclosure- and Column-Studies. Internal Report. FU Berlin, Germany.

Mentler, A., Parades, M., Fuerhacker, M., 2007. Adsorption of glyphosate to Cambisols, Podzol and silica sand. In: Chardon, W.J., Schoumans, O.F. (Eds.), COST Action 869, Mitigation Options for Nutrient Reduction in Surface Water and Ground Waters WG3-Mitigation Options Abstracts of the Workshop Mitigation Options: Framework, Effectiveness, and Interactions 27e29th November 2007, Local Organizers Phil Haygarth and Linda Jewell Proceedings. North Wyke, Devon, UK. Moore, M.T., Denton, D.L., Cooper, C.M., Wrysinski, J., Miller, J.L., Reece, K., Crane, D., Robin, P., 2008. Mitigation assessment of vegetated drainage ditches for collecting irrigation runoff in California. J. Environ. Qual. Vol. 37 (2), 486e493. OECD 106-Organization for Economic Co-operation and Development, 2000. Adsorptionedesorption Using a Batch Equilibrium Method. Published in OECD Guidelines for the Testing of Chemicals, Geneva. OECD 106. Post, B., Brauch, H.-J., Korpien, H., Lange, F.T., Herwig, T., Landrieux, T., Fien, R., 2000. Occurrence of Glyphosate and AMPA in the aquatic environment and their behaviour in water treatment. In: Schmidt, C.K., Lange, F.T. (Eds.), Determination of the Potential Purification Efficiency of Bank Filtration/Underground Passage with Regard to the Elimination of Organic Contaminants Under Site-specific Boundary Conditions, vol. 84. Technologiezentrum Wasser Karlsruhe, Karlsruher Straße, Karlsruhe, Germany, p. 76139. Reference No. 02WT0280. Schmidt, H., Boas, P., 2006. Accompanying experiments on weed control on public footways using the roller wiper ‘Rotofix’. Nachrichtenbl. Deut. Pflanzenschutzd. 58 (2), 46e49 (in German). Struger, J., Thompson, D., Statznik, B., Martin, P., McDaniel, T., Martin, C., 2008. Occurrence of glyphosate in surface water of Southern Ontario. Bull. Env. Cont. Tox. 80 (4), 378e384. Soerensen, S.R., Schultz, A., Jacobson, O.S., Aamand, J., 2006. Sorption, desorption and mineralisation of the herbizides glyphosate and MCPA in samples from two Danish soils and subsurface profiles. Environm. Poll. 141, 184e194. Strange-Hansen, R., Holm, P.E., Jacobsen, O.S., Jacobsen, C.S., 2004. Sorption, mineralization and mobility of N(phosphonomethyl)glycine (Glyphosate) in five different types of gravel. Pestic. Manage. Sci. 60 (6), 570e578. Stuyzand P.J., Juhasz-Holterman M.H.A. and De Lange W.J. (2004) Riverbank filtration in the Netherlands: well fields, clogging and geochemical reactions. Proceedings NATA Advanced Research Workshop: Clogging in Riverbank Filtration, Bratislava, 7e10 September 2004. Vereecken, H., 2005. A review e mobility and leaching of glyphosate. Pest Manag. Sci. 61, 1139e1151. Zippel, M., Hannappel, S., 2008. Evaluation of the groundwater yield of Berlin water works using regional numerical groundwater flow models (in German). Grundwasser, Zeitschrift der Fachsektion Hydrogeologie der Deutschen Geologischen Gesellschaft 4, 195e207. Springer Verlag.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Chromium removal from ion-exchange waste brines with calcium polysulfide Behrang Pakzadeh, Jacimaria R. Batista* Department of Civil and Environmental Engineering, University of Nevada, 4505 S. Maryland Parkway, Las Vegas, NV 890154-4015, USA

article info

abstract

Article history:

Chromium removal from ion-exchange (IX) brines presents a serious challenge to the

Received 2 October 2010

water industry. Although chromium removal with calcium polysulfide (CaS5) from drinking

Received in revised form

waters has been investigated somewhat, its removal from ion-exchange brines has not

1 February 2011

been evaluated to date. In this study, a Central Composite Design as well as experimental

Accepted 6 March 2011

coagulation tests were performed to investigate the influence of pH, CaS5/Cr(VI) molar

Available online 24 March 2011

ratio, alkalinity, and ionic strength in the removal of chromium from IX brines. The optimal pH range for the process was found to be pH 8e10.3 and brine alkalinity did not

Keywords:

affect coagulation. The efficiency of chromium removal improved only slightly when the

Chromium

ionic strength increased from 0.1 M to 1.5 M; no significant difference was observed for an

Ion-exchange

ionic strength change from 1.5 to 2.1 M. For chromium (VI) concentrations typically found

Reduction

in ion-exchange brines, a CaS5/Cr(VI) molar ratio varying from 0.6 to 1.4 was needed to

Calcium polysulfide

obtain a final chromium concentration <5 mg/L. Maximum efficiency for total chromium

Coagulation

removal was obtained when oxidation reduction potentials were between
0.1 and 0 (V). Solids concentrations (0.2e1.5 g/L) were found to increase proportionally with CaS5 dosage. The results of this research are directly applicable to the treatment of residual waste brines containing chromium. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

In the United States, chromium occurs naturally in groundwaters in the range of 0.0005e0.21 mg/L (U.S. EPA, 2002). However, chromium contamination in water is generally the result of improper discharge of wastewater from chromium-related industries (Shi et al., 2009; Zaroual et al., 2009; Mouedhen et al., 2009). Chromium in water occurs at two oxidation states: chromite [Cr(III)] and chromate [Cr(VI)]. Cr(VI) is highly soluble and ¨ lmez, 2009; toxic at any pH, and its speciation is pH-dependent (O Gode and Pehlivan, 2005; Ajouyed et al., 2010). Aqueous Cr(III) is less soluble and less toxic than Cr(VI) and can be precipitated as ¨ lmez, 2009; Zaroual chromium hydroxide, Cr(OH)3, at pH > 5.5 (O et al., 2009; Mouedhen et al., 2009). Cr (VI) concentrations up to

3 mg/L have been detected near industrial areas (NCASI, 1998). In drinking water applications, chromium occurs as a single contaminant or as a co-contaminant associated with arsenic, selenium, or nitrate (U.S. EPA, 2002). Chromium can be removed from waters using ionexchange (IX) (Shi et al., 2009; Gode and Pehlivan, 2005; Atia, 2006). Cr(VI) removal by ion-exchange is an efficient technology recommended by the United States Environmental Protection Agency (U.S. EPA) and has been used successfully by many water utilities to produce drinking water with chromium levels less than the U.S. EPA’s Maximum Contaminant Level (MCL) of 0.1 mg/L (Shi et al., 2009; Gode and Pehlivan, 2005; Atia, 2006; Alguacil et al., 2004). For the IX process to be economically viable, the spent IX resin beds must be

* Corresponding author. Tel.: þ1 702 895 1585; fax: þ1 702 895 3936. E-mail address: [email protected] (J.R. Batista). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.006

3056

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

regenerated, usually by using sodium chlorine brine. Regeneration of the IX resins leads to the generation of chromiumcontaining IX brines, which must be managed. The generated waste brine solutions contain high values of pH, alkalinity, initial Cr(VI) concentration, and ionic strength (Siegel and Clifford, 1988). Furthermore, chromium is found in waste brines of IX plants treating arsenic, selenium, or nitrate with chromium as a co-contaminant. Chromium levels greater than 5 mg/L in waste brine solutions are considered hazardous according to the U.S. CFR (2010). Therefore, chromium-containing brines must be treated prior to disposal. In recent years, the reducing agent calcium polysulfide has been used to remove Cr(VI) from groundwater (Freedman et al., 2005; Graham et al., 2006; Messer et al., 2003; Yu and Tremaine, 2002); it is also used for treatment of wastewater (Yahikozawa et al., 1978), and chromite ore processing residue (Graham et al., 2006; Wazne et al., 2007a, b; Moon et al., 2008). Calcium polysulfide has been reported to provide the highest chromium reduction efficiency in comparison with ferrous ammonium sulfate, sodium metabisulfate, and zero valent iron (Leoper et al., 2002). Previous studies have shown that the parameters that affect the process include initial Cr(VI) concentration, calcium polysulfide dosage, and pH (Graham et al., 2006; Wazne et al., 2007b). Calcium polysulfide reduces Cr(VI) to Cr(III) within 30 min (Yahikozawa et al., 1978) and Cr(III) precipitates as Cr(OH)3 at alkaline pH (Graham et al., 2006). In mixtures of calcium polysulfide, species that are present include polysulfide species (S52
and S42
), hydrosulfide (HS
), thiosulfate (S2O32
), dithionate (S2O62
), and hydrogen sulfide (H2S) (Yahikozawa et al., 1978; Kelsall and Thompson, 1993). The polysulfide species (S52
and S42
) and hydrosulfide (HS
) are the main sulfur species in the polysulfide mixture at concentrations approximately 10 times more than that of thiosulfate (S2O32
) or dithionate (S2O62
) (Kelsall and Thompson, 1993). Cr(VI) can be reduced to Cr(III) by HS
, S2O32
, or H2S; Cr (III) then precipitates as chromium hydroxide (Yahikozawa et al., 1978). Kim et al. (2001) reported a stoichiometry of 1.5 M CaS5 per M Cr (VI) at pH 8.2, assuming that elemental sulfur and chromium hydroxide are the products of Cr(VI) reduction when using hydrogen sulfide. The chromium removal process using CaS5 (Storch et al., 2002; Messer et al., 2003; Graham et al., 2006; Wazne et al., 2007a) has been described as follows:

remove chromium. These plants treat water for chromium, arsenic, selenium, or nitrate. Unlike drinking water, the IX brines have high ionic strength due to use of NaCl as a regenerant, high alkalinity, and high initial Cr(VI) concentration. To date, no study has been published to address chromium removal from high ionic strength IX brines. The increased use of IX as well as membrane technology to remove inorganic contaminants from waters has resulted in the generation of contaminated brines for which treatment technologies are lacking. Therefore, a need exists to investigate this process in connection with the treatment of highly toxic waste brines. The goal of this research was to evaluate the removal of chromium from synthetic ion-exchange brines using calcium polysulfide and to determine the effects of initial Cr(VI) concentration, calcium polysulfide dosage, pH, and ionic strength on the removal process. Ion-exchange brines from chromium treatment plants may also contain smaller amounts of other anionic metal co-contaminants, such as arsenic and selenium; the removal of these co-contaminants is not addressed by this research. A Central Composite Design (CCD) was utilized to investigate the interactive effects of pH, initial Cr(VI), and CaS5 dosage, as well as to plan the experimental setup. In addition, two series of batch experiments were performed to address the effects of ionic strength and alkalinity on the removal process.

2.

Material and methods

2.1.

Typical composition of actual IX waste brine

The typical composition of brines from IX plants that treat chromium was determined by collecting data from operating IX plants in the Southwest region of the United States. The residual IX brines usually have high concentrations of chromium (10e100 mg/L). In addition to high chromium levels, brines have high alkalinities (0.5e10 g/L as CaCO3), sulfate (4.8e48 g/L) and ionic strengths (0.8e2.1 M). Chromium removal experiments were designed based on the aforementioned data for brine composition. Synthetic brine was prepared with deionized water and stock solutions of sodium chloride, sodium sulfate, sodium bicarbonate, and potassium chromate.

2.2. 2CrO42
þ 3CaS5 þ 10Hþ / 2Cr(OH)3(s) þ 15S(s) þ 3Ca2þ þ H2O

(1)

This equation corresponds to 1.5 M CaS5 per M Cr(VI). For reduction of high concentrations of Cr(VI) (>30 mg/L), the CaS5/Cr(VI) molar ratio has been reported to be 1.66 and 1.5 by Graham et al. (2006) and Messer et al. (2003), respectively. Reduction of Cr(VI) by calcium polysulfide produces a stable sludge that passes the U.S. EPA Toxicity Characteristic Leaching Procedure (TCLP) test with a maximum limit of 5 mg/ L (Graham et al., 2006; Wazne et al., 2007a; Moon et al., 2008). Based on previous reports of chromium removal from wastewater, it is likely that calcium polysulfide also can be utilized to treat ion-exchange brines containing high levels of chromium. Chromium removal from brines is a major concern of many water treatment plants that use the IX process to

Central Composite Design (CCD)

Preliminary batch results for Cr(VI) removal with CaS5 indicated that pH, CaS5 dosage, and initial Cr(VI) concentration were the major parameters influencing Cr(VI) removal from IX brines. A Central Composite Design (CCD), a widely used type of response surface methodology, was used in this research. Surface response methodology has been successfully used by various authors to quantify the effects of initial concentration, pH, and coagulant dosage during electro-coagulation of ¨ lmez, 2009). In the CCD, chromium (Zaroual et al., 2009; O multiple-regression was used for fitting the quadratic prediction equation (2) (see below) to the experimental data. The statistical software, MINITAB, version 15, (Minitab Inc., State College, Pennsylvania, USA) was used to perform the required adjustments, calculate the coefficients, and perform analysis of variance (ANOVA) and lack-of-fit.

3057

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

Table 1 e Coded and uncoded levels of each factor for the Central Composite design. Independent variables (uncoded)

Unit

pH Initial Cr(VI) concentration CaS5 dosage

Value mg/L mM

Symbol

Coded level
1

0

þ1

X1 X2

4.7 9

8.6 46

12.5 93

X3

0.58

1.16

2.3.

2.32

In this study, a three-factorial and a three-level CCD with two replicate experiments for non-center points and six replicates for the center point were used, leading to a total number of 34 coagulation batch experiments. The purpose of the design was to understand the interactions between the major parameters and the chromium removal efficiency. The major independent variables (i.e., pH, initial Cr(VI) concentration, and CaS5 dosage) were designated as X1, X2, and X3, respectively. The coded and uncoded values of each variable for the Central Composite Design are presented in Table 1. The coded values were
1, 0, and þ1. Experimental data points that were used for the Central Composite Design are presented in Table 2. The goal of the design was to predict the response surface function (Y ), which is the total chromium removal efficiency in this research, based on the independent variables (X1, X2, and X3) according to equation (2): Y ¼ b0 þ b1X1 þ b2X2 þ b3X3 þ b12X1X2 þ b13X1X3 þ b23X2X3 þ b11X12 þ b22X22 þ b33X32

b11,.,b33 are the quadratic coefficients (Mathews, 2005). The prediction limits were chosen as 0% and 100%. These coefficients were calculated using the statistics software MINITAB.

Statistical analysis

For the CCD model developed and a second-order polynomial equation (equation (2)), the coefficients of determination values (R2) and adjusted R2 were found to be 96.55% and 93.44% respectively (Table 3). This indicates that only 7% of the total variation could not be explained by the quadratic model; this also shows agreement between the experimental and predicted data. The results of the analysis of variance (ANOVA) and lackof-fit between the experimental and predicted responses are depicted in Table 4. The calculated F-values of the model are greater than the critical F-value. The higher F-value indicates that the model is more accurate in predicting the variation with the regression equation. Therefore, the square variables are the most significant and the interactions of the variables are the least significant in the model. There is no evidence of lackof -fit (P ¼ 0.001 < 0.05). All the P values are below 0.05, which indicates that the statistical model is significant. Overall, by comparing the F, P, and the coefficient of determination values, the conclusion is that the model is statistically significant. This quadratic model along with the results of the coagulation batch experiments is utilized to predict the effects of pH, CaS5 dosage, and initial Cr(VI) concentration on chromium removal.

2.4.

Reagents

(2)

where b0 is the model constant coefficient; b1,.,b3 are the linear coefficients; b12,.,b23 are the crossproduct coefficients;

All solutions were prepared using high-grade deionized (DI) water produced by a Thermo Scientific Barnstead Nanopure System (Dubuque, IA). The glassware was soaked in an acidic

Table 2 e Experimental data points and responses used in Central Composite Design and predicted chromium removal efficiencies. Runa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

pH

Initial Cr(VI) (mg/L)

CaS5 dose (mM)

CaS5/Cr(VI) molar ratio

Experimental responseb

Predicted response


1 (4.7) þ1 (12.5)
1 (4.7) þ1 (12.5)
1 (4.7) þ1 (12.5)
1 (4.7) þ1 (12.5)
1 (4.7) þ1 (12.5) 0 (8.6) 0 (8.6) 0 (8.6) 0 (8.6) 0 (8.6) 0 (8.6) 0 (8.6) 0 (8.6) 0 (8.6) 0 (8.6)


1 (9)
1 (9) þ1 (93.2) þ1 (93.2)
1 (9)
1 (9) þ1 (93.2) þ1 (93.2) 0 (46) 0 (46)
1 (9) þ1 (93.2) 0 (46) 0 (46) 0 (46) 0 (46) 0 (46) 0 (46) 0 (46) 0 (46)


1 (0.74)
1 (0.74)
1 (0.74)
1 (0.74) þ1 (2.97) þ1 (2.97) þ1 (2.97) þ1 (2.97) 0 (1.48) 0 (1.48) 0 (1.48) 0 (1.48)
1 (0.74) þ1 (2.97) 0 (1.48) 0 (1.48) 0 (1.48) 0 (1.48) 0 (1.48) 0 (1.48)

4.3 4.3 0.4 0.4 17.2 17.2 1.7 1.7 1.7 1.7 8.6 0.8 0.8 3.4 1.7 1.7 1.7 1.7 1.7 1.7

19.2 68.4 1.0 17.7 11.9 60.0 19.3 68.5 15.8 62.6 94.0 53.2 74.7 98.7 95.0 96.0 96.0 97.0 92.0 99.0

23.1 65.1 0.0 17.3 9.6 67.3 19.9 61.9 23.8 65.8 88.5 69.8 84.5 100.0 92.1 92.1 92.1 92.1 92.1 92.1

a Two replicas were performed for runs 1e14. b Average of two replicas are presented for runs 1e14.

3058

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

Table 3 e Estimated regression coefficients of the quadratic model for the Central Composite design for chromium removal efficiency from brines using CaS5. Source Constant X1 X2 X3 X1X1 X2X2 X3X3 X1X2 X1X3 X2X3

Coefficient

Value

b0 b1 b2 b3 b11 b22 b33 b12 b13 b23

92.1399 21.0007
9.3756 7.7603
47.3858
12.977 0.0953
3.9211 3.9211 10.5997

R2 ¼ 0.9655, adjusted R2 ¼ 0.9344.

solution (2% Micro-90, International Products Corp., Burlington, New Jersey) for at least 24 h and completely rinsed with DI water. Reagent-grade (99.9%) NaCl crystals (EMD Chemicals Inc., USA) were added to deionized water to make a primary stock solution of 120 g/L NaCl. Sodium bicarbonate (NaHCO3) and sodium sulfate (NasSO4) (EMD Chemicals Inc., USA) were used to make a 61 g/L HCO3
and 100 g/L sulfate stock solutions, respectively. A primary stock solution of Cr(VI) [1000 mg/L] was prepared using potassium chromate crystals (K2CrO4, J.T. Baker Reagent Chemicals). A 1.85 M (371.2 g CaS5/ L) calcium polysulfide (CaS5) solution (Best Sulfur Products, Fresno, California) was used for the coagulation experiments. Previous reports (Moon et al., 2008; Wazne et al., 2007a, b) on the use of the same calcium polysulfide product have confirmed that the calcium polysulfide is in the form of CaS5. A chromium standard solution of 1000 mg/L (RICCA Chemicals Company, Texas) was used for equipment calibrations.

2.5. Coagulation batch experiments and analytical methods Coagulation batch experiments were performed using a PB-700 standard JarTester (Phipps & Bird, Richmond, Virginia). The batch experiments were performed by rapidly mixing 250 mL of synthetic brine, contained in open beakers, for 2 min at 12 rad/s (G ¼ 158 s
1) followed by 30 min of slow mixing at 3 rad/s (G ¼ 20 s
1) to induce flocculation. It has been reported that CaS5 decomposes slowly (e.g., 60 min) by aeration (Yahikozawa

Table 4 e Analysis of variance (ANOVA) of response surface quadratic model for Central Composite Design. Source Regression Linear Square Interaction Lack-of-fit a Fcritical ¼ <0.0001. b Pcritical ¼ >0.05.

Fa

Pb

31.07 24.13 64.4 4.69 29.33

<0.001 <0.001 <0.001 0.027 0.001

et al., 1978). The pH and oxidationereduction potential (ORP or Eh) of solutions were measured before and after adding CaS5 using an Orion 920A þ pH meter equipped with an Orion 8102BNUWP and Orion Ionalyzer 96-78-000 probes (Thermo Electron Corporation, USA), respectively. In this research, the Eh values are in volts as compared to the standard hydrogen electrode (SHE). The ORP meter was calibrated using a potassium iodide (KI) ORP standard (Orion 967961; Thermo Electron Corporation, USA), prior to each use and according to manufacturer instructions. After coagulation, 45 mL of the solution was collected and centrifuged for 1 h at 367 rad/s using a Sorvall Legend RT Bechtop centrifuge (Thermo Scientific Co., USA) to separate the liquid from the solids. To assure the quality of the measurements, the total chromium concentration in the separated liquid was measured using a DR 5000 spectrophotometer (Hach, Loveland, Colorado) following the standard procedures (Method No. 3500 Cr). Chromium concentrations were also measured with method 3111 (Standard Methods, 2005) utilizing an atomic adsorption spectrometer (PerkinElmer AAnalyst 100, USA). Three calibration standards were used for the measurements. Cr(VI) concentration was measured by a DR 5000 spectrophotometer following the standard procedures (Method No. 3500 Cr) (Standard Methods, 2005). The chromium in the brine before coagulation was in the form of Cr(VI) (chromate), therefore, the Cr(VI) concentration was equal to the total chromium concentration. However, the soluble chromium after coagulation can be in the Cr(VI) or Cr(III) form; thus, the total chromium concentration is equal to the concentrations of Cr(VI), Cr(III), Cr(II), and Cr(0). In this research, “total chromium removal” stands for the total chromium concentration, which is the concentration of Cr(VI) plus that of Cr(III). After coagulation, the total solids generated in the reduction and precipitation process were measured according to method 2540 B (Standard Methods, 2005) by using a 0.45 mm GF/C glass-fiber filter (Whatman Inc., Piscataway, NJ).

2.5.1. Batch experiments to evaluate the effects of pH, CaS5 dosage, and initial Cr(VI) concentration The results of CCD provided overall insight into the interactions between variables and assisted the planning of the experiments. The batch experiments described herein were planned and executed after statistical analysis of CCD. To evaluate the effect of pH, a series of coagulation batch experiments were performed for brines with an initial Cr(VI) concentration of 47 mg/L, 1.5 mM of CaS5, varying pH values from 2.2 to 12.4, and ionic strengths of 0.1, 0.8, or 1.5 M. The initial pH was adjusted using either hydrochloric acid or sodium hydroxide. For each ionic strength level, nine experiments were performed at nine different pH values; three of these experiments were performed in duplicate. In order to determine the required amount of calcium polysulfide (CaS5), nine batch experiments (three in duplicate) were conducted by adding varying amounts of calcium polysulfide to brines of 1.2 M ionic strength. Six initial Cr(VI) concentrations were used, varying from 9 to 93.2 mg/L. Preliminary testing indicated that sulfate concentrations up to 50 mg/L do not significantly affect chromium removal. Hence, in all batch tests performed, sulfate concentration was kept constant at 500 meq/L ¼ 24 g/L.

3059

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

The effect of ionic strength on the removal process was investigated by conducting a series of batch coagulation experiments for brines having ionic strength levels of 0.1, 0.8, 1.5, or 2.2 M, an initial Cr(VI) concentration of 46.0 mg/L, and varying dosages of CaS5. The effect of alkalinity was evaluated with a series of batch experiments for brines containing four levels of alkalinity (0.0024, 0.5, 2.5, and 5.0 g/L as CaCO3), an ionic strength of 0.8 M, and Cr(VI) concentration of 48.2 mg/L.

3.

Results and discussion

3.1.

Effect of ionic strength

3.2. Center Composite Design(CCD) prediction of the effects of pH and initial Cr(VI) concentration on total chromium removal As predicted by the CCD quadratic model developed, a contour plot of combined effects of initial Cr(VI) concentration and pH on total chromium removal (i.e., combination of the reduction of Cr(VI) to Cr(III) with precipitation of Cr(OH)3) is depicted in Fig. 2. The maximum removal (95%) was obtained at pH values between 8.8 and 10.5 for initial Cr(VI) concentrations of 20e42 mg/L. The chromium removal decreased when pH was increased above 10.5 or decreased below 8.8. Similar plots for various initial Cr(VI) concentrations (i.e., 9, 46, and 93.2 mg/L) and CaS5 dosages were developed, and it was revealed the optimum pH for chromium removal from IX brines is between 8.0 and 10.5.

3.3.

Effects of pH

The removal of chromium by calcium polysulfide addition occurs in two steps (Yahikozawa et al., 1978; Storch et al., 2002). Initially, Cr(VI) is reduced to Cr(III), with subsequent Cr (III) precipitation as chromium hydroxide (Cr(OH)3) (Yahikozawa et al., 1978; Storch et al., 2002). The two different steps of chromium removal by CaS5 are seen in Fig. 3a and b.

a 100 Total chromium removal, %

90 80 70 60

Final pH 8.5 ± 0.15 Initial Cr(VI) = 46 mg/L (0.88 mM)

50 I = 0.1 M

40 30

I = 0.8 M I = 1.5 M

20

I = 2.1 M

10 0 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 CaS5 dosage (mM)

b

8.0

0.46 0.36

6.0

I = 0.1 M I = 0.8 M I = 1.5 M I = 2.1 M

4.0

0.26 0.16

pE

Total chromium removal from the brine with increasing CaS5 addition for a fixed pH of 8.5, initial Cr(VI) concentration of 46 mg/L, and ionic strengths of 0.1, 0.8, 1.5, and 2.1 M is presented in Fig. 1a. The respective oxidation/reduction potentials and pE values appear in Fig. 1b. Chromium removal increased with increasing CaS5 dosage up to 1.5 mM (1.7 CaS5/ Cr(VI)) and remained constant with the addition of up to 3 mM (3.4 CaS5/Cr(VI)). After 3 mM CaS5 addition, chromium removal decreased slightly; this trend was observed for all ionic strengths investigated and is consistent with the reduction in potential depicted in Fig. 1b. The positive brine Eh of 0.44 V reduced sharply and stabilized at approximately
0.20 V following addition of CaS5. The CaS5 dosages used in this study were aimed at reducing Cr(VI) to levels that meet the drinking water MCL (0.1 mg/L) and the TCLP limit of 5 mg/L in support of optional disposal methods for the brine and to achieve maximum chromium removal efficiency. Total chromium removal for brines with 0.1 M, 0.8 M, 1.5 M, and 2.1 M ionic strengths were compared by paired t-test analysis. The small P values of 0.0012e0.0099 (<0.05) indicated that increasing the ionic strength of brines from 0.1 M to 0.8 M, 1.5 M, or 2.1 M or from 0.8 M to 1.5 M or 2.1 M promoted a statistically significant difference with increased ionic strength in the total chromium removal efficiencies. Notwithstanding its statistical significance, the actual total chromium removal increased only slightly (1.4%e4.4%) when the ionic strength was increased from 0.1 M to 0.8 M, 1.5 M or 2.1 M. There was no statistically significant difference (P value ¼ 0.4100 > 0.05) found between the total chromium removal for brines with 1.5e2.1 M ionic strength. A separate statistical analysis for data obtained at varying ionic strengths and with CaS5 concentrations below 3 mM showed no significant difference (P ¼ 0.15 > 0.05). This implies that ionic strength had no effects on Cr(VI) reduction. Pettine et al. (1994) reported that the rate constant of Cr(VI) reduction in NaCl media using hydrogen sulfide was insensitive to the change in ionic strength levels from 0.1 M to 1.42 M. However, for CaS5 above 3 mM, which results in more negative oxidationereduction potential and precipitation of chromium hydroxide, a significant difference (P ¼ 0.04 < 0.05) was observed. This indicated that ionic strength seems to favor the precipitation of chromium hydroxide causing a slight enhancement in chromium removal.

2.0 0.06

Eh (V)

2.5.2. Batch experiments to evaluate the effects of ionic strength and alkalinity

0.0 -0.04 -2.0

-0.14

-4.0

-0.24 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 CaS5 (mM)

Fig. 1 e (a) Chromium removal efficiency as a function of CaS5 dosage and different ionic strength levels at a constant initial Cr(VI) concentration of 46 mg/L (0.88 mM), pH of 8.5 ± 0.15, and alkalinity of 5 g/L of CaCO3. (b) Oxidation/reduction potential (Eh) and pE for experimental points shown in (a).

3060

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

Total chromium removal, %

Total chromium removal, % 60

12

2.9

50

80

9 90

8

60

7 70

40

6

50

5

20

30

10

20

40 50 60 70 Initial Cr(VI) Concentration, mg/L

80

Cr(VI) reduction efficiency, %

100 90 80

I = 0.1 M I = 0.4 M I = 0.8 M I = 1.5 M

70 60 50 40

T o ta l c h r o m iu m r e m o v a l, %

1 2

3

4 5

6

7 8 pH

100 90 80 70 60 50 40 30 20 10 0

9 10 11 12 13 14

I = 0.1 M I = 0.4 M I = 0.8 M I = 1.5 M 0

1

2

3 4

5

6

7 pH

8

91.9

9 10 11 12 13 14

Fig. 3 e (a & b) Cr(VI) reduction efficiency (a) and total chromium removal efficiency (b) as a function of pH and different ionic strength levels (0.1, 0.4, 0.8, and 1.5 M) at a constant initial Cr(VI) concentration of 47 mg/L (0.9 mM) and CaS5 dosage of 1.5 mM.

83.7

96.0

1.7 75.5

1.4 67.3

1.1 87.8

71.4 79.6

0.8 0.4

0.6

90

The efficiency of Cr(VI) reduction to Cr(III), based on the remaining Cr(VI) measurements, as a function of pH, ionic strength, a constant CaS5 dosage of 1.5 mM, and an initial Cr (VI) concentration of 47 mg/L (0.9 mM) obtained for coagulation batch experiments [CaS5/Cr(VI) ratio ¼ 1.7], is depicted in

0

87.8

2.0

0.2

30

Fig. 2 e Contour plot of predicted chromium removal efficiency as a function of initial Cr(VI) concentration and pH at a constant CaS5 dosage of 1.5 mM and ionic strength of 1.1 M.

a

2.3

59.1 63.2

0.8 1.0 1.2 1.4 Initial Cr(VI) concentration (mM)

1.6

1.8

Fig. 4 e Contour plot of predicted chromium removal efficiency as a function of initial Cr(VI) concentration and CaS5 dosage for brines with ionic strength of 1.1 M and alkalinity of 5 g/L of CaCO3 at final pH values of 8.6.

Fig. 3a. Complete Cr(VI) reduction from the brine was achieved in the pH range of 1.6e6.4. Cr(VI) reduction efficiency decreased from 100% to 96% when pH increased from 6.4 to 10.3, respectively (Fig. 3a). Kim et al. (2001) also reported a decrease in the reaction rate constant of Cr(VI) reduction using S2
when the pH was increased from 7.5 to 9.3. In this research, the Cr(VI) reduction efficiency strongly decreased from 96% to 72% when the pH increased from 10.3 to 12.4. When calcium polysulfide was applied to brines with a pH 12.4, the color of the brine became greenish. This is an indication of the presence of a soluble species of Cr(VI) in the brine that was not reduced at high pH. Total chromium removal, as a function of pH, ionic strength, a constant CaS5 dosage of 1.5 mM, and an initial Cr (VI) concentration of 47 mg/L (0.9 mM) is given in Fig. 3b. The highest total chromium removal of approximately 94% was achieved in the pH range of 6.5e10.3. Beszedits (1988) reported that the optimal pH for Cr(III) precipitation was between pH 8.5 and 9.0 following reduction of Cr(VI) to Cr(III) in industrial wastewaters. This is consistent with the finding of this research and with the fact that lower solubility of chromium hydroxide occurs at pH values between 8.6 and 8.9.

Total chromium removal, %

pH

70

95

10

CaS5 dosage (mM)

11

b

91.9

2.6

100 90 80 70 60

Cr(VI) = 9 mg/L Cr(VI) = 25.4 mg/L

50 40

Cr(VI) = 35.2 mg/L Cr(VI) = 46.3 mg/L

30

Cr(VI) = 74.2 mg/L Cr(VI) = 93.2 mg/L

20

pH = 8.4 ± 0.1

10 0 0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

8.0

9.0

10.0

CaS5 /Cr(VI) molar ratio

Fig. 5 e Total chromium removal as a function of CaS5/Cr (VI) molar ratio and different initial Cr(VI) concentrations for brines with ionic strength of 0.8 M, alkalinity of 5 g/L of CaCO3, and a constant pH of 8.4 ± 0.1.

3061

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

Table 5 e CaS5 dosages for brines with pH of 8.4 and ionic strength of 0.8 M needed to lower chromium to 5 mg/L, 0.1 mg/L, or < 0.05 mg/L.

9 mg/L (0.2 mM) 25.4 mg/L (0.5 mM) 35.2 mg/L (0.7 mM) 46.3 mg/L (0.9 mM) 74.2 mg/L (1.4 mM) 93.2 mg/L (1.8 mM)

Remaining Cr ¼ 5 (mg L
1)

Remaining Cr ¼ 0.1 (mg L
1)

Remaining Cr < 0.05 (mg L
1)

CaS5/Cr(VI)

CaS5/Cr(VI)

CaS5/Cr(VI)

0.6 0.8 1.2 1.5 1.9 1.4

3.7 2.2 2.2 2.0 2.0 1.7

4.2 2.3 2.2 2.0 2.0 1.7

Very poor total chromium removal is achieved at pH < 6.5 and at pH > 10.3 (Fig. 3b). Decreasing the pH from 6.5 to 4.5 drastically decreased the total chromium removal efficiency from 94% to 15% whereas total chromium removal efficiency (15%) remained constant at pH values between 1.5 and 4.5. The reason for the decrease may be explained by the presence of soluble forms of Cr(III) [Cr3þ, CrOH2þ, and Cr(OH)2þ] (Lai and McNeill, 2006) and the inability of chromium hydroxide to form at pH values below 5.1. Increasing pH values from 10.3 to 12.4 drastically decreased the total chromium efficiency from 94% to 71% because Cr(VI) reduction efficiency decreased at pH values above 10.3 (Fig. 3a) and chromium hydroxide transforms to soluble Cr(OH)4
above pH 10.6. At very high pH values, Cr(OH)3(s) does not form at all and Cr(OH)4
appears as the predominant species. Because the regulations for the disposal of chromiumcontaining wastes typically are based on total chromium, the effect of pH on total chromium removal (Fig. 3b) should be taken into consideration when treating these wastes. For pH values similar to those typically found in IX brines (i.e., pH ¼ 8e10), over 94% of Cr(VI) can be reduced to Cr(III) and precipitated as Cr(OH)3 in the natural pH range of brine. Therefore, no pH change is needed because the high pH of IX brines does not hinder their removal by CaS5. Indeed, the high typical pH of IX brines facilitates optimal precipitation of Cr (OH)3 and only slightly affects reduction of Cr(VI) to Cr(III); a significant advantage for the application of calcium polysulfide as compared to other treatment methods (Mouedhen et al., 2009) that require two pH adjustments, namely, a lowering of the pH to levels <6.5 for Cr (VI) reduction followed by a pH increase to above 8 to precipitate Cr (III) (Beszedits, 1988). An optimal pH range of 8e10.5 for chromium removal from brines with 1 M ionic strength and 10e93 mg/L of initial Cr(VI) concentration was estimated using response surface modeling. Experimentally, a pH range of 6.5e10.3 was found suitable for chromium removal in brines with 47 mg/L initial Cr(VI) concentration and 0.8e1.5 M ionic strength; hence, the upper pH values closely matched the value found with the model. However, for the lower pH range (i.e., 6.5e8), the model predicted lower chromium removal efficiencies than observed experimentally. Therefore, it is anticipated that the optimal pH range of 8e10.3 for chromium removal with CaS5 is valid for other brines with initial Cr(VI) concentrations between 9 and 94 mg/L. The results of the CCD model appear applicable to the treatment of IX brines contaminated with chromium.

3.4. CCD prediction for the effects of CaS5 dosage and initial Cr(VI) concentration on total chromium removal The predicted total chromium removal using the CCD model as a function of Cr(VI) concentration and CaS5 dosage is depicted in Fig. 4. The total chromium removal for brines with initial Cr(VI) concentrations of 0.2e1.8 mM (10e94 mg/L) increased when the CaS5 dosage varied from 0.8 to 2.9 mM. For brines containing 0.2e0.3 mM chromium (9.3e15.6 mg/L), the removal efficiency was maximum (91.9%) at 0.6 mM of applied CaS5 (2e3 CaS5/Cr(VI)). The total chromium removal decreased to 87.8% when CaS5 dosages were increased from 0.8 mM to 3.1 mM (10e15 CaS5/Cr(VI)). This result implies that although the removal efficiency increases by increasing CaS5 dosage, addition of CaS5 at CaS5/Cr(VI) molar ratios greater than 3 slightly reduces efficiency for brines containing 0.2e0.3 mM chromium.

3.5. Effects of calcium polysulfide dosage and initial chromium concentration Information from Fig. 4 was used to plan a series of coagulation experiments for investigating the effects of CaS5 dosage and initial Cr(VI) concentration. The results of batch experiments conducted to determine the total chromium removal as a function of CaS5 dosages for six different initial Cr(V) concentrations are presented in Fig. 5 and it was found that CaS5/Cr(VI) ratios of 1.7e2.0 were needed for optimal removal of chromium from IX brines. Only a slight benefit in removal

pE -4.2

-3.2

-2.2

-1.2

-0.2

0.8

1.8

2.8

3.8

4.8

5.8

6.8

7.8

100 Total chromium removal, %

Initial Cr(VI)

5

4

y = -6534.5x + 5589.2x +

80

3

2

449.94x - 922.7x - 131.62x + 96.695

60

2

R = 0.9418

40

20

0 -0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

Eh (V)

Fig. 6 e Total chromium removal as a function of oxidation/ reduction potential (Eh or pE).

3062

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

Table 6 e Summary of studies on Cr(VI) reduction by calcium polysulfide. Initial Cr(VI) (mg/L)

8.3 to 3.33

2.7 to 13.4

1.66 4.2e1.7 1.31

Final pH

Reference

100

9.9

Graham et al., 2006

26.8e67

98

9.9

Graham et al., 2006

9e93.2 3425

> 99.95 100

8.4 9.9

This study Graham et al., 2006

was achieved at CaS5/Cr(VI) ratios between 2 and 4. No benefit was observed for further addition of CaS5 at CaS5/Cr(VI) ratios greater than 4. Table 5 lists experimentally established CaS5/Cr(VI) ratios necessary to achieve residual chromium concentrations of 5 mg L
1, 0.1 mg L
1, and <0.05 mg L
1 for brines of pH ¼ 8.4, ionic strength of 0.8 M, and different initial Cr(VI) concentrations. In general, greater CaS5/Cr(VI) molar ratios are required to attain lower residual Cr(VI) concentrations. To achieve 5 mg/L remaining chromium concentration in brines containing 9e93.2 mg L
1 Cr(VI), CaS5/Cr(VI) molar ratios of 0.6e1.4 were required. For an approximate 10-fold increase in initial Cr(VI) concentration, only about a 2.3 times higher CaS5/ Cr(VI) molar ratio is needed to attain desired effluent quality. To achieve a residual Cr(VI) concentration of 0.1 mg/L, the necessary CaS5/Cr(VI) molar ratios are approximately 1.2e6.2 times higher than required to obtain a residual chromium concentration of 5 mg/L. As such, for brines with lower Cr(VI) concentration (i.e., 9 mg/L), approximately six times more CaS5/Cr(VI) is needed to achieve 0.1 mg L
1 remaining chromium than for a TCLP limit of 5 mg/L. For initial Cr(VI) concentrations between 46.3 mg L
1 and 93.2 mg L
1, which are typical of most ion-exchange brines, only about 1.25 times more CaS5 is needed to achieve a Cr(VI) residual of 0.1 mg/L as for a 5 mg L
1 residual. As expected, the lower the desired remaining Cr(VI) concentration, the more the CaS5 consumption. The final desired level of chromium in a brine treated with CaS5 will depend on approved disposal methods available in individual municipalities. In states where wastes with chromium concentrations lower than 0.1 mg/L can be discharged by alternative means (e.g., discharged into sewer lines), water utilities may consider applying higher CaS5/Cr(VI) ratios to obtain liquid wastes with lower chromium concentration without appreciable increase in CaS5 consumption. For a remaining Cr(VI) concentration of <0.05 mg/L and an initial Cr(VI) concentration of 9 mg L
1, the CaS5/Cr(VI) molar ratio needed is 4.2 which is 2.5 times greater than that needed for initial Cr(VI) concentration of 93.2 mg L
1 (Table 5). The reason for this can be explained with the oxidationereduction potential of brines. A CaS5 dosage that lowers the oxidationereduction potential to near zero results in a maximum removal (Fig. 6). Chromium removal efficiency versus the ORP of the treated brines in this research is displayed in Fig. 6. It was found that chromium removal efficiency correlates strongly with the final ORP of the brines. Lowering the ORP from positive values to between
0.1 and 0 maximized the chromium removal efficiency, for brines containing 9e93 mg/L of Cr(VI). Storch

% Cr(VI) removal

et al. (2002) also reported that about 200 mg/L of Cr(VI) was reduced to Cr(III) when the ORP of groundwater decreased from þ0.2 V to near 0 V following additions of CaS5 during geological fixation of groundwater. In previous studies, CaS5/Cr(VI) molar ratios required to remove 2.7e3425 mg/L of Cr(VI) from waters were found to be between 8.3 and 1.3, respectively (Table 6). In this research, the CaS5/Cr(VI) molar ratios of 4.2 to 1.7 were needed for removal of 9e93.2 mg/L of Cr(VI) from IX brines. The CaS5/Cr (VI) molar ratios required to remove low concentrations of Cr (VI) from water (<13 mg/L) were higher than those for high

a

100 90

Total chromium removal, %

Bench scale Synthetic water Bench scale Synthetic water IX Brine Groundwater

CaS5/Cr(VI)

80 70

Initial Cr(VI) = 48 mg/L

60 50

0.01 g/L as CaCO3

40

0.5 g/L as CaCO3

30 20

2.5 g/L as CaCO3

10

5.0 g/L as CaCO3

0 0

1

2

3

4

5

6

CaS5 /Cr(VI) molar ratio

b 12 10 8

pH

Type of study

Initial Cr(VI) = 48 mg/L 0.01 g/L as CaCO3

6

0.5 g/L as CaCO3

4

2.5 g/L as CaCO3 2

5.0 g/L as CaCO3

0 0

1

2

3

4

5

6

CaS5 /Cr(VI) molar ratio Fig. 7 e (a) Total chromium removal as a function of CaS5/ Cr(VI) molar ratio and different initial alkalinities for brines with ionic strength of 0.8 M and initial Cr(VI) of 48 mg/L. (b) The final pH values as a function of CaS5/Cr(VI)and different initial alkalinities for points presented in (a).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

concentrations (>30 mg/L) of Cr(VI). The stoichiometric relationship (CaS5/Cr(VI)) for reduction of higher concentrations of Cr(VI) (>30 mg/L) has been reported to be 1.66 and 1.5 by Graham et al. (2006) and Messer et al. (2003), respectively. This ratio is applicable when the contaminated water has no organic matter (Graham et al., 2006). The CaS5/Cr(VI) ratio required in this research for treatment of Cr(VI) concentrations of 9 (4.2) and 93.2 mg/L (1.7) are 2.8 and 1.1 times higher than the stoichiometric ratio of 1.5 found by Kim et al. (2001) for Cr(VI) reduction by hydrogen sulfide, respectively. The CaS5/Cr(VI) ratio of 8.3 found for removal of 2.7 mg/L Cr(VI) from water (Graham et al., 2006) is 5.5 times higher than the stoichiometric ratio of 1.5 while the CaS5/Cr(VI) ratio of 1.31 needed for removal of 3425 mg/L Cr(VI) from groundwater is 0.87 times less than the stoichiometric ratio of 1.5.

3.6.

3063

units. The pipes and mixing blades in the full-scale plants must be monitored and scale removal may be needed.

4.

Conclusions

In this research, calcium polysulfide was successfully utilized for Cr(VI) removal from IX brines. A Central Composite design and response surface methodology, along with a series of coagulation experiments, were utilized to investigate the effects of the major parameters, including initial Cr(VI) concentration, pH, and CaS5 dosages on the removal efficiency. In addition, two series of coagulation experiments were executed to investigate the effects of ionic strength and alkalinity on the removal process. The following conclusions were drawn from the results of this research:

Effect of alkalinity

Total chromium removal as a function of alkalinity and CaS5/ Cr(VI) molar ratio is observed in Fig. 7a. Cr(VI) removal for brines with 0.01, 0.5, 2.5, and 5.0 g/L (as CaCO3) alkalinities were compared by paired t-test analysis. The high P values of 0.13 and 0.50 (>0.05) indicated that increasing the alkalinity of brines from 0.01 to 5 g/L as CaCO3 promoted no statistical difference in the chromium removal efficiencies. The effect of final pH versus CaS5/Cr(VI) molar ratio on total chromium removal appears in Fig. 7b. For brines with 2.5 and 5 g/L of CaCO3, the final pH slightly increased from 8.2 to 8.7 when CaS5/Cr(VI) molar ratio increased to six. For brines with lower alkalinity (<0.5 g/L of CaCO3), the final pH increased more than the final pH of brines with high alkalinities (2.5 and 5 g/L of CaCO3). CaS5 solution is alkaline and will increase the pH of brine upon addition, thereby, favoring Cr(OH)3 precipitation. Therefore, calcium polysulfide can be successful used even in low alkalinity slightly acidic brines.

3.7. Solids generated in the removal of Cr by calcium polysulfide In this research, the total suspended solids (TSS) that were generated increased linearly with increasing CaS5 independent of the initial chromium concentration. Considering those batch experiments that resulted in the maximum total chromium removal, chromium removal from brines with 9e93.2 mg/L of initial chromium resulted in 0.2e1.5 g/L of sludge produced when CaS5 dosage increased from 0.8 mM (4.7 CaS5/Cr(VI)) to 2.7 mM (1.6 CaS5/Cr(VI)), respectively. According to equation (1) the sludge formed from the removal of chromium by calcium polysulfide contains is a combination of CaCO3, S(s), and Cr(OH)3 (Storch et al., 2002; Messer et al., 2003; Graham et al., 2006; Wazne et al., 2007a, b; Kelsall and Thompson, 1993). Addition of 1 M CaS5 results in generation of 5 M S(s), 1 M CaCO3(s), and 0.7 M Cr(OH)3(s) (equation (1)). There is a cost associated with sludge disposal. Depending on the brine characteristics, an economic evaluation is needed to determine the best remaining chromium concentration. In our batch experiments, it was observed that a significant amount of CaCO3 scale was formed. IX brines contain high alkalinities, which along with Ca2þ produce CaCO3. Formation of scale could cause problems for full-scale brine treatment

1. Cr(VI) reduction to Cr (III) was found to be 100%e93.5% for pH value of 1.6e10.3, respectively. However, the optimum pH range for maximum chromium removal (combined reduction/precipitation) was found to be 8e10.3. IX brines typically have pH values between 8 and 10. Because both Cr (VI) reduction and Cr(III) precipitation with calcium polysulfide can be achieved at the pH range of IX brines, CaS5 treatment technology does not require pH adjustment which constitutes a significant advantage in terms of cost and straightforward operation. 2. The CCD model developed to forecast chromium removal is a useful tool to predict the process performance and can be used to assist the water industry in determining CaS5 dosages needed to achieve desired chromium removals from ion-exchange brines. 3. The removal efficiency of chromium by CaS5 coagulation slightly increased when ionic strength of the brine was increased from 0.1 M to 0.8 M, 1.5 M or 2.1 M. No significant improvement in removal efficiency was observed for ionic strength increase above 1.5 M. Thus, the high ionic strength of brines does not influence the process. 4. Increasing alkalinity from 0.01 to 5 g L
1 as CaCO3 had no effect on the removal efficiency of chromium from IX brines. 5. To achieve the TCLP limit of 5 mg/L chromium in the treated IX brines, the CaS5/Cr(VI) molar ratios needed will increase slightly (i.e., 0.6e1.4) with a significant increase (e.g., 9 mg L
1e93.2 mg L
1) in initial Cr(VI) concentration. Higher CaS5/Cr(VI) molar ratios (i.e., 3.7e1.7) would be required to achieve the MCL of 0.1 mg/L. The higher the initial Cr(VI) concentration, the lower is the CaS5/Cr(VI) molar ratio. 6. The oxidation/reduction potential of brines was strongly correlated with total chromium removal. When the Eh values (ORP) were lowered to between
0.1 and 0 V, the brines were under reducing conditions, and maximum total chromium removal occurred. Addition of CaS5 at CaS5/Cr (VI) ratio of above 4 decreased the ORP to the lowest value of
0.32 V, while causing a slight decrease in total chromium removal. Therefore, addition of excessive dosages of CaS5 is not recommended, because it generates more sludge and lowers the reduction efficiency. 7. The amount of sludge solids generated in the coagulation process was directly proportional to the amount of CaS5 added and scale formation was observed.

3064

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 5 5 e3 0 6 4

8. The CCD model developed in this research provides a reasonable approximation to predict the performance of chromium removal using CaS5. It is a useful tool to support a technology for which very little is known.

Acknowledgements We would like to thank the Graduate and Professional Student Association (GPSA) of the University of Nevada, Las Vegas for the financial support. We also would like to thank Basin Water from Rancho Cucamonga, CA for their support in determining typical compositions of IX brines in actual IX plants that remove chromium. We are grateful for the careful editing of this manuscript performed by Dr. Richard Unz and Julie Longo. Finally, we greatly appreciate Tod Thurber for his invaluable support to this research.

references

Ajouyed, O., Hurel, C., Ammari, M., Allal, L.B., Marmier, N., 2010. Sorption of Cr(VI) onto natural iron and aluminum (oxy) hydroxides: effects of pH, ionic strength and initial concentration. Journal of Hazardous Materials 174 (1e3), 616e622. Alguacil, F.J., Alonso, M., Lozano, L.J., 2004. Chromium (III) recovery from waste acid solution by ion-exchange processing using Amberlite IR-120 resin: batch and continuous ionexchange modeling. Chemosphere 57 (8), 789e793. Atia, A.A., 2006. Synthesis of a quaternary amine anion-exchange resin and study its adsorption behaviour for chromate oxyanions. Journal of Hazardous Materials 137 (2), 1049e1055. Beszedits, S., 1988. Chromium removal from industrial wastewaters. In: Nriagu, O., Nieboer, E. (Eds.), Chromium in the Natural and Human Environments. Wiley, New York, pp. 232e263. Freedman, D.L., Lehmicke, L., Verce, M.F., 2005. Reductive dechlorination of tetrachloroethene following abiotic versus biotic reduction of hexavalent chromium. Bioremediation Journal 9 (2), 87e97. Gode, F., Pehlivan, E., 2005. Removal of Cr(VI) from aqueous solution by two Lewatit-anion-exchange resins. Journal of Hazardous Materials 119 (1e3), 175e182. Graham, M.C., Farmer, J.G., Anderson, P., Paterson, E., Hillier, S., Lumsdon, D.G., 2006. Calcium polysulfide remediation of hexavalent chromium contamination from chromite ore processing residue. Science of the Total Environment 364 (1e3), 32e44. Kelsall, G.H., Thompson, I., 1993. Redox chemistry of H2S oxidation by the British Gas Stretford process, Part II: electrochemical behavior of aqueous hydrosulphide (HS-) solutions. Journal of Applied Electrochemistry 23, 287e295. Kim, C., Zhou, Q., Deng, B., Thornton, E.C., Xu, H., 2001. Chromium(VI) reduction by hydrogen sulfide in aqueous media: stoichiometry and kinetics. Environmental Science and Technology 35 (11), 2219e2225. Lai, H., McNeill, L.S., 2006. Chromium redox chemistry in drinking water systems. Journal of Environmental Engineering 132 (8), 842e851. Leoper, J.M., Brown, R.A., Robinson, D., (2002). Bench scale evaluation of chemical reduction as a treatment technology for hexavalent chromium. In: The Second International Conference on Oxidation and Reduction Technologies for InSitu Treatment of Soil and Groundwater, Toronto, Ontario, Canada, November 17e21, 2002.

Mathews, P., 2005. Design of Experiments with MINITAB. American Society for Quality, Quality Press, Milwaukee, Wisconsin 53203, USA. Messer, A., Storch, P., Palmer, D., 2003. In situ remediation of a chromium-contaminated site using calcium polysulfide. Southwest Hydrology, 7e8. Moon, D.H., Wazne, M., Jagupilla, S.C., Christodoulatos, C., Kim, M.G., Koutsospyros, A., 2008. Particle size and pH effects on remediation of chromite ore processing residue using calcium polysulfide (CaS5). Science of the Total Environment 399 (1e3), 2e10. Mouedhen, G., Feki, M., De Petris-Wery, M., Ayedi, H.F., 2009. Electrochemical removal of Cr(VI) from aqueous media using iron and aluminum as electrode materials: towards a better understanding of the involved phenomena. Journal of Hazardous Materials. doi:10.1016/j.jhazmat.2009.02.117. National Council for Air and Stream Improvement, Inc. (NCASI), 1998. Technical and Economic Feasibility Assessment of Metals Reduction in Pulp and Paper Mill Wastewaters. In: Technical Bulletin, vol. 0756. National Council for Air and Stream Improvement, Inc., research Triangle Park, NC. ¨ lmez, T., 2009. The optimization of Cr(VI) reduction and removal O by electrocoagulation using response surface methodology. Journal of Hazardous Materials 162 (2e3), 1371e1378. Pettine, M., Millero, F.J., Passino, R., 1994. Reduction of chromium (VI) with hydrogen sulfide in NaCl media. Marine Chemistry 46 (4), 335e344. Shi, T., Wang, Z., Liu, Y., Jia, S., Changming, D., 2009. Removal of hexavalent chromium from aqueous solutions by D301, D314 and D354 anion-exchange resins. Journal of Hazardous Materials 161 (2e3), 900e906. Siegel, K.S., Clifford, D.A., 1988. Project Summary: Removal of Chromium from Ion-exchange Regenerant Solution. United States Environmental Protection Agency, Water Engineering Research Laboratory, Cincinnati, Ohio 45268. April, EPA-600S2-88-007, USA. Standard Methods, 2005. In: Greenberg, A.E., Clesceri, L.S., Eaton, A.D. (Eds.), Standard Methods for the Examination of Water and Wastewater USA. U.S. CFR, 2010. U.S. Code of Federal Regulations (CFR), Title 40: Protection of Environment, Chapter 1: Environmental Protection Agency, Part 261: Identification and Listing of Hazardous Waste, 40CFR261.24. U.S. Government Printing Office, USA. U.S. EPA, 2002. Occurrence Summary and Use Support Document for the Six-year Review of National Primary Drinking Water Regulations United States Environmental Protection Agency, Office of Water (4606), EPA-815-D-02e006, March, USA. Wazne, M., Jagupilla, S.C., Moon, D.H., Jagupilla, S.C., Christodoulatos, C., Kim, M.G., 2007a. Assessment of calcium polysulfide for the remediation of hexavalent chromium in chromite ore processing residue (COPR). Journal of Hazardous Materials 143 (3), 620e628. Wazne, M., Moon, D.H., Jagupilla, S.C., Jagupilla, S.C., Christodoulatos, C., Dermatas, D., et al., 2007b. Remediation of chromite ore processing residue using ferrous sulfate and calcium polysulfide. Geosciences Journal 11 (2), 105e110. Yahikozawa, K., Aratani, T., Ito, R., Sudo, T., Yano, T., 1978. Kinetic studies on the lime sulfurated solution (calcium polysulfide) process for removal of heavy metals from wastewater. Bulletin of the Chemical Society of Japan 51 (2), 613e617. Yu, G.H., Tremaine, J.M., (2002). Pilot test using CASCADE to treat Cr(VI) in groundwater of a carbonate aquifer. In: The Second International Conference on Oxidation and Reduction Technologies for In-Situ Treatment of Soil and Groundwater, Toronto, Ontario, Canada, November 17e21, 2002. Zaroual, Z., Chaair, H., Essadki, A.H., El Ass, K., Azzi, M., 2009. Optimizing the removal of trivalent chromium by electrocoagulation using experimental design. Chemical Engineering Journal 148 (2e3), 488e495.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Continuous water treatment by adsorption and electrochemical regeneration F.M. Mohammed a, E.P.L. Roberts a,*, A. Hill a, A.K. Campen b, N.W. Brown b a b

School of Chemical Engineering and Analytical Science, University of Manchester, The Mill, Oxford Road, Manchester M13 9PL, UK Arvia Technology Ltd, Liverpool Science Park Innovation Centre, 131 Mount Pleasant Liverpool L3 5TF, UK

article info

abstract

Article history:

This study describes a process for water treatment by continuous adsorption and elec-

Received 21 June 2010

trochemical regeneration using an air-lift reactor. The process is based on the adsorption of

Received in revised form

dissolved organic pollutants onto an adsorbent material (a graphite intercalation

16 November 2010

compound, Nyex
1000) and subsequent electrochemical regeneration of the adsorbent

Accepted 12 March 2011

leading to oxidation of the adsorbed pollutant. Batch experiments were carried out to

Available online 31 March 2011

determine the adsorption kinetics and equilibrium isotherm for adsorption of a sample contaminant, the organic dye Acid Violet 17. The adsorbent circulation rate, the residence

Keywords:

time distribution (RTD) of the reactor, and treatment by continuous adsorption and elec-

Adsorption

trochemical regeneration were studied to investigate the process performance. The RTD

Electrochemical regeneration

behaviour could be approximated as a continuously stirred tank. It was found that greater

Graphite intercalation compound

than 98% removal could be achieved for continuous treatment by adsorption and elec-

GIC

trochemical regeneration for feed concentrations of up to 300 mg L
1. A steady state model

Acid violet

has been developed for the process performance, assuming full regeneration of the

Residence time distribution

adsorbent in the electrochemical cell. Experimental data and modelled predictions (using parameters for the adsorbent circulation rate, adsorption kinetics and isotherm obtained experimentally) of the dye removal achieved were found to be in good agreement. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Organic pollutants such as dyes and pigments are problematic pollutants that are often discharged into wastewaters from industrial operations such as dye manufacturing, leather tanning, carpet, paper, food technology and the textile industry. Many of these dyes are toxic and can be carcinogenic (McKay et al., 1985). Therefore, it is necessary to remove them from liquid wastes to below concentrations accepted by national and international regulatory agencies before the wastes are discharged to the environment. Removal of dye compounds can be difficult and there are a number of processes used to reduce the concentration of dyes including

adsorption (Walker and Weatherley, 2000; McKay et al., 1985) filtration, (Mohan et al., 2002) chemical coagulation, (Vandevivere et al., 1998) and photo degradation (Chu and Ma, 2000). These processes can be very effective for the removal of organic pollutants such as dyes, but have the disadvantage that they produce secondary wastes. Adsorption processes are an attractive approach for water treatment, particularly if the adsorbent is cheap and does not require a pre-treatment step before its application (Wang et al., 2005). For many applications this process has been proven to be superior to other techniques for a variety of reasons (Sanghi and Bhattacharya, 2002; Meshko et al., 2001; Bulut and Aydin, 2006), including the simplicity of design,

* Corresponding author. Tel.: þ44 (0) 160 306 8849; fax: þ44 (0) 161 306 9321. E-mail address: [email protected] (E.P.L. Roberts). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.023

3066

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

low cost, high removal efficiency, ease of operation and availability. Activated carbon as widely used as the adsorbent material (McKay et al., 1985) as very low concentrations can be achieved and high loadings of pollutant on these adsorbents are possible. Adsorption processes are normally operated using a batch of adsorbent with sufficient capacity to operate for many months before reaching saturation. Once loaded the adsorbent must be disposed of or regenerated. Analysis of the whole life costs of water treatment by adsorption indicates that most of the treatment costs are associated with regeneration (EPA, 1989). In spite of this, most studies of adsorption have focused on the development of adsorbents with high capacity and very few on developing adsorbents that can be easily regenerated. Attempts have previously been made to regenerate activated carbon adsorbents electrochemically (Narbaitz and Cen, 1994; Garcı´a-Oto´n et al., 2005; Zhou and Lei, 2006; Narbaitz and Karimi-Jashni, 2009). However, due to its relatively low conductivity an electrolyte was required, the time and energy required for regeneration was prohibitive, and the degree of regeneration was often significantly less than 100%. Bench scale batch studies carried out by Brown et al. (2004a,b); Brown and Roberts (2007) have shown that graphite intercalation compounds (GICs) can be used to remove dissolved organic compounds from water, and that these adsorbents can be rapidly and fully regenerated by electrochemical oxidation. GICs are well known materials and their properties have been investigated (Enoki et al., 2003). In GICs the intercalated molecules form layers in the Van der Waals gaps of the graphite matrix. The use of such a material for adsorption significantly reduces both the time required to reach equilibrium, and the electrochemical regeneration time (Brown et al., 2004a). However, the specific surface area and adsorption capacity of these materials is around two orders of magnitude less than activated carbon, so on-site regeneration (to provide sufficient fresh adsorbent surface) would be essential for this process to be practical. Based on their previous batch studies (Brown et al., 2004a, 2004b; Brown and Roberts, 2007), Brown et al. (2007) have patented a continuous process with adsorption and electrochemical regeneration occurring in the same device. The device is similar to an annulus air-lift reactor, which includes an internal cylindrical draft tube which is enclosed in a cylindrical column (Sun et al., 2005, 2006). The device developed by Brown et al. (2007) is rectangular in construction, and air is used to lift the adsorbent into the adsorption zone where contacting of the graphitic adsorbent with the polluted water takes place. The adsorbent/water mix then flows into a quiescent zone where the adsorbent settles by gravity into a moving packed bed which gradually passes between the two electrodes of an electrochemical cell, where the current passing across the cell regenerates the adsorbent. Air injection close to the base of the bed leads to entrainment of the regenerated adsorbent back into the adsorption zone. The process can also be considered as a continuous flow reactor, the performance of which depends upon two key factors; the kinetics and the dispersion behaviour. In this paper we report for the first time the performance of the continuous treatment of a model effluent by adsorption with electrochemical regeneration, using a circulating

adsorbent system. In addition, the kinetics, bed flow rate and dispersion behaviour of the process have been investigated.

2.

Materials and methods

2.1.

Materials

2.1.1.

Sorbent

Fig. 1 shows a scanning electron microscope image of Nyex
1000 (a GIC) as supplied by Arvia Technology Ltd. The Nyex
particles have a characteristic flake like shape associated with the graphite precursor. The specification for Nyex
1000 provided by Arvia Technology Ltd indicates a carbon content of w95 wt%, a typical particle diameter of w360e500 mm (consistent with Fig. 1), with particle diameters ranging between 100 and 700 mm in size. The Brunauer Emmett Teller (BET) surface area of Nyex 1000
determined by nitrogen adsorption was found to be 1.0 m2 g
1, which is very small compared to activated carbon which may have a surface area up to 2000 m2 g
1 (Streat et al., 1995).

2.1.2.

Sorbate

Acid Violet 17 (AV17) is a powdered anionic tri-phenyl methane (TPM) dye which dissolves readily in hot water, chemical formula C41H44N3NaO6S2 and formula weight 761.92. TPM dyes are a large class of synthetic dyes that are widely used in the textiles industry and are also used to stain bacteria and tissue cultures (Venkataraman, 1952). Commercial dyes are typically prepared as a mixture of the dye compound and an inorganic salt. The dye used in this study was supplied by KEMTEX Educational Supplies Ltd under the trade name Kenanthrol Violet 2B. The supplier indicated that the AV17 content of the Kenanthrol Violet 2B was 22%, the remainder being an inorganic salt. The dye content was confirmed by total organic carbon (TOC) analysis of a sample of the dye dissolved in deionised water.

Fig. 1 e SEM micrograph of a sample of Nyex
1000, the GIC adsorbent used in this study.

3067

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

2.2.

a

Experimental methods

Analysis of AV17 was by UV/Vis spectroscopy, the dye having a lmax of 542 nm, this being consistent with values reported elsewhere (Chhabra et al., 2009). A range of known concentrations of AV17 solutions were measured in a spectrometer (Shimadzu, UK) at this wavelength and the resulting calibration curve gave an extinction coefficient of 6.3  104 cm
1 mol
1 at neutral pH, similar to that quoted elsewhere (Sigma Aldrich, 2010). The absorbance values were related to dye concentration using the BeereLambert Law, although the extinction coefficient was observed to be a function of the sample pH and calibration curves were prepared for each pH condition studied. Batch adsorption experiments were carried out by mixing 100 mL of dye solution (prepared by dissolving the dye in hot water) at a range of concentrations with a known mass of Nyex 1000
. The mixture was shaken in a 250 mL conical flask using a UNIMAX 1010 shaker (Heidolph, UK) at a speed of 385 rpm. Experiments were carried out at a temperature of 21  C and after agitation the solutions were filtered using a 0.45 mm syringe filter (Phenomenex, UK). The concentration of dye in the filtered sample was measured using UV/Vis spectroscopy as described above. From calibration tests the error in the analysis was estimated to be of 5% for concentrations of 5e210 mg L
1. Fig. 2 shows a schematic diagram of the experimental setup used for the air-lift continuous adsorption and electrochemical regeneration process. The reactor, shown in Fig. 3, was constructed from clear polycarbonate and the internal dimensions of the process unit were 35 cm wide, 2.2 cm deep and 147 cm tall. The volumetric capacity of the unit was approximately 9 L and a total of 2.5 kg of the GIC adsorbent (Nyex
1000) was added to the reactor. The principle of the design is to carry out the processes of adsorption and electrochemical regeneration in separate zones within the same device, as shown in Fig. 3a. Adsorption occurred in the two symmetrical side zones, each with a rectangular cross section of 10 cm by 2.2 cm. Air was injected at the bottom of these zones through 14 nozzles to generate mixing and to circulate the adsorbent. The injected air

Catholyte solution Tank

Outlet Continuous air lift adsorption and regeneration reactor

Brine / AV17 Storage Tank

Water Storage Tank

Air Peristaltic feed pumps

Fig. 2 e Schematic diagram of the experimental setup for continuous adsorption and electrochemical regeneration and RTD experiments.

Outlet for treated water Air disengagement

Air disengagement

Settlement zone Downcomer

Adsorption zones

134 cm

A 10 cm

12 cm

moving bed of adsorbent

Regeneration zone

5 cm

Water input

Water input

I7 I1 Air injection

A

I1 I7 Air injection

b hydrogen Adsorbent bed Catholyte out

Membrane Anode Cathode

Catholyte in

Fig. 3 e (a) Schematic diagram of the continuous air-lift adsorption/electrochemical regeneration reactor. (b) Schematic diagram of the electrochemical regeneration zone showing a cross section through line AeA in (a).

entrained the adsorbent from the bottom of the regeneration zone into the fluidised adsorption zones. The effluent to be treated was injected close to the bottom of the adsorption zones where the air injection generated significant mixing. Since very little of the water being treated flowed through the electrochemical cell, the dominant removal mechanism was by adsorption. However, some indirect oxidation may have occurred due to oxidising species carried into the adsorption zone with the adsorbent, or by electrochemical oxidation of dye in the small fraction of solution which flowed into the electrochemical cell. At the top of the adsorption zone the air bubbles disengaged and the treated water and adsorbent flowed into a quiescent settlement zone. The treated water overflowed at the top of the settlement zone and the high density adsorbent particles settled forming a bed. The bed was located in the anode compartment of an electrochemical cell (Fig. 3b), which formed the regeneration zone in the middle of the device.

3068

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

The adsorbent bed was drawn down under gravity as the injected air entrained and removed adsorbent from the bottom of the regeneration zone. On one side of the regeneration zone a dimensionally stable anode (DSA) current feeder was located, which was in contact with the moving bed of adsorbent forming the anode. The DSA used was a mixed metal oxide coated titanium (Electrode Products Technology Ltd., UK). On the other side of the regeneration zone was a microporous polyethylene membrane (Daramic
350, Grace GmbH, Germany) which separated the adsorbent from a perforated 316 L stainless steel cathode. The anode current feeder and cathode electrodes were both 12 cm wide by 60 cm in height, and the depth of the regeneration zone (between the anode current feeder and the membrane) was 22 mm. A separate tank containing 15 L of a 1% w/w sodium chloride electrolyte at pH 1 (adjusted using hydrochloric acid) was circulated through the cathode compartment, as shown in Figs. 2 and 3b. Addition of acid to the catholyte was required to decrease and stabilise the cell voltage. Hydrogen gas, generated at the cathode, was allowed to escape from the top of the catholyte compartment. The pH of the treated water was found to decrease from around 6.8 to between 3 and 4 due to a combination of diffusion of acid through the membrane and generation of acid at the anode. To start up the reactor it was necessary to build up a bed of adsorbent in the regeneration zone. Before each experiment, a feed tank of clean water was used to fill the reactor. Air was fed to the outer nozzles so that the adsorbent was circulated while no air was fed to the injection points close to the bed (I1 in Fig. 3a). When the bed had been formed in the cell, air was supplied to the I1 injection points close to the bed to enable bed movement. Constant conditions were then maintained until a steady bed movement was observed.

2.3.

Adsorbent circulation

The downward bed velocity was determined for a range of different air injection configurations and flow rates. The aim was to find a set of conditions which gave a suitable residence time for the adsorbent in the regeneration zone, and to determine the bed flow rate under these conditions. The nozzles were numbered from I1, the nozzles closest to the downcomer (i.e. closest to the regeneration zone), to I7, closest to the side wall as shown in Fig. 3a. In all cases the air was injected symmetrically so that the same flow rates were applied to the symmetrical nozzles on each side of the regeneration zone. The bed velocity was measured by inserting a particulate tracer into the bed, and determining the distance the tracer travelled in a fixed time interval. The particulate tracer used for this experiment was required to be both highly visible and to have similar settling characteristics to the adsorbent particles. Plastic beads of 6 mm diameter were found to meet these criteria and were used as the tracer in this study. The bed circulation was started up using a similar procedure to that described above, and circulation was maintained until a steady bed movement was observed before the beads were added to measure the bed velocity. Approximately 40e50 beads were added from the top of the adsorption zone. Some of the beads would fall into the centre of the bed and would not be visible, but it was found that under these conditions

some beads were visible through the Perspex walls. The distance a visible bead travelled during a fixed time interval was recorded in order to determine the velocity of the bed. Experiments were carried out using a range of air injection rates and nozzle configurations.

2.4.

Residence time distribution (RTD)

The RTD study was carried out with an impulse response test using sodium chloride as the tracer. The reactor was started up as described in Section 2.2 and air was fed at a rate of 0.8 L min
1 to four of the outer nozzles (I7 and I2 on each side) and to the two inner nozzles closest to the regeneration zone (I1). The feed pump was switched on with the required inlet flow rate. The brine storage tank valve was turned on to inject a concentrated brine solution (26 wt% NaCl) as a tracer material into the inlet of the reactor for two and half minutes at flow 320 mL/min. The outlet conductivity was recorded every 150 s using a conductivity meter (Sension 5, HACH, UK) at the outflow of the reactor until the conductivity of the solution had returned to close to its initial value. The conductivity of the sodium chloride solution was found to be a linear function of the sodium chloride concentration so that the concentration could be determined directly from the measured solution conductivity. As the experiment was carried out at constant temperature and this calibration was prepared at the same temperature, no temperature compensation was necessary.

2.5. Continuous adsorption and electrochemical regeneration In order to demonstrate continuous removal of AV17 the feed solution of AV17 (containing only the dye dissolved in clean water) with a concentration in the range 50e500 mg L
1 was fed to the reactor with a flow rate in the range 0.24e0.6 L min
1. The reactor was started up using clean water as described in Section 2.2. Once a steady bed movement was established, a constant current of 5 A (corresponding to a current density of approximately 7 mA cm
2) was applied across the moving bed of adsorbent and the catholyte pump was started. This current density was selected based on previous batch studies and was sufficient to ensure full regeneration of the adsorbent as it flowed through the electrochemical cell. Typical cell voltages of 5e6 V were obtained during operation. The feed was then switched from clean water to a tank containing a solution of AV17. Throughout each experiment the dye solution was supplied at a steady flow rate to the reactor. Samples were taken from the outflow at regular intervals and analysed for AV17 until the outlet concentration reached a steady value, corresponding to a typical experimental duration of 2e3 h.

3.

Results and discussion

3.1.

Kinetics

The kinetics of AV17 sorption on Nyex
1000 was studied at room temperature (21  C) using batch experiments with

3069

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

concentrations in the range 60e120 mg L
1 and an adsorbent dose of 20 g L
1 Fig. 4 shows the decrease in the concentration of AV17 following addition of the adsorbent. As the initial AV17 concentration was increased from 60 mg L
1 to 119 mg L
1, the percentage removal at equilibrium decreased from 96.3% to 67.3% whilst the adsorbate loading on the solid phase increased from 2.9 mg g
1 to 4.0 mg g
1 (see Fig. 5). It is also evident from Fig. 4 that most of the removal of dye occurred in the first 30 min after addition of the adsorbent. The adsorption can be considered to be rapid when compared to porous activated carbon adsorbents, which can take several days to reach equilibrium. Equilibrium is achieved after around 60 min, although a very gradual decrease in concentration is observed at long times. In order to fit kinetic models to the adsorption data, the variation of adsorbent loading (qt) with time was determined for the data plotted in Fig. 4. The pseudo first order (Equation (1)) and pseudo second order (Equation (2)) models were then fitted to the data. Each model was fitted to the four sets of data together, using a single value for the rate constant and different values for equilibrium loading for each initial concentration. Thus five parameters (four equilibrium loadings and the rate constant) were adjusted to obtain a maximum value of the correlation coefficient R2 for each model. Neither of the models gave a very good fit to the data, but the error obtained using the pseudo second-order model was less than half that obtained with the pseudo first-order model. Fig. 5 shows the variation of loading with time and the curve obtained from the pseudo secondorder model fitted to the data. The values of fitted rate constants and equilibrium loadings obtained from the kinetic models are shown in Table 1. The values of equilibrium loading obtained in the experiments (qe,exp) are closer to the values obtained with the pseudo second-order model than those from the pseudo first-order model. qt ¼ qe 1
e
k1 t qt ¼




(1)

q2e $k2 $t 1 þ qe $k2 $t

(2)

120

Initial concentration C0

100

119 mg/L 104 mg/L 84 mg/L 60 mg/L

Ct / mg L

−1

80

Fig. 5 e Variation in the adsorbent loading of AV17 during batch adsorption on Nyex
1000 calculated from data plotted in Fig. 4. The lines show the pseudo second-order kinetic model (Equation (2)) fitted to the data.

3.2.

Adsorption isotherms

Based on the kinetic data shown in Figs. 4 and 5, it was assumed that equilibrium was achieved during the batch studies after around 60 min. The equilibrium behaviour of the sorption of AV17 onto Nyex
1000 was thus studied in a batch process by mixing 2 g of sorbent with 100 mL of dye solution at a range of concentrations (20e210 mg L
1) for 60 min. Assuming that equilibrium was achieved, Fig. 6 shows the adsorption isotherm obtained at room temperature. The data was fitted to the Langmuir, Freundlich and RedlichePeterson equations (Langmuir, 1916; Dechow, 1989; Ho and McKay, 1999) by finding the parameter values which gave a maximum value of the correlation coefficient between the data (q and Ce) and the model in each case. It is clear from Fig. 6 and the values of the correlation coefficient R2 shown in Table 2 that the Langmuir and RedlichePeterson models (which are almost indistinguishable) give a much better fit to the data than the Freundlich model. The fitted parameter values for the three models are shown in Table 2. The value of bR fitted for the RedlichePeterson model is very close to one, which is consistent with the close agreement with the fitted Langmuir model plotted in Fig. 6. As expected, the maximum loading achieved is around two orders of magnitude less than

60

Table 1 e Kinetic rate constants for the adsorption of AV17 onto Nyex
1000 dose 20 g/L, where qe1 and qe2 are the fitted equilibrium loadings for the first-order and second-order models respectively.

40

20

0 0

50

100

105

200

Time / min

Fig. 4 e Variation in the concentration of AV17 during batch adsorption on Nyex
1000 at 23  C with an adsorbent dosage of 20 g LL1 for a range of initial concentration.

C0 mg L
1 60 84 104 119

qe,exp mg g
1

k1 min
1

qe1 mg g
1

k2 g mg
1 min
1

qe2 mg g
1

2.9 3.7 3.7 4.0

0.277

2.71 3.36 3.42 3.59

0.128

2.93 3.47 3.62 3.78

3070

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

5

4

3

Experimental Data

2

Redlich-Peterson model Freundlich model 1 Langmuir model

0 0

20

40

60

80

100

120

Fig. 6 e Isotherm for the sorption of AV17 onto Nyex
1000 at room temperature, where C₀ was determined after of mixing 100 mL of solution with 2 g of Nyex for 60 min.

that which can be achieved with activated carbon (e.g. Walker and Weatherley, 2000; Namasivayam et al., 2007).

3.3.

Fig. 7 e The effect of the rate of air injection to nozzle I1 (QI1) on bed velocity for a total air injection rate of 4.8 L minL1. For the data shown, equal flow rates of 0.4, 0.6 and 0.8 L minL1 was applied symmetrically to 12, 8 and 6 of the nozzles respectively. Error bars show the standard deviation from five measurements of bed velocity under each set of conditions.

Adsorbent circulation

The objective of the adsorbent circulation study was to establish bed flow conditions which would achieve complete regeneration. In addition the movement rate of the bed was required for the modelling study (see Section 3.5 below). The data obtained suggested that air injection to two nozzles closest to the bed (I1) had the strongest influence on the bed velocity. A steady and continuous bed movement was obtained for air injection rates to these nozzles I1 from 0.4 to 0.8 L min
1 Fig. 7 shows the measured bed velocity, ur cm s
1, at three different air injection rates to nozzles I1, with equal total air flow rate. For the data shown in Fig. 7, equal flow rates of 0.4, 0.6 and 0.8 L min
1 were applied symmetrically to 12, 8 and 6 of the nozzles respectively. Although further work is needed to investigate the behaviour of the bed flow rate under a wider range of operating conditions, the data obtained is sufficient to determine suitable operating conditions for RTD and water treatment experiments. Previous studies (Brown et al., 2004a,b and Brown, 2005) of batch electrochemical regeneration have indicated that a regeneration time of around 10 min should be sufficient to ensure complete electrochemical regeneration of the adsorbent. With 0.8 L min
1 of air supplied to nozzle I1, the bed velocity was around 0.12 cm s
1, giving a regeneration

time (given by the length of the anode divided by the bed velocity) of 8.3 min. Based on the bed movement studies, air was injected at 0.8 L min
1 to the two inner nozzles, I1, for the RTD and continuous treatment studies. In addition, to ensure good fluidisation and mixing in the adsorption zone 0.8 L min
1 was supplied to each of four outer nozzles (I2 and I7 in each adsorption zone), giving a total air injection rate of 4.8 L min
1.

3.4.

RTD behaviour

The RTD behaviour of the continuous water treatment process was found to be close to that of a CSTR, as might be expected given the intense mixing in the adsorption zones. Fig. 8 shows the measured exit age distribution Eq as a function of the normalized residence time q: q¼

t$Q V

Eq ¼ Et

(3)

V Q

(4)

where Q is the total volumetric flow rate of the feed (L min
1), V is the total volume of the adsorption zones, t is time and the

Table 2 e Langmuir, Freundlich and RedlichePeterson isotherm constants for sorption of Acid Violet 17 onto Nyex
1000, dosage 2 g/100 mL. Isotherm model

kL

bL

mg
1

mg g
1

Langmuir Freundlich RedlichePeterson

4.488

kF mg(1
1/n) L1/n g
1

1/n

0.978

0.302

kR L g
1

aR LbR mg
bR

bR

R2

0.99

0.968 0.855 0.967

0.068 0.312

0.073

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

RTD would essentially correspond to that of the adsorption zone. The inlet and outlet regions would also be expected to have some effect on the RTD, but the volume of these regions is relatively small compared to the volume of the adsorption zone. Thus the adsorption zone can be considered to behave as a CSTR.

1.2 Experimental data

n T =1 n T =2 n T =1.14

1

Eθ

0.8

0.6

3.5. Continuous adsorption and electrochemical regeneration

0.4

0.2

0 0

2

4

6

8

10

θ

Fig. 8 e The measured exit age distribution for the continuous treatment process. The exit age distribution obtained using the tanks in series model for values of nT of 1, 2 (Equation (5)) and 1.14 (Equation (6)) are also shown.

fraction of the fluid that spends a time t inside the reactor is given by: Et dt (Levenspiel, 1999). A range of models can be used to describe residence time distribution, but given that the observed RTD is qualitatively similar to that of a CSTR a dispersed plug flow model is unlikely to be appropriate. The Peclet number obtained from the variance of Eq (Levenspiel, 1999) was 0.43, confirming that the RTD behaviour is indicative of a mixed flow reactor. The ‘tanks in series’ model describes the flow in a well-stirred tank reactor by considering it to be discretised into a series of CSTRs, each of them having the same volume and being independent of those preceding or following it. Integration of a simple dynamic mass balance gives an RTD of (Levenspiel, 1999): T

Eq ¼

3071

nnT qðnT
1Þ e
nT q ðnT
1Þ!

(5)

The performance of the treatment process was determined for a range of inlet concentrations and flow rates. In each case the concentration at the exit was monitored until a steady value was obtained. Fig. 9 shows the percentage of AV17 removed: R¼

Cin
Cout $100% Cin

(7)

as a function of inlet concentration for a range of feed flow rates. The time required to reach steady state varied from 2 to 3 h, depended on the feed flow rate and concentration. In all cases the total air injection rate was 4.8 L min
1 and the rate of air injection to each of the nozzles adjacent to the adsorption zone (I1) was 0.8 L min
1, which gave a residence time for the adsorbent in the regeneration zone in the order of 8.3 min. Continuous removal of AV17 was achieved with removals of 98% or higher for inlet concentrations of up to 250 mg L
1. Even at inlet concentrations in the order of 500 mg L
1 95% removal could be achieved. When the feed flow rate Q was increased at a feed concentration of 500 mg L
1 the removal achieved decreased, probably due to the decreased residence time and the higher loading on the adsorbent. The results confirm that regeneration of the adsorbent was being achieved. In the absence of regeneration (when no current is applied to the electrochemical cell) the outlet concentration was found to approach the inlet concentration after a time of order 1 h, as expected due to the relatively low adsorption capacity of the adsorbent. In this study the effect

where nT is the number of tanks. The residence time distribution for nT ¼ 1 and nT ¼ 2 are shown in Fig. 8. The weakness of the tanks in series model associated with the quantization of the key parameter, nT, is evident as the experimental data appears to lie between nT ¼ 1 and nT ¼ 2. However Martin (2000) has shown that an extended tanks in series model with a non-integer value of nT can be described using the gamma distribution: T

Eq ¼

nnT qðnT
1Þ e
nT q GðnT Þ

(6)

which is equivalent to Equation (5) for integer values of nT. The value of nT to give the least squared error between Equation (5) and the experimental data was determined, and a best fit value of nT ¼ 1.14 was obtained for the case shown in Fig. 8. These results indicate that the reactor behaviour is very close to a that of a single CSTR. The high peak value of Eq suggests that there may be some short circuiting. While the measurements were obtained for the whole reactor, very little flow would be expected to go through the packed bed regeneration zone due to its high pressure drop and low velocity. Thus it can be expected that the measured

Fig. 9 e Performance of the continuous adsorption and electrochemical regeneration process for a range of inlet concentrations of AV17 and solution flow rates. The larger open symbols indicate the experimental data, while the smaller filled symbols show the predicted removal based on Equation (10).

3072

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

of the current density was not explored and this will be the subject of future work. Based on batch regeneration studies, it is expected that for the conditions studied the adsorbent was fully regenerated, and increasing the current density would not be expected to have a significant effect on the removal achieved. Although no attempt was made to determine the presence of any breakdown products in solution, the effluent was analysed for total organic carbon (TOC). Even for experiments where >90% removal of dye was achieved, the TOC removal was only around 20e30%, indicating the presence of organic breakdown products in the effluent. This is in contrast to previous batch studies of similar contaminants that have indicated that most of the breakdown products remain adsorbed until they are mineralised to carbon dioxide and water (Brown, 2005). Further work is clearly need to investigate the breakdown products occurring with the continuous treatment process.

3.6.

Reactor modelling

A model has been developed in order to predict the outlet concentration for a given set of operating conditions. In this study we will focus on the adsorption zone, as this is where the water treatment is carried out. It will be assumed that the electrochemical cell is operated under conditions which achieve 100% regeneration of the adsorbent. This assumption is justified based on previous studies of batch electrochemical regeneration of GIC adsorbents (Brown et al., 2004a, 2004b). Previous studies have developed models of continuous adsorption processes (e.g. Maji et al., 2007 and Najm, 1996), and here we modify these to account for the continuous supply of fresh adsorbent. The RTD study indicates that the adsorption zone can be considered to behave like a continuous stirred tank reactor, so that a material balance for the AV17 can be constructed as shown in Fig. 10. The feed flow rate of wastewater contaminated with the dye enters the tank continuously at a flow rate Q and a constant dye concentration Cin. At steady state, the outlet concentration reaches a value of Cout and a mass balance on the AV17 dye (solute) for the adsorption zone gives:

_ QCout þ qout m _ ¼0 QCin þ qin m

qout (mg g−1) • m (g min−1)

Electrochemical Regeneration Zone qin (mg g−1) • m (g min−1)

(8)

where m is the mass flow rate (g min
1) of adsorbent through the regeneration zone, and qin and qout are the loading of AV17 on the adsorbent (mg g
1) entering and leaving the adsorption zone respectively. Based on the batch adsorption experiments we assume second-order kinetics and a Langmuir isotherm for the adsorption equilibrium, so the rate of adsorption r (mg min
1 L
1) throughout the adsorption zone is given by: 2  bkL Cout
qout r ¼ k2 m 1 þ bCout

(9)

where m is the mass of adsorbent per unit volume of the adsorption zone (g L
1). Combining Equations (8) and (9) and assuming complete regeneration as discussed earlier (so qin ¼ 0) we obtain:  QðCin
Cout Þ ¼ k2 mV

2 bkL Cout Q
ðCin
Cout Þ _ 1 þ bCout m

(10)

Thus for a given flow rate and concentration of feed (Q and Cin), the outlet concentration Cout can be determined by solving Equation (10) if the values of the parameters V, k2, b, kL, _ are known. m, and m The values of all of these parameters have been measured or can be estimated with the exception of the adsorbent _ can be concentration in the adsorption zone, m. The value of m estimated from the bed velocity as follows: _ ¼ ur Ar rb m

(11)

where Ar is the cross sectional area of the regeneration zone (26.4 cm2) and rb is the bulk density of the adsorbent (0.368 g cm
3). Unfortunately it was not possible to accurately sample the solid liquid mixture in the adsorption zone in order to determine m in this study due to the relatively narrow width of the adsorption zone. However as a first approximation, assuming that the adsorbent were to travel with the water with no slip velocity, then m can be estimated from: m¼

_ m Q

(12)

Although the adsorbent might be expected to travel at a lower velocity than the water as it has a high density, it may also be attracted to the surface of the air bubbles rising though the water. Equation (10) was solved using the parameter values shown in Table 3. The value of Cout required to satisfy Equation (10) was found numerically using the Solver tool in Microsoft Excel
. Note that for a given set of conditions two

Q (L min−1) Cout( mg L−1)

Adsorption Zone

Table 3 e Values of the parameters used in Equation (10). Parameter

V (L) m (g L−1)

Q (L min−1) Cin (mg L−1)

Fig. 10 e Schematic diagram of the processes occurring in the continuous adsorption and electrochemical regeneration reactor.

Value

Source

V

6L

k2 b kL _ m

0.128 g mg
1 min
1 0.068 L mg
1 4.488 mg g
1 65.5 g min
1 _ m/Q

Volume of adsorption zone Table 1 Table 2 Table 2 Equation (11) Equation (12)

m

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

real solutions could be obtained, but the lower value of Cout was unfeasible as in this case the term inside the square brackets was negative (i.e. qe < qout). The removals obtained by finding the expected Cout from Equation (10) using the parameters shown in Table 3 are compared with the experimental data in Fig. 9. The results indicate that the model gives a good prediction of the outlet concentration for the range of conditions studied. The model described by Equation (10) may be suitable for design purposes or to investigate the effect of the key parameters on performance.

4.

Conclusion

This study has demonstrated for the first time that continuous treatment of water by adsorption and electrochemical regeneration can be effective for the removal and oxidation of dissolved organic contaminants. The treatment process was able to remove more than 98% of the AV17 from a feed solution containing up to 300 mg L
1 of AV17. Although it was not fully demonstrated in this study (since the water was not analysed for possible breakdown products), the process not only removes the contaminant, but also destroys it by anodic oxidation. The process requires minimal addition of chemicals (salt and acid for the catholyte) and does not produce any sludge or liquid waste streams. The results suggest that there is significant potential for the process in a wide range of water and wastewater treatment applications. Further work will be needed to demonstrate performance on real effluents which may contain a mixture of contaminants. In addition the duration of the experiments described in this study was a few hours, and the stability of operation over many thousands of hours would need to be demonstrated for practical application. However, the same batch of adsorbent was used throughout the study with no detectable change in performance or particle size distribution. The adsorption zone was found to behave as a continuous stirred tank. Using this observation, and measurements of the rate of circulation of the adsorbent, a simple model of the process has been developed which shows good agreement with the experimental results. The model is based on adsorption kinetic and isotherm data which would be relatively easy to obtain for real wastewaters using bench scale experiments such as those described in this study. Where a mixture of contaminants are present a combined measure such as COD could be used, although the simple model developed here would not be able to address selective removal from a mixture of contaminants. The system also operates at relatively low powers typically 25e30 W during the experiments described here, corresponding to 0.7 to 2.1 kWh per m3 for the range of flow rates studied. Further work is planned to use this model to investigate the effect of design and operating parameters on the process performance. In addition, the key parameters, including the adsorbent circulation rate and the concentration of adsorbent in the adsorption zone, will be measured for a range of operating conditions. Further work is also needed to investigate alternative geometries and flow arrangements in order to improve the process performance and to enable treatment of higher feed

3073

flow rates and/or concentrations. Options for scale-up include multiple pass systems (series and/or parallel), geometries using a larger adsorption zone and multiple electrochemical regeneration cells.

Acknowledgements The authors would like to acknowledge the financial and material support received from the UK Engineering and Physical Sciences Research Council, ARVIA Technology Ltd. and from the Iraqi Ministry of Higher Education and Scientific Research.

references

Brown, N.W., Roberts, E.P.L., Garforth, A.A., Dryfe, R.A.W., 2004a. Electrochemical regeneration of a carbon-based adsorbent loaded with crystal violet dye. Electrochimica Acta 49 (20), 3269e3281. Brown, N.W., Roberts, E.P.L., Chasiotis, A., Cherdron, T., Sanghrajka, N., 2004b. Atrazine removal using adsorption and electrochemical regeneration. Water Research 38 (13), 3067e3074. Brown, N.W., Roberts, E.P.L., 2007. Electrochemical pre-treatment of effluents containing chlorinated compounds using an adsorbent. Journal of Applied Electrochemistry 37 (11), 1329e1335. Brown N.W., Roberts E.P.L. and Eccleston, K.T., 2007. Apparatus for Electrochemical Regeneration of Adsorbents. PCT Int. Appl. WO 2007125334. Brown, N.W., 2005. Adsorption and electrochemical regeneration for wastewater treatment e development of a process. University of Manchester, PhD thesis. Bulut, Y., Aydin, H., 2006. A kinetics and thermodynamics study of methylene blue adsorption on wheat shells. Desalination 194, 259e267. Chhabra, M., Mishra, S., Sreekrishnan, T.R., 2009. Laccase/ mediator assisted degradation of triarylmethane dyes in a continuous membrane reactor. Journal of Biotechnology 143 (1), 69e78. Chu, W., Ma, C.W., 2000. Quantitative prediction of direct and indirect dye ozonation kinetics. Water Research 34 (2000), 3153e3160. Dechow, F.J., 1989. Separation and Purification Techniques in Biotechnology. Noyes publications, Park Ridge, New Jersey. Enoki, T., Suzuki, M., Endo, M., 2003. Graphite Intercalation Compounds and Applications. Oxford University Press, Oxford. EPA, 1989. Technologies for Upgrading Existing or Designing New Drinking Water Treatment Facilities United States EPA Technology Transfer Report EPA/625/4e89/023. Garcı´a-Oto´n, M., Montilla, F., Lillo-Ro´denas, M.A., Morallo´n, E., Vazquez, J.L., 2005. Electrochemical regeneration of activated carbon saturated with toluene. Journal of Applied Electrochemistry 35 (3), 319e325. Ho, Y.S., McKay, G., 1999. Batch lead(II) removal from aqueous solution by peat: equilibrium and kinetics. Process Safety and Environmental Protection 77 (B3), 165e173. Langmuir, I., 1916. The constitution and fundamental properties of solids and liquids part I. solids. Journal of the American Chemical Society 38 (11), 2221e2295. Levenspiel, O., 1999. Chemical Reaction Engineering, third ed. John Wiley and Sons, New York (Chapter 13).

3074

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 6 5 e3 0 7 4

Maji, S.K., Pala, A., Pal, T., Adak, A., 2007. Modelling and fixed bed column adsorption of As(III) on laterite soil. Separation and Purification Technology 56 (3), 284e290. Martin, A.D., 2000. Interpretation of residence time distribution data. Chemical Engineering Science 55 (23), 5907e5917. McKay, G., Otterburn, M.S., Aga, J.A., 1985. Fullers earth and fired clay as adsorbents for dyestuffs-equilibrium and rate studies. Water Air and Soil Pollution 24 (3), 307e322. Meshko, V., Markovska, L., Mincheva, M., Rodrigues, A.E., 2001. Adsorption of basic dyes on granular activated carbon and natural zeolite. Water Research 35 (14), 3357e3366. Mohan, D., Singh, K.P., Singh, G., Kumar, K., 2002. Removal of dyes from wastewater using flyash, a low-cost adsorbent. Industrial & Engineering Chemistry Research 41 (15), 3688e3695. Najm, I.N., 1996. Mathematical modelling of PAC adsorption processes. American Water Works Association Journal 88 (10), 79e89. Namasivayam, C., Sangeetha, D., Gunasekaran, R., 2007. Removal of anions, heavy metals, organics and dyes from water by adsorption onto a new activated carbon from jatropha husk, an agro-industrial solid waste. Process Safety and Environmental Protection 85 (B2), 181e184. Narbaitz, R.M., Cen, J.Q., 1994. Electrochemical regeneration of granular activated carbon. Water Research 28 (8), 1771e1778. Narbaitz, R.M., Karimi-Jashni, A., 2009. Electrochemical regeneration of granular activated carbons loaded with phenol and natural organic matter. Environmental Technology 30 (1), 27e36.

Sanghi, R., Bhattacharya, B., 2002. Review on decolorisation of aqueous dye solutions by low cost adsorbents. Coloration Technology 118 (5), 256e269. Sigma Aldrich, 2010. Sigma Aldrich On-line Catalogue. http:// www.sigmaaldrich.com/catalog/search/ProductDetail/ ALDRICH/210579 accessed November 2010. Streat, M., Patrick, J.W., Camporro Perez, M.J., 1995. Sorption of phenol and para-chlorophenol from water using conventional and novel activated carbons. Water Research 29 (2), 467e472. Sun, S.L., Bao, X.J., Liu, C.J., Xu, J., Wei, W.S., 2005. Hydrodynamic model for three-phase annulus airlift reactors. Industrial and Engineering Chemistry Research 44 (19), 7550e7558. Sun, S.L., Liu, C.J., Wei, W.S., Bao, X.J., 2006. Hydrodynamics of an annulus airlift reactor. Powder Technology 162 (3), 201e207. Vandevivere, P.C., Bianchi, R., Verstraete, W., 1998. Treatment and reuse of wastewater from the textile wet-processing industry: review of emerging technologies. Journal of Chemical Technology and Biotechnology 72 (4), 289e302. Venkataraman, K., 1952. The Chemistry of Synthetic Dyes, vol. II. Academic Press Inc., New York. Walker, G.M., Weatherley, L.R., 2000. Adsorption of dyes from aqueous solutiondthe effect of adsorbent pore size distribution and dye aggregation. Chemical Engineering Journal 83 (3), 201e206. Wang, S.B., Boyjoo, Y., Choueib, A., Zhu, Z.H., 2005. Removal of dyes from an aqueous solution using fly ash and red mud. Water Research 39 (1), 129e138. Zhou, M.H., Lei, L.C., 2006. Electrochemical regeneration of activated carbon loaded with p-nitrophenol in a fluidized electrochemical reactor. Electrochimica Acta 51 (21), 4489e4496.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Method for quantitative analysis of flocculation performance Ian C. Tse, Karen Swetland, Monroe L. Weber-Shirk*, Leonard W. Lion School of Civil and Environmental Engineering, Cornell University, Ithaca, NY 14853, USA

article info

abstract

Article history:

The sedimentation rate and the post-sedimentation residual turbidity of flocculated

Received 18 November 2010

suspensions are properties central to the design and operation of unit processes following

Accepted 13 March 2011

flocculation in a water treatment plant. A method for comparing flocculation performance

Available online 21 March 2011

based on these two properties is described. The flocculation residual turbidity analyzer (FReTA) records the turbidity of flocculent suspensions undergoing quiescent settling. The

Keywords:

fixed distance across which flocs must travel to clear the measurement volume allows

Tube flocculator

sedimentation velocity distributions of the flocculent suspension to be calculated from the

Sedimentation velocity

raw turbidity data. By fitting the transformed turbidity data with a modified gamma

Residual turbidity

distribution, the mean and variance of sedimentation velocity can be obtained along with the residual turbidity after a period of settling. This new analysis method can be used to quantitatively compare how differences in flocculator operating conditions affect the sedimentation velocity distribution of flocs as well as the post-sedimentation residual turbidity. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

The sedimentation velocity (Vs) of colloidal aggregates (flocs) formed in flocculation with hydrolyzing metal salts and their precipitates is an important parameter to consider in the design and operation of water treatment plants. Vs determines the design of sedimentation clarifiers and plate settlers. The Vs of a floc has been shown to increase with floc size (Tambo and Watanabe, 1979; Adachi and Tanaka, 1997). An ideal flocculator would produce flocs with high Vs and settled water with low residual turbidity after subsequent sedimentation processes. Floc Vs is typically measured in the laboratory using a settling column test (Metcalf and Eddy, 2003). In water treatment plants, coagulant doses are often determined by observing the residual turbidity of jar test samples to identify the dose that produces the most efficient floc sedimentation. Because of floc breakup and the formation of gelatinous precipitates, optical measurement techniques are preferred over particle counters to determine floc

size distributions (Ching et al., 1994). Gregory (1985) developed an optical technique based on measurement of turbidity fluctuations in flowing suspensions to monitor floc suspensions, and demonstrated that the ratio of the root mean square of the fluctuating turbidity signal to the mean value is roughly proportional to the size of the aggregates flowing through the detector and to the square root of their concentration. Two of the most informative parameters for plant designers and operators are floc sedimentation rates and residual turbidity after a period of settling. Thus, an apparatus capable of optically quantifying both Vs and residual turbidity as a method for comparing the performance of different flocculation conditions would be an extremely useful tool for researchers and plant operators alike. The following sections describe an experimental measurement apparatus and process for data analysis that is capable of providing the desired information. An analysis of flocs formed under different conditions is provided as an example application.

* Corresponding author. Tel.: þ1 607 255 8445; fax: þ1 607 255 9004. E-mail address: [email protected] (M.L. Weber-Shirk). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.021

3076

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

2.

Apparatus

2.1.

FReTA

from flocculator

The flocculation residual turbidity analyzer (FReTA) is a measurement apparatus designed at Cornell University that measures both the sedimentation velocity and the residual turbidity of the effluent from a flocculator (see Figs. 1 and 2). FReTA is capable of measuring floc Vs without affecting the structure of flocs that have been formed. FReTA consists of three primary components: an inline turbidimeter, a transparent glass column, and an electrically actuated ball valve. The interaction of these components, as well as the acquisition and the analysis of data were automated using Process Controller software created using LabVIEW by Weber-Shirk (2008). A modified HF Scientific MicroTOL 2 infrared inline nephelometric turbidimeter was used in the apparatus. The plastic housing of the turbidimeter was altered to allow a 2.54 cm (100 ) outer diameter, 2.06 cm (0.81200 ) inner diameter glass tube to fit vertically through the entire turbidimeter housing and through the measurement area. The glass column provided a quiescent chamber for flocs to settle as turbidity was measured over time. The glass column replaced the factory-standard measurement cuvette because the standard measurement cuvette had a restrictive inlet that disrupted flocs entering the chamber. A small diametersettling column was used to accommodate the diameter of the turbidimeter sample cell. Calculations using methods described in McNown and Malaika (1950) were performed to ensure that errors produced by wall effects were not significant. Wall effect errors were estimated to be much less than 1% in all cases.

clay

electrically actuated ball valve

z = 13.64 cm

NTU

infrared turbidimeter

aluminum frame

glass column

effluent solenoid valves

backwash

Fig. 2 e FReTA consists of an electrically actuated ball valve at the top and an IR nephelometric turbidimeter fitted with a glass tube and connected by fittings to an effluent line.

It was important that fluid motion inside the glass column be minimized once measurements had begun. A version of the MicroTOL 2 turbidimeter using a LED infrared light source

alum FReTA synthetic raw water

actuated ball valve

peristaltic pumps

NTU NTU

turbidimeter (feedback loop )

backwash effluent

solenoid valves

infrared turbidimeter

tight coil rapid mix settling column

tube flocculator

solenoid valve pressure sensor (measure head loss)

high-pressure backwash

Fig. 1 e Schematic of the complete experimental assembly.

effluent discharge

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

instead of an incandescent bulb was used to eliminate thermal convection currents that interfered with quiescent settling. The manufacturer-installed heat source and fan used to control condensation in the MicroTOL 2 were also disabled to minimize convection currents. The HF Scientific MicroTOL 2 turbidimeter was set at its minimum response time of 1 s, while the data was collected at a rate of 1 Hz. Prior to use, the turbidimeter was carefully calibrated using an HF Scientific, Inc. Primetime Calibration Standards kit, which uses a solution that is more stable and has a longer shelf life than Formazin. According to the manufacturer, the turbidimeter’s accuracy for readings below 40 NTU was 2% of readings or 0.02 NTU (whichever is greater) and for readings above 40 NTU, the accuracy was 5% of readings. The modifications made to the HF Scientific MicroTOL 2 turbidimeter to create FReTA did not affect the accuracy of the instrument. The 95% confidence interval was shown to be within 2% of the mean reading obtained from bootstrapping a data set containing 1000 turbidity measurements of a stable clay suspension. An electrically actuated ball valve (Gemini Valve model 630) attached to the top of the glass tube was used to seal the connection between the flocculator and the settling column. It prevented flocs in the flocculator above the valve from entering the settling column once measurements began. The valve defined the top of the settling column. A distance of 13.64 cm separated the bottom of the ball valve and the center of the 5 mm zone illuminated by the LED of the turbidimeter. This distance was used for the calculation of sedimentation velocities as discussed below. An elbow connected the bottom of the glass tube to an effluent discharge line. The quiescent settling test has been used for decades and the subsequent experimental results accepted for design of sedimentation tanks. The FReTA apparatus simply automates this test and the ensuing data analysis. The settling column and general experimental design have been validated by Adachi and Tanaka (1997), where a square glass settling tube (20  20  300 mm) was used to observe the settling of flocs through a microscope. The validity of this apparatus was confirmed by observing the sedimentation of standard latex spheres. A monodisperse suspension of flocs will settle at a single velocity and the time series of measured turbidity would start at some initial value and fall sharply to the residual turbidity of the supernatant as the entire suspension of flocs settled below the turbidity detector. In a sample with heterodisperse floc sizes, the distribution of sizes can be discretized into bins using average Vs. The time series of turbidity measurements for a heterodisperse sample has a more gradual decrease that asymptotically approaches a final residual supernatant turbidity. Since the maximum distance (z) a floc must settle in order to clear the measurement volume is the distance between the bottom of the ball valve to the measurement volume of the turbidimeter (13.64 cm in this case), an estimate of the sedimentation velocity of a bin of floc sizes can be made by dividing the distance (z) by the time elapsed since settling began (t). Vs ¼

z t

(1)

The maximum distance (13.64 cm), duration of the settle state (30 min), and turbidimeter response time (1 s) dictate

3077

that particles with settling velocities greater than 136.4 mm/s or less than 0.076 mm/s will not be detected. To minimize or eliminate differential settling, the distance between the top of the column to the illuminated zone was minimized to the shortest distance physically permitted by the valve and fittings. In addition, the distance (13.64 cm) used is much less than the 0.5 m interval between sampling ports used in conventional flocculent settling tests and discrete settling is assumed in analysis of data over this distance (Metcalf and Eddy, 2003). Therefore, the assumption of discrete settling in the analysis of FReTA data is consistent with the treatment of data in flocculent settling tests. In this study FReTA was located immediately downstream of the tube flocculator. Thus, the Vs distribution and residual turbidity values obtained by FReTA represent the initial characteristics of the flocculated particles, and do not represent the Vs distribution or residual turbidity that would be obtained deeper in a sedimentation tank after significant particle contact and aggregation through differential sedimentation had occurred. The choice of positioning FReTA immediately after the flocculator was based on a desire to characterize particles exiting this reactor; however, other points in a treatment process stream could also be used for sampling and analysis such as different positions within a sedimentation tank or subsequent to sedimentation.

2.2.

Tube flocculator

The complete experimental assembly consisted of three main parts: a synthetic raw water (SRW) metering system, followed by rapid mix and a tube flocculator (Fig. 3), and then FReTA. As a mixture of suspended clay and alum flow through the tube flocculator, velocity gradients in the tube cause particles to collide and form flocs. The tube flocculator consisted of a length of 9.5 mm (3/800 ) inner diameter transparent plastic tubing wrapped in a figure eight shape around two long parallel cylindrical prisms for structural support. A pressure sensor was attached at each end of the tube flocculator to monitor the pressure drop (head loss) across the flocculator. The length of the flocculator could be changed to accommodate different hydraulic residence times (q). Tube geometry was used for the flocculator because the velocity gradient (G) in laminar tube flow is well defined (Equation (2)) (Gregory, 1981). Gs ¼

8Q 3pr3

(2)

where: Q is the volumetric flow rate and r is the inner radius of the tube. The number of particle collisions per unit time in a laminar flow flocculator is proportional to G and the time available for collision is q, therefore the product Gq indicates the degree of flocculation that can be achieved (Cleasby, 1984). Initial calculations showed that the length of tubing needed to achieve adequate flocculation based on the suggested Gq value of 20,000 necessary for large floc formation (Camp and Stein, 1943) was roughly 28 m and was too long to maintain as an entirely straight segment, so the tube flocculator was initially arranged into a helical coil. The length of tube flocculator can be increased to two or three times this length, producing Gq values of 40,000 or 60,000.

3078

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

3/8” ID tubing

11 cm support cylinders

15 cm

influent effluent

head loss sensor

5 cm OD rapid mix

synthetic raw water & alum

Fig. 3 e Tube flocculator consists of a segment of 3/8” inner diameter clear plastic tubing wrapped in a Fig. 8 configuration.

The velocity gradient in straight laminar tube flow is axisymmetric about the centerline of the tube and increases linearly from the centerline where the velocity gradient is zero to the maximum value at the wall. While laminar flow can still be achieved in helically coiled tubes, the flow is no longer axisymmetric. The inertia of the fluid in the curved tube causes the highest velocity fluid at the center of the tube to move toward the wall farthest from the axis of curvature (Berger et al., 1983). The resulting flow pattern consists of two rotating cells with the line of symmetry being the radius of curvature of the coil. Because of the parabolic velocity profile found in straight laminar tube flow, both G and q are functions of radial position, creating a range of Gq values experienced by particles within the tube. Particles entering a coiled tube flocculator, however, do not maintain a constant radial position in the tube, consequently the distribution of Gq values experienced by particles is narrower than in a straight tube (Gregory, 1981). It was observed that particles preferentially aggregated inside the two vortical cells throughout the length of the coiled tube flocculator. Furthermore, one of these two cells consistently trapped larger sized flocs over the length of the flocculator than the other cell. Since particles were trapped spinning inside the vortical cells, the helical tube flocculator essentially acted like two separate flocculators. Reconfiguring the helical coils into a figure eight disrupted the two circulating cells and allowed particles to move around in the cross sectional plane. While the flow may still be laminar in a curved tube e in that streamlines are continuous and nonintersecting, the velocity gradients are not axisymmetric throughout the cross section and have a non-linear relationship with axial velocity. The axial velocity remains proportional to the flow rate through the tube, but the centrifugal force introduces velocity components perpendicular to the centerline axis. As described below, a correlation factor comparing the friction coefficients of a straight tube ( fs) to that of a curved tube ( fc) (Mishra and Gupta, 1979) was used in the calculation of the average velocity gradient in the curved tube, G.

Based on dimensional analysis, the velocity gradient G can be expressed as a function of the average energy dissipation rate (e) and kinematic viscosity of the fluid (n): G¼

rffiffiffi e n

(3)

Using conservation of energy, e can be expressed as kinetic energy loss over a period of time: e¼

ghL q

(4)

where: g is gravitational acceleration, hL is head loss and q is average hydraulic residence time. The head loss through a straight tube can, in turn, be defined as (Robertson et al., 1998):

h L ¼ fs

L U2 d 2g

(5)

where: L is the length of the flocculator and fs is the friction factor in a straight tube. For laminar flow, the friction factor fs ¼ 64/Red, and Red is the Reynolds number as defined as: Red ¼

Ud n

(6)

where: U is the average axial velocity and d is the tube inner diameter. The formulation for G derived by Gregory (1981)(see Equation (2)) can also be derived from algebraic rearrangement of Equations (3)e(6). A correlation factor (Mishra and Gupta, 1979) can be applied to Equation (5) to replace fs with fc (Equation (7)) and correct for the differences in head loss between straight and curved tubes. fc ¼ 1 þ 0:033logðDeÞ4 fs

(7)

where: De is the nondimensional Dean number and characterizes the effect of curvature on fluid flow:

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

De ¼

rffiffiffiffiffi r Red Rc

(8)

where: r is the inner radius of the tube and Rc is the radius of curvature. The average head loss measured as the pressure drop across the tube flocculator was within 2% of the head loss calculated using Equations (5) and (7) (Fig. 4). The figure eight coil configuration used in this research was different from the flow regime modeled by Mishra and Gupta. The fact that our data agrees with their model suggests that the change in direction of the coil had only a small effect on total head loss. The following G value obtained from combining Equations (3)e(8) was used to design the experimental runs.  1=2 Gc ¼ Gs 1 þ 0:033logðDeÞ4

(9)

The velocity gradient established in a tube is a function of the fluid flow rate and the cross sectional area of the tube. The cross sectional area of the tube can limit the largest size of flocs the flocculator can produce. The inner diameter of the tube flocculator was 9.5 mm (3/800 ). The expected diameter of the largest flocs was on the order of 1 mm, therefore an inner diameter of 9.5 mm was large enough to facilitate the formation of 1 mm flocs. The length of the tube flocculator and flow rate could be varied depending on the goals of a particular experiment. The data given in the paper corresponds to a flow rate of 5 mL/s, Reynolds number of 668, and a flocculator length of 56 m (any exceptions are noted).

2.3.

Raw water and coagulant metering system

The raw water metering system consisted of a concentrated stock suspension of kaolinite clay (R.T. Vanderbilt Co., Inc., Norwalk, CT) mixed with water to produce a feedback-regulated synthetic raw water (SRW) feedstock (see Fig. 5). The

Hagen-Poiseuille Mishra & Gupta 1979 measured from sensor

head loss (cm of water)

20

3079

concentrated stock and the SRW feedstock were each stirred by a variable speed mixer to keep the suspensions completely mixed. A Cole Parmer MasterFlex L/S digital controlled peristaltic pump provided a continuous stream of the SRW in a closed loop to an HF Scientific MicroTOL 2 turbidimeter to monitor its turbidity. If the turbidity reading of the SRW fell below the target turbidity for an experiment, a solenoid pinch valve regulating the flow between the concentrated clay suspension and the feedstock opened to gradually increase the turbidity of the SRW feedstock. A float valve regulated the flow of temperature controlled (25  C) tap water into the SRW tank to maintain a constant water level. Tap water characteristics were approximately: total hardness z 150 mg/L as CaCO3, total alkalinity z 113 mg/L as CaCO3, pH z8.05 and dissolved organic carbon z2.0 mg/L (Bolton Point Municipal Water System et al., 2010). The SRW was pumped into the tube flocculator using a peristaltic pump with multiple pump heads. An airtight 1-L flow accumulator between the pump and the tube flocculator was used to dampen the periodic pulses caused by the peristaltic pump rollers. Aluminum sulfate (alum) was metered into the SRW flow by a peristaltic pump upstream from the start of the tube flocculator. Flow through a 120 cm segment of 4.3 mm (0.1700 ) ID plastic tubing coiled around a cylinder with an outer diameter of 5 cm acted as a mixing unit to ensure that the alum was thoroughly mixed with the influent SRW stream. The Reynolds number (Red) in the mixing unit varied from 1200 to 4640 over the 4e15.75 mL/s range of flow rates used in experiments. The results from a dye study showed that adequate mixing was achieved at the lowest flow rate used in experiments. All information needed to replicate the experimental system is available online, including a materials list and link to Process Controller software. A complete materials list is available (https://confluence.cornell.edu/display/AGUACLARA/ FReTAþApparatusþMaterialsþList). The Process controller software used in conjunction with this apparatus is available (https://ceeserver.cee.cornell.edu/mw24/Software/Runtime8/ Process%2520Controller.htm). A description of the Process Controller configuration used to operate the FReTA apparatus is given in Tse (2009). Process controller communicates with the valves and pumps via a custom fabricated control box (https://confluence. cornell.edu/download/attachments/10948530/StampBox.pdf? version¼1). Similar results could be achieved using any number of interfaces between the computer and the hardware.

10

3.

0

4

6

8

10

Q (mL/s) Fig. 4 e Comparing head loss across a 18.64 m tube flocculator measured by a pressure sensor and values computed using the Mishra and Gupta (1979) correlation factor. HagenePoiseuille prediction for straight pipe flow is shown for comparison.

Software and operational controls

The apparatus assembly was controlled and monitored by Process Controller (Weber-Shirk, 2008), a software program written in LabVIEW for automated operation of experiments. Process Controller accepts user and sensor inputs to control the output devices such as valves and pumps. Process Controller is also able to compute logic commands to switch between states and can continuously run and log data from the experimental apparatus autonomously. The Process Controller method used to operate the entire tube flocculator/FReTA apparatus contained six operational states. Each of these states consisted of

3080

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

aluminum frame electric mixers

alum stock flow accumulator

clay stock

solenoid valve

peristaltic pump

raw water feed

peristaltic pump to flocculator

peristaltic pump

NTU

SRW

Fig. 5 e Synthetic Raw Water (SRW) and coagulant metering system.

a different set of inputs, commands and rules controlling the apparatus. The operation of the feedback-regulated loop used to maintain constant feedstock turbidity was present in all six of the operational states, because it was critical for the feedstock turbidity to be at steady state. The feedstock turbidity had a coefficient of variation less than 5%. The operational states and controls utilized to automate the entire experimental apparatus were as follows: The first state was “backwash” in which both FReTA and the tube flocculator were flushed with low-pressure tap water to purge the system in preparation for a new run. The backwash line was connected to the effluent tube originating from the bottom end of the FReTA glass settling column. Water flow was directed backwards through the FReTA apparatus, flocculator, and rapid mix tubing to dislodge clay and air bubbles trapped on the tube walls or in connectors. The backwash was discharged through a waste tube located between the rapid mix unit and the raw water metering system. The duration of the backwash state was set at two times the hydraulic residence time of the backwash stream inside the flocculator. The backwash state exited to the second state after the backwash state duration elapsed. The second state was the “loading” state in which the raw water and the coagulant were metered into the apparatus to be flocculated. The flow rates of the raw water and coagulant streams were set by user inputs. The user could choose to have the SRW flow rate and/or coagulant dose step up or down between each cycle of states to compare the results of varying these parameters. The duration of the loading state was twice the combined residence time of the flocculator and the rapid mix. The third state was the “pump ramp down” state in which the water flow in the apparatus was gradually slowed to a stop by incorporating both the deceleration of the raw water and coagulant pumps and a short period of stopped time. Sudden

stoppage of the pumps was observed to generate oscillatory flow caused by exchange of energy between the kinetic energy of the fluid in the flocculator and the pressure inside the flow accumulator. While rapid flow deceleration caused flow oscillation, excessively slow deceleration caused fluid entering the settling column to have experienced a significant part of the flocculator with a lower velocity gradient than the target value. Flow deceleration was controlled by a constant that corresponded to a desired rate of flow decrease. A deceleration of 1.25 cm/s2 was used in the tube flocculator. A stop time proportional to the length of the flocculator was experimentally determined based on data from a pressure sensor across the flocculator (Equation 10). TimeTotal ¼ 8sec 

LengthFlocculator m þ Q  0:01 2 28m s

(10)

The fourth state was the closing of the ball valve. The ball valve had an electric motor actuator that took 6 s to change between open and close states. The flow was completely stopped before this valve was closed to ensure that no flocs were broken by flow through a constricted orifice. The fifth state was the “settle” state in which the turbidity of the glass settling column was monitored under quiescent conditions. The duration of the settle state was determined by the desired range of sedimentation velocities. Increasing the duration of the settle state captured smaller sedimentation velocities, as slow settling flocs require more time to clear the measurement volume. Plate settlers used in sedimentation tanks are often designed with critical upward velocities of 0.12 mm/s (10 m/day). In order to measure particles with settling velocities of 0.12 mm/s, the settling duration was calculated using Equation (1) to be at least 23 min. Therefore, a “settle” state duration of 30 min was used. The sixth and last state was the reopening of the ball valve in preparation for backwash.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

4.

Data processing

One of the primary motivations for developing FReTA was the need for a more quantitative assessment of flocculation performance. The raw time series turbidity measurements of the settling suspension permitted qualitative comparisons between different runs. However, some data transformation and curve fitting was needed to permit quantitative comparisons. As noted above, floc settling velocity was calculated by dividing the 13.64 cm distance between the bottom of the ball valve and the center of the zone illuminated by the turbidimeter infrared LED by the time elapsed in settling (Equation (1)). Since this transformation is equivalent to taking the reciprocal of the time series, the transformed observations are more concentrated at lower velocities. The data in Figs. 6e10 were obtained from FReTA during the settle state of an experiment with 30 NTU influent turbidity (4.6% coefficient of variation) in a 56 m flocculator at Gq ¼ 40,000. Average background turbidity during backwash was 0.012 NTU. As shown in Fig. 6a, little additional information was obtained by observing turbidity changes in FReTA beyond 1500 s. Thus, recording sedimentation velocities much lower than 0.12 mm/ s was not considered to be worth the extended sample time required. Data smoothing and normalization were the next two transformations performed on the raw turbidity data. Dividing the raw turbidity data by the mean effluent turbidity during the loading state normalized the data sets to range between zero and one. This allowed comparison between data sets with differing initial turbidities. Turbidity fluctuations were observed when large flocs (high sedimentation velocities) moved past the measurement area and refracted more light into the light sensor. Cheng et al. (2010) confirms the correlation between turbidity standard deviation and floc diameter. The large fluctuations were often problematic for data fitting routines and required smoothing (Fig. 6a and b). As seen in Fig. 7, the raw data points were not equally distributed on the log of sedimentation velocity scale. If smoothed by averaging over a certain number of data points on this scale, the resulting values would not accurately represent the points that were smoothed. Therefore, we chose to smooth the raw data using an average with respect to the time scale (see Fig. 6b). The number of points that is averaged to create a single smoothed point can be input by the user, 36 points were chosen for the smoothed data in the figures in this paper because it produced a standard 50 points in each experimental data set. The smoothed data set was then converted to a log of sedimentation velocity scale. Fig. 7 shows the smoothed data with error bars representing one standard deviation in either direction. This smoothing technique allowed for the exclusion of outliers while preserving the shape characteristics of the distribution of sedimentation velocities. Figure illustrates the reproducibility of results obtained using FReTA under different experimental conditions. The normalized turbidity curve in Fig. 7 can be interpreted as a cumulative distribution function (CDF) of turbidity with

3081

a

b

Fig. 6 e a: Raw turbidity data and smoothed, normalized data plotted against time (on a log scale). The process for averaging data at regular time intervals is illustrated in Fig. 6b. Data are for a 30 NTU influent turbidity with an alum dose of 5.06 mg/L in a 56 m flocculator with Gq [ 40,000. b: A subset of raw turbidity data and smoothed, normalized data plotted against time to illustrate a sequence of average values calculated at 36 s intervals. Results are for 30 NTU influent turbidity at an alum dose of 5.06 mg/L in a 56 m flocculator with Gq [ 40,000.

respect to Vs. A CDF describes the probability that a variate is less than or equal to some value. Any point on the curve corresponds to a Vs on the abscissa and a value between 0 and 1 on the dependent axis. For instance, if one chooses the point on the curve corresponding to a Vs of 1 mm/s, one can see that it has a normalized value around 0.5 e meaning nearly 50% of the particles had a Vs less than or equal to 1 mm/s. Likewise, if one chooses the point on the curve corresponding to a Vs of 0.1 mm/s, it would mean that 15% of the particles had a Vs less than or equal to 0.1 mm/s.

3082

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

0.15

50

1 0.1

Turbidity (NTU)

Turbidity (NTU)

40

30

20

0.5

0.05

10 0

0 0.1

1

10

0 0.1

Sedimentation velocity (mm/s)

1

10

Sedimentation velocity (mm/s)

Fig. 7 e Smoothed data with error bars representing one standard deviation above and below the average. Standard deviations are calculated from the 36 data points that are averaged to create each smoothed point.

Gamma Fitted CDF (L axis) Smoothed and Normalized Data (L axis) Gamma Fitted PDF (R axis)

Fig. 9 e Smoothed/normalized data, fitted CDF and fitted PDF. While it is convenient to interpret the plot of normalized turbidity vs. Vs as a CDF, there is one aspect of the curve that deviates from the definition of a CDF: the lower bound of the normalized turbidity curve does not approach zero. It is expected that some colloids will never completely settle out even if an infinite amount of time had elapsed. In fact, the final residual turbidity is an important parameter that characterizes how effective flocculation was at sweeping up the raw water colloids. Therefore, in order to interpret the curves as CDFs, an offset equal to the residual turbidity was applied to the fitted distribution curve to allow the lower bound to approach a non-zero value.

This method of analysis becomes more robust if a known type of distribution is fit to the experimental data. Since the turbidity-Vs plot spans multiple orders of magnitude, the curve fit was performed on the base 10 logarithms of Vs in order to make it easier for a regression routine to converge. The gamma distribution was chosen because it provides a flexible shape that can fit many types of distributions with a minimal number of adjustable parameters. The gamma distribution probability density function (PDF) is defined as: f ðx; a; bÞ ¼ xa1

80

ex=b ab GðaÞ

(11)

40

30 40

20

Turbidity (NTU)

Turbidity (NTU)

60

1.5 mg/L 1.5 mg/L

20

0 mg/L

3.38 mg/L

10

3.38 mg/L

0 0.1

1.5 mg/L 2.25 mg/L

1

10

3.38 mg/L

Sedimentation velocity (mm/s) Fig. 8 e Evidence of reproducibility in FReTA experiments at two different alum doses that result in flocculating and non-flocculating data sets. 30 NTU influent water, 82 m flocculator, Gq [ 60,000.

5.06 mg/L

0 0.1

1

10

Sedimentation velocity (mm/s) Fig. 10 e Smoothed data sets with varying alum doses.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

3083

where: a is the shape parameter, b is the scaling parameter (both of which must be real and positive), x is the base 10 logarithms of the sedimentation velocity, and the gamma function is defined as ZN GðaÞ ¼

ta1 et dt

(12)

0

Therefore, the CDF of the gamma distribution is defined as: Zx Fðx; a; bÞ ¼ 0

xa1

ex=b dx ab GðaÞ

(13)

Equation (12) was further modified by an offset parameter (g) to account for the non-zero lower bound corresponding to non-zero residual turbidity: F0 ðx; a; b; gÞ ¼ ð1  gÞ

Zx 0

xa1

ex=b dx þ g ab GðaÞ

(14)

where log10 of the sedimentation velocity was used as the independent variable x. The derivative of the CDF of the gamma distribution provides a probability distribution of the particle population with respect to settling velocities (see Fig. 9). Curve fitting was performed using Mathsoft’s MathCAD 14.0, an engineering calculation program. MathCAD’s curve fitting function (genfit) is capable of fitting a user-defined equation to a set of data points using the optimized LevenbergeMarquardt method for minimization. MathCAD arrives at its best fit curve by optimizing the two variables (a, and b) in Equation (13). The value for residual turbidity, g, was fixed at the value of the last smoothed point, obtained by averaging the turbidity over the last 36 s of the data. The genfit function requires initial guess values for each of the parameters being fitted. A method for estimating a, and b by estimating the mean and variance of Vs was developed in order to provide the genfit function with guess values that would allow convergence. The mean and variance of a gamma distribution are defined as: E½x ¼ ab

(15)

Var½x ¼ ab2

(16)

Therefore, the values of a and b can be estimated by approximating the mean and variance of the normalized turbidity versus sedimentation velocity. The estimate of the mean of log (Vs) was obtained from the sedimentation velocity that corresponded to a normalized turbidity of 0.5. The variance was estimated by picking the two Vs data points with normalized turbidities of 0.25 and 0.75 respectively and computing the DVs spanned by those two points. When different data sets were analyzed, it was observed that the fitted CDF graph represented the data very well in some cases while fitting other data quite poorly, particularly at low sedimentation velocities. The occurrence of small flocs in the presence of large, well-formed, flocs that are produced by efficient flocculation is thought to be caused by floc breakup. Large flocs (i.e., those with high Vs) either have not undergone floc breakup or have regrown sufficiently after breakup, while the smaller colloids that have been sheared from the large flocs

Fig. 11 e Effect of increased Gq on flocculation. Data is smoothed and normalized. Alum dose is 5.06 mg/L in both cases. When Gq [ 40,000, mean Vs is 1.0 mm/s, residual turbidity 4.4 NTU. When Gq [ 60,000, mean Vs is 0.92 mm/ s, residual turbidity 2.3 NTU.

settle at much lower Vs. Thus, the result of significant floc breakup is to produce a bimodal distribution of settling velocities. A poor fit of a gamma distribution to the normalized and smoothed data was taken as an indication that the data set was not unimodal. The mean square error (MSE) was calculated in each data set to differentiate between unimodal and bimodal distributions. The MSE used for this distinction is a user selected parameter. Data sets identified as having bimodal distributions can be fit with a bimodal CDF fitting function. The data processing associated with bimodal fitting is complex and beyond the scope (and space constraints) of this paper. FReTA is capable of capturing the settling characteristics of different flocculent suspensions. Fig. 9 shows a very distinct quantitative difference in the settling characteristics of a flocculated 30 NTU raw water associated with different coagulant doses. It is easy to observe that the slight increase in alum dose from 2.25 mg/L to 3.38 mg/L causes rapid flocculation to occur, which is manifest in the significantly lower residual turbidity and higher mean Vs. Fig. 11 shows the effects of varying Gq on flocculation at a low coagulant dose. The suspension that experienced a lower Gq (mean Vs is 1.0 mm/s, residual turbidity 4.4 NTU) has a similar mean sedimentation velocity but a residual turbidity almost twice that of the high Gq case (mean Vs is 0.92 mm/s, residual turbidity 2.3 NTU). This data was processed using the algorithms introduced above to give a more quantitative comparison between the two cases.

5.

Conclusion

This work was motivated by a need to quantify the sedimentation properties of the effluent from flocculators. FReTA and

3084

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 7 5 e3 0 8 4

its accompanying data analysis methods were developed to quantify both Vs and residual turbidity as a method for comparing the performance of different flocculation conditions. The ability to identify and characterize sedimentation velocity distributions created by floc breakup is a demonstration of FReTA’s capabilities. Although FReTA was used here to analyze and compare laminar flow tube flocculator parameters, it could readily be used to compare the performance of full-scale turbulent flow flocculators. FReTA is anticipated to be a very useful tool for engineers and plant operators alike. Experiments using FReTA are currently being carried out to evaluate the contribution of velocity gradients, residence time, and coagulant dose in the formation of rapidly settling suspensions with low residual turbidity.

Acknowledgments This work was made possible through the generous support of the Sanjuan Foundation. Partial support was also obtained from the National Science Foundation under Grant CBET0604566. Initial designs for FReTA were developed by student members of the AguaClara project at Cornell University’s School of Civil and Environmental Engineering. Additional information on AguaClara can be found at: http://aguaclara. cee.cornell.edu

references

Adachi, Y., Tanaka, Y., 1997. Settling velocity of an aluminumkaolinite floc. Water Research 31 (3), 449e454. Berger, S.A., Talbot, L., Yao, L., 1983. Flow in curved pipes. Annual Review of Fluid Mechanics 15, 461e512.

Bolton Point Municipal Water System, et al., 2010. Drinking Water Quality Report 2010. http://energyandsustainability.fs.cornell. edu/file/RPT-2010-WaterQuality.pdf. Camp, T.R., Stein, P.C., 1943. Velocity gradients and internal work in fluid motion. Journal of the Boston Society of Civil Engineers 30, 219. Cheng, W., Chang, J., Chen, P., Ruey Yu, R., Huang, Y., Hsieh, Y., 2010. Monitoring floc formation to achieve optimal flocculation in water treatment plants. Environmental Engineering Science 28 (6), 523e530. Ching, H., Tanaka, T.S., Elimelech, M., 1994. Dynamics of coagulation of kaolin particles with ferric chloride. Water Research 28 (3), 559e569. Cleasby, J., 1984. Is velocity gradient a valid turbulent flocculation parameter? Journal of Environmental Engineering 110 (5), 875e897. Gregory, J., 1981. Flocculation in laminar tube flow. Chemical Engineering Science 36 (11), 1789e1796. Gregory, J., 1985. Turbidity fluctuations in flowing suspensions. Journal of Colloid and Interface Science 105 (2), 357e371. McNown, J.S., Malaika, J., 1950. Effects of particle shape on settling velocity at low Reynolds numbers. Transactions e American Geophysical Union 31, 74e81. Metcalf and Eddy, 2003. Wastewater Engineering, Treatment and Reuse. McGraw-Hill, New York. 1819. Mishra, P., Gupta, S.N., 1979. Momentum transfer in curved pipes. 1. Newtonian fluids. Industrial & Engineering Chemistry Process Design and Development 18 (1), 130e137. Robertson, J.A., Cassidy, J.J., Chaudhry, M.H., 1998. Hydraulic Engineering. John Wiley & Sons, Inc. Tambo, N., Watanabe, Y., 1979. Physical characteristics of flocs. I. The floc density function and aluminum floc. Water Research 13 (5), 409e419. Tse, I.C., 2009. Method for Quantitative Analysis of Flocculation Performance. M.S. Thesis. Cornell University. pp. 92. Weber-Shirk, M.L., 2008. An Automated Method for Testing Process Parameters. https://confluence.cornell.edu/display/ AGUACLARA/ProcessþControllerþBackground.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Activated sludge pilot plant: Comparison between experimental and predicted concentration profiles using three different modelling approaches Y. Le Moullec, O. Potier, C. Gentric, J.P. Leclerc* Laboratoire Re´action et Ge´nie des Proce´de´s, UPR 3349 CNRS, Nancy-Universite´, ENSIC, 1 rue Grandville B.P 20451 F-54001 Nancy, France

article info

abstract

Article history:

This paper presents an experimental and numerical study of an activated sludge channel

Received 3 June 2010

pilot plant. Concentration profiles of oxygen, COD, NO3 and NH4 have been measured for

Received in revised form

several operating conditions. These profiles have been compared to the simulated ones

2 December 2010

with three different modelling approaches, namely a systemic approach, CFD and

Accepted 12 March 2011

compartmental modelling. For these three approaches, the kinetics model was the ASM-1

Available online 12 April 2011

model (Henze et al., 2001). The three approaches allowed a reasonable simulation of all the concentration profiles except for ammonium for which the simulations results were far

Keywords:

from the experimental ones. The analysis of the results showed that the role of the kinetics

Wastewater treatment plant

model is of primary importance for the prediction of activated sludge reactors perfor-

Experiments

mance. The fact that existing kinetics parameters in the literature have been determined

Modelling

by parametric optimisation using a systemic model limits the reliability of the prediction of

Pollution removal

local concentrations and of the local design of activated sludge reactors. ª 2011 Elsevier Ltd. All rights reserved.

Activated sludge CFD Compartmental model Systemic model

1.

Introduction

The pollutant removal by micro-organisms is an essential step in the wastewater treatment plants -WWTPs. These biological reactions often take place in large air-wastewater cross flow activated sludge reactors. There are several designs to carry out these reactions (Brannock et al., 2010a; Schmit et al., 2009; Tizghadam et al., 2008); one of the most common reactor type is the “channel reactor”. Channel reactors present a very long length compared to their width and depth, landscape availability being a strong constraint for their design. A large number of studies shows the importance of the reactor hydrodynamics for the WWTP reactor modelling.

Authors still develop new methods and tools (e.g. tracing methods, velocity measurements, etc.) (Ahnert et al., 2010; Schubert et al., 2010) to characterize the flow patterns in the different types of reactors (Diamantis et al., 2010; Gresch et al., 2010). Nowadays, WWTP reactors are commonly designed through simulations carried out with commercial softwares. The models used in these softwares are based on the systemic approach, which represents the hydrodynamics of the water treatment reactor with the continuous stirred tank reactors (CSTR) in series model (Makinia and Wells, 2000). The number of reactors depends on the gas and liquid flowrates and on the reactor geometry (Potier et al., 2005; Le Moullec et al., 2008a).

* Corresponding author. E-mail address: [email protected] (J.P. Leclerc). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.019

3086

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

Gas-liquid mass transfer is frequently estimated by empirical correlations. The most common biological kinetics models are the ones developed by the International Water Association (IWA): the series of Activated Sludge Models (ASM). These simulations allow describing the main trends of these reactors, i.e. the global oxygen and energy consumptions, the produced biomass, and the pollution removal efficiency. As an example, recently Yu et al. (2010) developed a model of anaerobic hydrolysiseaerationesedimentation treatment series in a fullscale textile dyeing wastewater treatment plant. They compared successfully experimental data and simulations with a modified activated sludge model ASM-1. Another interesting work (Zima et al., 2008, 2009) combines the one-dimensional advectionedispersion equation with a simple bio-kinetics model. These types of models are very useful tools for process control and allow a good global simulation of the reactor. However, for reactor design and optimisation, these kinds of models are unsuitable to model the influence of the reactor geometry (length/width ratio, presence of baffles, wastewater inlet device), of the aeration process (sparging device, gas fraction field) and of the resulting local mixing. In order to design new wastewater aerated reactors in a more efficient way, computational fluid dynamics (CFD) is a powerful tool, which allows studying the influences of the operating parameters and the phenomena at a local scale (Le Moullec et al., 2008b; Cockx et al., 2001; Glover et al., 2006; Wang et al., 2009; Laborde-Boutet et al., 2009; Brannock et al., 2010b). Recent works integrate the kinetics and the mass transfer in CFD modelling (Le Moullec et al., 2010b; De Bonis et al., 2010). However, it is really compute-intensive (memory and time consuming) for such a complex system as an industrial wastewater treatment reactor. That is why a compartmental model, derived from CFD hydrodynamics calculations, has been developed and tested (Le Moullec et al., 2010a). Few proposed models in the literature have been validated with experimental data. This paper describes an experimental work carried out in a large aerated sludge pilot reactor with a large number of data from several operating conditions in order to validate the models, not only with outlet concentrations predictions but also with local concentration profiles along the whole reactor.

Firstly, the pilot plant, wastewater composition and experimental procedure are presented. Then, the three models: systemic, CFD and compartmental are briefly described. Finally, experimental results are discussed and compared to simulation results with the final objective of a better understanding of this complex reactor.

2.

Experiments

2.1.

Pilot plant description

The studied reactor is a gas/liquid reactor with a very long length compared to its height and width (channel reactor). In this type of reactor, water mainly flows along the length and gas sparging takes place at the bottom. Since the biological kinetics involved are well represented by Monod equations with apparent reaction orders greater than zero, the pollutant removal efficiency depends on the hydrodynamics (Levin and Gealt, 1993). The total length of this pilot scale reactor is 3.6 m with a rectangular section of width and height respectively equal to 0.18 m and 0.2 m. One side of the walls is fitted at its bottom with stainless-steel tubes where 1 mm holes have been drilled every centimetre for air sparging. The mixed liquor was partially recycled at the inlet. A settler of 0.88 m3 was used to clarify the mixed liquor and to produce sludge which was also partially recycled at the reactor inlet. The experimental set-up is presented in Fig. 1. Real wastewater is difficult to handle in laboratory since its composition changes over days, whereas a steady composition is necessary to perform reproducible experiments. Therefore synthetic wastewaters are often used to carry out reliable experiments. Such synthetic wastewaters are frequently composed of a mixture of sugar (glucose and saccharose), ethanol, and acetate or of Viandox (a synthetic substrate of vegetal amine extract, meat extract and sugar) as the readily degradable substrate. Ammonium chlorine as ammoniacal nitrogen and potassium dihydrogenophosphate as phosphor source are also added. Some authors add mineral compounds such as MgSO4, FeSO4, CaCl2 or NaHCO3 (Wang et al., 2006; Beun

Fig. 1 e Experimental set-up.

3087

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

et al., 1999; Orhon and Artan, 2006; Lobos et al., 2007). Typically, a real wastewater has a COD/N/P ratio of around 100/5/1. In the present study, the synthetic wastewater composition was based on Viandox with ammonium sulphate and potassium dihydrogenophosphate: 1.36 L of Viandox, 32 g of ammonium sulphate, 8 g of potassium dihydrogenophosphate and 500 L of water. The proportion of COD/N/P was thus 100/15/1. The behaviour of this synthetic wastewater has been tested through respirometric tests following the method proposed by Le Bonte´ et al. (2005) and compared with the wastewater of the WWTP located in Nancy-Maxe´ville (France). Fig. 2 shows the results of the respirometric tests carried out with the synthetic wastewater and with the wastewater from the industrial plant. Sludge has been acclimated to the Viandox substrate during approximately one month before these experiments. This substrate seems more difficult to assimilate than real wastewater. Initial rate of reaction is lower and the total COD consumed is 70% of the real wastewater COD consumption (Fig. 2). Nevertheless, Viandox substrate is a good approximation of real wastewater. Saccharose could be a good additive to this substrate in order to improve its behaviour. The behaviours of the two substrates are close enough to consider the synthetic one as representative of the urban wastewater.

For each experiment, measurements of the suspended solid concentration and of the gas-liquid mass transfer coefficient kLa have been carried out. Nitrate, ammonium, soluble COD and oxygen concentrations have been measured at the reactor inlet, every 30 cm in the reactor and at the settler outlet. Nitrates concentrations have been measured by ionic chromatography, ammonium concentration and soluble COD by standard HACH protocols and oxygen concentration with a standard oxygen probe. Measurement uncertainties have been evaluated from the experimental protocol, the measurement techniques and the dilution ratio. Table 1 summarizes these uncertainties. The kLa value has been determined experimentally with the method proposed by Le Bonte´ et al. (2005). A defined quantity of biomass was taken from the steady state reactor and introduced into a watertight perfectly mixed respirometer equipped with an oxygen probe. The evolution of the oxygen concentration in this respirometer allowed the determination of the instantaneous oxygen uptake rate which is equal to the oxygen transfer rate induced by gas bubbling in the reactor.

2.3.

Description of experiments

Experiments have been carried out in four main Phases:

2.2.

Experimental parameters

Liquid residence time was set to 140 min for all experiments (a minimum of 120 min is necessary to allow a good development of biomass) which is equivalent to a flowrate of 1.545
105 m3/s. Recycling rates were set to 4 for the mixed liquor and 1 for the sludge. Treated wastewater flowrate was therefore 0.262
105 m3/s. These choices have been made to have a standard reactor behaviour and to ensure the presence of a significant amount of biomass in the reactor. The sludge retention time was set to approximately 10 days by a purge of the sludge outlet of the decanter and by the analytic sampling. The temperature of the reactor was the same as ambient temperature: approximately 20  C, during all experiments.

- The first Phase was biomass acclimatization to substrate: biomass has been acclimatized to the artificial substrate in three steps. Firstly, the reactor was loaded with biomass taken from the Nancy-Maxe´ville wastewater treatment plant; it was aerated during three days without any additional substrate. Secondly, the wastewater was circulated with half of the nominal pollution charge during one week. Finally, operating conditions were stabilized during another week before experiments began. This stabilization Phase allowed checking that the pilot plant worked properly with reasonable performances. - The second Phase consisted of an experiment at low COD concentration in the wastewater. Because air flowrate was purposely too high in the first Phase to accelerate the biomass development, it has been reduced by half. experimental measurements of the concentration profiles began after one week of stabilization to assume steady state conditions.

Table 1 e Uncertainties estimation of experimental measurements.

Fig. 2 e Respirogram of real wastewater and Viandox based synthetic wastewater.

Measurement

Range

Uncertainty

COD (mgO2/L) COD (mgO2/L)

0e150 0e750

16% 8%

Total nitrogen (mg/L) Ammonium (mg/L)

0e150 0e20

12% 14%

Nitrate (mg/L) Suspended solids (g/L)

0e100 0.5e50

10% 0.1 g/L

Dissolved oxygen (mgO2/L)

0e10

0.4 mg/L

3088

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

- The third Phase consisted of an experiment at high COD concentration, twice the one of Phase 2. Once again, experimental measurements began after one week of stabilization to get steady state conditions. - Finally, the fourth Phase consisted in carrying out experiments with an anoxic zone located in the first third of the reactor. The COD concentration was of the same order of magnitude as in Phase 3. Here again, measurements began after one week of stabilization. The optimal stabilization time to ensure a steady state operation is around 5 times the sludge retention time. In our experiment, this led to unmanageable synthetic wastewater storage, therefore a shorter stabilization time of one week, close to the sludge retention time, has been chosen. This ensured an almost steady operation during a few days (i.e. the amount and nature of biomass did not evolve significantly).

3.

Modelling aspects

3.1.

Kinetics modelling

The biological kinetics model chosen for this study is the ASM-1 model. This model is frequently adopted to simulate or predict performances of biological reactors. It is suitable to simulate carbone oxidation, nitrification and denitrification in the aerobic and anoxic zones of activated sludge reactors. It considers 12 different components and 8 kinetics processes. The biomass is composed of heterotrophic, autotrophic and inert biomass. Heterotrophic biomass grows with the consumption of soluble biodegradable pollution and oxygen (aerobic processes) or nitrate (anoxic process). Autotrophic biomass grows with the consumption of ammonium and oxygen and produces nitrate. Both heterotrophic and autotrophic biomasses decay into inert biomass, particulate biodegradable pollution and particulate nitrogen pollution. Both particulate biodegradable pollution and particulate organic nitrogen are hydrolysed in the presence of heterotrophic biomass into, respectively, soluble biodegradable pollution and soluble organic nitrogen. Finally soluble organic nitrogen is transformed into ammonium by the action of heterotrophic biomass. The considered components are briefly described in Table 2. The 8 processes are summarized in Table 3. The standard IWA advised values are used for all stœchiometric and kinetics coefficients. The fractionation of the synthetic influent can be found in Table 4. It can be noticed that with this kind of synthetic substrate the particulate substrate XS and XI is soluble and therefore the filtered COD measurements represent the sum of SS þ SI þ XS þ XI. XS is soluble but behaves like a slowly biodegradable substrate. SI and XI behave exactly the same way, that explains why XI ¼ 0 in our influent representation. A few modified ASM-1 models have been proposed in the literature either to extend the reactions taken into account or to refine the model such as the ASM-2 (phosphor chemistry taken into account), ASM-3 (better representation of COD consumption mechanism) or BNRM-1 (better representation of flocs) models (Henze et al., 2001; Seco et al., 2004). In this work, we focused on the effects of the hydrodynamics

Table 2 e Brief description of the ASM-1 model components. Notation SI SS

XI XS

Description

Process(es) in which each component is involved

Soluble inert pollution Soluble biodegradable pollution

none

Particulate inert pollution Particulate biodegradable pollution

none

Aerobic and anoxic growth (r1 and r2), Hydrolysis (r7)

Decay (r4 and r5), Hydrolysis (r7)

XB;A

Heterotrophic Aerobic and anoxic growth (r1 and r2), Decay(r4) biomass Autotrophic biomass Aerobic growth (r3), Decay (r5)

XP SO

Inert biomass Dissolved oxygen

Decay (r4 and r5) Aerobic growth of XB;H and XB;A (r1 and r3)

SNO

Nitrate and nitrite

SNH

Ammonium

Anoxic growth of XB;H and aerobic growth of XB;A (r2, r3) Aerobic growth (r3), Ammonification ( r6)

SND

Soluble organic nitrogen Particulate organic nitrogen

XB;H

XND

Ammonification ( r6), Hydrolysis (r8) Decay (r4 and r5), Hydrolysis (r8)

interaction with biochemistry, we chose to keep the biochemistry model as simple as possible. Therefore, the simple ASM-1 model has been chosen as a reference without any modifications of its constants for all the simulations. However, some possible modifications of the model parameters are detailed in the discussion part.

3.2.

Systemic approach

The liquid phase hydrodynamics in this reactor is well modelled either by the plug flow with axial dispersion model (Pons et al., 1996) or by the equivalent continuous stirred tank reactors (CSTRs) in series model (Le Moullec et al., 2008a). The number of CSTRs (or equivalently, the axial dispersion coefficient) depends on the gas and liquid flowrates and on the reactor geometry. It has been shown that a series of CSTRs with backmixing presents the advantage to take into account the variation of the number of CSTRs with the liquid flowrate that occurs along the day by simply adjusting the backmixing rate without modifying the reactors number (Potier et al., 2005): this approach was adopted in the present study. The number of mass balance equations solved ranged between 5 and 20 times the number of compounds. The dispersion coefficient of the reactor has been determined with the correlation developed in a previous work (Le Moullec et al., 2008a). The oxygen transfer was modelled with a volumetric gas-liquid mass transfer

3089

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

Table 3 e Brief description of the processes of the ASM-1 model. Description Aerobic growth of heterotrophic biomass Anoxic growth of heterotrophic biomass Aerobic growth of heterotrophic biomass

Balance equation

Kinetics expression


1 1  YH SS þ SO þ iXB SNH /XB;H YH YH

r 1 ¼ mH

1 1  YH SS þ SNO þ iXB SNH /XB;H 2:86YH YH

r 2 ¼ m H hg


iXB þ

 1 4:57  YH 1 SO /XB;A þ SNO SNH þ YH YA YA

SS KS þ SS



r 3 ¼ mA




SS K S þ SS

XB;H /ð1  fp ÞXS þ fp XP þ ðiXB  fp :iXP ÞXND

r4 ¼ bH
XB;H

Decay of autotrophic biomass

XB;A /ð1  fp ÞXS þ fp XP þ ðiXB  fp :iXP ÞXND

r5 ¼ bA
XB;A

Ammonification

SND /SNH

r6 ¼ ka :SND
XB;H

Particulate biodegradable pollution hydrolysis

XS /SS

r7 ¼ kh

Particulate organic nitrogen hydrolysis

XND /SND

r8 ¼

coefficient (kLa) which has been determined experimentally by the method proposed by Le Bonte´ et al. (2005) previously described in the experimental part of this paper.

3.3.

CFD approach

The CFD study has been carried out with the FLUENT package using an EulereEuler approach. A drag coefficient suitable for bubbles, bubble induced turbulence source terms and a degassing boundary condition have been implemented via user-defined functions. The simulated liquid phase velocity fields and turbulence characteristics have been compared to laser Doppler velocimetry measurements and the gas fraction field has been compared to optical probe measurements, with a satisfying agreement. The oxygen gas-liquid mass transfer coefficient was calculated using the Higbie’s film penetration theory. Each of the 12 components concentrations involved in the kinetics process were simulated by a transport equation coupled with a source term for the biological reaction. This hydrodynamics/kinetics coupling is satisfactory because the smallest kinetics timescale is much larger than the hydrodynamics timescales. This model has been more thoroughly

Table 4 e Synthetic influent fractionation in the ASM-1 model. Notation

Value

SI SS

50 mg/L 127 mg/L

XI XS

0 mg/L 393 mg/L

SNO SNH

0 mg/L 64 mg/L

SND XND

8 mg/L 19 mg/L




SNH KNH þ SNH

Decay of heterotrophic biomass




 SO XB;H KO;H þ SO KO;H KO;H þ SO




XS =XB;H KX þ XS =XB;H




 SNO XB;H KNO þ SNO

 SO XB;A KO;A þ SO

"


SO KO;H þ SO




þ hh

KO;H KO;H þ SO




SNO KNO þ SNO

# XB;H

XND r XS 7

detailed by Le Moullec et al. (2010b). The number of cells ranged between 50000 and 200000. In each of these cells, mass, momentum, turbulence energy and turbulence dissipation rate balance must be solved in combination with an additional mass balance equation for each compound.

3.4.

Compartmental approach

The compartmental modelling describes the reactor as a network of functional compartments spatially localized (Rigopoulos and Jones, 2003; Guha et al., 2006; Le Moullec et al., 2010a). It is based on the determination of volumes in which the physicalechemical properties are homogeneous within a given tolerance. In our case, we based the choice of the pertinent properties, and the determination of the compartments, on both process knowledge and hydrodynamics simulations using the CFD code without reactions. The interested reader can find detailed information on the determination of the compartmental model in Le Moullec et al. (2010a); thus, only the basic principles are summarized here. For the channel reactor, the flow is invariant along the length. Therefore, the reactor was split into slices of equal sizes along its length. The number of slices depends mostly on the residence time distribution of the reactor in order to reproduce correctly its hydrodynamic behaviour; a minimum of 5 slices is necessary in the present case. Each slice was divided into different compartments. The number, the shape and the connectivity of these compartments were determined using a detailed analysis of three key parameters (previously determined using CFD calculations): gas fraction, liquid velocity field and liquid turbulence characteristics. The flowrates between two adjacent compartments due to convective transport were computed from CFD simulations. Turbulent dispersion between two adjacent compartments was computed with the method presented in Le Moullec et al. (2010a). The final structure of the compartment model is presented in Fig. 3. The number of differential equations

3090

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

solved for this modelling approach, ranged between 50 and 500 times the number of compounds in the liquid and gas phases.

3.5.

Modelling of the settler

The behaviour of the reactor is linked to the behaviour of the settler. Therefore the modelling of the settler can impact the results for the reactor. In order to avoid any interference between these two parts of the process, the separation efficiency of the settler was considered identical for each model. Experiments allowed an estimation of the settler efficiency in steady state which is close to 100% with a relative error of 1%. The efficiency of our experimental settler calculated using the IWA model (Copp, 2001) is 99.6%, which corresponds to our experimental results within the experimental uncertainty range. Thus, the quality of separation of particulate compounds has been considered equal to 99.6% for the three models. This assumption is reasonable and allows the comparison of the three models of the aerated sludge reactor without any influence or interference with the settling process.

4.

Results and discussion

4.1.

Experimental results

Table 5 summarizes the main results of the different experiments Phases. The purpose of the first experiment was to acclimate the biomass to the substrate, to be sure of the correct behaviour of the reactor and the settler. The pollutant removal performance

is 80.9% for COD and 84.6% for Kjeldhal nitrogen. Biomass concentration is about 1.1 g/L, which is low compared to the industrial values. Approximately 14% of the nitrogen is consumed in the reactor: this low value is probably due to measurement uncertainty or stagnant zones inside the reactor inducing anoxic fermentation. These results are summarized in Table 5. Oxygen concentration within the reactor is very high: between 7 and 8 mg/L, whereas the saturation concentration is 9.6 mg/L (Fig. 4a). The oxygen uptake rate (OUR) illustrates the bio-activity along the reactor more clearly (Fig. 4a). The COD evolution along the reactor is consistent with the expected normal behaviour with a regular decrease along the reactor length (Fig. 4b). Ammonium concentration is stable along the reactor and nitrate concentration raises slightly. Equilibrium exists between the consumption and production of ammonium because of the high oxygen concentration. Nitrates are not consumed because there is no dedicated anoxic zone in this experiment (Fig. 4c). This experiment validates the experimental procedure and experiments but it has not been taken into account for model validation because of the lack of anoxic zone. After this first Phase, the second series of experiments was carried out with half of the air flowrate used in the first experiment. Biomass concentration has risen to 1.45 g/L, substrate and oxygen consumption have, therefore, increased as shown in Fig. 5a and b. COD removal performance is approximately 87.3% and Kjeldhal nitrogen removal is 79.5%. It should be noticed that the measurement error on nitrogen is about 9%, which needs attention for the experiments interpretation. Evolution of COD concentration along the reactor is relatively stiff (from 60 mg/L to 30 mg/L) as illustrated in Fig. 5b. This important pollutant abatement will allow evaluating the

Fig. 3 e Structure of the compartmental model.

3091

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

Table 5 e Summary of experimental results. Phase

1

Experiment

1

2

3

4

5

6

7

8

9

140 38.6

140 21.4

140 21.4

140 21.4

140 21.4

140 21.4

140 21.4

140 21.4

140 21.4

9.5 6.3 2.2

5.5 4.7 2.7

5.5 4.7 2.7

5.5 4.7 2.7

5.5 4.7 2.7

5.5 4.7 2.7

5.5 4.7 2.7

6.7 e e

6.7 e e

570 28 64

570 28 64

Liquid residence time (min) Air flowrate (L/min) kLa (h1) Axial dispersion coefficient (104 m2/s) Equivalent number of CSTR Inlet COD (mg/L) Inlet organic nitrogen (mg/L) Inlet ammonium (mg/L)

285 14 32

2

285 14 32

3

285 14 32

285 14 32

570 28 64

570 28 64

4

570 28 64

Dry matter concentration (g/L) Turbidity of cleaned water (TBU)

1.1 13.0

1.3 23.8

1.5 28.1

1.5 24.0

1.7 25.6

1.6 24.2

1.8 29.9

2.5 31.1

3.2 28.7

COD removal (%) Kjeldhal nitrogen removal (%) Total nitrogen removal (%)

80.9 84.6 13.9

88.1 82.6 11.0

88.9 78.1 5.9

84.9 77.9 12

89.7

90.0 85.4 24.6

89.6 85.7 25.7

85.6 86.2 37.8

85.8 84.7 39.3

ability of the different models to reproduce rapid variations of pollutant concentrations along the reactor. Average oxygen concentration is 5.5 mg/L (Fig. 5a). Evolutions of ammonium and nitrate concentrations are opposite as shown in Fig. 5c and d. The high depollution performance obtained during this experiment despite a lower oxygen concentration than in the previous one confirms that the air injection flowrate was too high in the previous experiment. The third experiment was carried out with twice the COD concentration of the second experiment. The concentration of the correctly fed biomass is approximately 1.7 g/L. The COD removal and the Kjeldhal nitrogen removal are, respectively, 90% and 85%. The removal of nitrogen is approximately 25% with an uncertainty of 10%. As already mentioned before, there are probably some anoxic reactions in a dead zone or in the settler. Average oxygen concentration is 4 mg/L (Fig. 6a). This high value of air flowrate was necessary to overcome the pressure drop along the air injection device. Evolution of COD concentration is even stiffer than in the second experiment (from 130 mg/L to 60 mg/L) (Fig. 6b). Despite the high variability and/or uncertainty of nitrate measurement, evolutions of ammonium and nitrate concentrations are logically opposite (Fig. 6c and d). Finally, the fourth experiment was carried out with an anoxic zone located in the first third of the reactor. The same air flowrate as in the former experiments was sparged in the two last thirds of the reactor which induced a raise of the kLa and of the dispersion coefficient. The COD removal and the Kjeldhal nitrogen removal are, respectively, 86% and 85%. The nitrogen removal is approximately 40%. The oxygen profile illustrates the biological activity of the reactor (Fig. 7a). This low nitrogen removal is probably due to a very low concentration of denitrifying biomass. COD, nitrogen and ammonium concentration profiles seem invariant along the reactor (Fig 7b, c and d). As planned, the global performance of the pilot plant reactor in terms of COD removal is consistent with the one of a real wastewater treatment plant. The pilot scale reactor has

been dimensioned in order to reproduce the hydrodynamics, mass transfer and depollution efficiency of the Nancy-Maxe´ville municipal plant. The observed positive (or negative) total nitrogen removal during Phases 2, 3 and 4 can be explained by experimental uncertainties (approximately 12%) or by the presence of an anoxic zone in the settler. The performance of nitrogen removal during anoxic experiments seems very low. This is probably due to the quality of biomass in the reactor; the synthetic wastewater acclimatization may have induced a sludge evolution detrimental to nitrogen removal.

4.2.

Comments on the experiments

Measurements made along the reactor are quite reproducible from one experiment to another. However, the evolution of concentrations along the reactor can sometimes be quite low. Previous hydrodynamic studies on this reactor have shown that its hydrodynamics without recycling is equivalent to 2.2 or 3 continuous stirred tank reactors in series (Potier et al., 2005). Because of the high recycling rate, the overall behaviour of the system tends towards much more stirred conditions (perfectly stirred tank would correspond to an infinite recycling rate) and lead to flatter concentration profiles for the less eliminated compounds. During the experiments, the reactor has been considered in stationary regime. In practice, biomass concentration grows during successive experiments even if the substrate feeding rate is kept constant. But the characteristic times of oxygen and substrate consumption as well as the mixing time are much shorter than the characteristic time of biomass growth. Thus, the slow biomass concentration evolution can be neglected during each experiment. The high uncertainty of nitrate measurement is not only due to the analytical method but also to two specific effects:  The very high dilution needed for the analysis tends to increase its uncertainty,

3092

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

a

0,02 8,0

6,0

0,012 4,0

0,008

2,0

OUR (mg/L/s)

O2 concentration (mg/L)

0,016

0,004 O2 experiment 1 OUR experiment 1

0

0,0 0

60

120

180

240

300

360

Coordinate along the reactor (cm)

b

60,0

COD concentration (mg/L)

COD experiment 1

40,0

20,0

0,0 0

60

120

180

240

300

360

NH4 concentration (mg/L)

c

20,0

50,0

16,0

40,0

12,0

30,0

8,0

20,0

4,0

10,0

NO3 concentration (mg/L)

Coordinate along the reactor (cm)

NH4 experiment 1 NO3 experiment 1

0,0

0,0 0

60

120

180

240

300

360

Coordinate along the reactor (cm)

Fig. 4 e a: Oxygen concentration and OUR profiles along the reactor for Phase 1. b: COD concentration profile along the reactor for Phase 1. c: Ammonium and nitrate concentration profiles along the reactor for Phase 1.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

3093

Fig. 5 e a: Oxygen concentration along the reactor for Phase 2. b: COD concentration profile along the reactor for Phase 2. c: Nitrate concentration profile along the reactor for Phase 2. d: Ammonium concentration profile along the reactor for Phase 2.

 There was some delay between sampling and analysis; during this delay, samples have been filtered on a 22 mm filter before being stocked in cold room for approximately one week.

4.3. Comparison between experimental and simulated results It should be pointed out that all the simulations have been carried out with the standard parameters of the ASM-1 model without any adjustment. Oxygen concentration profiles are overestimated by the systemic model by approximately 25%, well represented by the compartmental model and underestimated by CFD at the inlet of the reactor by 60% (Fig. 5a). It can be noticed that the shape of the oxygen profile simulated with CFD is significantly different from the oxygen profiles obtained experimentally and with other models (Figs. 5a and 6a). One possible explanation for this shape may be a partial mixing near the reactor inlet for the CFD simulation, whereas the mixing is complete in the first CSTR for the other two models. This explanation is supported by the concentration patterns obtained with CFD near the inlet. The fact that the discrepancy is clearer for the high COD experiments is consistent with this hypothesis. COD and nitrate concentration profiles are well modelled for the phase 3 experiments (Fig. 5b and c) and quite well modelled for the Phase 4 experiments (Fig. 6b and c). Total soluble COD is the central variable of the ASM-1 model and is the result of numerous processes

(Table 3), therefore errors on different effects may compensate each other, which explains that it is well simulated despite other variables, while linked to COD, are less accurately predicted. Ammonium concentration profiles are not well predicted by any of the three models (Figs. 5d, 6d and 7d). This will be discussed in the next paragraph. All the three models have very similar responses regarding experimental parameters. One of the major reasons is that they all use the same biological kinetics laws and only differ by the modelling of phases hydrodynamics. Moreover, the three models reproduce the same global residence time distribution so the impact of the variation of hydrodynamics on the simulated results is less important than the one of the biochemistry model.

4.4.

Comments on the modelling

4.4.1.

Compartmental model structure

Concentrations of all the reactants in three of the four compartments of a slice of the compartmental model (see Fig. 3) appear to be equal. Only the sparger zone exhibits a different concentration distribution, in particular a higher level of oxygen. The dead zone and the central zone have only little influence on the model results. The dead or stagnant zone may have an effect on the nitrification and denitrification reactions. This point has been pointed out during the experimental results description, where it has been shown that the level of ammonium is sometimes far from what can be expected. However, the ASM-1 model seems to be very

3094

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

Fig. 6 e a: Oxygen concentration along the reactor for Phase 3. b: COD concentration profile along the reactor for Phase 3. c: Nitrate concentration profile along the reactor for Phase 3. d: Ammonium concentration profile along the reactor for Phase 3.

sensitive to the initial heterotrophic to nitrifying biomass ratio for the prediction of the ammonium concentration as detailed in section 4.4.3. Therefore, the presence of the stagnant zone taken into account in the hydrodynamic model does not affect the prediction because the error due to the estimation of nitrifying biomass is larger than the effect of low oxygen concentration in the stagnant zone. Moreover, even if the stagnant zone is modelled in the sense of hydrodynamics, the specific long time of biochemistry taking place in very low oxygen concentration volumes (such as anaerobic or specific anoxic reactions) is not modelled. Therefore, with the current level of knowledge, it is possible to simplify the compartmental model with only two compartments per slice. In this configuration, the model is similar to the one developed by Rigopoulos and Jones (2003).

4.4.2.

4.4.3.1. Determination of the growth rate of biomass. Respirometric measurements allow the determination of the maximal growth rate of heterotrophic biomass. In our case, the couple biomass/wastewater is correctly represented with a maximal heterotrophic biomass growth rate of 3.5
105 s1 whereas the default ASM-1 value is 6.94
105 s1. For the couple biomass/synthetic substrate, this maximal growth rate is 4.106 s1. This variability illustrates the difficulty to obtain a reliable kinetics constants set.

Difficulties to simulate anoxic experiments with CFD

In the anoxic zone, the sludge is maintained in suspension by an inclined Rushton turbine. The CFD modelling of an open gas-liquid mechanically mixed reactor is difficult and time consuming. It was not covered by the scope of this work. Therefore, CFD has not been used to simulate the fourth Phase experiments.

4.4.3.

wastewater characteristics. Moreover, these parameters have been experimentally determined by global measurement methods, such as respirometry; therefore it is possible that they are not adapted to a local approach such as computational fluid dynamics. Two examples can highlight these considerations:

Complexity of kinetics modelling

All the kinetics constants used in the ASM-1 model have been kept to their default values proposed by IWA. These parameters are clearly dependent on the reactor biomass and on the

4.4.3.2. Influence of the fraction of nitrifying biomass on ammonium concentration. The poor estimation of ammonium concentration can be due to an error in the autotrophic biomass initial concentration estimation. In fact, ammonium concentration is very dependent on this autotrophic biomass. This nitrifying biomass needs a sludge retention time ranging from 10 to 25 days to develop. In our experiment, the sludge retention time is at the minimum boundary of this range. It is, therefore, probable that the hypothesis of 10% of autotrophic biomass is an overestimation. With 5% of autotrophic biomass, ammonium concentration would be correctly

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

3095

Fig. 7 e a: Oxygen concentration along the reactor for Phase 4. b: COD concentration profile along the reactor for Phase 4. c: Nitrate concentration profile along the reactor for Phase 4. d: Ammonium concentration profile along the reactor for Phase 4.

simulated. Nitrate concentration is also influenced by this percentage and the hypothesis of 5% of autotrophic biomass induces an underestimation of nitrate concentration. Thus, even if a correction on autotrophic biomass estimation corrects 90% of the simulation errors, the last 10% of errors can only be corrected with a good estimation of the kinetics parameters. Two others ASM-1 parameters having a great influence on ammonium concentration are the kinetics constant of ammonification ka and the half saturation constant of ammonium KNH. The tuning of these two constants could lead to a good representation of both ammonium and nitrate concentration.

5.

Conclusion

The objective of this work was to develop an experimental study of an aerated sludge channel reactor in order to validate several types of models, not only for global predictions but also for the local estimation of the concentrations. Such studies should allow to improve the detailed design of aerated reactors in wastewater treatment plants (gas distribution system, baffles location.). The obtained results have two different levels of interest: - From the wastewater treatment point of view, we demonstrated that it is possible to predict the major pollutants concentrations not only with a detailed

approach such as CFD but also with less time consuming models such as a compartmental approach (1 week of calculation for the CFD approach compared to a few minutes for the compartmental model). If oxygen and nitrate concentration and carbon reactions predictions seem to be reasonable, this is not the case for ammonium predictions. This is due to the high sensitivity of the existing kinetics model ASM-1 to initial nitrifying biomass. This value is difficult to estimate a priori and it has a strong influence on ammonium concentration predictions. We showed that the way to determine the kinetics model and/or the initial composition has a strong influence on the concentration profiles predictions. As a matter of fact, if the systemic approach gives reasonable predictions, this is also due to the fact that this type of model has been used for the determination of the kinetics model by parameters optimisation: it takes into account in an implicit manner the flow behaviour and mass transfer globally as in systemic models. Researches to improve or to extend the kinetics scheme are thus of primary importance (Henze et al., 2001; Secco et al., 2004). - At a second level, it is one of the first times that the compartmental approach has been validated in the case of real industrial reactions. This is an innovative methodology with a much lower calculation time than CFD. Unfortunately, it still remains necessary to have a more detailed kinetics model to take all the advantages of this method for the design of industrial reactors.

3096

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

Acknowledgments Authors would like to thank the instrumentation and mechanics staffs and especially Pascal Beaurain and Christian Blanchard for the construction and adjustments of the pilot plant and Irene Ramirez for her work on respirometry and her help during the pilot experiments. Experiments have been carried out with the advices of Marie-Noe¨lle Pons.

Nomenclature

ka kLa KNH SI SND SNH SNO SO SS XBA XBH XI XND XP XS

kinetics constant of ammonification (m3 kg1 s1) volumetric gas-liquid mass transfer coefficient (s1) half saturation constant of ammonium (mgNH4 L1) soluble inert pollution (mgDCO L1) soluble organic nitrogen (mgN L1) ammonium (mgNH4 L1) nitrate and nitrite(mgNO2 L1) dissolved oxygen(mgO2 L1) soluble biodegradable pollution (mgDCO L1) autotrophic biomass (mg L1) heterotrophic biomass(mg L1) particulate inert pollution(mgDCO L1) particulate organic nitrogen(mgN L1) inert biomass (mg L1) particulate biodegradable pollution(mgDCO L1)

Abbreviations ASM Activated Sludge Model CFD Computational Fluid Dynamics COD Chemical Oxygen Demand CSTR Continuous Stirred Tank Reactor OUR Oxygen Uptake Rate WWTP WasteWater Treatment Plant

references

Ahnert, M., Kuehn, V., Krebs, P., 2010. Temperature as an alternative tracer for the determination of the mixing characteristics in wastewater treatment plants. Water Research 44 (6), 1765e1776. Beun, J.J., Hendriks, M.C., Van Loodsdrecht, M.C.M., Morgentorh, E., Wilderer, P.A., Heijnen, J.J., 1999. Aerobic granulation in sequencing batch reactor. Water Research 33 (10), 2283e2290. Brannock, M., Leslie, G., Wang, Y., Buetehorn, S., 2010a. Optimising mixing and nutrient removal in membrane bioreactors: CFD modelling and experimental validation. Desalination 250 (2), 815e818. Brannock, M.W.D., Wang, Y., Leslie, G., 2010b. Evaluation of fullscale membrane bioreactor mixing performance and the effect of membrane configuration. Journal of Membrane Science 350 (1e2), 101e108. Cockx, A., Do-Quang, Z., Audic, J.M., Line´, A., Roustan, M., 2001. Global and local mass transfer coefficients in waste water treatment process by computational fluid dynamics. Chemical Engineering and Processing 40, 187e194. Copp, J.B., 2001. COST ’Simulation Benchmark’ Manual. IWA Publishing.

De Bonis, M.V., Ruocco, G., 2010. Heat and mass transfer modeling during continuous flow processing of fluid food by direct steam injection. International Communications in Heat and Mass Transfer 37 (3), 239e244. Diamantis, V., Papaspyrou, I., Melidis, P., Aivasidis, A., 2010. High aeration rate enhances flow stratification in full-scale oxidation ditch. Bioprocess and Biosystems Engineering 33 (2), 293e298. Glover, G.C., Printemps, C., Essemiani, K., Meinhold, J., 2006. Modelling of wastewater treatment plants e how far shall we go with sophisticated modelling tools? Water Science and Technology 53 (3), 79e89. Gresch, M., Braun, D., Gujer, W., 2010. The role of the flow pattern in wastewater aeration tanks. Water Science and Technology 61 (2), 407e414. Guha, D., Dudukovic, M.P., Ramachandran, P.A., Mehta, S., Alvare, J., 2006. CFD-based compartmental modelling of single phase stirred-tank reactors. AIChE Journal 52 (5), 1836e1846. Henze, M., Gujer, W., Mino, T., van Loosedrecht, M., 2001. Activated Sludge Models ASM1, ASM2, ASM2d and ASM3. IWA Publishing. Laborde-Boutet, C., Larachi, F., Dromard, N., Delsart, O., Schweich, D., 2009. CFD simulation of bubble column flows: Investigations on turbulence models in RANS approach. Chemical Engineering Science 64 (21), 4399e4413. Le Bonte´, S., Potier, O., Pons, M.N., 2005. Toxic event detection by respirometry and adaptive principal components analysis. Environmetrics 16, 589e601. Le Moullec, Y., Potier, O., Gentric, C., Leclerc, J.-P., 2008a. A general correlation to predict axial dispersion coefficients in aerated channel reactors. Water Research 42 (6e7), 1767e1777. Le Moullec, Y., Potier, O., Gentric, C., Leclerc, J.-P., 2008b. Flow field and residence time distribution simulation of a crossflow gas-liquid wastewater treatment reactor using CFD. Chemical Engineering Science 63 (9), 2436e2449. Le Moullec, Y., Gentric, C., Potier, O., Leclerc, J.-P., 2010a. Comparison of systemic, compartmental and CFD modelling approaches: application to the simulation of a biological reactor of wastewater treatment. Chemical Engineering Science 65 (1), 343e350. Le Moullec, Y., Gentric, C., Potier, O., Leclerc, J.-P., 2010b. CFD simulation of the hydrodynamics and reactions in an activated sludge channel reactor of wastewater treatment. Chemical Engineering Science 65 (1), 492e498. Levin, M.A., Gealt, M.A., 1993. Biotreatment of Industrial and Hazardous Waste. MCGraw-Hill, Inc. 71e72. Lobos, J., Wisniewski, C., Heran, M., Grasmick, A., 2007. Membrane bioreactor performances: effluent quality of continuous and sequencing systems for water reuse. Desalination 204, 39e45. Makinia, J., Wells, S.A., 2000. A general model of the activated sludge reactor with dispersive flow-I. Model development and parameter estimation. Water Research 34 (16), 3987e3996. Orhon, D., Artan, N., 2006. Nutrient removal performance of a five-step sequencing batch reactor as a function of wastewater composition. Process Biochemistry 41 (1), 216e220. Pons, M.N., Pereira, L., Roche, N., Weiss, B., Prost, C., Corriou, J.P., 1996. Methodology for the development of a wastewater treatment plant simulator. Computers Chemical Engineering 20, 1395e1400. Potier, O., Leclerc, J.P., Pons, M.N., 2005. Influence of geometrical and operational parameters on the axial dispersion in an aerated channel reactor. Water Research 39, 4454e4462. Rigopoulos, S., Jones, A., 2003. A hybrid CFD-reaction engineering framework for multiphase reactor modelling: basic concept and application to bubble columns. Chemical Engineering Science 58, 3077e3089.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 8 5 e3 0 9 7

Schmit, C.G., Jahan, K., Schmit, K.H., Debik, E., Mahendraker, V., 2009. Activated sludge and other aerobic suspended culture processes. Water Environment Research 81 (10), 1127e1193. Schubert, M., Kryk, H., Hessel, G., Friedrich, H.-J., 2010. Residence time measurements in pilot-scale electrolytic cells: application of laser-induced fluorescence. Chemical Engineering Communications 197 (9), 1172e1186. Seco, A., Ribes, J., Serralta, J., Ferrer, J., 2004. Biological nutrient removal model no1. Water Science and Technology 50 (6), 66e78. Tizghadam, M., Dagot, C., Baudu, M., 2008. Wastewater treatment in a hybrid activated sludge baffled reactor. Journal of Hazardous Materials 154 (1e3), 550e557. Wang, C., Li, J., Wang, B., Zhang, G., 2006. Development of an empirical model for domestic wastewater treatment by biological aerated filter. Process Biochemistry 41 (4), 778e782.

3097

Wang, Y., Brannock, M., Leslie, G., 2009. Membrane bioreactors: overview of the effects of module geometry on mixing energy. Asia-Pacific Journal of Chemical Engineering 4 (3), 322e333. Yu, J., Gu, G., Wang, S., Zhang, Z., Sun, L., 2010. Modeling and simulation of dissolved oxygen concentration in a full-scale textile dyeing wastewater treatment plant. International Journal of Chemical Reactor Engineering Vol. 8 A29. Zima, P., Makinia, J., Swinarski, M., Czerwionka, K., 2008. Effects of different hydraulic models on predicting longitudinal profiles of reactive pollutants in activated sludge reactors. Water Science and Technology 58 (3), 555e561. Zima, P., Makinia, J., Swinarski, M., Czerwionka, K., 2009. Combining computational fluid dynamics with a biokinetic model for predicting ammonia and phosphate behavior in aeration tanks. Water Environment Research 81 (11), 2353e2362.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Proliferation of antibiotic resistance genes in microbial consortia of sequencing batch reactors (SBRs) upon exposure to trace erythromycin or erythromycin-H2O Caian Fan, Jianzhong He* Department of Civil and Environmental Engineering, National University of Singapore, Block E2-02-13, 1 Engineering Drive 3, Singapore 117576, Singapore

article info

abstract

Article history:

A variety of antibiotics and their metabolites at sub-inhibitory level concentrations are

Received 13 January 2011

suspected to expand resistance genes in the environment. However, knowledge is limited

Received in revised form

on the causal correlation of trace antibiotics or their metabolites with resistance prolifer-

14 March 2011

ation. In this study, erythromycin (ERY) resistance genes were screened on microbial

Accepted 14 March 2011

consortia of sequencing batch reactors (SBRs) after one year acclimation to ERY (100 mg/L)

Available online 21 March 2011

or dehydrated erythromycin (ERY-H2O, 50 mg/L). The identified esterase gene ereA explains that ERY could be degraded to six products by microbes acclimated to ERY (100 mg/L).

Keywords:

However, ERY could not be degraded by microbes acclimated to ERY-H2O (50 mg/L), which

Antibiotics

may be due to the less proliferated ereA gene. Biodegradation of ERY required the presence

Erythromycin (ERY)

of exogenous carbon source (e.g., glucose) and nutrients (e.g., nitrogen, phosphorus) for

Dehydrated erythromycin

assimilation, but overdosed ammoniumeN (>40 mg/L) inhibited degradation of ERY. Zoo-

(ERY-H2O)

gloea, a kind of biofilm formation bacteria, became predominant in the ERY degradation

Biodegradation

consortia, suggesting that the input of ERY could induce biofilm resistance to antibiotics.

Resistance

Our study highlights that lower mg/L level of ERY or ERY-H2O in the environment encour-

Esterase

ages expansion of resistance genes in microbes. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Antibiotics are detected in the environment such as hospital effluent, municipal wastewater, surface water, and groundwater (Kummerer, 2009a). The most frequently detected antibiotics and their derivatives include trimethoprim, tetracycline, norfloxacin, penicillin G, cefalexin, cefotaxim, and dehydrated erythromycin (ERY-H2O), which are usually persistent in wastewater treatment plants (WWTPs) (Kummerer, 2009a). The persistence of antibiotics may accelerate the development of resistance genes and resistance bacteria by exerting selective pressure on microbes in the environment. The occurrence and

transfer of new combination of resistance genes are predicted to be much more frequent in the compartments with higher diversity and abundance of microorganisms (Murray, 1997), such as WWTPs that receive various antibiotics and their metabolites discharged from residential or hospital areas (Gros et al., 2007). So far, it is not clear and far from complete whether antibiotics at concentrations as low as detected in hospital effluents (in the higher mg/L range) or in the aquatic environment (in the lower mg/L range) are important for the expansion of resistance in microbes (Kummerer, 2009b). The correlation of input of antibiotics at lower environmental concentrations with the development or occurrence of antibiotic resistance genes is

* Corresponding author. Tel.: þ65 6516 3385; fax: þ65 6774 4202. E-mail address: [email protected] (J. He). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.025

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

short of support with experimental data (Kummerer, 2009b). Other findings indicate that continuous input of resistant bacteria and resistance genes rather than the presence of antibiotics at sub-inhibitory concentrations may be more important for keeping resistance in the environment (Ohlsen et al., 2003, 1998). All these uncertainties have inspired researchers to investigate the fate of antibiotics and the occurrence of resistance in WWTPs and downstream natural water regions in the past decade (Le-Minh et al., 2010). However, these investigations have been becoming increasingly difficult due to the lack of reference WWTPs free from input of resistance bacteria and genes. Without a proper reference from an environmental system against resistance input, influence of antibiotics on resistance proliferation is inconclusive. Among the antibiotics studied, erythromycin (ERY) has received less attention compared with other antibiotics in WWTPs since ERY is sensitive to pH. At the operational pH ranges (6.5e8) of most municipal WWTPs, active ERY co-exists with its dehydrated-form ERY-H2O (Le-Minh et al., 2010). Erythromycin-H2O is removed mainly by sorption to sewage sludge in WWTPs via hydrophobic interactions and cation exchanges due to its surfactant-like structure, but its lower removal efficiency (9e19%) results in its higher concentration up to 6 mg/L in the effluent of WWTPs (Kummerer, 2009a; Le-Minh et al., 2010). Moreover, ERY-H2O was reported to induce bacterial resistance as ERY does (Fan et al., 2009; Majer, 1981). Accordingly, the occurrence of ERY derivatives other than ERY-H2O could also be possible to introduce microbial resistance. Up to now, knowledge is still scarce on the contribution of ERY or other ERY derivatives (e.g., ERY-H2O) at mg/L levels in wastewater treatment systems to the amplification of resistance genes. Microbial resistant mechanisms to ERY include excretion of ERY by efflux pumps (e.g., efflux genes mefA/E and msrA/B), alteration of the target site to avoid binding of ERY (e.g., erythromycin ribosomal methylase genes erm), and destruction of ERY directly (e.g., esterase genes ereA and ereB, macrolide-20 phosphotransferase gene mphA) (Amin et al., 2006; Wright, 2005). Among these resistance genes, erm genes (A, B, C, E, F, T, V, and X), mef genes (A, E, and I), msrA, ereA/B, and mphA genes have been detected in wastewater and activated sludge of WWTPs, and ermB is the most prevalent gene in the environmental samples (Szczepanowski et al., 2009; Zhang et al., 2009). Actually, all modes of ERY-related resistance genes can be found in both gram-positive and gram-negative bacteria (http:// www.ncbi.nlm.nih.gov/), but intrinsic possession of efflux resistance on gram-negative bacteria’s membrane results in less sensitivity of them to ERY than gram-positive bacteria (Pechere, 2001). Another mode of ERY resistance, destruction or the so-called biodegradation of antibiotics can be used as an indicator to interpret the function of antibiotic resistance genes. Until recently, a few studies have tackled the possible biodegradation of ERY at different concentrations (Alexy et al., 2004; Gartiser et al., 2007). A closed bottle test exhibited that ERY at concentrations of 2.46 mg/L (equal to the theoretical oxygen demand (ThOD) of 5 mg/L) was not readily biodegradable by activated sludge (Alexy et al., 2004). Erythromycin at 167 mg/L (equal to 100 mg total organic carbon (TOC) per litter) could not be degraded completely and had evident inhibition on carbon removal (Gartiser et al., 2007). Noteworthy, the inocula utilized by the above studies were taken from WWTPs not receiving

3099

effluents from hospitals and were assumed to be less-adapted to antibiotics. Biodegradation of antibiotics would rarely occur by non-selected microbes, that is, in the absence of resistance genes (Ding and He, 2010; Kummerer, 2009b). In order to highlight effects of low concentrations of ERY and ERY-H2O on amplification of resistance genes, it is of significance to investigate biodegradation of ERY by microorganisms acclimated to ERY or ERY-H2O (in the mg/L range). The aim of this study is to identify the development of resistance genes and to investigate biodegradation of ERY with microbial consortia that have been acclimated to ERY (100 mg/L) or ERY-H2O (50 mg/L) for over one-year running with synthetic wastewater free from resistant bacteria and resistance genes input. Findings of this study will provide significant information for the inadequate data on effects of trace antibiotics to promote resistance genes development in the aquatic and terrestrial environment.

2.

Materials and methods

2.1.

Startup and operation of SBRs

Three SBRs (4 L) were started up and operated in exactly the same conditions, including seeding sludge, feeding synthetic wastewater (theoretical chemical oxygen demand (COD), NH4þeN, and PO43eP of 600, 60, and 15 mg/L, respectively), and an 8-h operating batch mode (Fig. S1) as described previously (Fan et al., 2009). However, the SBRs differed from each other in terms of antibiotics spiked, ERY-H2O of 50 mg/L (R1), ERY of 100 mg/L (R2), and no antibiotics (R3). As reported previously (Fan et al., 2009), in order to minimize residue antibiotics and resistant materials, an 8-month pretreatment with the synthetic wastewater was applied on the seeding sludge in a mother reactor (MR) before being inoculated into the three SBRs. Erythromycin (potency  850 mg/mg) and all other chemicals were supplied by SigmaeAldrich (Singapore). Erythromycin stock solution and ERY-H2O were prepared as described previously (Abuin et al., 2006; Fan et al., 2009; McArdell et al., 2003).

2.2.

Batch experiments

Biodegradability of ERY (10 mg/L, simulation of peak concentration in the effluent of pharmaceutical production plants) was tested in 250 ml flask bottles with inocula from steady state R1 (ERY-H2O), R2 (ERY), and R3 (control) after running over one year (the performance of the SBRs were reported previously (Fan et al., 2009)). The inocula were freshly applied to short-term batch experiment. Before being inoculated to flask bottles, the activated sludge of 1 ml withdrawn from the three steady state SBRs were washed three times with the same medium used in the batches. Negative controls were prepared with autoclaved sludge to indicate possible ERY removal by non-biotic processes. Bottles were filled with 50 ml of medium similar to the synthetic wastewater described above except for various concentrations of COD (contributed by glucose), NH4þeN, and PO43eP in different batches as described below (Table 1). (1) Effects of inocula source on degradation of ERY: ERY degradation capability was tested

3100

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

Table 1 e e Batch experiments to study effects of inocula source, glucose (calculated as COD), NH4DeN, and PO43LeP on biodegradation of ERY (10 mg/L). Factors

Inocula source

COD (mg/L)

NH4þeN (mg/L)

PO43eP (mg/L)

Inocula source Glucose Ammonium Phosphate

R1, R2, and R3 R2 R2 R2

600 600 and 0 600 600

30 30 30, 40, 50, and 60 30

6 6 6 6, 13, 20, and 26

with inocula from R1 (ERY-H2O), R2 (ERY), and R3 (control) by using medium containing COD, NH4þeN, PO43eP, and ERY of 600, 30, 6, and 10 mg/L (COD:N:P ¼ 100:5:1, as assimilation required), respectively. (2) Effects of glucose on degradation of ERY: ERY (10 mg/L) degradation by inocula from R2 (ERY) was tested in medium with glucose equal to 600 mg/L COD or without glucose. Other elements in the medium were exactly the same as above mentioned (e.g., NH4þeN of 30 mg/L and PO43eP of 6 mg/L). (3) Effects of ammonium and phosphate on degradation of ERY: Degradation of ERY (10 mg/L) was tested with inocula from R2 (ERY) by using medium containing 600 mg/L COD, 6 mg/L PO43eP and gradient concentrations of NH4þeN (30, 40, 50, and 60 mg/L), or 30 mg/L NH4þeN and gradient concentrations of PO43eP (6, 13, 20, and 26 mg/L). All the above experiments were performed in triplicates. The flask bottles were placed on a shaker for mixing and aeration at room temperature (25e27  C). The pH ranged from 6.9 to 8.1 during one batch of experiments without adjustment. Samples from batch conical flasks were collected daily and centrifuged at 14,000 rpm for 15 min at 25  C. Therefore, the measured phosphate and all kinds of nitrogen were in dissolved forms throughout this study.

2.3.

Analytical methods

Dissolved phosphorus (PO43eP) and nitrogen in forms of NO3eN, NO2eN, and NH4þeN were measured on an ion chromatography (DIONEX DX500 chromatography system, USA). Erythromycin and ERY-H2O were detected on a high performance liquid chromatography (Agilent 1100 Series, Agilent Technologies, Germany) equipped with electro-spray tandem mass spectrometry (API 2000, Applied Biosystems/ MDS Sciex, USA; LC-MS-MS). A reverse phase Xbridge Phenyl column (50  2.1 mm id, 3.5 mm, Waters Inc., USA) was used according to the method reported previously (McArdell et al., 2003). Erythromycin and ERY-H2O were detected by using multiple reaction monitoring (MRM) pairs (m/z 734.5/ 158.2 amu for ERY; m/z 716.5/158.2 amu for ERY-H2O). Detection limits were 10 mg/L for ERY and 8 mg/L for ERY-H2O. Products of ERY were full-scanned by LC-MS over the m/z range of 100e1000 amu.

2.4. (PCR)

DNA extraction and polymerase chain reaction

The genomic DNA of 1 ml of mixed liquid was extracted and purified by using DNeasy Tissue Kit (QIAGEN GmbH, Germany) with modified method (Fan et al., 2009). The obtained DNA was quantified on a Nanodrop-1000 (NanoDrop Technologies Inc., USA).

PCR (Eppendorf, Germany) amplification of 16S rRNA genes was performed by using universal eubacterial primers 8F (Zhou et al., 1995) and 1392R (Lane et al., 1985). The fluorescently labeled forward primer 8F-cy5 was used to amplify 16S rRNA genes for terminal restriction fragment length polymorphism (T-RFLP). Erythromycin resistance genes, including esterase genes ereA and ereB, phosphotransferase gene mphA, efflux genes mefA/E and msrA/B, and methylase genes ermA, ermB, and ermC, were PCR (Eppendorf, Germany) amplified by using primers described previously (Sutcliffe et al., 1996). PCR products were subsequently visualized on a Molecular Imager Gel Doc XR System (Bio-Rad, USA).

2.5.

T-RFLP

The fluorescently labeled 16S rRNA genes were digested with 5 U of restriction endonucleases HhaI, MspI, or RsaI (NEB, USA) as described previously (He et al., 2003). The terminal restriction fragments (T-RFs) were determined on a CEQ 8000 automated sequencer (Beckman Coulter, USA) by using GenomeLab Fragment Analysis Kit Insert (608113-AH) and internal standard (DNA Size Standard Kit e 600, p/n 608095) (Beckman Coulter, USA). Individual T-RFs were normalized as a percentage of the total peak area.

2.6.

Clone library and sequencing

A 16S rRNA gene clone library was established by using TOPOTA cloning kit (Invitrogen, USA) according to the manufacturer’s recommendation. The 16S rRNA gene inserts were PCRamplified with TA primers (Zhou et al., 1997) and then digested with enzyme HhaI or MspI (NEB, USA). Plasmid DNA was purified with Qiagen Miniprep kit (Qiagen GmbH, Germany) and subsequently sequenced on an ABI 3100 Sequencer (Applied Biosystems, USA) by using primers M13F-20, M13R-24, 533F, 529R, 907F (http://www.genomics.msu.edu). Sequences were aligned and analyzed by using BioEdit assembly software (http://www.mbio.ncsu.edu/BioEdit/bioedit.html) and BLAST (http://www.ncbi.nlm.nih.gov/), respectively.

3.

Results

3.1. Effects of ERY and ERY-H2O on expansion of resistance genes Erythromycin resistance genes were screened on microbial consortia in the SBRs (Table 2 and Fig. 1). Esterase genes ereA and ereB, efflux gene mefA/E, and methylase gene ermA were detected in the microbes of MR (before 8-month pretreatment). After 8-month pretreatment, ereA, ereB, and ermA

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

Table 2 e e Resistance genes detected in MR, R1 (50 mg/L of ERY-H2O), R2 (100 mg/L of ERY) and R3 (control). Gene

Month Function

ereA Erythromycin esterase ereB mphA Macrolide20 phosphotransferase mefA/ Macrolide efflux E msrA/ Macrolide efflux B ermA Erythromycin ermB ribosomal methylase ermC

8 0

12

12

12

MR MR R1 (ERY- R2 R3 H2O) (ERY) (control) þa þ e

b e e

þ e e

þþc e e

e e e

þþ þ

þ

þ

þ

e

e

e

e

e

e þ e

e e e

e e e

e e e

e e e

a þ detected. b  not detected. c þþ relatively higher concentrations of PCR products detected.

genes disappeared in the MR and the latter two genes did not recover in any reactors regardless of the one year addition of ERY or ERY-H2O; mefA/E always appeared at a lower level in the MR and kept in all three SBRs (R1, R2, and R3) in the following one more year operation. Interestingly, though ereA gene was not detected in the MR after 8-month pretreatment, it bounced back in R1 (ERY-H2O) and R2 (ERY), but was not detected in R3 (control). No any other factors were different in the three SBRs except the presence/absence of ERY-H2O or ERY. Thus, both ERY-H2O (50 mg/L in R1) and ERY (100 mg/L in R2) exhibited significant effects on proliferation of the resistance gene ereA.

3.2.

Biodegradation of ERY

Microbial consortia containing esterase gene ereA may be capable of esterifying ERY. Therefore, microorganisms from R1 (ERY-H2O), R2 (ERY), and R3 (control) were tested on their capability to degrade ERY (10 mg/L) in batch bottles with a medium containing COD, NH4þeN, and PO43eP of 600, 30, and 6 mg/L (COD:N:P ¼ 100:5:1, as assimilation required), respectively. A remaining percentage of ERY (concentration of ERY in the tested bottles compared to that in negative controls with autoclaved inocula) was used to indicate biodegradation

3101

of ERY. During five days incubation, remaining percentage of ERY was constantly w100% in the batches with inocula from R1 (ERY-H2O) or R3 (control), indicating that ERY-H2O (50 mg/L in R1 running over one year) could not acclimate microbes to degrade ERY (Fig. 2). In contrast, ERY in batches with inocula from R2 (ERY) was completely removed within 3 days, indicating that ERY (100 mg/L in R2 running over one year) induced microbes to degrade ERY (Fig. 2). In addition, similar batches were setup to test biodegradation of ERY-H2O, but results showed that ERY-H2O was persistent regardless of the inoculum source (data not shown). To investigate the degradation products of ERY, LC-MS-MS was used to monitor the precursor ion of ERY with a mass-tocharge ratio m/z 734.5/158.2 amu. The LC-MS-MS chromatograms demonstrated that two main peaks were eluted at a retention time of 5 min and 10 min with the disappearance of ERY in the batches with inocula from R2 (ERY at 100 mg/L), suggesting that at least two products (product I and II) with the same precursor ion m/z ratio as ERY were produced (Fig. 3a). Furthermore, biodegradation products of ERY were fully scanned using LC-MS (m/z 100e1000 amu), the chromatogram of which exhibited six products (Fig. 3b). According to the mass spectra of these six products, four products with the same precursor ion m/z ratio (735 amu) as ERY appeared at a retention time of 4.6 (product II), 9.2, 9.6, and 10.3 min (product I); and the other two products with lower m/z ratios (718 amu and 720 amu) of precursor ions than ERY eluted at 9.9 min and 10.6 min, respectively (Fig. S2). The six products, possessing either the same or 15e17 amu lower m/z ratios of precursor ions (735, 720, and 718 amu) than ERY (m/z 735 amu), are good matches of the downstream products of esterified ERY as reported previously by isolates of Escherichia coli, Providencia stuartii, Staphylococcus aureus, and Pseudomonas sp. (Wright, 2005). Based on the degradation products and previous reported esterase mechanism of ERY (Wright, 2005), the possible pathway for ERY biodegradation was shown in Fig. 4. The formation of two main products I and II (m/z 735 amu) from ERY may follow four steps: 1) esterase enzyme cleaves the macrocycle ester via adding one H2O molecule, 2) non-enzymatic intramolecular hemiketal formation (one ERY eOH group transformed), 3) internal dehydration to form enol ether product I, or a second internal cyclization event via intramolecular condensation and dehydration to form product II (one ERY eOH group transformed), 4) Product I transferred to

Fig. 1 e Detection of esterase genes ereA and ereB in the microbes of mother reactor (MR), R1 (ERY-H2O), R2 (ERY), and R3 (control). Lane 1, Generuler 100 bp Plus ladder (Fermentas); lanes 2e7, 420 bp PCR products of ereA in the microbes of MR (month e 8 and 0), R1, R2 and R3 (month 12), and negative control (NC); lanes 8e13, 546 bp PCR products of ereB in the microbes of MR (month e 8 and 0), R1, R2 and R3 (month 12), and NC.

3102

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

120

R1 R3 R2

100

ERY [%]

80

600COD 30N,6P

60 40 20 0 0

1

2 3 Time [days]

4

5

Fig. 2 e Degradation of ERY in the batches with inocula from R1 (ERY-H2O), R2 (ERY), and R3 (control). A percentage of ERY is determined by the concentration of ERY in the tested bottles compared to that in the negative control bottles with autoclaved inocula. The values represent an average (n [ 3), and the standard deviations (less than 6%) were not shown. product II (one ERY eOH group transformed). The other two products with precursor ion m/z ratios (735 amu) may share the same formation mechanism with the products I and II at the first 2 steps, but differ in dehydration positions at the third step. The two products with precursor ion m/z ratios (720 and 718 amu) may be dehydrated twice at the third step (at most two ERY eOH groups transformed) and followed by saturation with several eH. Overall, biodegradation of ERY seemed to transform 1e3 eOH groups. As most of eOH groups of ERY are active groups for antibiotic activity (Schlunzen et al., 2001), the biodegradation products of ERY may lose or weaken their antibiotic effects as ERY-H2O does.

3.3. Effects of glucose, ammonium, and phosphate on biodegradation of ERY Since most of ERY esterase-producing microbes, such as E. coli and S. aureus (Wondrack et al., 1996), are known as heterotrophic bacteria, it is necessary to assess the influence of exogenous carbon source (e.g., glucose) on ERY biodegradation. The medium containing NH4þeN (30 mg/L) and PO43eP (6 mg/L) was prepared in two conditions with or without glucose (equal to COD of 600 mg/L) and received inocula from R2 (ERY). In the medium without glucose, 10 mg/L ERY did not show any obvious decrease compared with negative control. In the medium containing glucose, however, ERY decreased quickly from 10 mg/L to w1 mg/L within 3 days (Fig. 5a), suggesting that the exogenous carbon source (e.g., glucose) was necessary for degrading ERY. To evaluate effects of ammonium and phosphate on ERY biodegradation, the medium containing different initial concentrations of NH4þeN (30, 40, 50, and 60 mg/L) and PO43eP (6, 13, 20, and 26 mg/L) was tested on inocula from R2 (ERY). In batches with different phosphate concentrations (NH4þeN of 30 mg/L and COD of 600 mg/L), ERY degradation was similar for all batches (Fig. 5b), excluding the effects of

Fig. 3 e Biodegradation products of ERY. a e The LC-MS-MS chromatograms (734.5/158.2 amu) exhibit the degradation products of ERY in the batches of R2 (ERY) (shown in Fig. 2) after incubation for 0 day, 2 days and 3 days; b e The LCMS chromatograms (full-scan with m/z 100e1000 amu) exhibit the degradation products of ERY (shown in Fig. 2) after incubation of 2 days.

phosphate on the biodegradation of ERY. Ammonium, however, was shown to greatly inhibit the biodegradation of ERY (Fig. 5c). When inocula were fed with NH4þeN of 30 mg/L (theoretically just sufficient for assimilation of 600 mg/L COD), ERY of w9 mg/L (90%) was biodegraded; whereas with higher concentrations of NH4þeN, ERY of 7 mg/L (70%, NH4þeN of 40 mg/L), and 3 mg/L (30%, NH4þeN of 50 and 60 mg/L) was biodegraded, respectively.

3.4. Shift of microbial communities due to ERY biodegradation T-RFLP and clone library were employed to find out the shift of microbial communities due to ERY biodegradation.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

3103

Fig. 4 e Reaction and downstream products of ERY esterases.

3.4.1. Change of community structure due to ERY biodegradation The DNAs extracted from batch samples that ERY (10 mg/L) was completely biodegraded by inocula from R2 (ERY) were applied to T-RFLP analysis. The T-RFLP results showed that microbial community changed during ERY biodegradation process (Fig. 6). At the end of ERY degradation, T-RFLP distinguished three dominant terminal restriction fragments (T-RFs) with the sizes of 206 base pairs (bp), 492 bp and 121 bp digested by the enzymes HhaI, MspI, and RsaI, respectively.

Sequence results of the two patterns revealed that they were reversely complementary. Therefore, 36% of the clones represented one dominant species. The sequence results also showed that the dominant species possess 98% similarity of the 16S rRNA gene sequence of Zoogloea (DQ413151.1), a well-known producer of extracellular polymeric substances (EPS) that is vital in maintaining activated sludge floc. Consequently, bacteria in floc or biofilm increase antibiotic resistance 10 to 1000 folds comparing to their planktonic formation (Anderson and O’Toole, 2008). Also, the dominant species exhibit the same T-RFs as those dominant T-RFs in the T-RFLP results based on the analysis of sequence by BioEdit assembly software.

3.4.2. Identification of the predominant species by clone library To identify the species represented by the predominant T-RFs, a 16S rRNA clone library was established with the same genomic DNA as that in the T-RFLP. Restriction analysis of amplified ribosomal DNA (by MspI and HhaI) generated two dominant patterns (16 and 10 clones from 72 clones, respectively).

4.

Discussion

The presence of antibiotics and the continuous input of resistant bacteria and genes are considered as two main factors for the amplification of resistance (Kummerer, 2009b).

3104

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

12

a

10

ERY [mg/L]

8 6

NC 0COD,30N,6P 600COD,30N,6P

4 2 0 0

1

2

3 4 5 Time [days]

6

7

12

b

10 8

ERY [mg/L]

8

NC 30N,6P 30N,13P 30N,20P

6 4

30N,26P

2 0 0

1

2

3 4 5 Time [days]

6

7

12 NC

ERY [mg/L]

10 8

8

c

60N,6P 50N,6P

6 4 40N,6P 2

30N,6P

0 0

1

2

3 4 5 Time [days]

6

7

8

Fig. 5 e The effects of a e glucose, b e phosphate, and c e ammonium on the biodegradation of ERY. The values represent the means ± standard deviations (n [ 3). NC means negative control. Here in the absence of continuous resistant microorganisms and genes input via using synthetic wastewater, proliferation of esterase gene ereA as well as esterification of ERY to 6 products was observed with microbes exposed to ERY (100 mg/ L) for over one year, indicating that long-term exposure to ERY can induce resistance development in microbes. Therefore, the causal relationship of antibiotics at concentrations as low as found in the environment with expansion of resistance genes is strongly supported in this study by experimental data

rather than by hypothesis in previous studies (Kummerer, 2009b). This discovery is inconsistent with other findings that ERY (up to 100 mg/L) is too low to sustain antibiotic resistance and continuous introduction of resistance genes may be more important for resistance expansion in the sewage (Ohlsen et al., 2003, 1998). The difference may be due to an over-simplified microbial community (two kinds of bacteria, one donor and one recipient for resistance genes) being operated over short time (10 days) in those studies, which may underestimate the transfer frequency and sustain of resistance genes in the ecosystem. The ERY derivatives (e.g., ERY-H2O) can be very persistent (Kummerer, 2009a), which makes complete mineralization of ERY difficult (Alexy et al., 2004; Gartiser et al., 2007). The intermediates during antibiotics degradation was usually reported to possess negligible antibacterial activity due to that bacteria can directly destruct antibiotic functional groups or modify the macrolide antibiotics to gain resistance (Schlunzen et al., 2001; Wright, 2005). However, they may also induce microbial resistance to the parent drugs (Fan et al., 2009; Majer, 1981). Current study on ERY-H2O further proved the above resistance development principle, although the developed resistance gene ereA in this study cannot induce enzyme immediately to degrade ERY. Similar as ERY-H2O, other biodegradation products of ERY may also show no antibiotic activity due to the lost of several antibiotic functional eOH groups but may still induce microbial resistance to ERY (Schlunzen et al., 2001). Further study is needed to find out whether the degradation products of other antibiotics can still induce resistance to their parent drugs as ERY-H2O does to ERY (Fan et al., 2009; Majer, 1981). Antibiotics or their structure closely related derivatives at sub-inhibitory concentrations may act as signals to trigger specific response in micro-ecosystem to resistant antibiotics (Fajardo and Martinez, 2008). Erythromycin at sub-inhibitory concentrations has been reported to activate expression of specific gene encoding for polysaccharide (one of important components of EPS) intercellular adhesion in Staphylococcus (Rachid et al., 2000). Thus, ERY and Zoogloea may follow similar interactions as shown in the above study. Erythromycin may trigger the gram-negative Zoogloea to become dominant and to form protective biofilms, since the presence of intrinsic efflux resistance genes in gram-negative bacteria can make them more impermeable and less sensitive to ERY (Pechere, 2001). Bacteria containing ereA gene are speculated to be located at the outer edges of Zoogloea formed biofilm to eliminate antibiotics from harming the inner microbes. In turn, both microbes containing ereA gene and biofilm cells work together to provide sufficient antibiotic resistance in the microbial communities. Moreover, long-term treatment of bacteria with lower antibiotic concentrations has been discovered to result in higher resistant capability of biofilm populations of E. coli than planktonic bacteria (Harrison et al., 2005). Further study is needed to clarify gene-based biofilm resistance to antibiotics at low concentrations (e.g., in WWTPs). Notably, fullscale WWTPs may show more frequent resistance gene development and transfer than in this study, which could be caused by (1) wastewater propertiesecomplex wastewater versus synthetic wastewater; (2) diversity of microbial communitiesecomplex versus enriched microbial consortia; (3)

3105

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

456 472 474 475

166

97

118 119 120 121

96

117

533 534

percentage [%]

491 492

RsaI

121

30

124

10

475

20

103

percentage [%] 482

72 73

10

83 84 90 132 149 157 162 192 235 433

20

c'

40

59 67 70 79 90 94

MspI

30

58

501 498

367

281 294 295 299 300

205

percentage [%]

206

20

Size [bp]

493

481 432

83 81 82

50

b'

HhaI

0

10

Size [bp]

40

30

RsaI

0

50

a'

68 101 173 174 175

percentage [%]

10

c

20

Size [bp]

40

10

20

0

Size [bp]

50

MspI

192 197 302 366 423

0

30

b 70 71

percentage [%]

567570

367 322 368 370

10

206 208 214

20

30 HhaI

195 204

a

63 68 85 101

percentage [%]

30

0

0 Size [bp]

Size [bp]

Fig. 6 e T-RFLP results for the samples in the degradation batches of R2 (ERY). a, b and c e the sample on day 0 before ERY degradation; a0 , b0 and c0 e the sample on the day that ERY was completely degraded. Peaks less than 1% were not shown.

frequent resistant bacteria and genes input versus free of these two sources. Also, since ERY is sensitive to pH, it may be dehydrated into ERY-H2O in wastewater collection pipelines before entering WWTPs. Consequently, ERY does not have a chance to be esterified in WWTPs, but ERY-H2O becomes prevalent there as reported previously (Le-Minh et al., 2010). Higher concentrations of ammonium (NH4þeN > 40 mg/L) were found to inhibit ERY biodegradation for more than 30% (Fig. 5), which may be explained by the toxicity of free ammonia to most microorganisms. Considering NH4þeN of 60 mg/L in the synthetic wastewater was diluted to a final concentration of 30 mg/L in the influent of R2 (ERY of 100 mg/L, a dilution factor of 2) (Fan et al., 2009), this concentration of ammonia did not inhibit the ERY biodegradation. However, in the sewage treatment plants (STPs) receiving antibiotics discharged from both hospitals and households, ammonium concentrations fluctuate during the day (could be >40 mg/L) and dilution factors are process-dependent (usually <2), which is more likely to inhibit the biodegradation of ERY. Moreover, the even higher concentrations of ammonium in separate sewage system and in pharmaceutical wastewater will definitely make ERY biodegradation more difficult to occur. The untreated ERY, in turn, will inhibit ammonium oxidization (Alexy et al., 2004; Nimenya et al., 1999). Awareness is needed to optimize STPs to cope with this problem.

5.

Conclusions

In conclusion, both ERY-H2O (50 mg/L) and ERY (100 mg/L) can encourage development of ERY esterase gene ereA under conditions of free input of resistant bacteria and genes. Dehydrated erythromycin acclimated microbes cannot esterify ERY as ERY acclimated ones do, which may be due to the less proliferated ereA gene. This study suggests that the

presence of ERY and ERY-H2O at concentrations as low as found in the environment can enhance establishment of antibiotic resistance, and will provide important information to substantiate correlation between resistance proliferation and antibiotics at sub-inhibitory concentrations.

Acknowledgements This research was supported by the Public Utility Board (PUB) and American Water Works Association Research Foundation (AwwaRF) under project No. 4116.

Appendix. Supplementary data Supplementary data related to this article can be found online at doi:10.1016/j.watres.2011.03.025.

references

Anderson, G.G., O’Toole, G.A., 2008. Bacterial Biofilms. SpringerVerlag Berlin, Berlin, pp. 85e105. Abuin, S., Codony, R., Compano, R., Granados, M., Prat, M.D., 2006. Analysis of macrolide antibiotics in river water by solid-phase extraction and liquid chromatography-mass spectrometry. Journal of Chromatography A 1114 (1), 73e81. Alexy, R., Kumpel, T., Kummerer, K., 2004. Assessment of degradation of 18 antibiotics in the closed bottle test. Chemosphere 57 (6), 505e512. Amin, M.M., Zilles, J.L., Greiner, J., Charbonneau, S., Raskin, L., Morgenroth, E., 2006. Influence of the antibiotic erythromycin on anaerobic treatment of a pharmaceutical wastewater. Environmental Science & Technology 40 (12), 3971e3977.

3106

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 0 9 8 e3 1 0 6

Ding, C., He, J.Z., 2010. Effect of antibiotics in the environment on microbial populations. Applied Microbiology and Biotechnology 87 (3), 925e941. Fajardo, A., Martinez, J.L., 2008. Antibiotics as signals that trigger specific bacterial responses. Current Opinion in Microbiology 11 (2), 161e167. Fan, C.A., Lee, P.K.H., Ng, W.J., Alvarez-Cohen, L., Brodie, E.L., Andersen, G.L., He, J.Z., 2009. Influence of trace erythromycin and erythromycin-H2O on carbon and nutrients removal and on resistance selection in sequencing batch reactors (SBRs). Applied Microbiology and Biotechnology 85 (1), 185e195. Gartiser, S., Urich, E., Alexy, R., Kummerer, K., 2007. Ultimate biodegradation and elimination of antibiotics in inherent tests. Chemosphere 67 (3), 604e613. Gros, M., Petrovic, M., Barcelo, D., 2007. Wastewater treatment plants as a pathway for aquatic contamination by pharmaceuticals in the Ebro river basin (Northeast Spain). Environmental Toxicology and Chemistry 26 (8), 1553e1562. Harrison, J.J., Ceri, H., Roper, N.J., Badry, E.A., Sproule, K.M., Turner, R.J., 2005. Persister cells mediate tolerance to metal oxyanions in Escherichia coli. Microbiology-SGM 151, 3181e3195. He, J.Z., Ritalahti, K.M., Yang, K.L., Koenigsberg, S.S., Loffler, F.E., 2003. Detoxification of vinyl chloride to ethene coupled to growth of an anaerobic bacterium. Nature 424 (6944), 62e65. Kummerer, K., 2009a. Antibiotics in the aquatic environment e a review e part I. Chemosphere 75 (4), 417e434. Kummerer, K., 2009b. Antibiotics in the aquatic environment e a review e part II. Chemosphere 75 (4), 435e441. Lane, D.J., Pace, B., Olsen, G.J., Stahl, D.A., Sogin, M.L., Pace, N.R., 1985. Rapid-determination of 16S ribosomal-RNA sequences for phylogenetic analyses. Proceedings of the National Academy of Sciences of the United States of America 82 (20), 6955e6959. Le-Minh, N., Khan, S.J., Drewes, J.E., Stuetz, R.M., 2010. Fate of antibiotics during municipal water recycling treatment processes. Water Research 44 (15), 4295e4323. Majer, J., 1981. In vitro introduction of resistance to erythromycin by its metabolite. Antimicrobial Agents and Chemotherapy 19 (4), 628e633. McArdell, C.S., Molnar, E., Suter, M.J.F., Giger, W., 2003. Occurrence and fate of macrolide antibiotics in wastewater treatment plants and in the Glatt Valley Watershed, Switzerland. Environmental Science & Technology 37 (24), 5479e5486. Murray, B.E., 1997. Advances in Internal Medicine, vol. 42. MosbyYear Book Inc., St Louis, pp. 339e367. Nimenya, H., Delaunois, A., La Duong, D., Bloden, S., Defour, J., Nicks, B., Ansay, M., 1999. Short-term toxicity of various pharmacological agents on the in vitro nitrification process in

a simple closed aquatic system. ATLA Alternatives to Laboratory Animals 27 (1), 121e135. Ohlsen, K., Ziebuhr, W., Koller, K.P., Hell, W., Wichelhaus, T.A., Hacker, J., 1998. Effects of subinhibitory concentrations of antibiotics on alpha-toxin (hla) gene expression of methicillinsensitive and methichillin-resistant Staphylococcus aureus isolates. Antimicrobial Agents and Chemotherapy 42 (11), 2817e2823. Ohlsen, K., Ternes, T., Werner, G., Wallner, U., Loffler, D., Ziebuhr, W., Witte, W., Hacker, J., 2003. Impact of antibiotics on conjugational resistance gene transfer in Staphylococcus aureus in sewage. Environmental Microbiology 5 (8), 711e716. Pechere, J.C., 2001. Macrolide resistance mechanisms in Grampositive cocci. International Journal of Antimicrobial Agents 18, S25eS28. Rachid, S., Ohlsen, K., Witte, W., Hacker, J., Ziebuhr, W., 2000. Effect of subinhibitory antibiotic concentrations on polysaccharide intercellular adhesin expression in biofilmforming Staphylococcus epidermidis. Antimicrobial Agents and Chemotherapy 44 (12), 3357e3363. Schlunzen, F., Zarivach, R., Harms, J., Bashan, A., Tocilj, A., Albrecht, R., Yonath, A., Franceschi, F., 2001. Structural basis for the interaction of antibiotics with the peptidyl transferase centre in eubacteria. Nature 413 (6858), 814e821. Sutcliffe, J., Grebe, T., TaitKamradt, A., Wondrack, L., 1996. Detection of erythromycin-resistant determinants by PCR. Antimicrobial Agents and Chemotherapy 40 (11), 2562e2566. Szczepanowski, R., Linke, B., Krahn, I., Gartemann, K.H., Gutzkow, T., Eichler, W., Puhler, A., Schluter, A., 2009. Detection of 140 clinically relevant antibiotic-resistance genes in the plasmid metagenome of wastewater treatment plant bacteria showing reduced susceptibility to selected antibiotics. Microbiology-SGM 155, 2306e2319. Wondrack, L., Massa, M., Yang, B.V., Sutcliffe, J., 1996. Clinical strain of Staphylococcus aureus inactivates and causes efflux of macrolides. Antimicrobial Agents and Chemotherapy 40 (4), 992e998. Wright, G.D., 2005. Bacterial resistance to antibiotics: enzymatic degradation and modification. Advanced Drug Delivery Reviews 57 (10), 1451e1470. Zhang, X.X., Zhang, T., Fang, H., 2009. Antibiotic resistance genes in water environment. Applied Microbiology and Biotechnology 82 (3), 397e414. Zhou, J.Z., Fries, M.R., Cheesanford, J.C., Tiedje, J.M., 1995. Phylogenetic analysis of a new group of denitrifiers capable of anaerobic growth on toluene and description of Azoarcusetolulyticus sp-nov. International Journal of Systematic Bacteriology 45 (3), 500e506. Zhou, J.Z., Davey, M.E., Figueras, J.B., Rivkina, E., Gilichinsky, D., Tiedje, J.M., 1997. Phylogenetic diversity of a bacterial community determined from Siberian tundra soil DNA. Microbiology-UK 143, 3913e3919.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Application of robust statistical methods to background tracer data characterized by outliers and left-censored data Malcolm S. Field* U.S. Environmental Protection Agency, National Center for Environmental Assessment (8623P), 1200 Pennsylvania, Ave., N.W. Washington, D.C. 20460, USA

article info

abstract

Article history:

Accurate analysis of tracer-breakthrough curves is dependent on the removal of measured

Received 10 November 2010

background concentrations from the measured tracer recovery data. Background concen-

Received in revised form

trations are commonly converted to a single mean background concentration that is

26 January 2011

subtracted from tracer recovery data. To obtain an improved estimate for the mean

Accepted 13 March 2011

background concentration, a statically-robust procedure addressing left-censored data and

Available online 21 March 2011

possible outliers in background concentration data is presented. A maximum likelihood estimate and other robust methods coupled with outlier removal are applied. Application

Keywords:

of statically-robust procedures to background concentrations results not only in better

Tracer-breakthrough curves

estimates for mean background concentration but also results in more accurate quanti-

Outliers

tative analyses of tracer-breakthrough curves when the mean background concentration is

Left-censored data

subtracted. Published by Elsevier Ltd.

Mean Maximum likelihood estimation Median

1.

Introduction

Hydrologic tracer studies are influenced by background tracer concentrations that must be properly accounted for in every tracer test. The occurrence of background concentrations can be significant and results from a variety of naturally occurring and man-made substances. Even low background concentrations of the tracer can have profound effects on the detection and analysis of the collected samples and any subsequent numerical analyses of the tracer-test results. Unfortunately, although it is generally well recognized that background concentration measurements are important, basic tracer-test methods textbooks (e.g., Leibundgut et al., 2009; Ka¨ß, 1998; Gaspar, 1987a,b; Davis et al., 1985) only briefly mention background concentrations. Other technical reports also mention

* Tel.: þ1 703 347 8601; fax: þ1 703 347 8692. E-mail address: [email protected]. 0043-1354/$ e see front matter Published by Elsevier Ltd. doi:10.1016/j.watres.2011.03.018

background-concentration measurements (e.g., Wilson et al., 1986; Mull et al., 1988; Kilpatrick and Wilson, 1989; Field, 2002, 2003; Schudel et al., 2002; Benischke et al., 2007), but do not emphasize the need to collect more than a single background measurement. Collecting only a single background measurement suggests no temporal variability in the flow system. Background tracer concentrations are caused by a variety of factors. For fluorescent dyes, background signals may develop from natural substances such as dissolved organic carbon (DOM). Fluorescent properties associated with DOM develop from highly reactive, oxygen-rich functional groups that are accompanied by phenols, amines, alcohols, inorganic iron and aluminum species, and other organic acids (Brown, 2009, p. 15). Anthropogenic sources of fluorescence typically

3108

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

detected in background samples include optical brighteners used in laundry detergents, antifreeze colored with uranine dye, mineral oil, hydraulic fluids, polycyclic aromatic hydrocarbons, landfill leachate, some agricultural chemicals and pharmaceuticals, and wood preservatives (Brown, 2009, pp. 20e21), Other commonly-used tracing agents, such as salts, can be useful by increasing the electrical conductivity. However, most cations (e.g., Liþ, Naþ, Kþ, Mg2þ, Ca2þ, and Sr2þ) are prone to ion exchange, but this is not necessarily so with anions (Benischke et al., 2007). Anions, such as Br and Cl, are significantly affected by such factors as variances in atmospheric concentrations, temporal, seasonal, and high and low precipitation, dry deposition, run-on and runoff processes, soil characteristics, evapotranspiration, and land-use changes (Guan et al., 2010; Hrachowitz et al., 2010). In addition, widespread application of NaCl by such activities as road salting (Leibundgut et al., 2009, p. 105) cause elevated background concentrations of Cl. Because hydrologic tracer tests are adversely affected by high background concentrations, accurate and precise background concentration measurements are of critical importance. Hydrologic tracer tests consist of the measurement of tracer concentrations in water samples from which background concentrations are subtracted. However, separation of representative background concentrations from an injected tracer substance is difficult if not impossible after injection of the tracer substance. In general, after a tracer substance has been injected, separation of the sample signal from the background signal cannot be accomplished (Brandt, 1999, p. 169). According to Brandt the expected number of signals in an experiment is a Poisson distribution in which l ¼ lS þ lB where l represents signal measurements. To obtain information about the sought after parameter for the number of signal events lS the parameter for the background events lB must be known. This is problematic for accurate tracer-breakthrough curve (BTC) analysis because of the near impossibility of separating the two signals. Typically, for hydrologic tracer tests, several background measurements are taken prior to tracer injection of any tracer substance. After tracer injection, background concentration samples can no longer be collected because it is impossible to know if the measurement represents a background signal, or an injection signal. These multiple background measurements may also be problematic because of temporal differences between the time that the background measurements were taken and the time of tracer injection. In addition, there is no consensus method by which the measured background concentrations may be utilized when subtracting from measured tracer concentrations. Most commonly, the arithmetic mean of the background concentrations is calculated and this parameter is then subtracted from the measured tracer recovery data. Unfortunately, the sample mean is a very non-robust statistic and as such is adversely affected by outliers (Daszykowski et al., 2007). Whether a result of instrumental errors, transcription errors, or some other cause, outliers are a serious problem. Incorporating one or more outliers in the calculation of a mean background concentration (MBC) leads to an excessively large estimate for the MBC. Subtracting an overly large MBC from measured tracer recovery data may

actually result in some negative concentration values, an obvious impossibility. Resulting negative concentrations are usually converted to zero concentrations as a matter of convenience, not for any valid scientific or statistical reason. In the statistical sense, the term, outliers, is an ill-defined concept without clear boundaries, but is useful when regarded as a continuous transition to ordinary observations (Hampel, 2001). Outliers may be a result of gross errors in analysis (e.g., transcription errors, laboratory-analysis errors, or computer-controlled-equipment problems) or they may be a result of anomalous field conditions such as an unusual spike in background concentrations. A common reaction to outliers according to (Hampel, 2001) is to subjectively or objectively reject outliers when, in principle, outliers should be given separate treatment. An additional problem with the MBC occurs when some measured concentrations are greater than or equal to zero, but less than the method detection limit (DL). The DL is the minimum concentration of a substance that can be measured and reported with 99% confidence that the analyte concentration is greater than zero when determined from analysis of a sample in a given matrix containing the analyte (USEPA, 2009). Because the DL is a statistical concept it is quite possible that a substance can be detected at concentrations below the DL (Ripp, 1996). According to Ripp, censoring data below some unspecified or non-statistical reporting limit severely biases data sets and restricts its usefulness. This censoring can lead to erroneous decisions when calculating sample means or mass balances. A number reported as less than with no corresponding information is very difficult to interpret, and often must be discarded even though it may have some validity. The lower the DL, the more likely the analyte of interest will be detected in a sample. Measured concentration data that are below the detection limit (BDL) are commonly termed left-censored data (Antweilier and Taylor, 2008) or Type I censored data (Kroll and Stedinger, 1996), and have been a source of controversy for many years. Most recently, environmental regulators have struggled with development of an appropriate methodology for incorporating BDL environmental samples into site pollution assessments. For example, commonly-used approaches entail deleting the censored data (Helsel, 2005, p. 11) or multiplying the DL by some constant falling within the interval [0,1] the result of which is then substituted for all BDL data. Typically, the chosen constant takes a value of zero, 0.5, or 1.0 (El-Shaarawi and Esterby, 1992; Singh and Nocerino, 2001). Other substitution methods replace left-censored data with the DL multiplied by 0.75 as has been done with pffiffiffi geochemical data or by DL= 2 as is sometimes done for airquality data and industrial-hygiene chemistry (Hewett and Ganser, 2007; Helsel, 2005, p. 56). However, there is no scientific basis to support such a replacement methodology. Such data fabrication reportedly can produce seriously biased estimates (Gleit, 1985; Gilliom and Helsel, 1986; Haas and Scheff, 1990; Helsel, 2006), although a reportedly accurate substitution method was recently developed (Ganser and Hewett, 2010). There will always be questions regarding the reliability of background concentration measurements and analysis methods, but appropriate statistical methods have the advantage

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

of repeatability. Subjective methods for dealing with background measurements, based on experience and/or judgment may arguably be a justifiable approach in some circumstances by some experienced individuals, but such an approach may not be as defensible as a valid statistical approach. The purpose of this paper is to investigate and develop a statistically-valid approach for assessing the MBC subjected to outliers and BDL data as it affects the measured BTC.

2.

Robust statistical methods

For every hydrologic tracer test, an adequate number of background samples are needed to properly characterize background variability. Obtaining a reasonably large background sample set makes various forms of data plotting possible, which allows trends in the data to be observed, and may help in determining appropriate statistical routines to employ. The sample mean xb and standard deviation sb for background concentrations is calculated from, respectively, xb ¼

n 1X xi ; n i¼1

(1)

and sb ¼

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n 1 X ðxi  xb Þ2 ; n  1 i¼1

(2)

where n is the sample size and xi is a sample measurement. Eqs. (1) and (2) consider all measured values equally so undue influence may be imparted by one or more extreme low or high values, especially for small sample sizes, unless varying weights are assigned to the measured values. Unfortunately, it is very difficult to assign adequate weights to the measurements if specific details about each measurement are lacking. When weights cannot be applied to the data, alternative robust statistical methods need to be employed. A measure of the robustness of a statistical estimator against outliers is the breakdown value, which may be defined as an indicator of the smallest fraction of contaminants (errors) in a sample that causes the estimator to breakdown (i.e., to take on values that are arbitrarily bad or meaningless) (Hubert and Debruyne, 2009). The closer the breakdown value to 100%, the better. Another measure of robustness, the influence function, measures the effect of a small number of outliers (Rousseeuw et al., 2006). The influence function is defined as the influence of an infinitesimal proportion of bad observations on the value of the estimate (Cowell and Victoria-Feser, 1993). It is essentially the first derivative of an estimator (Hampel, 1974) that measures the effect or influence of a single observed value on an estimator of a particular parameter. The breakdown value for the sample mean and standard deviation is 1/n, or 0% for large n. The influence function for the mean is unbounded and thus reflects the nonrobustness of the sample mean (Rousseeuw et al., 2006). For these two reasons alone, the sample mean should be accepted as representative of background tracer concentrations only after careful consideration and assessment.

2.1.

3109

Outlier determination

Outliers in a set of data are observations that are far from the bulk of the data (Olive, 2008, p. 4). Because data outliers can be very problematic in data analysis it is critical that they be accurately identified and properly addressed. Background data plotting can be advantageous for identifying outliers. Constructing a simple scatter plot, histogram, dot plot, boxplot, quantile plot, or stem and leaf plot may allow detection of possible outliers (Chambers et al., 1983, pp. 11e29). For example, a typical boxplot will include whiskers representing the data points at the tenth and ninetieth percentiles of the data. Any data plotted beyond these two whiskers may be data outliers. Unfortunately, robust statistical methods can only detect certain configurations of outliers and the ability to detect outliers rapidly decreases as the sample size and the number of predictors increase (Olive, 2008, p. 9). Because of the difficulty associated with outlier determination, improvements in detection methods continue to be investigated (e.g., Hubert and Van der Veeken, 2008). If any outliers are believed to exist based on one or more of the data plotting methods, then appropriate statistical methods for dealing with the outliers need to be considered. It is possible to remove outliers based on personal experience and judgment, but such an action often lacks statistical validity. Alternative actions consider outlier accommodation and removal by statistical determination.

2.1.1.

Outlier accommodation

Accommodating outliers generally is not a common practice, but some studies suggest that this may be a more reliable method for addressing outliers than is removal (AMC, 1989). The approach recommended by the (AMC, 1989) consists of a set of iterative calculations for estimating the population standard deviation sb alongside the population mean mb. At each iteration, pseudo-values x~i are formed and their mean xb and variance s2b computed. This allows for basic calculations that are repeated until the values stabilize.

2.1.2.

Outlier removal

There are several methods for statistically removing outliers when calculating a mean and standard deviation. The easiest but least acceptable method is to just arbitrarily remove any apparent outliers from a data set. A more acceptable method is to employ some statistical procedure to remove outliers after a scientific determination has been taken regarding the appropriateness of all of the data. Statistical determination and removal of outliers can be accomplished using any one of several robust methods. One powerful method developed at NASA (Swaroop and Winter, 1971) identifies outliers by computing, for a given significance level, a critical value D* for the data set and a positive real number Di for each observation vector, xi where

Di ¼

ðxi  xÞT ðxi  xÞ; S

(3)

D ¼

pðn  1Þ2 Fa;p;np1 ; nðn  p  1Þ þ npFa;p;np1

(4)

3110

S¼

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

n 1 X ðxi  xÞðxi  xÞT ; n1 1

(5)

Fa; p,np1 is the significance level (either 5 percent or 1 percent) of the F-distribution with p and (np1) degrees of freedom, p is the number of variables, and T is the transpose of the matrix. If Di > D*, observation xi is identified as an outlier and removed from further analysis. Another common approach is to calculate the median MED and median absolute deviation MAD from, respectively (Olive, 2008, p. 27)



MED ¼

x½ðnþ1Þ=2 xðn=2Þ þ x½n=ð2þ1Þ 2

if n is odd if n is even

;

  MAD ¼ MED xj  MED ðxi Þ; j¼1;.;n

i¼1;.;n

(6)

(7)

and a robust standard deviation estimate MADE from (Burke, 2001) MADE ¼ 1:4826 MAD:

(8)

Because MED measures the exact middle of a set of data it is insensitive to outliers. Replacing a single value by some arbitrary value does not result in much of a change in the MED. Significantly, the breakdown value of the median is 50%, which means that it can resist up to 50% outliers and its influence function is bounded. However, the median is a less efficient estimator than is the mean for the normal model (Rousseeuw et al., 2006) because the MED is adversely affected by heavytailed distributions close to the normal (AMC, 1989), which renders the MED and its associated estimators problematic. A different approach when 15e50% of the data are nondetects is to calculate a trimmed mean xt or winsorized mean xw of the data as suggested by USEPA (2000, pp. 4-42e4-44) and Navy (1999, pp. 11e15). The trimmed mean xt and associated standard deviation st are given by 1 X xi ; n  2j i¼jþ1 nj

xt ¼

(9)

and vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u nj X u 1 st ¼ t ðxi  xt Þ2 ; n  2j  1 i¼jþ1

(10)

and the winsorized mean xw and associated standard deviation sw are given by (Irwin, 2005) 3 2 nj1 X 14 xw ¼ xi þ ðj þ 1ÞxðnjÞ 5; ðj þ 1Þxðjþ1Þ þ n i¼jþ2

(11)

and 2

X 2 nj1  1 ðxi  xw Þ2 ðj þ 1Þ xðjþ1Þ  xw þ sw ¼4 n1 i¼jþ2 30:5  2 þ ðj þ 1Þ xðnjÞ  xw 5 ;

ð12Þ

where j represents the number of trimmed or winsorized observations.

The trimmed mean approach removes the least and greatest values and the descriptive statistics are calculated from what is left while the winsorized mean approach replaces the least and greatest values with each’s nearest neighbors. Both approaches are robust and generally result in better approximations for the data mean and standard deviation, but the trimmed mean is also insensitive to small numbers of gross errors and is not adversely affected by heavy-tailed distributions close to the normal (AMC, 1989).

2.2.

Left-censored data

Whether any outliers exist or not, most background concentration data sets contain a significant number of values that equal zero (a data floor) and a significant number of values that are greater than zero but less than the DL (i.e., leftcensored data). Such concentrations result in positively skewed data. Measured zero concentrations are still real measurements and need to be included in any statistical calculations, but left-censored data require evaluation using appropriate statistical techniques. A comprehensive study on censored data analysis methods (EFSA, 2010) reported that there are three appropriate methods for handling left-censored data: parametric maximum likelihood estimation (MLE) models, the log-probit regression method, and the non-parametric KaplaneMeier method. In this study, only the MLE was evaluated for background concentration data.

2.2.1.

Maximum likelihood estimation

The MLE is reported to be the gold standard for evaluating leftcensored data (Hewett and Ganser, 2007; Ganser and Hewett, 2010) and has been shown to be perhaps the best method to choose (Kroll and Stedinger, 1996), despite some suggestions that it may not be quite as good as generally accepted (Thompson and Nelson, 2003). For example, although the MLE is considered the best of the methods usually applied to censored data, Helsel (2005, p. 13) reported that the MLE is only valid for data sets with 50 observations and is sensitive to outliers. However, the MLE is still preferred over other non-parametric methods where either the percent of censored data is small or the shape of the distribution can be well characterized. Ganser and Hewett (2010) recently showed that the MLE can be applied to sample sizes as small as n ¼ 3 provided there are at least two uncensored values, but they also found that for data sets of n  10, the estimates for the sample geometric mean and geometric standard deviation can be severely biased, particularly when the actual percent censored data is large. For the MLE to work, the data are assumed to follow some statistical distribution and the parameters of the distribution are estimated to compatibly work with the percentage of BDL data to best fit the distribution of the observed values above the DL. The estimated parameters are the ones that maximize the likelihood function (EFSA, 2010). The MLE is commonly applied to the log-normal for concentration values that, when log-transformed approximates a normal distribution, the Weibull distribution, and the gamma distribution. Analyses described by the EFSA (2010) suggest that for left-censored data, a gamma distribution may inflict the least bias when the MLE method is used.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

3111

The sample geometric mean xg and geometric standard deviation sg are the values that maximize the likelihood function (LF) obtained from (Hewett and Ganser, 2007) LF ¼

n Y

k      Y  PDF ln xi ln xg ;ln sg  CDF ln xj ln xg ;ln sg ;

i¼kþ1

j¼1

(13)

where k is the number of censored data, PDF is the probability density function, CDF is the cumulative distribution function, and the xg and sg are calculated from, respectively

xg ¼

n Y xi i¼1

!1=n ¼ exp

! n 1X ln xi ; n i¼1

(14)

and sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi! 2 Pn  i¼1 ln xi  ln xg : sg ¼ exp n1

3.

(15)

Fig. 2 e Data plot of combined pre-injection and postinjection concentration data. Synthetic pre-injection background concentration data include two apparent outliers. Note that zero time represents time of tracer injection.

Quantitative example

To demonstrate the effect of outliers and left-censored data on descriptive statistics of background concentration data, a synthetic set of background concentration data values (Fig. 1) were applied to a selected measured BTC (Fig. 2) that was slightly modified for this analysis. The synthetic background data consists of two apparent outliers, several instances of zero concentration, and eight left-censored concentration values based on a DL ¼ 10 ng L1. Calculating the MBC from the synthetic data set should equal 0.01 mg L1 (representing the background events signal lB) for subtraction from the measured BTC (representing the signal measurements l) if the resulting BTC (representing signal events lS) is to be obtained.

The BTC depicted in Fig. 2 was developed on top of a constant background concentration equal to 0.01 mg L1 (i.e., xb ¼ 0:01 mg L1 and sb ¼ 0.0 mg L1). Subtracting 0.01 mg L1 from each measured tracer concentration results in the actual recovered tracer data. Evaluating the BTC shown in Fig. 2 using Qtracer2 (Field, 2002, pp. 25e30) resulted in a w96% tracermass recovery. A perfect 100% tracer-mass recovery almost never occurs because of errors in discharge measurements, tracer sorption, and tracer decay (Field, 2002, pp. 25e30). The combined plot of the pre-injection (synthetic background data) and post-injection data shown in Fig. 2 suggests the existence of the two apparent background data outliers. However, if the peak recovery concentration was significantly greater than that shown in Fig. 2 (e.g., 2 the peak), the existence of the outliers would be less apparent. The basic statistical distribution for the synthetic background data is depicted in Fig. 3, which shows just how extreme the two outliers in the synthetic data set really are. No lower whisker was calculated because a data floor representing zero concentration measurements occur in the synthetic background data. The dashed line in Fig. 3 is the sample mean, xb ¼ 0:0192, which denotes a bias toward the stronger concentrations even though nearly half of the data equals zero concentration. The interquartile range (IQR ¼ 0.0244), however, clearly shows that the median value (MED ¼ 0.0042) tends to the lower end of the boxplot because of the influence of the number of data values equal to zero concentration.

3.1. Fig. 1 e Data plot of pre-injection concentration data. Preinjection background concentration data include two apparent outliers, several instances of zero concentration, and eight left-censored concentration values (concentrations greater than zero, but less than the method detection limit).

Descriptive statistics estimates

Estimates for xb and sb were initially developed for the full data set along with the MED, MAD, and MADE. The trimmed and winsorized descriptive statistics were also calculated. All calculations were then repeated after outlier removal according to the method developed by Swaroop and Winter (1971) at the 1 percent significance level. The results for the

3112

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

3.2.

Fig. 3 e Boxplot of synthetic pre-injection background concentration data. Note the two apparent outliers. Dashed line represents the calculated mean background concentration.

descriptive statistics estimates are listed with outliers included in Table 1 and with outliers deleted in Table 2. A performance index for the mean estimates may be determined using the root mean square error rMSE (modified from Helsel and Hirsch, 2002, p. 358). sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðxb  mb Þ2 =mb ; rMSE ¼ n

(16)

where xb represents the background estimate for all methods described and mb represents the true background concentration. The nearer rMSE to zero, the better the performance index. Calculation of the rMSE for the various background estimates are also shown in Tables 1 and 2. However, a perhaps more useful performance index for the background estimates is subtraction of each estimate from the measured tracer BTC. Results of subtracting the background estimates from the measured BTC are shown in Table 3, the effects of which are listed in Table 4 and plotted in Fig. 4.

Assessment of the estimation methods

The significance of using estimates for the background mean that deviate from the expected 0.0100 mg L1 is evident when the rMSE for each background estimate listed in Tables 1 and 2 are examined. On inspection of both Tables 1 and 2, it is initially apparent that removing outliers from the background data (Table 2) results in rMSE calculations that are closer to zero than when outliers are included (Table 1). Removal of outliers appears to be appropriate in this instance (the MLE estimate of 0.0100 mg L1 results in a rMSE ¼ 0.01), but such an action is still dependent on a careful assessment of the validity of the data outliers. Background estimates ranged from a low of 0.0 mg L1 for a median calculation for instances in which BDL data are deleted (or substitution of the DL multiplied by zero) for either instance of outliers inclusion or exclusion to a high of 0.1176 mg L1 for an MLE with outliers included (Tables 1 and 2). This suggests that the median is strongly affected by a large number of background measurements equal to zero concentration while the MLE is strongly affected by outliers in the data. Methods resulting in a rMSE ¼ 0.01 (i.e, background estimate ¼ 0.0100 mg L1) occurs when the MLE is applied to the data after deleting the outliers, but also for instances when the DL  1.0 is substituted for BDL data for the median estimate regardless of whether outliers are included or excluded. Application of the MLE to data that includes outliers results in a background estimate that represents more than a five-fold increase over the next largest background estimate (0.0217 m g L1), which suggests that blind use of the MLE could result in extremely serious errors in background estimates. However, data fabrication methods, such as substituting various fractions of the DL for BDL data for MBC estimation, is unsupported by any scientific or statistical basis and cannot be assured to produce acceptable results when applied to other data sets. If outliers are accommodated the estimated mean ranges from a low of 0.0071 mg L1 with a standard deviation of 0.0062 mg L1 (method A15 in AMC (1989)) to a high of 0.0119 mg L1 with a standard deviation of 0.0164 mg L1 (method H15 in AMC (1989)). Although neither estimate for the

Table 1 e Descriptive statistics for synthetic background concentration data with outliers included.a Analysis Method Censoring Method

Data Medianb

Arithmetic Mean

Trimmed Mean

Winsorized Mean

xb

sb

rMSE

xt

st

rMSE

xw

sw

rMSE

MED MADe rMSE

0.0192 0.0217 0.0183 0.0191 0.0195 0.0199 0.0194

0.0374 0.0403 0.0377 0.0374 0.0372 0.0371 0.0373

0.0130 0.0181 0.0117 0.0128 0.0134 0.0140 0.0133

0.0110 0.0130 0.0099 0.0109 0.0114 0.0119 0.0113

0.0144 0.0167 0.0149 0.0144 0.0142 0.0141 0.0142

0.0014 0.0047 0.0002 0.0012 0.0019 0.0026 0.0018

0.0135 0.0151 0.0126 0.0134 0.0138 0.0142 0.0137

0.0130 0.0151 0.0136 0.0130 0.0129 0.0128 0.0129

0.0049 0.0079 0.0036 0.0048 0.0053 0.0059 0.0052

0.0042 0.0000 0.0000 0.0050 0.0075 0.0100 0.0071

0.0062 0.0000 0.0000 0.0074 0.0111 0.0148 0.0105

MLE xb

sb

rMSE

0.0083 0.1176 0.2516 0.1522 0.0154 e e e 0.0141 e e e 0.0071 e e e 0.0035 e e e 0.0000 e e e 0.0041 e e e

a All background value estimates and standard deviation estimates have units of mg L1. b Although the median cannot be construed to be the same as the mean, it is appropriate to compare the sample median MED with the true background mean mb here because the sample median is being considered for subtraction from the measured breakthrough curve.

3113

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

Table 2 e Descriptive statistics for synthetic background concentration data with outliers deleted.a Analysis Method Censoring Method

Data Medianb

Arithmetic Mean

Trimmed Mean

Winsorized Mean

xb

sb

rMSE

xt

st

rMSE

xw

sw

rMSE

MED MADe rMSE

0.0124 0.0138 0.0115 0.0123 0.0127 0.0132 0.0127

0.0171 0.0184 0.0176 0.0171 0.0170 0.0169 0.0170

0.0035 0.0060 0.0021 0.0033 0.0039 0.0045 0.0038

0.0098 0.0109 0.0087 0.0097 0.0102 0.0107 0.0101

0.0133 0.0146 0.0138 0.0133 0.0131 0.0131 0.0131

0.0002 0.0013 0.0019 0.0005 0.0003 0.0001 0.0002

0.0119 0.0132 0.0110 0.0118 0.0122 0.0126 0.0122

0.0123 0.0132 0.0127 0.0122 0.0121 0.0120 0.0121

0.0028 0.0050 0.0014 0.0026 0.0032 0.0038 0.0031

0.0030 0.0000 0.0000 0.0050 0.0075 0.0100 0.0071

0.0045 0.0000 0.0000 0.0074 0.0111 0.0148 0.0105

MLE xb

sb

rMSE

0.0101 0.0100 0.0129 0.0000 0.0158 e e e 0.0144 e e e 0.0072 e e e 0.0036 e e e 0.0000 e e e 0.0042 e e e

a All background value estimates and standard deviation estimates have units of mg Le1. b Although the median cannot be construed to be the same as the mean, it is appropriate to compare the sample median MED with the true background mean mb here because the sample median is being considered for subtraction from the measured breakthrough curve.

Table 3 e Number and location of data points less than zero after subtracting background concentration value. Outliers Included

Outliers Excluded Data Points Affected

Data Points Affected

Background Data Pts. Less Early Time Late Time Background Data Pts. Less Early Time Late Time Analysis Method Value, mg L1 Than Zero Data Pts. Data Pts. Value, mg L1 Than Zero Data Pts. Data Pts. Data Mean

0.0192 0.0217 0.0183 0.0191 0.0195 0.0199 0.0194

15 15 14 15 15 15 15

Trimmed Mean

0.0110 0.0130 0.0099 0.0109 0.0114 0.0119 0.0113

4 7 0 3 5 7 5

Winsorized Mean

0.0135 0.0151 0.0126 0.0134 0.0138 0.0142 0.0137

Data Median

1e8 1e8 1e5, 7e8 1e8 1e8 1e8 1e8

62e68 62e68 62e68 62e68 62e68 62e68 62e68

0.0124 0.0138 0.0115 0.0123 0.0127 0.0132 0.0127

7 8 5 7 7 8 7

1e3, 7 1e3, 7 1e3 1e3, 7 1e3, 7 1e3, 7 1e3, 7

1e3 1e3, 7 e 1e2 1e3 1e3, 7 1e3

67 63, 67e68 e 67 67e68 63, 67e68 67e68

0.0098 0.0109 0.0087 0.0097 0.0102 0.0107 0.0101

0 3 0 0 2 2 1

e 1e2 e e 1e2 1e2 1

8 10 7 8 8 8 8

1e3, 1e3, 1e3, 1e3, 1e3, 1e3, 1e3,

63, 65, 67e68 63, 65e68 63, 67e68 63, 65, 67e68 63, 65, 67e68 63, 65, 67e68 63, 65, 67e68

0.0119 0.0132 0.0110 0.0118 0.0122 0.0126 0.0122

7 8 4 6 7 7 7

1e3, 7 1e3, 7 1e3 1e3 1e3 1e3 1e3, 7

0.0042 0.0000 0.0000 0.0050 0.0075 0.0100 0.0071

0 0 0 0 0 0 0

e e e e e e e

0.0030 0.0000 0.0000 0.0050 0.0075 0.0100 0.0071

0 0 0 0 0 0 0

e e e e e e e

e e e e e e e

Max. Likelihood Estimate Left-Censored Data 0.1176

32

1e10

0.0100

0

e

e

7 7e8 7 7 7 7 7

e e e e e e e

47e68

63, 67e68 63, 65, 67e68 67e68 63, 67e68 63, 67e68 63, 65, 67e68 63, 67e68

e 67 e e e e e

63, 63, 67 63, 63, 63, 63,

67e68 65, 67e68 67e68 67e68 67e68 67e68

3114

Table 4 e Effect of subtracting background value estimates on percent mass recovered and time of travel from the measured breakthrough curve. Outliers Included Negative Concentrations Retained

Outliers Excluded Negative Concentrations Converted to Zero

Negative Concentrations retained

Negative Concentrations Converted to Zero

Background Value, mg L1

% Mass Recovered

Mean Travel Time, h

% Mass Recovered

Mean Travel Time, h

Background Value, mg L1

% Mass Recovered

Mean Travel Time, h

% Mass Recovered

Mean Travel Time, h

Data Mean

0.0192 0.0217 0.0183 0.0191 0.0195 0.0199 0.0194

93.8 93.2 94.1 93.9 93.8 93.7 93.8

9.86 9.79 9.88 9.86 9.85 9.84 9.85

94.2 93.7 94.4 94.3 94.2 94.1 94.2

9.90 9.86 9.92 9.90 9.90 9.89 9.90

0.0124 0.0138 0.0115 0.0123 0.0127 0.0132 0.0127

95.7 95.3 95.9 95.7 95.6 95.5 95.6

10.03 10.00 10.06 10.04 10.03 10.01 10.03

95.7 95.4 96.0 95.8 95.7 95.6 95.7

10.04 10.01 10.06 10.04 10.03 10.02 10.03

Trimmed Mean

0.0110 0.0130 0.0099 0.0109 0.0114 0.0119 0.0113

96.1 95.5 96.4 96.1 96.0 95.8 96.0

10.07 10.02 10.10 10.07 10.06 10.05 10.06

96.1 95.6 96.4 96.1 96.0 95.9 96.0

10.07 10.02 10.10 10.07 10.06 10.05 10.06

0.0098 0.0109 0.0087 0.0097 0.0102 0.0107 0.0101

96.4 96.1 96.7 96.4 96.3 96.2 96.3

10.10 10.07 10.13 10.10 10.09 10.08 10.09

96.4 96.1 96.7 96.4 96.3 96.2 96.3

10.10 10.07 10.13 10.10 10.09 10.08 10.09

Winsorized Mean

0.0135 0.0151 0.0126 0.0134 0.0138 0.0142 0.0137

95.4 95.0 95.6 95.4 95.3 95.2 95.3

10.01 9.97 10.02 10.01 10.00 9.99 10.00

95.5 95.1 95.7 95.5 95.4 95.3 95.4

10.01 9.98 10.03 10.01 10.01 10.00 10.01

0.0119 0.0132 0.0110 0.0118 0.0122 0.0126 0.0122

95.8 95.5 96.1 95.9 95.8 95.6 95.8

10.05 10.01 10.07 10.05 10.04 10.03 10.04

95.9 95.6 96.1 95.9 95.8 95.7 95.8

10.05 10.02 10.07 10.05 10.04 10.03 10.04

Data Median

0.0042 0.0000 0.0000 0.0050 0.0075 0.0100 0.0071

97.9 99.1 99.1 97.7 97.0 96.4 97.1

10.24 10.34 10.34 10.22 10.16 10.09 10.17

97.9 99.1 99.1 97.7 97.0 96.4 97.1

10.24 10.34 10.34 10.22 10.16 10.09 10.17

0.0030 0.0000 0.0000 0.0050 0.0075 0.0100 0.0071

98.3 99.1 99.1 97.7 97.0 96.4 97.1

10.27 10.34 10.34 10.22 10.16 10.09 10.17

98.3 99.1 99.1 97.7 97.0 96.4 97.1

10.27 10.34 10.34 10.22 10.16 10.09 10.17

Max. Likelihood Estimate Left-Censored Data

0.1176

67.1

6.25

78.0

8.60

0.0100

96.4

10.09

96.4

10.09

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

Analysis Method

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

3115

Fig. 4 e Effects of subtracting an arithmetic mean background value ðxb [0:0192 mg LL1 Þ and a maximum likelihood estimate (MLE [ 0.0100 mg LL1) from the measured breakthrough curve. Early time, peak time, mean time, and late-time plots are shown enlarged to emphasize the effect of subtracting an overly large background value ðxb [0:0192 mg LL1 Þ causing negative concentration values in the early time and late-time plots and the importance of correct the MLE estimate of 0.0100 mg LL1.

3116

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

mean is exactly 0.0100 mg L1, both are sufficiently close to 0.0100 mg L1 to be considered acceptable. A rMSE ¼ 0.0041 for the A15 estimate and 0.0027 for the H15 estimate implies that these two estimates underestimate the expected value of 0.0100 mg L1. Removing the outliers results in an A15 ¼ 0.0055 mg L1 with a standard deviation of 0.0044 mg L1 and an rMSE ¼ 0.0065. A corresponding H15 ¼ 0.0100 mg L1 with a standard deviation of 0.0139 mg L1 results in a rMSE equal to zero. This suggests that the method developed by the (AMC) is also affected by data outliers, but less so than other methods such as the sample mean and the MLE.

outliers so the implied physical problems associated with tracer-test design, implementation, and subsequent data collection is incorrect. Application of the estimated transport parameters in a theoretical model might correct the apparent error, but such a correction should not be expected. Worse, if one or more of the parameters in error is very far from the true value, it is likely that the theoretical transport model will isolate a local minimum rather than continuing to the global minimum and the final assessment regarding solute-transport processes could be drastically wrong.

3.2.1.

4.

Effect of background estimates on breakthrough curves

Subtracting background estimates greater than 0.0100 mg L1 results in a significant number of data points less than zero in both the early-time and late-time data (Table 3). For example, subtracting the sample mean using all the data (xb ¼ 0:0192 mg L1 ) from the measured BTC causes the first eight data points and last seven data points to be less than zero for a total of 15 data points that are less than zero. Traditionally, these 15 negative concentration values would be converted to zero concentration. The net effect of retaining the negative concentrations or conversion to zero concentrations will be to influence basic numerical analyses and overall quantitative assessment of the tracer test. To emphasize this fact, the percent tracer-mass recovered and mean travel time for the example data set (Fig. 2) after subtracting the estimated background value are shown in Table 4. In this instance the traditional mean estimate of 0.0192 mg L1 will result in 93.8% of the mass being recovered and a mean travel time of 9.86h whereas the correct values are 96.4% of the mass recovered and a mean travel time of 10.09 h. Fig. 4 shows how some early- and late-time data points will be less than zero when the traditional mean value of 0.0192 mg L1 is subtracted from the BTC. Fig. 4 also shows how it is possible to obtain an incorrect percent mass recovery when the early-, peak-, and late-time data are examined. The mean time may not appear greatly affected for this particular BTC, but mean travel time errors become more significant when tailing or right-skew is more severe. Overall differences may not appear extreme for this particular data set, but very large differences could occur with a different data set and different sample mean.

3.2.2. Potential for calculation errors and incorrect site assessments The potential for a large difference in percent mass recovered and mean travel time becomes apparent when the background value estimated by the MLE with the outliers included is applied to the BTC. In this instance, only 67.1% of the mass is recovered, but now the mean travel time is estimated to take only 6.25 h. Such a poor mass recovery typically implies serious data collection and analysis errors, inaccurate discharge estimates, possible tracer migration to unmonitored locations, and/or the assumption that tracer transport is nonconservative. The apparent faster mean travel time also may result in an incorrect sampling frequency for possible pollutants. However, the real source of recovery error in this instance was caused by use of the MLE with inclusion of

Conclusions

Accurate calculation of a single background concentration for subtraction from a measured tracer-breakthrough curve (BTC) is essential for the calculation of such basic parameters as percent tracer-mass recovered, mean travel time, and related solute-transport parameters. Typical use of the arithmetic mean can often lead to an excessively large background value for subtraction from the measured BTC if one or more outliers are included in the background data set. The creation of negative concentrations that are usually treated as zero concentrations in the resultant BTC routinely occur when the arithmetic mean is used as the background concentration value. Below detection limit data, conventionally known as left-censored data, can also adversely affect background value estimation. Proper evaluation and possible removal of outliers can improve background value estimation. In addition, application of statistical techniques designed to address censored data have been shown to better represent estimates than the arithmetic mean. However, use of one or more of the common data fabrication methods in which some fraction of the method detection limit is substituted for data below the detection limit has no basis in fact scientifically or statistically. Application of statistical methods designed for censored data, such as the maximum likelihood estimation (MLE), have been shown in the past to generally provide better descriptive statistics than does the arithmetic mean. In particular, when applied to background data, the MLE was found to produce excellent estimates for the descriptive statistics after any outliers were first removed. However, although the MLE appears to have worked well with the synthetic background data set provided, a major problem with the MLE as applied to the background data set is the assumption that the data conforms to a log-normal distribution, which may not necessarily be the case. Although the MLE produced the expected estimate for the sample mean, in reality, the background concentration data appears to conform to a gamma distribution so use of the MLE method in this paper should be viewed with some skepticism. Future work should focus on developing an MLE for gamma-distributed data. Alternatively, outliers may be accommodated during data analysis, but the possibility that the median value may adversely affect the analysis should be considered. It is still appropriate to carefully evaluate any potential outliers to determine if they should be included or excluded in any outlier accommodation analysis because the outliers may or

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

may not represent actual data measurements. However, whether outliers are accommodated in background data analysis or not, methods that appropriately consider leftcensored data in which numerous zero concentration values are included need to be addressed.

Disclaimer The views expressed in this paper are solely those of the author and do not necessarily reflect the views or policies of the U.S. Environmental Protection Agency.

Acknowledgments The author would like to thank Dr. John Fox of the National Center for Environmental Assessment of the U.S. Environmental Protection Agency for suggesting an approach for dealing with left-censored data with a data floor and Dr. Leonid Kopalev of the National Center for Environmental Assessment of the U.S. Environmental Protection Agency for his review and constructive criticisms of the manuscript. The author would also like to thank two anonymous reviewers for their comments and suggested improvements to the manuscript.

references

AMC, 1989. Robust statisticsdhow not to reject outliers: Part 1. Basic concepts. Analytical methods committee. Analyst 114, 1693e1697. Antweilier, R.C., Taylor, H.E., 2008. Evaluation of statistical treatments of left-censored environmental data using coincident uncensored data sets: I. Summary statistics. Environmental Science and Technology 42 (10), 3732e3738. Benischke, R., Goldscheider, N., Smart, C., 2007. Tracer techniques. In: Goldscheider, N., Drew, D. (Eds.), Methods in Karst Hydrogeology, Vol. 26. Taylor & Francis, London, UK, pp. 148e170. Brandt, S., 1999. Data Analysis: Statistical and Computational Methods for Scientists and Engineers, third edition. SpringerVerlag, New York. Brown, T.L., 2009. Fluorescence characterization of karst aquifers in east Tennessee. Master’s thesis, University of Tennessee, Knoxville, Tenn., 151 p. Burke, S., 2001. Missing values, outliers, robust statistics & nonparametric methods. Statistical data analysis. LCeGC European Online Supplement 59, 19e24. URL. http://www. lcgceurope.com/lcgceurope/data/articlestandard/lcgceurope/ 502001/4509/article.pdf (accessed 15.07.10). Chambers, J.M., Cleveland, W.S., Kleiner, B., Tukey, P.A., 1983. Graphical Methods for Data Analysis. Brooks/Cole Publishing Company, Pacific Grove, Calif. Cowell, F.A., Victoria-Feser, M.P., 1993. Robustness properties of inequality measures: The influence function and the principle of transfers. Tech. Rep. DARP 01. Suntory and Toyota International Centres for Economics and Related Disciplines (STICERD), London School of Economics and Political Science, London, U.K. http://eprints.lse.ac.uk/2141/1/Robustness_ Properties_of_Inequality_Measures_The_Influence_Function_ and_the_Principle_of_Transfers.pdf (accessed 05.01.11).

3117

Daszykowski, M., Kaczmarek, K., Vander Heyden, Y., Walczak, B., 2007. Robust statistics in data analysisda review: basic concepts. Chemometrics and Intelligent Laboratory Systems 85, 203e219. doi:10.1016/j.chemolab.2006.06.016. Davis, S.N., Campbell, D.J., Bentley, H.W., Flynn, T.J., 1985. Ground Water Tracers. National Water Well Association, Worthington, Ohio. EFSA, 2010. Management of left-censored data in dietary exposure assessment of chemical substances. European Food Safety Authority. EFSA Journal 8 (3), 96. doi:10.2903/j.efsa.2010. 1557. URL. http://www.efsa.europa.eu/en/scdocs/doc/1557.pdf (accessed 17.07.10). El-Shaarawi, A.H., Esterby, S.R., 1992. Replacement of censored observations by a constant: an evaluation. Water Research 26 (6), 835e844. Field, M.S., 2002. The Qtracer2 Program for Tracer-breakthrough Curve Analysis for Hydrological Tracer Tests. Tech. Rep. EPA/ 600/R-02/001 and EPA/600/CR-02/001. U.S. Environmental Protection Agency, Washington, D.C. URL. http://cfpub.epa.gov/ ncea/cfm/recordisplay.cfm?deid¼54930 (accessed 21.07.10). Field, M.S., 2003. Tracer-test Planning Using the Efficient Hydrologic Tracer-test Design (EHTD) Program. Tech. Rep. EPA/600/R-03/ 034 and EPA/600/CR-03/034. U.S. Environmental Protection Agency, Washington, D.C. URL. http://cfpub.epa.gov/ncea/risk/ recordisplay.cfm?deid¼66073, 175 pp., (accessed 12.01.11). Ganser, G.H., Hewett, P., 2010. An accurate substitution method for analyzing censored data. Journal of Occupational and Environmental Hygiene 7 (4), 233e244. doi:10.1080/ 15459621003609713. Gaspar, E., 1987a. Tracer Hydrology, Vol. I. CRC Press, Boca Rato, Fla. Gaspar, E., 1987b. Tracer Hydrology, Vol. II. CRC Press, Boca Rato, Fla. Gilliom, R.J., Helsel, D.R., 1986. Estimation of distributional parameters for censored trace level water quality data, 1. Estimation techniques. Water Resources Research 22 (2), 135e146. Gleit, A., 1985. Estimation for small normal data sets with detection limits. Environmental Science and Technology 19 (12), 1206e1213. Guan, H., Love, A.J., Simmons, C.T., Hutson, J., Ding, Z., 2010. Catchment conceptualization for examining applicability of chloride mass balance method in an area with historical forest clearance. Hydrology and Earth System Sciences 14, 1233e1245. Haas, C.H., Scheff, P.A., 1990. Estimation of averages in truncated samples. Environmental Science and Technology 24 (6), 912e919. Hampel, F.R., 1974. The influence curve and its role in robust estimation. Journal of the American Statistical Association 69 (346), 383e393. Hampel, F.R., 2001. Robust Statistics: a Brief Introduction and Overview. Tech. Rep. 94. Eidgeno¨ssische Technische Hochschule (ETH), CH-8092 Zu¨rich, Switzerland. URL. ftp://129.132.148.131/ sfs/pub/Research-Reports/94.pdf (accessed 22.10.10). Helsel, D.R., 2005. Nondetects and Data Analysis: Statistics for Censored Environmental Data. John Wiley & Sons, New York. Helsel, D.R., 2006. Fabricating data: how substituting values for nondetects can ruin results, and what can be done about it. Chemosphere 65, 2434e2439. doi:10.1016/j.chemosphere.2006. 04.051. Helsel, D.R., Hirsch, R.M., 2002. Statistical Methods in Water Resources, Techniques of Water-resources Investigations of the United States Geological Survey, Book 4, Hydrological Analysis and Interpretation. Tech. Rep. Chapter A3. U.S. Geological Survey, Washington, D.C. URL. http://pubs.usgs. gov/twri/twri4a3/pdf/twri4a3-new.pdf (accessed 22.07.10). Hewett, P., Ganser, G.H., 2007. A comparison of several methods for analyzing censored data. The Annals of Occupational Hygiene 51 (7), 611e632. doi:10.1093/annhyg/mem045.

3118

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 0 7 e3 1 1 8

Hrachowitz, M., Soulsby, C., Tetzlaff, D., Malcolm, I.A., Scoups, G., 2010. Gamma distribution models for transit time estimation in catchments: physical interpretation of parameters and implications for time-variant transit time assessment. Water Resources Research 46 (W10536). doi:10. 1029/2010WR009148. Hubert, M., Debruyne, M., 2009. Breakdown value. Computational Statistics 1 (3), 296e302. doi:10.1002/wics.034. URL. http://online library.wiley.com/doi/10.1002/wics.34/pdf (accessed 11.06.10). Hubert, M., Van der Veeken, S., 2008. Outlier detection for skewed data. Journal of Chemometrics 22 (3/4), 235e246. doi:10.1002/ cem.1123. Irwin, M.E., 2005. Summary statistics in SAS. URL. http://www. markirwin.net/stat135/Lecture/Lecture27.pdf (accessed 19.07.10). Ka¨ß, W., 1998. Tracing Technique in Geohydrology. A.A. Balkema, Rotterdam, The Netherlands. Kilpatrick, F.A., Wilson Jr., J.F., 1989. Measurement of Time of Travel in Streams by Dye Tracing. Tech. Rep. Techniques of Water-Resources Investigations of the U.S. Geological Survey, Book 3, Chapter A9. U.S. Geological Survey, Washington D.C. Kroll, C.N., Stedinger, J.R., 1996. Estimation of moments and quantiles using censored data. Water Resources Research 32 (4), 1005e1012. Leibundgut, C., Ma1oszewski, P., Ku¨lls, C. (Eds.), 2009. Tracers in Hydrology. John Wiley & Sons Ltd., West Sussex, U.K. Mull, D.S., Liebermann, T.D., Smoot, J.L., Woosley Jr., L.H., 1988. Application of Dye-tracing Techniques for Determining Solute-transport Characteristics of Ground Water in Karst Terranes. Tech. Rep. EPA/904/9-88-001. U.S. Environmental Protection Agency, Region IV, Atlanta, Ga. Navy, 1999. Handbook for Statistical Analysis of Environmental Background Data. Tech. rep. Department of the Naval, Southwest Division, Naval Facilities Engineering Command, San Diego, Cal. https://portal.navfac.navy.mil/portal/page/ portal/navfac/navfac_ww_pp/navfac_nfesc_pp/ environmental/erb/lab/hndbk-sw.pdf (accessed 22.08.10). Olive, D.J., 2008. Applied Robust Statistics. Preprint M-02e006 URL. http://www.math.siu.edu/olive/ol-bookp.htm (accessed 15.07.10). Ripp, J., 1996. Analytical Detection Limit Guidance & Laboratory Guide for Determining Method Detection Limits. Tech. Rep. PUBL-TS-056e96. Laboratory Certification Program, Wisconsin

Department of Natural Resources, Madison, Wis. URL. http:// www.dnr.state.wi.us/org/es/science/lc/outreach/publications/lod (accessed 07.01.11). Rousseeuw, P.J., Debruyne, M., Engelen, S., Hubert, M., 2006. Robustness and outlier detection in chemometrics. Analytical Chemistry 36 (3), 221e242. doi:10.1080/10408340600969403. Schudel, B., Biaggi, D., Dervey, T., Kozel, R., Mu¨ller, I., Ross, J.H., Schindler, U., 2002. Utilisation des traceurs artificiels en hydroge´ologie: Guide pratique. Tech. Rep. BWGG-3-F. Groupe de travail Trac¸age de la Socie´te´ suisse d’hydroge´ologie (SSH), Berne, Switzerland. URL. http://www.bafu.admin.ch/ publikationen/publikation/00439/index.html?lang¼fr (accessed 11.01.11). Singh, A., Nocerino, J., 2001. Robust Estimation of Mean and Variance Using Environmental Data Sets with below Detection Limit Observations. Tech. Repp. 01e062. U.S. Environmental Protection Agency, National Exposure Research Laboratory, Las Vegas, Nev. URL. http://www.epa.gov/esd/tsc/images/ robust_estim.pdf (accessed 13.07.10). Swaroop, R., Winter, W.R., 1971. A Statistical Technique for Computer Identification of Outliers in Multivariate Data. Tech. Rep. NASA TN D-6472. NASA Flight Research Center, Washington, D.C. URL. http://ntrs.nasa.gov/archive/nasa/casi. ntrs.nasa.gov/19710021895_1971021895.pdf (accessed 20.07.10). Thompson, M.L., Nelson, K.P., 2003. Linear regression with Type 1 interval-and left-censored response data. Environmental and Ecological Statistics 10 (2), 221e230. USEPA, 2000. Guidance for Data Quality Assessment: Practical Methods for Data Analysis (EPA QA/G-9, QA00 Update). Tech. Rep. EPA/600/R-96/084. U.S. Environmental Protection Agency, Washington, D.C. URL. http://www.clu-in.org/conf/tio/pasi_ 121603/g9-final.pdf (accessed 22.10.2010). USEPA, 2009. Definition and procedure for the determination of the method detection limitdrevision 1.11. 40 CFR Part 136, Appendix B. In: Code of Federal Regulations. U.S. Government, Washington, D.C., pp. 343e346. URL. http://edocket.access.gpo. gov/cfr_2009/julqtr/pdf/40cfr136AppB.pdf (accessed 07.01.11). Wilson Jr., J.F., Cobb, E.D., Kilpatrick, F.A., 1986. Fluormetric Procedures for Dye Tracing. Tech. Rep. In: Techniques of Water-Resources Investigations of the U.S. Geological Survey, Vol. 3. U.S. Geological Survey, Washington D.C. Chapter A12.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Short-term and long-term effects on carbon storage of pulse feeding on acclimated or unacclimated activated sludge gın a,d,*, Derin Orhon a,b, Simona Rossetti c, Mauro Majone d Aslı Seyhan C¸ıg a

Faculty of Civil Engineering, Environmental Engineering Department, Istanbul Technical University, 34469 Maslak, Istanbul, Turkey Turkish Academy of Sciences, Piyade sokak No. 27, 06550 C¸ankaya, Ankara, Turkey c Istituto di Ricerca Sulle Acque C.N.R., Via Salaria Km 29,300, 00016 Monterotondo, Italy d Department of Chemistry, Sapienza University of Rome, P.le Aldo Moro 5, 00185 Rome, Italy b

article info

abstract

Article history:

This study was aimed to investigate the effect of different feeding patterns on the physi-

Received 5 December 2010

ological state of the activated sludge and related microbial composition in an SBR (SRT of 2

Received in revised form

days, acetate as the sole carbon source, aerobic conditions). The activated sludge was

17 February 2011

acclimated to two subsequent feeding patterns, namely to continuous feeding throughout

Accepted 14 March 2011

the reaction phase and then to pulse feeding. FISH and microscopy staining procedure (Nile

Available online 21 March 2011

blue) were used to investigate the microbial composition, in combination with quantitative determination of storage. At steady state, storage was significant even under continuous

Keywords:

feeding whereas pulse feeding brought a strong increase of both rate and yield of storage.

Activated sludge

Short-term and long-term effects were clearly distinguishable: the immediate adaptation

PHA storage

of biomass coming from continuous feeding to a single spike accounted for a significant

Microbial composition

portion of the overall increase of both rate and yield of polymer storage that was obtained

FISH

after long acclimation to pulse feeding. On the contrary, after either type of feeding, both

Feeding pattern

cultures were mainly constituted from the members of Thauera/Azoarcus group. Thus, the same dominant group preferably consumed the acetate via storage or growth depending on acclimation conditions. Our study clearly showed that a progressive increase of storage capacity is not necessarily due to a shift of microbial composition. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Extensive researches have been conducted for the mechanistic description of biological processes occurring in activated sludge systems, in order to obtain best practices for their design and operation. This effort led to recognize the storage phenomena as an important mechanism for the carbon source removal, particularly when the activated sludge experiences highly dynamic conditions (aeration tanks with plug-flow configuration, selectors for bulking control, contact-

stabilization processes, and sequencing batch reactors), so called the feast and famine conditions (van Loosdrecht et al., 1997). Storage is explained as a way that the microorganism can use for face against the highly dynamic environment in activated sludge systems, while maintaining favourable growth conditions. Indeed, biomass growth is unbalanced due to the gradients in the substrate availability. Thus, cells are not able to instantaneously adapt their growth rate to the changing substrate concentration because it would require also for

* Corresponding author. Faculty of Civil Engineering, Environmental Engineering Department, Istanbul Technical University, 34469 Maslak, Istanbul, Turkey. Tel.: þ90 532 515 6328; fax: þ90 212 2277296.  gın), [email protected] (D. Orhon), [email protected] (S. Rossetti), mauro.majone@uniE-mail addresses: [email protected] (A.S. C¸ıg roma1.it (M. Majone). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.026

3120

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

Nomenclature COD DO FISH F/M HRT OLR PHA Qf Qeff Qw qp qp qp/qs qs

chemical oxygen demand dissolved oxygen fluorescence in situ hybridization food to microorganism ratio hydraulic retention time (d) organic loading rate (mgCOD/L$d) polyhydroxyalkanoate feed flow rate (L/d) supernatant flow rate (L/d) mixed liquor withdrawal flow rate (L/d) PHA production rate (mgCOD/mgCOD$h) PHA consumption rate (mgCOD/mgCOD$h) Substrate fraction used for storage (adimensional) specific substrate removal rate (mgCOD/ mgCOD$h)

maintaining the relative synthesis rate of all cellular components. Therefore, under transient conditions, microorganisms can more quickly activate other mechanisms of substrate uptake like storage of substrate into specialized internal polymers, usually polyhydroxyalkanoate (PHA), especially with volatile fatty acids as substrates. Initially, the role of storage on population dynamics in activated sludge has been studied in detail for filamentous bulking control, where selection of floc-formers against filaments is exerted by selectors or other process configurations that introduce a substrate concentration gradient. Because the storage response is faster than growth response (less adaptation is required), the more the microorganisms are able to store substrates during imposed transients and subsequently reuse them for growth, the more they would have a competitive advantage and the resulting enriched activated sludge should have a higher storage response (Majone et al., 1996). Generally, storage has been reported as the dominant mechanism when the length of feast phase became a lower fraction of the overall reaction length (Dionisi et al., 2001). However, there is no agreement whether a lower storage capacity is truly a general feature of filamentous microorganisms and bulking sludge with respect to floc-formers and wellsettling sludge, respectively (Martins et al., 2004). Beccari et al. (1998) observed that feast and famine conditions created a bulking sludge with a high storage capacity. Martins et al. (2003) also found that there were no differences in PHB storage rates or yields of bulking and well-settling sludge and observed that filamentous bacteria are usually no more than 20% of the bacterial population in bulking sludge and therefore small kinetic differences may be undistinguishable in mixed activated sludge populations. Moreover, when dealing with mixed cultures, the type and extent of biomass response to the transient conditions can depend not only on microbial composition of the consortium but also on the physiological state of any microorganisms in the consortium, which is also affected by the operating conditions (e.g. sludge residence time). A similar short-term effect was also observed for mixed cultures; van Loosdrecht et al. (1997) tested pulse feeding of acetate on the system previously operated at steady state with

S S0 SBR SRT SS SVI V VSS X

effluent COD (mgCOD/L) feed COD (mgCOD/L) sequencing batch reactor sludge retention time suspended solid sludge volume index (mL/g) SBR liquid volume during the reaction phase (L) volatile suspended solid (mgCOD/L) biomass concentration in the stirred mixed liquor (gVSS/L) biomass concentration in the supernatant after Xe the settling phase (gVSS/L) average observed yield (gCOD/gCOD) Yobs average storage yield of PHA on acetate YSTO (gCOD/gCOD) qC (SRT) sludge retention time (d)

continuous feeding and observed immediate PHA formation. Similarly, Martins et al. (2004) has operated a continuously fed SBR under anoxic conditions and observed PHA storage when pulse feeding was applied in a single cycle. In addition, pure culture studies have clearly shown that the role of storage (yield and rate) is also depending on operating conditions (organic load rate and culture residence time), even with well-adapted microorganism (Dionisi et al., 2005; Majone et al., 2007). The identification of microorganisms responsible for the storage gained increasing interest in recent years, mainly in the frame of PHA production from modified activated sludge processes (i.e. under highly dynamic conditions to favour as much storage as possible), Thauera spp. and Azoarcus genus within Betaproteobacteria phylum were often dominant in the PHA storing biomass (Dionisi et al., 2005; Lemos et al., 2008; Serafim et al., 2006). Because the main objective was to evaluate the microbial composition in the presence of the storage, the identification of the microbial composition under less dynamic conditions, i.e. in the absence of storage, received minor interest. Thus, the aim of this paper is to better understand the mechanism by which the presence of more or less dynamic conditions can affect the mechanism of substrate removal (i.e. the relative role of storage and growth) in acclimated or unacclimated mixed cultures. In particular, the experimental activity was designed to understand whether the increase of storage under dynamic conditions is induced either by changes on microbial composition or by physiological adaptation of the given microbial composition to the dynamic conditions. Moreover, the latter mechanism has been further investigated to understand how much the physiological adaptation is a short-term effect or it requires longer acclimation to dynamic feeding. Therefore, the effect of a shift from continuous to dynamic conditions has been investigated by two subsequent runs in a sequencing batch reactor (SBR, aerobic conditions) with two different feeding patterns, namely by continuous feeding throughout the reaction phase and then by pulse feeding. Moreover, single disturbance of usual feeding conditions was imposed to each acclimated biomass to understand short-term effects and distinguish

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

them from long-term effects after long acclimation. In situ detection methods (FISH) and microscopy staining procedure (Nile blue) were employed to investigate the microbial composition, in combination with quantitative determination of rate and yield of storage phenomena, under all investigated conditions.

2.

Materials and methods

2.1.

SBR operation

Experiments were conducted using a laboratory-scale sequencing batch reactor (SBR) that was inoculated with the activated sludge from the Roma Nord domestic wastewater treatment plant, Italy. This is a conventional and large treatment plant (more than 800.000 inhabitants), usually operated at medium-low organic load with the sludge residence time more than 10 days. The SBR was operated at a Sludge Retention Time (SRT) of 2 days, in sequence with two different feeding patterns: (i) a slow and long feeding almost throughout the reaction phase (namely, “continuous feeding”) and (ii) a quick and short feeding for a 1 min (namely, “pulse feeding”). SBR operation involved 6 cycles a day, each one made by: 10 min of idle phase, 5 min of nutrient addition, 165 min of reaction phase, 1 min of waste sludge withdrawal, 30 min of settling and 30 min of effluent discharge. At the start of the reaction phase, the necessary amount of acetate was added as the only carbon source, at a constant flow rate of 1.3 ml/min for 150 min or at 195 ml/min for 1 min, for continuous or pulse feeding, respectively. The substrate feed was prepared by diluting an acetate stock solution of sodium acetate trihydrate (CH3COONa$3H2O) in distilled water. The nutrient feed was obtained by diluting in distilled water a nutrient stock solution, which composition was 120 g/L NH4Cl, 160 g/L KH2PO4, 320 g/L K2HPO4, 15 g/L MgSO4$7H2O, 0.5 g/L FeSO4$7H2O, 2 g/L CaCl2$7H2O, 0.5 g/L MnSO4$H2O and 0.5 g/L ZnSO4$7H2O. For each SBR run, the dilution ratio of the nutrient stock solution was adjusted so to have enough buffer capacity and to supply nitrogen and phosphorus at a non-limiting concentration with respect to the carbon source. The total working volume of the reactor was 1.2 L (at the end of the fill), with an exchange volume ratio of 0.6 L/cycle. Aerobic conditions were maintained by mechanical stirring and air bubbling throughout the idle, feeding and reaction phases. Dissolved oxygen was continuously monitored and was never below 2 mg/L during the reaction phase. The temperature of the SBR was kept at 25  C and the pH was kept at around 7.0  0.5. The SRT was adjusted according to the daily sludge production, by taking into account the amount of volatile suspended solid (VSS) discharged with both waste sludge withdrawal and suspended solids in the effluent at each cycle. In order to verify the establishment of steady state conditions, the SBR performance was monitored by measuring VSS, polyhydroxyalkanoate (PHA), and chemical oxygen demand (COD) at the end of the cycle. After steady state was observed under continuous feeding, the SBR performance was further

3121

characterized during typical cycles, by taking several samples throughout the reaction phase for COD, acetate and PHA analyses (so called “kinetic” experiments, three replicates). Additional kinetic experiments were conducted by applying a single pulse feeding (three replicates), for providing information on the dynamic response of the biomass acclimated under continuous feeding. Enough time was allowed between replicates of dynamic experiments with pulse feeding, to avoid excessive disturbance of the steady state. Subsequently, at 35th day, the SBR feeding pattern was changed from continuous to pulse feeding, all other conditions remaining the same. The same experimental program was adopted as for monitoring the steady state and performing the kinetic experiments. As for the latter, three replicates of kinetic experiments were performed with sampling throughout the reaction phase at pulse and continuous feeding (i.e. usual and changed feeding condition, respectively).

2.2.

Analytical methods

Samples taken for COD and acetate analyses were filtered through 0.45 mm PVDF syringe filters. The acetate samples were analyzed by gas chromatograph (stationary phase Carbowax 20 M 4% on CarboPak B-DA on a PerkineElmer 8410 instrument). The COD samples were measured with the Spectroquant Kit (Merck). For PHA measurements, the sludge samples were preserved immediately after sampling with a NaClO solution (7% of active Cl2). PHA contents of the samples were determined by gas chromatograph after extracted, hydrolyzed, and esterified to 3-hydroxyacyl methyl esters (Braunegg et al., 1978). Suspended solids (SS) and VSS were determined in the aerobic slurry and in the effluent supernatant by using the procedure defined in Standard Methods (1995). When necessary, analytical results were converted into COD units according to conversion factors of 1.067 mgCOD/mgacetate, 1.60 mgCOD/mgPHB and 1.42 mgCOD/mgVSS.

2.3.

Molecular analysis

Fluorescence in Situ Hybridization (FISH) was performed on paraformaldehyde-fixed biomass samples for the identification of the main microbial components of SBRs operated at steady state conditions. Group specific oligonucleotide probes were always applied with probes EUB338, EUB338-II and EUB338-III combined in a mixture (EUB338mix) and DAPI for the detection of most Bacteria and the total cell, respectively. Details on oligonucleotide probes are available at probeBase (Loy et al., 2007). DAPI was added to hybridization buffer at a final concentration of 1 mg/ml. All the probes were synthesized with FITC and Cy3 labels purchased from MWG AG Biotech (Germany). Slides were examined with an epifluorescence microscope (Olympus BX51) at 1000 magnification and images were captured with Olympus F-View CCD camera and handled with AnalySIS software (SIS, Munster, Germany). Post-FISH Nile Blue staining was applied for the identification of the cells containing intracellular polymeric storage compounds (PHA). For this purpose, FISH was carried out and representative images captured, then the cover slip was removed from the slide. The mount washed off with distilled

3122

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

water and dried again by compressed air. Then chemical staining was carried out by Nile Blue according to the method developed by Ostle and Holt (1982). Fields photographed in FISH were relocated and rephotographed with the chemical stain. For the quantification, the amount of the cell hybridized with specific probe, EUBmix and DAPI were counted on at least twenty different fields of view selected randomly for each probe.

2.4.

Calculations

The following relationships were used in order to define SBR operating conditions: OLR ¼

qc ¼

Qf S0 V

(1)

VX Qw X þ Qeff Xe

(2)

where OLR ¼ Organic loading rate (mgCOD/L$d); qC ¼ SRT ¼ solid retention time (d); V ¼ SBR liquid volume during the reaction phase (L); Qf ¼ feed flow rate (L/d); S0 ¼ feed COD (mgCOD/L); X ¼ biomass concentration in the stirred mixed liquor (gVSS/L); Xe ¼ biomass concentration in the supernatant after the settling phase (gVSS/L); Qw ¼ mixed liquor withdrawal flow rate (L/d); Qeff ¼ supernatant flow rate (L/d) and Qf ¼ Qw þ Qeff

(3)

The average observed yield Yobs (gCOD/gCOD) in the SBR was calculated according to the following relationship: Yobs ¼

Qw X þ Qeff Xe HRT$X ¼ QðS0  SÞ qc ðS0  SÞ

(4)

HRT ¼ V/Q ¼ hydraulic retention time (d); S ¼ effluent soluble COD (mgCOD/L) The average storage yield of PHA on the substrate (YSTO, gCOD/gCOD) in the SBR was calculated from the ratio between the amount of PHA formed per cycle and the organic load of acetate given per cycle (because all the acetate was consumed during any cycle). The amount of PHA formed per cycle was calculated from the difference between the PHA content in the mixed liquor at the end of the feast phase and at the start of the feeding.

a

During the reaction phase, the specific acetate removal rate (qs, mgCOD/mgCOD$h) was calculated in different ways, depending on pulse or continuous feeding. Under pulse feeding, the specific rate was calculated by linear regression of experimental acetate profile and by dividing the obtained slope for the initial biomass concentration. Under continuous feeding, no acetate profile was detected and so the specific rate was obtained by dividing the organic loading per cycle for the feeding length and for the initial biomass concentration. Under both feeding conditions, the specific PHA production and consumption rates (qp and qp, mgCOD/mgCOD$h) were calculated by linear regression of experimental PHA profiles and by dividing the obtained slope for the initial biomass concentration. Under continuous feeding, the experimental PHA profile was also corrected by taking into account the progressive increase of volume during the reaction phase. The ratio between the rates of PHA formation and of acetate consumption was also calculated (qp/qs, adimensional), that also corresponds to another way to calculate the storage yield during the initial part of experiments where linear regression was applied to.

3.

Results and discussion

3.1. Comparison of the steady state performances under different feeding conditions Time profiles of suspended solids (SS), volatile suspended solids (VSS) and polyhydroxyalkanoates (PHAs) during SBR operation are given in Fig. 1, as a function of SBR operation time starting from the inoculum with activated sludge. Left part of the figure refers to continuous feeding whereas second part refers to pulse feeding. The shift from continuous to pulse feeding was performed at the 35th day of operation, while all other conditions were maintained same. After nearly 10 days of operation from SBR start-up, a steady state was reached under continuous feeding; average concentration of VSS and PHA was 1380 mgVSS/L and 108 mgCOD/L, respectively. The shift from continuous to pulse feeding caused a new transient period of about 10 days where VSS concentration decreased, whereas PHA concentration increased. After a new steady state was reached at day 45, average concentrations of VSS and PHA were determined as

b

Fig. 1 e Time profiles of (a) SS, VSS and VSS/SS ratio and (b) PHA.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

Table 1 e Steady State Performances under different acclimation conditions.

Acclimation condition SRT (d) Average OLR (mgCOD/L$d) (mgCOD/L$cycle) Average VSS (mgVSS/L) YOBS (gCOD/gCOD) Feast phase (min) Average amount of Stored PHA (mgCOD/L$cycle) YSTO (gCOD/gCOD) COD Removal (%)

Run 1

Run 2

Continuous 2 1560  12 280 1380  10 0.55 150 83  1.6

Pulse 2 1554  10 259 1050  14 0.36 45 141  0.7

0.32 78  2

0.54 90  1

1050 mgVSS/L and 170 mgCOD/L, respectively (i.e. a 24% decrease and a 63% increase, respectively). During operation under continuous feeding, sludge settling worsened and the sludge volume index (SVI) strongly increased to nearly 500 mL/g, although the SVI of the inoculum was lower than 120 mL/g. After the shift from continuous to pulse feeding, the sludge settling property quickly improved (SVI < 120 mL/g). The average data obtained for different feeding patterns under steady state conditions are summarized in Table 1. Based on the higher VSS concentration, the observed yield was also higher under continuous feeding than under pulse feeding. As previously reported in a pure culture study with Paracoccus pantotrophus (Van Aalst-van Leeuwen et al., 1997), the presence of the storage is thermodynamically unfavourable compared to balanced growth response; on the other hand, the estimated decrease of Yobs was about 6%, quite less than here observed. This evidence suggests a better adaptation to dynamic growth conditions of well-storing P. Pantotrophus with respect to microorganims in the present microbial consortium, possibly also due to other different environmental conditions (i.e. a lower culture residence time and a higher temperature in the cited reference). As expected, based on higher PHA concentration, the amount of the carbon source used for PHA storage was also higher under pulse feeding than under continuous feeding. In addition, the amount of residual soluble COD under pulse feeding was less than under continuous feeding.

3123

Table 1 also shows that residual COD is higher under continuous feeding than under pulse feeding. Because the acetate was fully removed in both cases, the residual COD is likely due to cellular decay, that causes the release of soluble that are not degradable in the allowed residence time. Thus, a lower COD release suggests that less cellular decay is observed under pulse feeding than under continuous feeding; this finding could be explained from the higher PHA content that minimize the need for maintenance and endogenous metabolism.

3.2. Comparison of the dynamic behaviour of mixed culture under different feeding conditions After the establishment of steady state conditions under continuous feeding, kinetic experiments were conducted by monitoring relevant parameters during a typical SBR cycle. Experiments were performed by applying either usual continuous feeding which SBR biomass was already acclimated to or a single disturbance of usual conditions by using pulse feeding. The average profiles (three replicates) observed during both types of kinetic experiments are given in Fig. 2. As expected, during experiments with the usual continuous feeding, COD remained almost constant and acetate was always very low. Thus, the acetate removal rate was fully controlled with the acetate loading rate, i.e. with feed flow rate. In spite of the fact that acetate was removed at a low concentration and rate, a gradual increase of PHA concentration was observed during the long feeding period as well as a corresponding decrease observed during the short nofeeding period. The PHA concentration shifted from 106 mgCOD/L to 189 mgCOD/L, corresponding to about one third of acetate load being converted to PHA. On the other hand, when pulse feeding was applied for just one cycle to the biomass acclimated to continuous feeding conditions; typical “feast and famine” conditions were created. A peak substrate concentration was observed in a few minutes, which were totally removed in about 90 min. The relative ratio of feast and famine period was 11 under continuous feeding as dictated by substrate loading, although this ratio become 1 when pulse feeding was applied for just one cycle as also validated with the dissolved oxygen profiles given in Fig. 3.

Fig. 2 e Dynamic behaviour of biomass acclimated to continuous feeding when applying (a) usual continuous feeding and (b) single disturbance by pulse feeding.

3124

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

Fig. 3 e Dissolved oxygen profile of biomass acclimated to continuous feeding when applying (a) usual continuous feeding and (b) single disturbance by pulse feeding.

Accordingly, more PHA storage was observed when single pulse feeding was applied instead of the usual continuous feeding (119 mgCOD/L after 90 min versus 83 mgCOD/L after 150 min, respectively). It is interesting to observe that PHA consumption was also active for a longer period; thus, residual PHA at the end of reaction phase was quite similar under both conditions. The ability of microorganisms to quickly increase their acetate removal rate and shift it towards more storage has been largely reported for several microorganisms in pure culture studies (Blackall et al., 1991; Frigon et al., 2006; Pagni et al., 1992; van den Eynde et al., 1983; van Niekerk et al., 1987; van Niel et al., 1995). A short-term dynamic effect has been clearly observed by applying a substrate spike to aerobic cultures, previously grown under balanced conditions. In most cases, microorganisms were shown to be able to immediately increase their substrate uptake rate and the quick storage of substrate was considered as the reason for the observed phenomenon (even though not always measured). For each microorganism, the short-term response was a function of its previous “history”, e.g. hydraulic residence time in the chemostat. At the same time, the short-term response was found to be different from species to species (Majone and Tandoi, 2002), so potentially affecting the microbial population dynamics in the activated sludge. Similar kinetic experiments were performed after the SBR biomass was acclimated to pulse feeding and a new steady state was reached. As above reported (Fig. 1), the new steady

state was characterized from a lower VSS and a higher PHA concentration compared to the previous acclimation. The average results (three replicates) of kinetic experiments obtained by carrying out either pulse feeding or continuous feeding with the biomass acclimated to pulse feeding is given in Figs. 4 and 5. In this case, pulse feeding was the usual one whereas continuous feeding was applied as a single disturbance of previous steady state. Kinetic experiments conducted under pulse feeding conditions resulted in similar “feast and famine” profiles with biomass acclimated to either continuous or pulse feeding (Figs. 2b and 4a, respectively); however, the storage yield (YSTO) further increased from 46% to 54% when the biomass had been already acclimated to pulse feeding. Correspondingly, the duration of the feast period decreased from 70 to 45 min after adaptation to the pulse feeding as also observed from dissolved oxygen profiles (Figs. 3b and 5a). This is in agreement with the different length of the feast phase for the different feeding patterns, which the biomass was acclimated to. Indeed, under continuous feeding, the length of feast phase corresponds to the length of feeding phase (150 min out of 180 min of the reaction phase) whereas under pulse feeding, the length of feast phase is quite shorter, as determined from the substrate removal rate after the initial concentration peak. On the contrary, when the biomass acclimated to pulse feeding was disturbed by using continuous feeding, a continuous decrease in PHA concentration was observed (around

Fig. 4 e Dynamic behaviour of biomass acclimated to continuous feeding when applying (a) usual pulse feeding and (b) single disturbance by continuous feeding.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

3125

Fig. 5 e Dissolved oxygen profile of biomass acclimated to pulse feeding when applying (a) usual pulse feeding and (b) single disturbance by continuous feeding.

37 mgCOD/L) although the culture was simultaneously using acetate (Fig. 4b). This behaviour can be explained by considering that the biomass is likely to be acclimated under pulse feeding; hence, its specific substrate uptake rate is high during the feast phase as well as PHA consumption is fast during the famine phase for growth metabolism. Thus, in the presence of slow feeding of acetate, it is reasonable that PHA consumption is also activated to maintain the previously higher level of growth metabolism. Accordingly, because acetate partially supports the metabolism, the consumption rate of the storage product during acetate feed is slower than during the last part of the reaction phase after the feed ends up (10.23 mgCOD/ gCOD$h and 80.47 mgCOD/gCOD$h, respectively). Ciggin et al. (2007) have reported similar observation for the biomass acclimated in similar conditions under anoxic conditions. In this study, the optimization of the growth mechanism was observed when the carbon source was provided continuously at lower concentration to the activated sludge culture acclimated to pulse feeding. A summary of the biomass response obtained from dynamic experiments is given in Table 2. Short-term effects are shown in Table 2 by comparing column 1 versus column 2 and column 3 versus column 4, i.e. the same biomass is experiencing a sudden shift from the usual feeding to the opposite one. It is evident that the

Table 2 e Dynamic behaviour under different feeding conditions.

Acclimation Condition

Run 1

Run 2

Continuous

Pulse

Feeding pattern during Continuous Pulse Pulse Continuous dynamic experiments F/M (mgCOD/mgCOD$d) 0.80 0.80 1.04 1.06 Feast Phase length (min) 150 70 45 150 0.55 0.52 0.36 0.37 YOBS (gCOD/gCOD) 0.32 0.46 0.54 e YSTO (gCOD/gCOD) 53 114 232 70 Specific Substrate Uptake Rate, qs (mgCOD/gCOD$h) Specific PHA Production 16 52 123 e Rate, qp (mgCOD/gCOD$h) 0.31 0.45 0.53 e qp/eqs

substrate removal rate was always higher under pulse feeding than under continuous feeding; accordingly, both storage rate (qp) and storage yield (YSTO) or qp/qs were higher under pulse feeding than under continuous feeding. Short-term effects due to a single spike have been also reported for mixed cultures (Albuquerque et al., 2010) as well as for several microorganisms in pure culture studies where microorganisms had been previously cultured a steady state in a chemostat (A summary of literature data with pure cultures is given in Majone and Tandoi, 2002). On the other hand, by comparing columns 2 and 3 with column 1, it is clear that the short-term effects of pulse feeding (columns 2 versus 1, single disturbance on unacclimated biomass) are lower than the long-term effects (column 3 versus 1, 10-day acclimation of the biomass under pulse feeding). Dealing with PHA storage rate, short-term response to single pulse feeding increases the PHA production rate from 16 to 52 mgCOD/gCOD$h whereas long-term acclimation further increases it to 123 mgCOD/gCOD$h. In other words, when a continuously fed acclimated biomass shifts to a pulse fed acclimation (all other conditions being the same), the full span of PHA storage rate is increased from 16 to 123 mgCOD/ gCOD$h (107 mgCOD/gCOD$h more, a 769% increase). However, about one third of this span is immediately observed with a single disturbance (36 mgCOD/gCOD$h more, a 325% increase). It is noteworthy that neither full adaptation nor significant change of microbial composition can be supposed to occur in a single disturbance by pulse feeding. A similar difference between short-term and long-term effects is also observed as for storage yield YSTO (or also for qp/ qs). A single pulse feeding on unacclimated biomass brings to a short-term increase of PHA storage yield from 0.32 to 0.46 gCOD/gCOD (a 45% increase) whereas long-term acclimation further increases it up to 0.54 gCOD/gCOD (a 71% increase). Even though the overall span for storage yield is obviously less than for kinetics, it is noteworthy that almost two thirds of overall span is due to the short-term effect whereas long-term acclimation only brings to one third of overall effect. Similarly, long-term effects have been reported in pure culture studies (Aulenta et al., 2003; Majone et al., 2007) showing that the adaptation of the microorganism to different dynamic conditions (e.g. frequency of feed as well as hydraulic residence time) also affected its kinetics and yield of PHA storage.

3126

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

Fig. 6 e Summary of relative contribution of a) storage yield YSTO and b) storage rate (qP) to respect to overall acetate removal.

It is also noteworthy that PHA storage remains significant even if biomass is fully acclimated to the continuous feeding. Under less favourable conditions, that causes both slow acetate uptake rate and slow PHA production rate, PHA storage accounts in any case for 32% of COD removal (YSTO in Table 2), that can be considered as a “background” level of storage for acetate removal (at least for the investigated experimental conditions). A summary of relative contribution of different effects to determine the overall contribution of PHA storage to acetate uptake is given in Fig. 6. Based on results from Table 2, storage contribution has been corrected with taking into account that a fixed amount of acetate has to be oxidized to provide energy for storage. The overall acetate consumption has been calculated as DPHA/YSTO where YSTO is the thermodynamic yield of storage on substrate, taken at 0.73 gCOD/gCOD (Van Aalst-van Leeuwen et al., 1997). In Fig. 6, “no storage” indicates the remaining fraction of acetate removal that is not related to storage phenomena. It is evident that only 10e15% of the acetate removal was not related to storage metabolism after that “long” acclimation to pulse feeding. Thus, even though we cannot exclude a further effect after a longer acclimation, its potential increase in storage is anyway rather low and we would argue that pulse feeding under applied operating conditions was enough strong to create full adaptation of the microbial consortium. Because this long-term effect is clearly distinguishable from a short-term effect, the question arises whether the additional effect of the long-term acclimation is due to slow adaptation of microorganisms, change of consortium microbial composition, or combination thereof.

3.3.

total) hybridized with probes specific to Alphaproteobacteria, Gammaproteobacteria, Actinomycetes, Firmicutes, Chloroflexi and CytophagaeFlavobacteriaeBacteroidetes groups. Therefore, only cell hybridized with the probe specific to Betaproteobacteria phylum were quantified as a proportion of EUB338mixbinding cells and DAPI-binding cells. As the Thauera/Azoarcus spp. was identified as main group responsible for the PHA storage within Betaproteobacteria in previous studies (Ciggin et al., submitted; Lemos et al., 2008; Serafim et al., 2006), FISH was also performed with probes specific to members of Thauera/Azoarcus group (probes AZA645 and THAU832). The analysis showed that the Thauera/ Azoarcus group comprised nearly 52% and 54% of the total bacteria and almost 57% and 58% of the Betaproteobacteria for the biomass acclimated to continuously feeding or pulse feeding, respectively (Fig. 7). FISH images of cells belonging to Thauera/Azoarcus group detected in the inoculum and in the SBRs are shown in Fig. 8. Post-FISH Nile Blue staining was performed on the samples hybridized with Thauera/Azoarcus group in order to combine the molecular identification by FISH with the proof of their capability to store intracellular polymers. Post-FISH Nile Blue staining showed that all cells binding to the Thauera/Azoarcus probes were involved in PHA storage in pulse fed SBR although

Biomolecular characterization

In order to answer the question whether the shift from continuous to pulse feeding causes a shift of the microbial composition of the aerobic consortium, fluorescence in situ hybridization (FISH) was carried out for detection of the main phyla within Bacteria domain for samples from each feeding pattern. For both feeding conditions, Bacteria were nearly 85% of the total DAPI stained cells (EUB338mix/DAPI). In both cases, the Betaproteobacteria were detected as most abundant phylum, whereas small amounts of bacterial cells (<5% in

Fig. 7 e FISH quantification of the main microbial components of biomass acclimated to either continuous or pulse feeding conditions.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

3127

Fig. 8 e FISH images of Thauera/Azoarcus-binding cells (red) and DAPI-binding cells (blue) in (a) inoculum, (b) continuously fed SBR and (c) pulse fed SBR. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

some of the Thauera/Azoarcus-binding cells contained PHA inclusion in continuously fed system. No Nile Blue staining was observed on other cells which indicated that the Thauera/ Azoarcus group was the only one which was responsible for storage. Although continuous versus pulse feeding conditions caused strong differences as for PHA storage rate and yield, no significant differences were observed in terms of microbial composition of respective steady state acclimated microbial consortia. Hence, in this specific study the observed long-term effect cannot be attributed to a slow change of microbial composition during the shift from one feeding pattern to the other one. Based on microbial characterization by FISH analyses, the same group was dominant (members of floc-former Thauera/ Azoarcus group) and preferably consumed the acetate via storage or growth mechanisms depending on acclimation to the feeding pattern, as well as it was also able to quickly adapt itself to single disturbances, even though partially. Similarly, no differences in microbial composition were also observed in a previous study (Ciggin et al., submitted) where two SBRs were operated in parallel under different feeding patterns, both started up by using the same fresh inoculum. However, in that case filamentous bacteria belonging to Alphaproteobacteria were detected as main responsible for PHA storage. As the only difference between these two sets of experiments was the inoculum utilized for the start-up of the SBRs (coming from different plants in different countries), it can be argued that the inoculum played a major role on population dynamics whereas the feeding pattern was not able to alter microbial composition by itself. Thus, in both cases, it was demonstrated the clear effect of feeding pattern on substrate removal mechanism is mostly due to the adaptation of same microorganisms more than to any shift of microbial composition. We do not claim that this is a general feature, because several literature evidences exist that continuous versus pulse feeding can also alter microbial composition. The presence or not of such an additional effect will also depend on several other parameters, including substrate composition, organic loading rate and culture residence time (Martins et al., 2003). More experimentation is probably necessary to ascertain whether microbial change can also occur on a longer time.

4.

Conclusions

The results of the experimental studies conducted by using different feeding patterns in an aerobic SBR have mainly shown that the feeding pattern strongly affects the physiological state of the biomass which in turn affects the substrate removal pathway: pulse feeding brings to a strong increase of acetate uptake rate, mostly due to an increase of both rate and yield of PHA storage. Short-term and long-term effects are clearly distinguishable, where immediate adaptation (single spike) accounts for a significant portion of overall effect that can be obtained by full acclimation (one third and two thirds of overall increase for rate and yield of storage, respectively). In this specific study, the observed long-term effect cannot be attributed to a slow change of microbial composition during the shift from one feeding pattern to the other one. Based on microbial characterization by FISH analyses, the same group was dominant (members of floc-former Thauera/ Azoarcus group) and preferably consumed the acetate via storage or growth mechanisms depending on acclimation to the feeding pattern as well as it was also able to quickly adapt itself to single disturbances, even though partially. We do not claim that this is a general result, because several literature evidences exist that continuous versus pulse feeding can also alter microbial composition. On the other hand, our study clearly shows that a progressive increase of storage capacity is not necessarily due to a shift of microbial composition and should not interpreted by itself as an evidence of it. It is also relevant that storage is significant even under almost continuous feeding and “steady state” operation, at least for the chosen substrate (acetate). In the present study, the feed lasted for 150 min out of 180 min reaction phase and the cycle also included 1 h settling and idle phase. In our opinion, this minimum disturbance of a true steady state is practically unavoidable with any plant configuration including a settler and excess sludge recycle. Thus, it is not surprising that a mixed culture always has a certain potential for storage and this explains why shortterm effects are so significant (again, at least when using the acetate as carbon source and for the chosen experimental conditions).

3128

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 1 9 e3 1 2 8

It is authors’ opinion that a more general understanding of observed effects and a more detailed distinction among short-term effects, long-term effects and microbial population dynamics can significantly help for both design and control of activated sludge processes and for PHA production processes.

Acknowledgements  The research activities of Aslı S. C¸IGGIN at the Sapienza University of Rome were supported by the grant from The Scientific and Technical Research Council of Turkey (TUBITAK).

references

Albuquerque, M.G.E., Torresa, C.A.V., Reis, M.A.M., 2010. Polyhydroxyalkanoate (PHA) production by a mixed microbial culture using sugar molasses: effect of the influent substrate concentration on culture selection. Water Research 44 (11), 3419e3433. Aulenta, F., Dionisi, D., Majone, M., Parisi, A., Ramadori, R., Tandoi, V., 2003. Effect of periodic feeding in sequencing batch reactor on substrate uptake and storage rates by a pure culture of Amaricoccus kaplicensis. Water Research 37 (11), 2744e2772. Beccari, M., Majone, M., Massanisso, P., Ramadori, R., 1998. A bulking sludge with high storage response selected under intermittent feeding. Water Research 32 (11), 3403e3413. Blackall, L.L., Tandoi, V., Jenkins, D., 1991. Continuous culture studies with Nocardia amarae from activated sludge and their implications for Nocardia foaming control. Research Journal of Water Pollution Control Federation 63 (1), 44e50. Braunegg, G., Sonnleitner, B., Lafferty, R.M., 1978. A rapid gas chromatographic method for the determination of polyb-hydroxybutyric acid in microbial biomass. European Journal of Applied Microbiology and Biotechnology 6 (1), 29e37. ¨ , Orhon, D., 2007. Effect of feeding pattern  gın, A.S., Karahan, O C¸ıg on biochemical storage by activated sludge under anoxic conditions. Water Research 41 (4), 924e934. Ciggin, A.S., Rosetti, S., Majone, M., Orhon, D., Dynamic feeding of carbon source in sequencing batch reactor affects the substrate removal mechanism more than the microbial composition of activated sludge. submitted for publication. Dionisi, D., Majone, M., Ramadori, R., Beccari, M., 2001. The storage of acetate under anoxic conditions. Water Research 35 (11), 2661e2668. Dionisi, D., Beccari, M., Di Gregorio, S., Majone, M., Petrangeli Papini, M., Vallini, G., 2005. Storage of biodegradable polymers by an enriched microbial community in a sequencing batch reactor operated at high organic load rate. Journal of Chemical Technology and Biotechnology 80 (11), 1306e1318. Frigon, D., Muyzer, G., van Loosdrecht, M.C.M., Raskin, L., 2006. rRNA and poly-b-hydroxybutyrate dynamics in bioreactors subjected to feast and famine cycles. Applied and Environmental Microbiology 72 (4), 2322e2330. Lemos, P.C., Levantesi, C., Serafim, L.S., Rossetti, S., Reis, M.A.M., Tandoi, V., 2008. Microbial characterization of

polyhydroxyalkanoates storing populations selected under different operating conditions using a cell-sorting RT-PCR approach. Applied Microbiology and Biotechnology 78 (2), 351e360. Loy, A., Maixner, F., Wagner, M., Horn, M., 2007. ProbeBase an online resource for rRNA-targeted oligonucleotide probes: new features 2007. Nucleic Acids Research 35 (Suppl. 1), 800e804. Majone, M., Massanisso, P., Carucci, A., Lindrea, K., Tandoi, V., 1996. Influence of storage on kinetic selection to control aerobic filamentous bulking. Water Science and Technology 34 (5), 223e232. Majone, M., Beccari, M., Dionisi, D., Levantesi, C., Ramadori, R., Tandoi, V., 2007. Effect of periodic feeding on substrate uptake and storage rates by a pure culture of Thiothrix (CT3 strain). Water Research 41 (1), 177e187. Majone, M., Tandoi, V., 2002. Storage polymers: role in the ecology of the activated sludge. In: Bitton, G. (Ed.), Encyclopaedia of Environmental Microbiology. Wiley, New York, pp. 3004e3014. Martins, A.M.P., Heijnen, J.J., van Loosdrecht, M.C.M., 2003. Effect of feeding pattern and storage on the sludge settleability under aerobic conditions. Water Research 37 (11), 2555e2570. Martins, A.M.P., Pagilla, K., Heijnen, J.J., van Loosdrecht, M.C.M., 2004. Filamentous bulking sludgeda critical review. Water Research 38 (4), 793e817. Ostle, A.G., Holt, J.G., 1982. Nile Blue A as a fluorescent stain for poly-b-hybroxybutyrate. Applied and Environmental Microbiology 44 (1), 238e241. Pagni, M., Beffa, T., Isch, C., Aragno, M., 1992. Linear growth and poly(b-hydroxybutyrate) synthesis in response to pulse-wise addition of the growth-limiting substrate to steady state heterotrophic continuous cultures of Aquaspirillum autotrophicum. Journal of General Microbiology 138 (3), 429e436. Serafim, L.S., Lemos, P.C., Rossetti, S., Levantesi, C., Tandoi, V., Reis, M.A.M., 2006. Microbial community analysis with a high PHA storage capacity. Water Science and Technology 54 (1), 183e188. Standard Methods, 1995. Standard Methods for the Examination of Water and Wastewater, nineteenth ed.. American Public Health Association/American Water Works Association/Water Environment Federation, Washington, DC. Van Aalst-van Leeuwen, M.A., Pot, M.A., Van Loosdrecht, M.C.M., Heijnen, J.J., 1997. Kinetic modeling of poly (-b-hydroxybutyrate) production and consumption by Paracoccus pantotrophus under dynamic substrate supply. Biotechnology Bioengineering 55 (5), 773e782. van den Eynde, E., Geerts, J., Maes, B., Verachtert, H., 1983. Influence of the feeding pattern on the glucose metabolism of Arthrobacter sp. and Sphaerotilus natans, growing in chemostat culture, simulating activated sludge bulking. European Journal of Applied Microbiology and Biotechnology 17 (1), 35e43. van Loosdrecht, M.C.M., Pot, M.A., Heijnen, J.J., 1997. Importance of bacterial storage polymers in bioprocesses. Water Science and Technology 35 (1), 41e47. van Niekerk, A.M., Jenkins, D., Richard, M.G., 1987. The competitive growth of Zooglea ramigera and type 021N in activated sludge and pure cultureda model for low F: M bulking. Research Journal of Water Pollution Control Federation 59 (5), 262e273. van Niel, E.W.J., Robertson, L.A., Kuenen, J.G., 1995. Rapid short term poly-b-hydroxybutyrate production by Thiosphaera pantothropha in the presence of excess acetate. Enzyme and Microbial Technology 17 (11), 977e982.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 2 9 e3 1 3 6

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Quantification of groundwater infiltration and surface water inflows in urban sewer networks based on a multiple model approach Christian Karpf*, Peter Krebs Dresden University of Technology, Institute for Urban Water Management, 01062 Dresden, Germany

article info

abstract

Article history:

The management of sewer systems requires information about discharge and variability of

Received 7 December 2010

typical wastewater sources in urban catchments. Especially the infiltration of groundwater

Received in revised form

and the inflow of surface water (I/I) are important for making decisions about the reha-

28 February 2011

bilitation and operation of sewer networks. This paper presents a methodology to identify

Accepted 14 March 2011

I/I and estimate its quantity. For each flow fraction in sewer networks, an individual model

Available online 21 March 2011

approach is formulated whose parameters are optimised by the method of least squares. This method was applied to estimate the contributions to the wastewater flow in the sewer

Keywords:

system of the City of Dresden (Germany), where data availability is good. Absolute flows of

Infiltration

I/I and their temporal variations are estimated. Further information on the characteristics

Inflow

of infiltration is gained by clustering and grouping sewer pipes according to the attributes

I/I

construction year and groundwater influence and relating these resulting classes to infil-

Modelling

tration behaviour. Further, it is shown that condition classes based on CCTV-data can be

Parameter fitting

used to estimate the infiltration potential of sewer pipes.

Sewer rehabilitation

1.

Introduction

Infiltration of groundwater and inflow of drainage and surface water e also referred to as infiltration/inflow (I/I) e significantly influence costs and operation of both sewer system and wastewater treatment plants (WWTP). I/I increase hydraulic loading and reduce wastewater treatment efficiency, thus resulting in additional costs and a potential deterioration of the receiving water (Ellis, 2001). Furthermore, I/I can cause heavier floods in urban areas and thus a deterioration of urban infrastructure through flood events. Groundwater infiltration may also accelerate pipe ageing and increase the endangerment of adjacent infrastructure due to backfill material flushing around pipe leaks. I/I also have some positive effects on urban infrastructure and the operation of sewer networks.

ª 2011 Elsevier Ltd. All rights reserved.

Due to I/I the total flow rate is higher and causes improved flushing, lower concentrations, and e depending on the conditions e a significant decrease of sewage temperature and an increase in nitrate in the systems. The consequences are increased transport capacity and reduced sediment build-up, a decrease of anaerobic processes, smell development and corrosion of sewer pipes. Furthermore, the drainage of groundwater might prevent the groundwater table from rising and wetting urban areas (Gustafsson, 2000), e.g. in former mining areas or where groundwater abstraction has been reduced. For the assessment of operation strategies, for the future development of the systems and for the design of sewage structures it is necessary to analyse I/I. A balancing approach is usually applied to identify and estimate the flow

* Corresponding author. Tel.: þ49 351 46333430; fax: þ49 351 46337204. E-mail addresses: [email protected] (C. Karpf), [email protected] (P. Krebs). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.022

3130

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 2 9 e3 1 3 6

components sewage, rainwater and I/I flow (De Benedittis and Bertrand-Krajewski, 2005). These methods require measurements of flow, sewage flow (drinking water consumption) and rain intensities. Further flow separation methods (Vaes et al., 2005; Wittenberg and Brombach, 2002) were applied in sewer systems to identify the base flow which consists of I/I and sewage flow. Kracht et al. (2003) introduced an I/I identification approach based on the analysis of isotope information, while Kracht and Gujer (2005) identified I/I based on the analysis of on-line wastewater component signals, e.g. continuous spectroscopic measurements of COD values. Generally it can be stated that localising I/I-sources by measurements requires high spatial resolution of measurement points or experiments. Franz and Krebs (2006) introduced a statistical approach to position measurement locations most efficiently thus minimising their number while still generating reliable information about the spatial distribution of I/I. Franz (2007) also proved a correlation between attributes (structural data) of sewer systems (e.g., diameter, material, distance to surface water) and infiltration rates. However, despite the confirmation of relationships between structural parameters and I/I, a quantification according to structural data was not possible. Modelling approaches, which describe the I/I processes mathematically, are based on conceptual (Belhadj et al., 1995; Gustafsson et al., 1999) and mechanistic modelling approaches (Gustafsson et al., 1997; Rodriguez et al., 2004). The main advantage of applying I/I-models is that these approaches can be used to estimate and assess the spatial distribution of I/I with a reasonable number of measurement points in the sewer network. It is a drawback that the set-up and the calibration of model parameters are very labour-intensive and require data recorded over a long period of time. On the basis of past studies and method developments, which are summarised in Franz (2007), we identify two main drawbacks of the methods introduced above for I/I quantification and localisation. On the one hand, methods (with the exception of model approaches) that can be used to estimate catchment-wide I/I distributions are not suited to distinguish between single I/I components, i.e. infiltration and inflow. On the other hand, methods and models which support a detailed assessment of the I/I situation in a catchment are very costly. Therefore, it was our goal to develop an I/I estimation method that can be used to generate well-resolved information about sources of I/I and their location with moderate work input. The method to be introduced is based on I/I-modelling according to essential boundary conditions. Groundwater and surface water data of a test catchment are combined with sewage flow data in an integrated I/I-model whose parameters can be relatively easily optimised by the method of least squares. Furthermore it is shown how essential information on localising groundwater infiltration can be provided by classifying sewer pipes.

2.

Method

The method is based on the combination of models for each component of the dry-weather flow (DWF) in an urban sewer

system. The approaches for each component of I/I are combined in a quasi-linear model and parameters are identified by a least-square optimisation. The time resolution of the model depends on the availability of input data and amounts to one day. In order to get more information about the infiltration potential of pipes and the spatial variability of infiltration parameters, the sewer pipes are clustered and grouped on the basis of two characteristic attributes.

2.1.

The model approaches

2.1.1.

Infiltration of groundwater

The approach to describe groundwater infiltration is based on Darcy’s law (Eq. (1)). Dhi Qin;i ¼ kf ;i $Aleak;i $ Dli

(1)

For the case of groundwater infiltration Qin,i in sewer pipe i, the potential head Dhi, which is the difference between groundwater level and water level in the pipes, can be calculated when groundwater information is available. The conductivity kf,i of the surrounding soil, the thickness Dli of the infiltration layer and the cross section area of the leaks Aleak,i are very uncertain and are crudely estimated. Further, it must be stated that other factors like leak shape and the flow conditions near the leaks (3-dimensional groundwater flow) affect the infiltration process significantly (Karpf and Krebs, submitted for publication). In order to solve the discrepancy between data availability and physical process modelling, the approach was simplified. It is assumed that the hydraulic processes near leaks can be approximated as stable during a certain period. The parameters kf,i, Aleak,i, Dli are lumped into one factor kin,i (Eq. (2)), which represents the integrative infiltration potential of a pipe (i.e. “infiltration conductivity”). Besides the parameters of Eq. (1), the factor also integrates leak shape and hydraulic conditions in the vicinity. Since kin represents the integrative infiltration conductivity of a pipe and its neighbouring conditions, the focus is not on the process around a single leak but on the specific infiltration per unit pipe length, assuming homogeneous distribution of leakage along the considered pipe length. Thus, the infiltration conductivity is multiplied with the potential head Dhi as the driving force of the process, and the infiltration rate is related to the length Li (Eq. (2)). Qin;i ¼ kin;i $Dhi $Li

2.1.2.

(2)

Inflow of surface water

For the estimation of permanent and temporary surface water inflows, two approaches were used depending on data availability. The calculation of surface water inputs during flood events via inverse flow through combined sewer overflow (CSO) structures or flooded manhole covers is described with Eq. (3) and based on the approach of Toricelli. The temporal inflow of surface water Qflood,j depends on the pressure head Dhflood,j, the area Aj of the decisive cross section and a coefficient mj that describes the shape of the openings. Since the individual description of the parameters Aj and mj is difficult, they are

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 2 9 e3 1 3 6

again integrated into a mean flood factor kflood which directly relates the inflow due to flood to the driving force of the process, which is the square root of the pressure head Dhflood,j1/2. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Qflood ¼ mj $Aj $ 2$g$Dhflood;j ¼ kflood $ Dhflood;j

(3)

The total flood-induced inflow is obtained by summing up the pressure heads of all individual inflows i.e. S(Dhflood,j1/2). Permanent inflows of surface water courses QSW,in are modelled by a simple conceptual approach (Eq. (4)). It consists of a coefficient kSW that relates the permanent inflow to the runoff in a local creek QSW, QSW;in ¼ kSW $QSW

2.2.

(4)

Multiple linear model and parameter optimisation

The I/I-models introduced above were combined into an integrated model (Eq. (5)) thus summing up to the dry-weather flow QDW as a dependent variable. Independent variables are the infiltration and inflow components. Further, a constant value, which represents the wastewater discharge, was implemented. In order to assess groundwater infiltration in more detail, the infiltration component was divided into sub-components whose identification is based on a classification according to pipe attributes. These attributes are examined and verified to be essential to realistically describe the infiltration process as a whole. The estimation of the coefficients kin1,.kin,p, kflood, kSW is based on the method of least squares (Sachs and Hedderich, 2006), using the deviation between predicted and observed values of the dry-weather flow QDW as variable whose sum of squares is to be minimised. The sewage flow QS (Eq. (5)) is assumed constant, which is justified because the diurnal variation is not represented in a calculation time-step of one day. The data sets of the dependent and independent variables shown in Eq. (5) are the input for the optimisation by the method of least squares. These data sets cover a wide range of process conditions and phenomena which result from fluctuating water levels of aquifer and surface waters and the reaction of the sewer system. The optimisation of the parameters was realised by the lm-function of the statistical software R (R Development Core Team., 2008) and the goodness-of-fit measure was the Nash-Sutcliffe coefficient NSE (Nash and Sutcliffe, 1970).

2.3.

Test system

The Dresden catchment covers an area of 98 km2 with approximately 470,000 inhabitants and industrial areas contributing significantly to the wastewater discharge. The sewer system consists of 900 km of combined sewers, 380 km of foul sewers and 340 km of stormwater pipes. The city is situated along the river Elbe with a mean flow rate of 327 m3 s1. During flood events the river water may enter the sewer system via flooded manholes and leaky combined sewer overflow gates (CSO-gates), which should cut off the sewer system from the river when the water level in the latter is higher than that in the sewer. Parts of the sewer system are temporarily or permanently influenced by the aquifer. Some creeks and springs are connected to the sewer system.

3131

For our investigations we used flow data of the WWTP inflow, rain intensity, air temperature, groundwater levels, water levels of the river Elbe and flow measurements in a local creek. The data were available from 1995 to 1999 and from 2005 to 2007 with a resolution of 1 day except for groundwater levels which were measured every 8 days. The data from 1995 to 1999 are used to optimise the parameter. Data sets from 2005 to 2007 are used for parameter validation.

2.4.

Pre-processing of data for parameter optimisation

Pre-processing of the data according to Eq. (5) includes two main steps: 1. formation of infiltration groups 2. preparation of data sets of the dependent and independent variables (Eq. (5))

2.4.1.

Formation of infiltration groups

Within the test case, the grouping was realised by defining group limits and by cluster analysis. We used the k-means-method as clustering algorithm (Stoyan et al., 1997). It was our goal to group sewers with significantly different infiltration behaviour on the basis of two available attributes of sewer pipes. Due to the primary importance of groundwater conditions for pipe deterioration and namely infiltration (Franz, 2007), the groundwater influence e parameterised as number of days per year when the groundwater table is above the sewer invert ewas used as the grouping criterion. Furthermore, the year of construction is used as an additional attribute. According to results of a multi-dimensional scaling (Franz and Krebs, 2006), the year of construction is of high importance for the infiltration potential of pipes. According to a variance analysis, Rutsch (2007) furthermore identified a significant influence of the construction period on the extent of damages in the Dresden catchment. Fig. 1 and Table 1 shows the different group configurations which were separated by constant limits (configurations IeIV) and by a cluster analysis (configurations V and VI). The configurations are obtained by dividing the population of sewer pipes into two (configurations I, III and V) and into three (configurations II, IV and VI) infiltration groups, where each grouping is based on one attribute “groundwater influence” (configurations I and II) on the one hand, and on two attributes “groundwater influence” and “construction year” (configurations III, IV, V, VI) on the other hand. Each infiltration group is considered as an individual infiltration term in Eq. (5). In order to test the influence of grouping on the model quality and the parameter plausibility, the optimisation was performed for the six group configurations (Fig. 1,Table 1) and an additional configuration with no grouping (configuration VII).

2.4.2. Preparation of dependent and independent variable data sets The parameter optimisation by the method of least squares according to Eq. (5) requires the calculation of some (dependent and independent) variables as input.

3132

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 2 9 e3 1 3 6

Fig. 1 e Grouping of sewer pipes by constant limits (configurations I e IV) and by clustering (conf. V and VI).

The determination of DWF (QDW) as dependent variable is based on flow measurements at the WWTP inlet. Further information was used to check whether the measured value can be accepted as a dry-weather flow rate. Dry-weather days are defined by a maximum rainfall intensity of 0.3 mm d1 on the considered day and the day before in order to exclude the contribution of stormwater runoff to the measured flow rate QDW. Furthermore, three days before and on the respective DWF-day, the air temperature should not be in the rangee2  C to þ2  C. If this condition is satisfied, the runoff of snow-melt water in the sewer system is excluded. The sewage flow QS

was derived from measurements of the drinking water supply. These measurements were provided by the water supply company of the city of Dresden. The calculation of the infiltration variables (potential head Dhi, see Eq. (2) and (5)) requires the estimation of the groundwater level near the sewer system elements and of the wastewater level in the pipes. Groundwater levels at the sewer pipes were obtained by spatiotemporal, linear interpolation from the measurement network, whereas surface water levels are used as an additional boundary condition. Wastewater levels in the sewer pipes were obtained from hydrodynamic

Table 1 e Attributes of infiltration groups. Configuration I II

III IV

V VI

VII

Group no.

Time below groundwater level (da1)

Mean year of constuction

Sewer length (m)

Grouping method

1 2 1 2 3 1 2 1 2 3 1 2 1 2 3 1

318 40 361 277 46 361 73 360 80 102 37 311 26 46 319 113

1919 1930 1923 1914 1929 1914 1929 1904 1901 1959 1931 1918 1994 1909 1918 1927

43397 143528 24639 13764 148522 21676 165249 14991 102974 68960 140606 46319 23520 120509 42895 186925

By constant limits

Cluster analysis

No grouping

3133

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 2 9 e3 1 3 6

simulations for each pipe, namely by simulations of the minimum and maximum water levels and the linear interpolation of values in the range between these limits, depending on the value of the dry-weather flow. Surface water levels, which are necessary to calculate surface water variable 1 (Eq. (5)), are deduced by a linear interpolation of surface water levels in the catchment. For the surface water variable 2 (Eq. (5)), data of a local creek are used.

3.

Results

3.1.

Models and parameters

The method of the least squares was applied to optimise the parameter sets of seven configurations (introduced above), where the configurations differ regarding the approach of the infiltration component. An overview of the parameter optimisation results is given in Table 2. The optimisation gives plausible results except for configuration VI, where the negative value of the infiltration factor kIN,1 is not explainable. Thus configuration VI was excluded for further analysis. The Nash-Sutcliffe model efficiency illustrates an acceptable goodness-of-fit.

3.2. Validation of the model and plausibility of infiltration parameters The models were validated with data from 2005 to 2007 (except for model configuration VI). Predicted and observed values are illustrated in Fig. 2. It can be seen that values up to 10,000 m3h1 assume a relatively even distribution around the measured values (diagonal line). Flows above 10,000 m3h1 are overestimated. A plausibility check, which ensures whether the introduction of infiltration sub-components is justified, was carried out by comparing the determined coefficients kin,i and the pipe

condition classes according to the official sewer classification of the Dresden sewerage company. However, the correlation of the infiltration factors kin,i and the pipe condition classes PC, which are available for about 28% of the sewer pipes and calculated as mean value of the class of all pipes referring to the group of pipes included in the respective model configurations according to Fig. 1 except for configuration VI, yielded a clear trend (Fig. 3). The increase of the condition class (value 1: worst condition, value 6: best condition) correlates with the decrease of the infiltration coefficient. If the outlier (marked point in Fig. 3) is excluded, a rather significant linear regression function can be identified that allows to estimate infiltration factors from the pipe condition classes. The outlier represents not a point which is completely implausible, but it is assumed that the location of the outlier point outside the main trend depend on the uncertainty of the pipe condition data. The intersection point with the X-axis at a mean condition class of about 5.3 is plausible since class 5 or better indicates a very good status and thus the absence of leakage.

3.3.

Assessment of the temporal variability of I/I

The variability of the DWF-fractions and coefficients is illustrated in Fig. 4 for the period 1995 to 1999. For the assessment of dynamics and variability, regression configuration II was used since the Nash-Sutcliffe coefficient of the model indicates the best goodness-of-fit (Table 1). With an average fraction of 85% of the I/I-volume, groundwater infiltration is the major contributor to I/I in the test catchment. Permanent surface water input generates about 14% and temporal inflows of surface water induced by flood contribute less than 1% of the I/I-volume over the 5-year period. However, in a shortterm period after a flood event (mostly in the spring season), flood-induced surface water inflow also represents an important contribution to I/I. This is indicated by the highest maximum flow rate in the 5-year period.

Table 2 e Results of the parameter optimisation based on Eq. (5) for the seven infiltration configurations defined in the Table footnote. Parameter

Configuration I

QS (m3s1)h kSW () kflood (m5/2 s1) kin,1 (ms1) kin,2 (ms1) kin,3 (ms1) NSEi

a

0.813 0.333 0.010 8.4E-06 8.0E-06 0.84

II

b

0.813 0.278 0.0139 7.4E-06 2.0E-05 4.9E-06 0.85

c

III

IVd

Ve

VIf

VIIg

0.813 0.332 0.009 8.3E-06 8.6E-06

0.813 0.356 0.009 1.4E-05 1.2E-05 1.8E-06 0.84

0.813 0.333 0.011 7.9E-06 8.4E-06

0.813 0.253 0.029 2.1E04 4.0E-05 8.7E-06 0.84

0.813 0.333 0.009 8.4E-06

0.84

0.84

0.84

a 2 infiltration sub-components classified according to the groundwater influence by the definition of group limits. b 3 infiltration sub-components classified according to the groundwater influence by the definition of group limits. c 2 infiltration sub-components classified according to the groundwater influence and construction year by the definition of group limits. d 3 infiltration sub-components classified according to the groundwater influence and construction year by the definition of group limits. e 2 infiltration sub-components classified according to the groundwater influence and construction year by a cluster analysis. f 3 infiltration sub-components classified according to the groundwater influence and construction year by a cluster analysis. g 1 infiltration component (no classification). h Sewage flow is a default value. i Nash-Sutcliffe model efficiency.

3134

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 2 9 e3 1 3 6

Fig. 2 e Observed values vs. values predicted with the regression model from 2005 to 2007.

The formation of three classes of infiltration-relevant pipes makes it possible to further interpret the characteristics of the infiltration components. The contribution of the two infiltration groups 1 and 2, representing pipes with infiltration over relatively long periods (Fig. 4), to the total groundwater infiltration volume is more than 90% with only 21% of the groundwater-influenced sewer pipes (Fig. 4). The two reasons for this are the permanent character of infiltration in these pipes and the higher infiltration coefficients, that has been determined by the regression analysis. Further, it can be seen, that the peak flows of the three infiltration groups are nearly equal. This phenomenon can be explained by the sewer length of each group (Fig. 4). During the period of maximum infiltration of group 3 the total length of the infiltration-relevant pipes is much higher than that of group 1 and 2 and thus, despite a lower infiltration factor, the infiltration rate is increased.

4.

Discussion

The validation of the approach introduced above, including calibration of an integrated I/I-model showed that the method is valuable and effective. The model introduced above provided highly plausible coefficients in the test case. It allows to estimate the dynamics of dry-weather flow e including extreme conditions e and its components. This is an important information for developing efficient investment and operation strategies and estimating the associated costs. The characteristics and the consequences of more or less permanent groundwater infiltration differ significantly from those of temporary and dynamic surface water inflow. Clustering and classification of pipes demonstrate the further potential of the method. The significant correlation

Fig. 3 e Mean pipe condition class and infiltration coefficients (The regression result is obtained without the outlier.)

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 2 9 e3 1 3 6

3135

Fig. 4 e Variability of I/I components from 1995 to 1999 in the test system (City of Dresden).

between infiltration potential and condition classes of pipes can be used to relate pipe ageing models (Baur and Herz, 2002) to infiltration predictions and, in turn, CCTV-based sewer classification can be used to prioritise rehabilitation activities with the goal to reduce groundwater infiltration. In the Dresden study, 79% of the mean I/I flow rate is expected in 3% of the total length of combined and foul sewers. However, the introduced method has some limitations, mainly with regard to the data and information required. The I/I approaches, which are linked in a multiple model, must be reduced to linear or quasi-linear models in order to support optimisation by the lm-function (linear model efunction) in the software R. On the one hand, the investigation shows that such simplified model approaches produce satisfactory results regarding the estimation of I/I components. On the other hand, a detailed process representation is impossible since the number of parameters that are necessary to feed the optimisation procedure is limited. The grouping of sewer pipes did not significantly improve the model result since the differences between the sewer pipe groups are obviously not distinct enough. Further, it is assumed that the result of the test case can be improved by a more detailed modelling of permanent and temporary inflows, if data are available. The validation of the model shows higher simulated rates. It is assumed that these higher flow rates are caused by the overestimation of temporary surface water inflows during flood events. This phenomenon can be explained by backwater effects and overloaded pipes, which are not adequately represented by detailed hydrodynamic flow simulations. An important requirement of the described methodology is the availability of groundwater and surface water data. Furthermore, flow data of the sewer network, at least from the WWTP inflow, rain intensity and temperature measurements in the catchment are necessary to identify DWF. Data are needed for periods of low, high and e if possible e extreme

boundary conditions (dry periods, floods). Normally a period of one year will be sufficiently for the model calibration. The necessary data are not available everywhere. Especially groundwater information is rarely available in a sufficiently high spatial resolution to interpolate levels near pipes, as a result uncertainty increases significantly when the infiltration rate is estimated with the introduced approaches. In such cases, conceptual approaches have to be applied for infiltration modelling. However these approaches have the drawback that infiltration parameters cannot be linked to pipe characteristics and hence a transfer to other catchments is impossible.

5.

Conclusions

The combination of model approaches to simulate I/I processes and the calibration of the parameters with the method of least squares offer opportunities to quantify the dynamics of infiltration and inflow (I/I) and their components. The main findings of the study are summarised in the following. - The simultaneous calibration of I/I approaches by the method of the least squares includes various processes and boundary conditions and a fractionation of I/I related to their sources. The parameter calibration by the least-square method does not depend on subjective estimates thus making it possible to compare different catchments. - Due to the application of mechanistic approaches, determined parameters can be interpreted in relation to structural data. In the test case, a relationship between infiltration potential and pipe condition was detected. Thus, the determination of the infiltration coefficient according to the pipe condition is reliable. This supports the transfer and

3136

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 2 9 e3 1 3 6

the comparison of results to other sewer pipes and other catchments. - In the study the groundwater influence and the year of construction were identified as meaningful indicators to estimate the infiltration potential of sewer pipes. - The multiple model and the parameters can be used for further hydrodynamic modelling and the investigation of present, past and future hydrological scenarios. Besides the quantification of I/I components the dry-weather dynamic can be assessed.

Acknowledgements This work was supported by the German Ministry of Education and Research (BMBF) within the project “Development of a 3-zone model for groundwater and infrastructure management after extreme flood events in urban areas” (FKZ: 02WH0558) and by the German Research Foundation (DFG) within the project “Modelling of sewerage exfiltration with indicator parameters” (GZ: KR 2337/3-2). Also, the support of the City of Dresden is gratefully acknowledged.

references

Baur, R., Herz, R., 2002. Selective inspection planning with ageing forecast for sewer types. Water Science and Technology 46, 389e396. Belhadj, N., Joannis, C., Raimbault, G., 1995. Modelling of rainfall induced infiltration into separate sewerage. Water Science and Technology 32, 161e168. De Benedittis, J., Bertrand-Krajewski, J.L., 2005. Infiltration in sewer systems: comparison of measurement methods. Water Science & Technology 52 (3), 219e228. Ellis, J.B., 2001. Sewer Infiltration/exfiltration and Interactions with Sewer Flows and Groundwater Quality. Interactions between Sewers, Treatment Plants and Receiving Waters in Urban Areas e INTERUBA II, pp. 311e319, Lisbon. Franz, T. (2007). Spatial classification methods for efficient infiltration mesurements and transfer of measuring results. Dissertation, Dresden University of Technology, Institute for Urban Water management.

Franz, T., Krebs, P., 2006. Statistical methods towards more efficient infiltration measurements. Water Science & Technology 54, 153e160. Gustafsson, L.-G., Winberg, S., Refsgaard, A., 1997. Towards a distributed physically based model description of the urban aquatic environment. Water Science and Technology Vol. 36, 89e93. Gustafsson, L.G., 2000. Alternative Drainage Schemes for Reduction of Inflow/infiltration e Prediction and Follow-up of Effects with the Aid of an Integrated Sewer/aquifer Model. In: 1st International Conference on Urban Drainage via Internet. Gustafsson, L.G., Hernebring, C., Hammarlund, H., 1999. Continuous Modelling of Inflow/infiltration in Sewers with MouseNAM e 10 years of Experiences. Third DHI Software Conference. Kracht, O., Gresch, M., deBenedittis, J., Prigiobbe, V., Gujer, W., 2003. Stable isotopes of water as natural tracer for infiltration into urban sewer systems. Geophysical Research Abstracts 8, 07852. Kracht, O., Gujer, W., 2005. Quantification of infiltration into sewers based on time series of pollutant loads. Water Science & Technology 52 (No. 3), 209e218. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models part I e A discussion of principles. Journal of Hydrology 10, 282. R Development Core Team., 2008. R: A Language and Environment for Statistical Computing Vienna, Austria, ISBN 3-900051-07-0. Rodriguez, F., Morena, F., Andrieu, H., 2004. Developement of a Distributed Hydrological Model Based on Urban Databanks e Production Processes of URBS. In: Conference on Urban Drainage Modelling 2004, pp. 561e570 (Dresden). Rutsch, M. (2007). Assessment of sewer leakage by means of exfiltration measurements and modelling tests. Dissertation. Dresden University of Technology, Institute for Urban Watermanagement. Sachs, L., Hedderich, J., 2006. Angewandte Statistik. (Applied Statistics), twelfth ed. Springer-Verlag. ISBN-10 3-540-32160-8. Stoyan, D., Stoyan, H., Jansen, U., 1997. Umweltstatistik. (Environmental Statistics). Teubner Verlagsgesellschaft, Leipzig. ISBN 3-8154-3526-9. Vaes, G., Willems, P., Berlamont, J., 2005. Filtering Method for Infiltration Flow Quantification. In: 10th International Conference on Urban Drainage 2005 Copenhagen/Denmark, 21e26.8.2005. Wittenberg, H., Brombach, H., 2002. Hydrological Determination of Groundwater Drainage by Leaky Sewer Systems. In: Int. Conf. on Water Resources and Environment Research. Dresden University of Technology, pp. 138e143.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Kinetic modelling of nitrogen and organics removal in vertical and horizontal flow wetlands Tanveer Saeed*, Guangzhi Sun Department of Civil Engineering, Building 60, Monash University, Clayton, Victoria 3800, Australia

article info

abstract

Article history:

This paper provides a comparative evaluation of the kinetic models that were developed to

Received 29 November 2010

describe the biodegradation of nitrogen and organics removal in wetland systems. Reaction

Received in revised form

kinetics that were considered in the model development included first order kinetics,

7 March 2011

Monod and multiple Monod kinetics; these kinetics were combined with continuous-

Accepted 15 March 2011

stirred tank reactor (CSTR) or plug flow pattern to produce equations to link inlet and outlet

Available online 23 March 2011

concentrations of each key pollutants across a single wetland. Using three statistical parameters, a critical evaluation of five potential models was made for vertical and hori-

Keywords:

zontal flow wetlands. The results recommended the models that were developed based on

Constructed wetlands

Monod models, for predicting the removal of nitrogen and organics in a vertical and

Denitrification

horizontal flow wetland system. No clear correlation was observed between influent

Kinetics

BOD/COD values and kinetic coefficients of BOD5 in VF and HF wetlands, illustrating that

Nitrification

the removal of biodegradable organics was insensitive to the nature of organic matter.

Organics removal

Higher effluent COD/TN values coincided with greater denitrification kinetic coefficients,

Wetland modelling

signifying the dependency of denitrification on the availability of COD in VF wetland systems. In contrast, the trend was opposite in HF wetlands, indicating that availability of NO3-N was the main limiting step for nitrogen removal. Overall, the results suggested the possible application of the developed alternative predictive models, for understanding the complex biodegradation routes of nitrogen and organics removal in VF and HF wetland systems. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Subsurface flow wetlands are engineered systems, which mostly employ gravel as substrate to support the growth of plants, and wastewater flows vertically or horizontally through the substrate where it comes into contact with microorganisms, living on the surfaces of plant roots and substrate (Cooper et al., 1996; Kadlec and Knight, 1996). As such, subsurface flow constructed wetlands are further divided into two groups, according to the flow direction inside the packed media: (1) vertical flow, and (2) horizontal flow systems.

A large number of physical, chemical and biological processes are involved in these systems influencing each other (Langergraber et al., 2009), which are not fully understood to date due to lack of appropriate models. The most widely employed modelling equations (such as, Kickuth equation and KeC* models) give only an exponential profile of inlet and outlet pollutant concentrations (Rousseau et al., 2004), without considering the full range of pollutant variability of engineered wetlands (Garcia et al., 2010). The incapability of the traditional first order models, for capturing the diversity encountered in wetland systems could

* Corresponding author. Tel.: þ61 399055577; fax: þ61 39904944. E-mail address: [email protected] (T. Saeed). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.031

3138

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

be attributed to their over simplified postulations such as: (a) the concentration of reactant (i.e. pollutant) is limited, and the presence of catalysts (i.e. microorganisms) is in excess (Mitchell and McNevin, 2001), and (b) wastewater is assumed to follow plug flow approach in these systems, packed with substrates. These basic assumptions of first order models often do not fit with the complex operating conditions of subsurface flow wetlands. For example classic denitrification route, which is still relied upon as the major mechanism for nitrogen removal (Bachand and Horne, 2000; Reilly et al., 2000) in wetland systems, exhibits complicated interrelation between denitrifier microorganisms, availability of NO3-N, and organic carbon (Hamersley and Howes, 2002; Lavrova and Koumanova, 2010). The critical interdependency between limiting reactants (i.e. NO3-N, organic carbon) and the catalysts (i.e. biomass) demonstrate that presence of excess biomass may not be a prevalent condition in these systems for nitrogen removal, particularly at ‘under capacity’ conditions. Subsequently, the assumption of idealized plug flow condition may not be appropriate for VF wetlands. These systems are usually batch fed (in flushes) with wastewater, which is likely to be diverged from the bulk flow direction by the unsaturated packed media during downward flow. To overcome the inadequacy of first order models, a number of mechanistic models have been developed recently   (Langergraber and Sim unek, 2005; Mayo and Bigambo, 2005; Wang et al., 2009; Giraldi et al., 2010). Nevertheless, the presence of numerous empirical parameters in these models might lead to inefficient removal prediction (Rousseau et al., 2004), under different circumstances. Efficient performance of the predictive models predominantly depends on linking two important factors, which are believed to have played a major role on pollutant removal performances in wetland systems: (a) the kinetics of biological degradation, and (b) the hydrodynamic behaviour. Monod kinetics, providing a close interrelation on substrate availability and growth of biomass, could be an alternative solution for developing more realistic wetland predictive models. Recent research studies by Saeed and Sun (2011b) and Sun and Saeed (2009) provided a critical evaluation of Monod kinetics, for predicting dynamics of nitrogen and organics in wetland systems. These studies combined CSTR and plug flow pattern with Monod approach, for correlating inlet and outlet nitrogen and organics values in VF and HF systems, respectively. The developed alternative predictive models matched predicted values and measured data more precisely, compared with traditional first order kinetics for both wetland systems. However, the proposed alternative models need to be verified by substantial amount of wetland data, before being accepted as general predictive tools for wetland systems. Subsequently, the efficacy of the proposed models, in terms of predicting NO3-N and COD removal in HF wetlands has not been explored to date, despite these systems are effective for removing these target pollutants (Garcia et al., 2004; Vymazal et al., 2006). The acceptance of the alternative models (i.e. Monod models) is inherently dependent on its preciseness, in terms of predicting the removal pathways of all the target pollutants in constructed wetlands. In addition, comparative

performance evaluations of VF and HF wetlands (on the context of kinetic models) are scarce in literatures, thereby hindering effective implementation and arrangements of these systems for wastewater treatment. The study has been planned to provide a comparative evaluation between first order and Monod kinetics, combined with two different hydrodynamics pattern (i.e. CSTR and plug flow pattern) in VF and HF wetlands, respectively. As such, the objectives of this study are two-folds: (a) to explore the precision of the Monod approach for modelling the dynamics of nitrogen and organics removal in VF and HF wetland systems, and (b) to observe the performance differences between the two systems on the context of kinetic modelling.

2.

Model development

2.1.

Configurations of the wetland systems

2.1.1.

Lab-scale systems

The experimental data for the verification of VF kinetic models was collected from first and third stage VF wetlands (in total eight VF wetland units) of six hybrid systems, employed for the treatment of synthetic domestic wastewater. All the wetland columns were made of Perspex glass. The height and diameter of each vertical flow column were 995 mm and 90 mm, respectively. These wetlands had different substrates such as gravel, organic wood mulch, mixture of gravelewood mulch and zeolite. For the verification of HF kinetic models, data had been collected from the second stage HF wetlands of these hybrid systems (in total three HF wetland units), with gravel substrate. The length, width and height of each horizontal flow wetland (made of Perspex glass) were 600 mm, 200 mm and 800 mm, respectively. All the wetland columns were planted with young Phragmites Australis. Average hydraulic retention time (HRT) in VF wetland columns ranged between 0.35 d and 1.14 d; for HF wetland columns, HRT varied between 6.6 d and 14 d. A more detailed description of substrates, media depth and compositions of the synthetic wastewater is available in Saeed and Sun (2011a).

2.1.2.

Full-scale systems

For the verification of the predictive models developed for HF systems, data had also been collected from 80 full-scale HF systems in the UK, treating domestic wastewater. Details of these wetlands configuration, substrate, plantation and wastewater characteristics have been provided in Sun and Saeed (2009).

2.2.

Target pollutants

Kinetic models have been developed to link the inlet and outlet concentrations of four target pollutants across a vertical flow wetland, such as NH4-N, NO3-N, BOD5, and COD. For horizontal flow wetlands, models have been developed to link the inlet and outlet concentrations of NO3-N, BOD5, and COD.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

3139

hinder complete mixing inside the porous media. This is also supported by Sun and Saeed (2009), where the authors compared plug flow and CSTR hypothesis (for modelling BOD removal in HF systems), and concluded the former approach to be more effective for matching the performance data. For these reasons, the mixing of wastewater in the packed media (of HF systems) had been assumed to follow ideal plug flow pattern. Fig. 1 gives the schematic illustrations of two alternative modelling approaches that were undertaken for VF wetlands after combining biological degradation kinetics and hydrodynamic pattern. Fig. 2 illustrates the two alternative modelling approaches for the HF wetlands.

2.4.

Kinetic models for vertical flow wetlands

2.3.

Model assumptions

Three alternative kinetic models have been developed to link the inlet and outlet concentrations of NH4-N, NO3-N, BOD5 and COD across a single VF wetland. The models have been developed by combining first order, Monod and multiple Monod kinetics with CSTR flow pattern.

2.3.1.

Biological degradation kinetics

2.4.1.

Fig. 1 e Modelling approaches for predicting pollutant dynamics in VF wetlands.

To describe pollutant degradation kinetics in VF and HF wetlands, two approaches had been adopted, and their accuracies compared: approach 1 e assuming that the degradations of target pollutants follow first order kinetics; and approach 2 e assuming the degradations follow Monod or multiple Monod kinetics.

2.3.2.

Model 1: first order kinetics with CSTR flow pattern

First order kinetics in a reactor is expressed as: dC ¼ kv Cout dt

(1)

where, Cout ¼ outlet pollutant concentration (mg/L), kv ¼ volumetric rate constant (d1). CSTR flow pattern in a reactor can be expressed as:

Hydrodynamic pattern

Since VF wetland systems are batch fed (in flushes) with wastewater, local flow direction of the wastewater at any random position is likely to diverge rapidly (by the saturated and unsaturated matrix) from the bulk flow direction during downwards flow; this resembling CSTR flow pattern, rather than plug flow (Kadlec, 1994; Wynn and Liehr, 2001; Saeed and Sun, 2011b). In this study, a VF wetland had been assumed to function as a single CSTR reactor. For HF wetland systems, large amount of wastewater is usually applied which flows horizontally through the saturated substrates. It is likely that the flow inside the saturated media could be divided into several laminar streamlines, due to longer flowing distance. Such flow characteristics might

dC 1 1 þ Cin ¼ Cout dt s s

(2)

where, Cin ¼ inlet pollutant concentration (mg/L), s ¼ hydraulic retention time (days). Combining Eqs. (1) and (2) results simplified first order kinetics combined with CSTR flow pattern (in terms of areal rate constant K1, m/d) as expressed in Eq. (3), for correlating influent and effluent nitrogen and organics values in the VF wetlands. K1 ¼

qðCin  Cout Þ Cout

where, q ¼ hydraulic loading rate (m/d).

Fig. 2 e Modelling approaches for predicting pollutant dynamics in HF wetlands.

(3)

2.4.3. Model 3: multiple Monod kinetics combined with CSTR flow pattern Multiple Monod kinetics assumes that more than one substrate can limit the rate of pollutant degradation. The multiple Monod kinetics can be expressed in the following equation (Bitton, 1994; Henze et al., 1995; Wong et al., 2003). dC Cout1 Cout2 ¼ Kmax Chalf 1 þ Cout1 Chalf 2 þ Cout2 dt

(8)

In Eq. (8) Cout1 and Cout2 represent the outlet concentrations (mg/L) of two substrates that may limit the biodegradation

39.6 24.0 54.0 30.0 684.0 3.2 12.8 18.0 18.6 120.0 0.3 6.0 9.8 9.6 74.2 0.01 1.7 5.1 0.2 37.5

75% percentile Median 25% percentile Minimum

0.0 0.1 0.4 0.0 9.5 104 99 83 37 89 22.2 1.7 26.5 34.6 85.8 80.0 12.1 84.0 102.0 1516.0 34.0 1.9 35.4 49.5 79.0 20.6 0.9 24.0 32.0 54.0

Eq. (7) can be used to predict nitrogen and organics degradation in VF columns. For the first step of NH4-N transformation (i.e. from NH4-N to NO2-N) during nitrification, the half saturation constant value (Chalf for Nitrosomonas degradation) has been measured as 0.05 mg/L (Verstraete and Vaerenbergh, 1986). As such, 0.05 mg/L can be used as the Chalf value in Eq. (7) for nitrification. For the denitrification of NO3-N, half saturation nitrate constant in the Monod kinetics has been reported as 0.14 mg/L (Wiesmann, 1994). For heterotrophic biodegradable organics (BOD5) removal, half saturation degradation constant is recommended as 60 mg/L for wastewater treatment (Metcalf and Eddy, 2003). For COD, half saturation COD constant can be taken as 20 mg/L, as suggested by Vaccari et al. (2006) for sewage treatment.

3.6 0.1 8.8 9.4 33.3

(7)

0.0 0.0 0.7 0.2 3.1


 qðCin  Cout Þ Chalf þ Cout Cout

86 99 83 27 71

K2 ¼

N

Combining volumetric maximum pollutant removal rates (Kmax, g/m3/d) with h and e results areal maximum pollutant removal rates (K2, g/m2/d) as expressed in Eq. (7), to correlate inlet and outlet pollutant concentrations (Saeed and Sun, 2011b).

Mean

(6)

Maximum

Ahe Q

75% percentile

s¼

Median

Eq. (5) is expressed in terms of hydraulic retention time s (units in days), and Kmax represents maximum pollutant mass removal per m3 of wetland volume per day. The relation between hydraulic retention time, area (A, m2), depth (h, m), porosity (e) of packed media and inlet discharge (Q, m3/d) can be expressed in Eq. (6).

25% percentile

(5)

Table 1 e Statistical analyses of the pollutant concentration profile across experimental VF wetland columns.

Cin  Cout Cout ¼ Kmax s Chalf þ Cout

Minimum

Combining CSTR flow pattern (Eq. (2)) with Monod kinetics (Eq. (4)) gives Eq. (5), which links inlet and outlet pollutant concentrations.

Effluent pollutant concentration (mg/L)

(4)

Influent pollutant concentration (mg/L)

dC Cout ¼ Kmax Chalf þ Cout dt

Maximum

Monod kinetics, which contains the parameters of substrate concentration (Cout), half saturation constant of limiting substrate (Chalf, mg/L) and maximum volumetric pollutant removal rates (Kmax, g/m3/d), can be expressed as:

NH4-N NO3-N TN BOD5 COD

Mean

2.4.2. Model 2: Monod kinetics combined with CSTR flow pattern

2.4 7.6 12.8 10.9 118.0

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

N

3140

3141

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

Table 2 e Statistical analyses of the NO3-N and COD concentration profile across experimental HF wetland columns. Influent pollutant concentration (mg/L) N Minimum COD 47 NO3-N 44

Effluent pollutant concentration (mg/L)

25% Median 75% Maximum Mean N Minimum 25% Median 75% Maximum Mean percentile percentile percentile percentile

9.8 0.4

55.5 2.6

106.0 9.9

214.0 14.9

684.0 23.5

159.0 47 9.7 44

rate of target pollutant, while Chalf1 and Chalf2 are the half saturate constants (mg/L) for these substrates, respectively. Combining Eqs. (2) and (8) gives Eq. (9), which link inlet and outlet pollutant concentration and K3 value (maximum areal pollutant removal rate K3, g/m2/d), as expressed by Saeed and Sun (2011b). K3 ¼



 qðCin  Cout Þ Cout1 þ Chalf 1 Cout2 þ Chalf 2 Cout1 Cout2

(9)

2.4.3.1. Equation for nitrification (NH4-N removal). When Eq. (9) is used to predict nitrification in VF wetlands, the limit of nitrification rate by NH4-N concentration and dissolved oxygen value can be considered as two limiting substrates. Eq. (8) becomes: dC CNH4out CDOout ¼ Kmax Chalf þ CNH4out CDOout þ CDOhalf dt

(10)

where, CNH4out ¼ effluent NH4-N concentration (mg/L), CDOout ¼ effluent DO concentration (mg/L), CDOhalf ¼ half saturation constant of DO (mg/L). Combining Eqs. (10) and (2) gives:

10.1 0.0

24.0 0.18

33.3 1.2

52.3 2.8

349.0 5.6

54.2 1.6

account of the NO3-N level, (2) using COD as an indicator of organic carbon for denitrification, and (3) taking into account of the inhibitory oxygen constant DOi (denitrification ceases above this constant value as this process occurs in anoxic/ anaerobic environment). The interrelation of these factors can be expressed through the following equation (Wong et al., 2003). CDOi dC CNO3 out CCODout ¼ Kmax CNO3 out þ Chalf CDOi þ CDOout CCODout þ CCODhalf dt

(12)

where, CNO3out ¼ effluent NO3-N concentration (mg/L), CDOi ¼ inhibitory oxygen constant (mg/L), CCODout ¼ effluent COD concentration (mg/L), and CCODhalf ¼ half saturation COD constant (mg/L). Combining multiple Monod kinetics (Eq. (12)) with CSTR flow pattern (Eq. (2)) gives Eq. (13), which includes half saturation constant for nitrate as 0.14 mg/L (Wiesmann, 1994), half saturation COD constant for denitrification (as carbon source) as 5 mg/L (Henze et al., 1995), and inhibitory oxygen constant as 0.2 mg/L (Gujer et al., 1999).

  q CNO3 in  CNO3 out ðCCODout þ 5ÞðCDOout þ 0:2Þ CNO3out þ 0:14 K3 ¼ CNO3 out CCODout 0:2 (13)


 q CNH4 in  CNH4 out ðCNH4 out þ 0:05ÞðCDOout þ 0:25Þ K3 ¼ CNH4 out CDOout

(11)

where, CNH4in ¼ influent NH4-N concentration (mg/L). In Eq. (11), the half saturation oxygen constant ranges from 0.25 to 1.3 mg/L for nitrification. Half saturation constant of NH4-N transformation has been taken as 0.05 mg/L (Verstraete and Vaerenbergh, 1986), and half saturation DO constant being 0.25 mg/L (Hawkes, 1983; Verstraete and Vaerenbergh, 1986; Stenstrom and Song, 1991), as given in Eq. (11).

2.4.3.2. Equation for denitrification (NO3-N removal). For denitrification in experimental VF wetland columns, the limiting factors could be: (1) availability of nitrate, (2) availability of organic carbon, and (3) dominancy of anoxic/ anaerobic conditions. These three limiting factors can be expressed in the multiple Monod kinetics by: (1) taking into

where, CNO3in ¼ influent NO3-N concentration (mg/L).

2.4.3.3. Equation for BOD5 removal. For biodegradable organics, the dependency of biodegradable organics (BOD5) degradation on the presence of oxygen is accounted for in Eq. (14). dC CBOD5 out CDOout ¼ Kmax Chalf þ CBOD5 out CDOout þ CDOhalf dt

(14)

where, CBOD5out ¼ effluent BOD5 concentration (mg/L). For the modelling of BOD5 removal in VF wetlands, combining Eqs. (14) and (2) gives Eq. (15), which takes into account of the values of half saturation constant for biodegradable organics (60 mg/L) and half saturation oxygen constant (0.2 mg/L) for heterotrophic organic degradation (Vaccari et al., 2006).

Table 3 e Statistical analyses of input hydraulic load (q, m3/m2d) across experimental VF and HF wetland columns. VF wetland columns N Minimum q 66

0.20

HF wetland columns

25% Median 75% Maximum Mean N Minimum 25% Median 75% Maximum Mean percentile percentile percentile percentile 0.29

0.33

0.48

0.90

0.40

65

0.01

0.02

0.02

0.03

0.06

0.02

3142

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

Table 4 e Statistical analyses of the BOD5 concentration profile across 80 HF wetland systems in the UK. Influent BOD5 concentration (mg/L) N Minimum BOD5 80

K3 ¼

2.3

Effluent BOD5 concentration (mg/L)

25% Median 75% Maximum Mean N Minimum 25% Median 75% Maximum Mean percentile percentile percentile percentile 12.0

53.4

153.0

332.1


 
q CBOD5 in  CBOD5 out CBOD5 out þ 60 ðCDOout þ 0:2Þ CBOD5 out CDOout

90.0 80

2.5.2. (15)

2.4.3.4. Equation for COD removal. For COD modelling in VF wetlands, Eq. (15) can be expressed as following, after considering half saturation COD values (20 mg/L) and half saturation oxygen values (0.2 mg/L).
 q CCODin  CCODout ðCCODout þ 20ÞðCDOout þ 0:2Þ K3 ¼ CCODout CDOout

(16)

Model development for horizontal flow wetlands

Two kinetic models have been developed for correlating inlet and outlet NO3-N, BOD5 and COD concentrations across a single HF wetland. Such development of the predictive models includes combination of First order, Monod kinetics with plug flow pattern. It should be noted that, nitrate and organics (BOD5 and COD) modelling with multiple Monod kinetics (i.e. incorporating the effect of oxygen) does not suit the HF wetlands, due to strong reducing conditions in these systems (Garcia et al., 2010).

2.5.1.

Model 4: first order kinetics with plug flow pattern

The application of idealized plug flow conditions in Eq. (1) yields: ZCout

dC ¼ Kv C

Zt

Cin

dt

(17)

0

Arranging Eq. (17) in terms of areal rate constant (K4, m/d) results first order plug flow model (i.e. Kickuth equation) for correlating influent and effluent nitrate and organics values across HF systems, as expressed in Eq. (18). Cin Cout 1=q

ln K4 ¼

10.6

32.0

75.0

20.0

Model 5: Monod kinetics with plug flow pattern

 Chalf þ Cout dC ¼ Kmax dt Cout

(19)

Applying the boundary conditions of idealized plug flow pattern in the above equation yields:  Zt ZCout  Chalf þ Cout dc ¼ Kmax dt Cout

Cin

(20)

0

Arranging Eq. (20) in terms of areal maximum pollutant removal rates (K5, g/m2/d) yields Eq. (21). The equation expresses the combination of Monod kinetics with plug flow pattern, to correlate inlet and outlet pollutant values in HF wetlands (Sun and Saeed, 2009).

where, CCODin ¼ influent COD concentration (mg/L).

2.5.

3.4

Eq. (4) can be rearranged in the form of Eq. (19), for developing Monod plug flow model. 

where, CBOD5in ¼ influent BOD5 concentration (mg/L).

0.6

(18)

Cin Cin  Cout þ Chalf ln Cout K5 ¼ 1=q

(21)

Eq. (21) may be used to correlate inlet and outlet NO3-N, BOD5 and COD values in HF wetlands. The half saturation constant for these target pollutants can be used as 0.14, 60 and 20 mg/L, respectively, as discussed previously.

2.6.

Cautions notes

It should be noted that the development of models 2, 3 and 5 from Monod kinetics entails significant simplifications. Firstly, the combination of bacterial biomass concentration and the maximum substrate utilization rate in classic Monod equation, into a single parameter Kmax in Monod CSTR and plug flow models, requires the presence of a stable microbial population in the treatment wetlands. As such, these alternative models may not exhibit satisfactory results in these engineered systems, where nutrient uptake by plants contribute to significant removal (over biodegradation), particularly in temperate climate conditions (Tanner, 2001). Secondly, the half saturation constants of the models are typical values for wastewater treatment and might have to be calibrated, for modelling pollutant removal dynamics in the wetland systems that treat high strength wastewater (such as strong industrial and agricultural effluents).

Table 5 e Statistical analyses of input hydraulic load (q, m3/m2d) across 80 HF wetland systems in the UK. HF wetland columns

q

N

Minimum

25% percentile

Median

75% percentile

Maximum

Mean

80

0.03

0.06

0.12

0.25

1.5

0.22

3143

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

3.

Model comparison and verification

Table 6 e Statistical parameters used in evaluating model performance.

3.1.

The collection of data for model verification

Parameter R2

For VF wetland models, NH4-N, NO3-N, BOD5, COD and hydraulic loading (q) data were collected from eight VF wetland columns. These data (Tables 1 and 3) were used to compare the accuracy of the developed models (models 1, 2 and 3). For HF wetland modelling (models 4 and 5), NO3-N, COD and hydraulic loading (q) data (Tables 2 and 3) were collected from three experimental gravel HF columns. BOD5 data from experimental HF columns was not incorporated in the models due to inadequate sampling numbers. In order to investigate the accuracy of the developed models for BOD5 prediction in HF systems, hydraulic loading and BOD5 data have been collected from 80 full-scale HF wetland systems operated in the UK (Sun and Saeed, 2009). Among the 80 systems, the performances of 76 systems were recorded in a database established by Constructed Wetland Association in the UK (CWA, 2008). The performance data of the other four systems were collected from the literature (Coombes and Collett, 1995; Green et al., 1997; Cooper, 2001). For each HF wetland, the mean of BOD5 performance data (for the whole sampling period) was used in model verification. As such, each system gave a single data point and all together 80 data points were collected for the HF modelling work. All the 80 HF wetlands were operated for the treatment of domestic sewage in the UK. Tables 1e3 summarize the experimental data (hydraulic load, influent and effluent concentration of nitrogen, organics) collected from eight VF and three HF wetland columns. These tables indicate the range of experimental data to be used for the comparison of the kinetic models (from two approaches). Tables 4 and 5 summarize BOD5 data and hydraulic loading profile across 80 HF beds (operated for domestic sewage treatment), for the evaluation of BOD5 prediction by the models.

3.2.

Comparison of the five kinetic models

The accuracy and reliability of the five predictive models, for correlating influent and effluent nitrogen and organics values, had been compared using three statistical parameters: coefficient of determination (R2), relative root mean square error (RRMSE), and model efficiency (ME). The description of these parameters is illustrated in Table 6. To allow statistical analyses, the three predictive models for VF columns (Eqs. (3), (7) and (9)) had been arranged into the general form of Eqs. (22) and (23).

K¼

f ðCin ; Cout ; qÞ Cout

(22)

K¼

f ðCin ; Cout ; qÞ Cout1  Cout2

(23)

Subsequently, the two predictive models ((18) and (21)) for HF systems had been arranged in the general form of Eq. (24), for statistical analyses.

RRMSE

ME

Description

Mathematical definitiona

P Coefficient of ½ N ðX  XÞðY  YÞ2 R2 ¼ PN i¼1 i 2 PN i 2 determination: i¼1 ðXi  XÞ i¼1 ðYi  YÞ ranging 0e1. R2 measures the extent of linear correlation between two sets of data (Xi, Yi). A higher R2 value corresponds to a stronger linear vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u N relation between u1 X t b i Þ2 ðY  Y these two datasets. N i¼1 i Relative root mean RRMSE ¼ Y square error: ranging 0N. Measuring the differences between the values predicted by the model and observed values. RRMSE near 0 corresponds to close match. P b 2 Model efficiency: ME ¼ 1  N i¼1 ðYi  Y i Þ = N ranging N to 1. X ðYi  YÞ2 Measuring the i¼1 variations accounted for by the model. A higher ME value corresponds to closer match between predicted and observed values.

a Xi, Yi ¼ two sets of data; X, Y ¼ mean values of X, Y data series; b i ¼ predicted value. Yi ¼ observed value; Y

K¼

f ðCin ; Cout Þ 1=q

(24)

The general form of Eqs. (22)e(24) allowed comparison between the actual performance data (from the VF and HF wetland systems) and the data predicted by the five predictive models, with the statistical parameters (R2, RRMSE, ME) indicating the overall deviations between the experimental and predicted data. For general Eq. (22), the value of f ðCin ; Cout ; qÞ and (Cout) can be calculated from its performance data for each wetland; as such, three data are obtainable from each bed for each parameter: (Cout) (i.e. Xi in Table 6), f ðCin ; Cout ; qÞ (i.e. Yi) and K b i ). These data have been used to calculate the (Cout) (i.e. Y values of R2, RRMSE and ME in Table 6, to examine the accuracy of each design equation. Similar procedure had been adapted for general Eqs. (23) and (24). Subsequently, the slope of the linear regression line allows the calculation of areal rate constants, and maximum areal pollutant removal rates.

3.3.

Model verification results: VF systems

3.3.1.

Model 1: first order CSTR model

When CSTR flow pattern was combined with first order kinetics, Eq. (22) could be expressed in explicit form as Eq. (3).

3144

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

R2=0.1 RRMSE=13.1 ME=-203.3

First order kinetics for VF columns

40 q*(CNO3in-CNO3out )

q*(CNH4in-CNH4out )

60

40

20

R2=0.01 RRMSE=1.2 ME=-0.8

30

First order kinetics for VF columns

20

k1=0.5 m/d

10

k1=0.9 m/d 0

0

10

20

30

40

0

50

0

5

10

CNH4out

R2=0.6 RRMSE=0.7 40 ME=0.6

First order kinetics for VF columns

40 q*(CBOD5in-CBOD5out )

q*(CCODin-CCODout )

50

30 20 10

k1=0.3 m/d 50

20

25

First order kinetics for VF columns

2

R =0.2 RRMSE=0.9 ME=0.1

30 20 10

k1=0.6 m/d

0 0

15

CNO3out

100

0

150

0

10

20

CCODout

30

40

CBOD5out

Fig. 3 e Regression of first order CSTR model for correlating inlet and outlet nitrogen and organics values in VF wetland columns. The dotted lines are 95% confidence band, indicating the band contains true regression fit line.

500

800

q*(CNH4in-CNH4out )* (CNH4out +0.05)

q*(CNO3in-CNO3out )* (CNO3out +0.14)

Monod kinetics

2

R =0.9 RRMSE=0.8 600 ME=0.9

for VF columns

400

200

Monod kinetics 2 R =0.3 for VF columns RRMSE=1.2 ME=0.3

400

300 2

k2=6.2g/m /d 200

100

2

k2=14.2 g/m /d

0 0

10

20

30

40

0 0

50

5

10

6000

ME=0.7

Monod kinetics

q*(CBOD5in-CBOD5out )* (CBOD5out +60)

RRMSE=0.8

for VF columns

4000

2000

2

k2=29.8 g/m /d

0

25

Monod kinetics 2

R =0.4 RRMSE=0.7 2000 ME=0.3

for VF columns

1000

2

k2=54.3g/m /d

0 0

20

3000

2

R =0.7 q*(CCODin-CCODout )* (CCODout +20)

15

CNO3out

CNH4out

50

100

CCODout

150

0

10

20

30

40

CBOD5out

Fig. 4 e Regression of Monod kinetics with CSTR flow pattern for correlating inlet and outlet nitrogen and organics values in VF wetland columns. The dotted lines are 95% confidence band, indicating the band contains true regression fit line.

3145

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

Fig. 5 e Regression of multiple Monod CSTR model for correlating inlet and outlet nitrogen and organics values in VF wetland columns. The dotted lines are 95% confidence band, indicating the band contains true regression fit line.

3

R2 =0.38 RRMSE=0.56 ME=0.36

30 20 10

k4=0.05 m/d

0 0

20

R2 =0.01 RRMSE=0.80 ME=-0.20

HF Columns lnCCODin-lnCCODout

lnCNO3in-lnCNO3out

40

40

60

80

HF Columns

2

1

k4 =0.02m/d

0

100

0

20

40

1/q

60

80

100

1/q

lnCBOD5in-lnCBOD5out

4 R2 =0.14

80 HF systems in the UK

RRMSE=0.62 ME=-1.22

3 2 1

k4 =0.09 m/d

0 0

10

20

30

1/q Fig. 6 e Regression of first order plug flow model for correlating inlet and outlet NO3-N, COD and BOD5 values in lab-scale and 80 full-scale HF wetlands, respectively. The dotted lines are 95% confidence band, indicating the band contains true regression fit line.

3146

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

and BOD5 values in VF wetland columns (indicated by R2 and ME parameters), compared with the simple first order reaction model (Fig. 3).

Fig. 3 shows the regression of f(Cin, Cout, q) vs. Cout, to derive the values of K1 when Eq. (3) was used for correlating inlet and outlet concentrations of NH4-N, NO3-N, COD and BOD5, as a wastewater passed through a VF wetland. The values of statistical parameters R2, RRMSE and ME were embedded in Fig. 3, to indicate the deviations between experiment data and values predicted by Eq. (3). The dotted lines embedded below and above of regression lines are 95% confidence band, containing true regression fit line. According to Fig. 3, first order CSTR model (Eq. (3)) was inefficient (as indicated by the statistical parameters), for matching NH4-N, NO3-N and BOD5 performance data across VF wetland columns. However, the model illustrated better correlation for matching inlet and outlet COD data across VF columns.

3.3.2.

3.3.3.

Model 3: multiple Monod CSTR model

Fig. 5 demonstrates the accuracy of combining multiple Monod kinetics with CSTR flow pattern (Eq. (9)), for correlating influent and effluent nitrogen and organics values in experimental VF wetland columns. As observed in Fig. 5, multiple Monod kinetics showed higher accuracy (over Monod kinetics) as indicated by the statistical parameters, in terms of matching NO3-N, COD and BOD5 profiles across VF columns. However, lower statistical correlation values were observed for predicting nitrification in VF wetland columns modelled with multiple Monod kinetics, compared to Monod CSTR model.

Model 2: Monod CSTR model

20 15 10 5

k5 =0.25 g/m2 /d

0 0

20

40

60

80

100

1/q

500

Model verification results: HF systems

3.4.1.

Model 4: first order plug flow model (Kickuth equation)

Fig. 6 shows the correlation of f (Cin, Cout) vs. 1/q, when Eq. (24) was represented in explicit form by Eq. (18) (i.e. first order plug flow model). As observed in Fig. 6, all the statistical parameters showed very weak correlation, for matching inlet and

HF columns

R2 =0.55 RRMSE=0.47 ME=0.64

3.4.

CCODin-CCODout +20ln(CCODin/CCODout )

25

CBOD5in-CBOD5out +60ln(CBOD5in/CBOD5out )

CNO3in-CNO3out +0.14ln(CNO3in/CNO3out )

Fig. 4 shows the correlation of f (Cin, Cout, q) vs. Cout, when Eq. (22) was represented in explicit form by Eq. (7) (resulted from Monod CSTR model). Maximum pollutant removal rates (K2, g/m2/d) were derived from the slope of the regression fit line in Fig. 4. According to Fig. 4, Eq. (7) provided better accuracy for matching experimental and predicted NH4-N, NO3-N, COD

600

HF columns

400

200

k5 =3.8g/m2 /d

0 0

20

40

60

1/q

80 HF systems in the UK

R2 =0.45 RRMSE=0.56 ME=0.20

400

R2 =0.30 RRMSE=0.70 ME=0.30

300 200 100 k5 =12.48 g/m2 /d

0 0

10

20

30

1/q

Fig. 7 e Regression of Monod kinetics with plug flow pattern for correlating inlet and outlet NO3-N, COD and BOD5 values in lab-scale and 80 full-scale HF wetlands, respectively. The dotted lines are 95% confidence band, indicating the band contains true regression fit line.

0.4 0.5

0.2 0.5 0.4

0.3 0.7 0.3

This research Vaccari et al. (2006) Sun and Saeed (2009) Metcalf and Eddy (2003) This research Wiesmann (1994)

0.5 0.7 0.5 This research Metcalf and Eddy (2003) Vaccari et al. (2006)

0.8 0.9 0.8

0.8 1.0

0.9 0.8 0.9

NO3-N

BOD5

COD Single HF Wetland

BOD5

COD

ME RRMSE

0.8 NO3-N

Coefficient sources

This research Verstraete and Vaerenbergh (1986) This research Wiesmann (1994) Henze et al. (1995) Gujer et al. (1999) This research Vaccari et al. (2006) NH4-N Single VF Wetland

In order to investigate the difference between nitrogen and organics kinetic coefficients (g/m2/d) derived from Monod (model 2) and multiple Monod kinetics (model 3) for VF columns, statistical analyses have been carried out using software GraphPad Prism (version 5.03). Various statistical tests (KolmogoroveSmirnov test, D’Agostino and Pearson omnibus normality test, and ShapiroeWilk normality test) were performed to check the distribution of data approximated normality (data was log-transformed if necessary). The results (data approximated to normality) were accepted when a ¼ 0.05. The statistical analyses in this paper include paired t test and ManneWhitney test. Detailed description of these tests is given in Table 8. The choice of these two tests was dependent on whether the data approximated to normality. If data approximated normality, kinetic coefficients derived from Monod and multiple Monod kinetics across VF columns were analysed through paired t test, to illustrate the statistical difference ( p < 0.05) between the two groups for a given parameter (for example NO3-N). If data did not approximate normality, ManneWhitney test (instead of paired t test) was performed, to illustrate the difference of kinetic coefficients between two groups (i.e. resulted

qðCin  Cout ÞðChalf þ Cout Þ KNH4-N ¼ 14.2 g/m2/d KNH4 N ¼ Cout Chalf ¼ 0.05 mg/L qðCNO3 ;in  CNO3 ;out ÞðCCODout þ CCODhalf ÞðCDOout þ CDOi ÞðCNO3 ;out þ Chalf Þ KNO3 N ¼ KNO3-N ¼ 300.0 g/m2/d CNO3 ;out CCODout CDOi Chalf ¼ 0.14 mg/L CCODhalf ¼ 5 mg/L CDOi ¼ 0.2 mg/L qðCin  Cout ÞðCout þ Chalf ÞðCDOout þ CDOhalf Þ KCOD ¼ 29.2 g/m2/d Korganics ¼ Cout CDOout Chalf ¼ 20 mg/L CDOhalf ¼ 0.2 mg/L KBOD5 ¼ 65.7 g/m2/d Chalf ¼ 60 mg/L CDOhalf ¼ 0.2 mg/L Cin Cin  Cout þ Chalf ln Cout KNO3 N=organics ¼ KCOD ¼ 3.8 g/m2/d 1=q Chalf ¼ 20 mg/L KBOD5 ¼ 12.48 g/m2/d Chalf ¼ 60 mg/L KNO3-N ¼ 0.25 g/m2/d Chalf ¼ 0.14 mg/L

Statistical analyses of the kinetic coefficients

Pollutant

3.6.

Table 7 e Summary of best fit equations and values of coefficients.

Overall, Figs. 3e5 indicate that multiple Monod kinetics with CSTR flow pattern (model 3) correlated inlet and outlet NO3-N, BOD5 and COD degradation closely in VF wetland columns. However, Monod CSTR model (i.e. model 2) matched NH4-N performance data in VF columns more closely. For the HF wetland, Monod kinetics with plug flow pattern showed better performances over first order plug flow kinetics for matching denitrification and organics data, as observed in Figs. 6 and 7. Table 7 summarize the best fit equation for predicting outlet pollutant concentrations from inlet concentrations and associated coefficient values. All equations in Table 7 need to be solved numerically, or by a simple optimization program, when they are used in method design.

Model equation

Coefficient values

3.5. Overall verification results and summary of coefficients

R2

Model 5: Monod plug flow model

Fig. 7 indicates the correlation of NO3-N, COD and BOD5 performance data (by Monod kinetics with plug flow pattern) in HF columns and 80 HF wetland systems (Sun and Saeed, 2009). According to Fig. 7, Monod plug flow model (i.e. model 5) matched denitrification and organics data more closely (indicated by the statistical parameters), compared to first order plug flow model (Fig. 6).

Systems

3.4.2.

Statistical correlation

outlet COD and BOD5 values in HF columns and 80 full-scale HF wetlands (Sun and Saeed, 2009), respectively. These results indicate the inefficiency of first order plug flow model (i.e. model 4), for predicting organics removal in HF systems. Subsequently, the model showed better correlation, in terms of matching denitrification performance data (compared to organics removal).

0.6

3147

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

3148

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

Table 8 e Description of the statistical analyses. Statistical Analyses Paired t test

ManneWhitney test

Definition Paired t test is used to compare means on the same subject under different circumstances. In this paper, paired t test had been used to compare the kinetic coefficients (obtained from Monod/multiple Monod kinetics) of the VF wetland columns, to observe the impact of single and multiple limiting substrates on same subjects. ManneWhitney test is a non-parametric test, which had been performed when the assumptions of paired t test was not satisfied. It is usually used to compare two population means that come from the same population.

from Monod and multiple Monod kinetics) according to the process of paired t test. Table 9 summarizes statistical analyses (ManneWhitney test and paired t test) of the kinetic coefficients (g/m2/d), derived from Monod kinetics and multiple Monod kinetics in experimental VF columns (Figs. 3e5). As observed in Table 9, significant difference was observed ( p < 0.05) between the kinetic coefficients (obtained from the two models) of NO3-N, BOD5, COD degradation in VF columns. These statistical differences further support the difference of the two models in terms of predicting pollutant removal in VF wetland columns.

4.

Discussion

4.1. Interrelation between loading and removal rates: dependency on limiting substrates The correlation between nitrogen, organics loading and their removal rates in VF and HF wetland systems has been illustrated in Figs. 8 and 9. Figs. 8 and 9 denote increase of removal rate with loadings in VF and HF wetlands, and are in agreement with the results of the previous research studies (Albuquerque et al., 2009; Calherios et al., 2009). The positive correlation between loading increment and removal rates coincides with the assumption of both traditional first order and Monod kinetics. However, the dependency of biomass growth rate on either single or multiple limiting substrates (i.e. assumption of Monod approach), fits more precisely with such correlation

Table 9 e Statistical analyses of the kinetic coefficients. *p values from ManneWhitney test and paired t test KNH4-N KNO3-N KCOD KBOD5

NS <0.0001 0.0017 <0.0001

*NS: No significant difference and p < 0.05 Significant difference.

between pollutant loading and removal rates as a result of intensified biological metabolism at greater loading. The influence of single or multiple limiting substrates on nitrogen and organics removal is further confirmed through the matching of nitrification and organics performance data, by Monod (Fig. 4) and multiple Monod CSTR models (Fig. 5), respectively, in VF wetland columns. These results illustrate the dependency of nitrification on the availability of NH4-N, whereas organics removal was dependent on the availability of both organic matter and DO. The dissimilarity between the kinetics of nitrification and heterotrophic organics removal (in terms of oxygen utilization) in VF wetland columns could be attributed to the growth metabolism of the microorganisms, associated with these biochemical reactions. The specific growth rate of autotrophic bacteria is reported to be lower than heterotrophic microorganisms, due to their restricted energy yielding metabolism (Grady et al., 1999). Subsequently, higher specific growth rate is associated with greater oxygen utilization (by heterotrophic microbes), which could be higher than the atmospheric oxygen diffusion rate in the wetland media, causing the organics degradation rate to be limited by the transfer of oxygen flux through the void spaces of the media. Closer matching of denitrification performance data by multiple Monod CSTR model (in VF columns) (Fig. 5), indicates the complex interrelation between denitrifiers, availability of NO3-N, organic carbon (Lavrova and Koumanova, 2010), and anoxic/anaerobic environments (Kadlec and Knight, 1996; Cerezo et al., 2001). For HF systems, Monod plug flow model (Fig. 7) showed better performances to correlate inlet and outlet COD, NO3-N and BOD5 values in HF columns and 80 HF full-scale systems in the UK (Sun and Saeed, 2009). These observations elucidate the accuracy of the alternative model for predicting denitrification and organics removal in HF systems, compared with the widely implemented Kickuth equation (Fig. 6). It should be noted that, all the statistical parameters (R2, RRMSE, ME) illustrated better performances for predicting NO3N and COD removal in experimental VF columns (Figs. 4 and 5), when compared with experimental and full-scale HF systems (Figs. 6 and 7). Such performance disparity between VF and HF systems (operated for the treatment of domestic wastewater) illustrates more intensified microbiological degradation in the former wetlands, for the removal of pollutants (Figs. 4 and 5).

4.2. Applicability of the alternative predictive models for modelling the complex biodegradation routes of nitrogen and organics removal in wetland systems Better statistical correlation indicates the applicability of the developed Monod models, in terms of capturing the diverse interaction between different chemical and environmental parameters (Figs. 4, 5 and 7) that usually occurs in wetland systems. Subsequently, the major drawbacks of first order models such as, simple correlation of inlet and outlet pollutant concentration, and dependency on steady-state conditions (Langergraber et al., 2009; Rousseau et al., 2004) hinder these equations to capture pollutant removal diversity encountered in wetland systems (Garcia et al., 2010). This is also supported by this research study, where the first order models illustrated

3149

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

40 TN Removal Rates g/m2/d

R2=0.9

30 g/m2/d

NH4-N Removal Rates

40

20 10 0 0

10

20

30

R2=0.7

30 20 10 0

40

0

10

Input NH4-N (g/m2/d) 40

R2=0.9

BOD5 Removal Rates g/m2/d

COD Removal Rates g/m2/d

40

20

30

40

Input TN (g/m2/d)

30 20 10

R2=0.8

30 20 10 0

0 0

20

40

60

80

0

100

10

Input COD (g/m2/d)

20

30

40

50

2

Input BOD5 (g/m /d)

Fig. 8 e Correlation of input nitrogen and organics loading vs. removal rates in VF wetland columns.

25

HF columns

1.0

0.5

0.0 0.0

2

R =0.9 COD Removal Rates g/m2/d

R2 =0.96

g/m2/d

NO3-N Removal Rates

1.5

HF columns

20 15 10 5 0

0.5

1.0

0

1.5

10

30

2

Input COD (g/m /d)

Input NO3-N (g/m2/d) 25

80 HF wetland systems in the UK

R2=0.9

20 g/m2/d

BOD5 Removal Rates

20

15 10 5 0 0

10

20

30

2

Input BOD5 (g/m /d) Fig. 9 e Correlation of input NO3-N, COD and BOD5 loading vs. removal rates in HF wetland systems.

3150

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

250

80

VF Columns

100

2 g/ m /d

150

40 20

50 0 0.0

HF wetlands in the UK

60 KBOD5

2 g/ m /d

KBOD5

200

0.2

0.4

0.6

0.8

0 0.3

1.0

0.4

0.5

0.6

0.7

Influent BOD/COD value

Influent BOD/COD value 2

Fig. 10 e Plot of BOD5 kinetic coefficients (g/m /d) from alternative approach against organic loading in experimental VF columns and 80 HF wetlands in the UK.

poor performances in terms of matching inlet and outlet nitrogen and organics data (Figs. 3 and 6), despite of higher removal rates at greater loading, as observed in Figs. 8 and 9. It should be noted that the developed alternative predictive models have certain advantages over other wetland modelling approaches (developed through general modelling software), for example MAPLE 5.1 (Wang et al., 2009), STELLA II (Wynn and Liehr, 2001), and biokinetic models (Sklarz et al., 2010) due to: (a) reduced number of guessed parameters, and (b) their potential application for modelling the dynamics of nitrogen and organics removal in both VF and HF wetland systems.

4.3. Inadequacy of the first order rate constants for the design of wetland systems The BOD5 rate constants (KBOD5) currently employed for the design of HF wetland systems in the UK range from 0.067 to 0.1 m/d (Cooper, 1990). The value of KBOD5 (from first order Kickuth equation e model 4) derived from 80 HF wetland systems in the UK (Sun and Saeed, 2009) falls within this range. However, weaker statistical correlation, as illustrated by the Kickuth equation over Monod plug flow model (Figs. 6 and 7) for matching BOD5 performance data in these systems, indicates the inefficiency of the rate constants in terms of estimating effective bed area. Such inadequacy of the

4.4.

Effect of BOD/COD ratio on organics degradation

The ratio BOD/COD is an indicator of organic matter degradability. A ratio of 0.5 or greater indicates that the organic matter is easily biodegradable, whereas, a ratio lower than 0.3 signifies that the organic matter is difficult to degrade in wastewater treatment (Metcalf and Eddy, 2003). Fig. 10 shows such plot of influent BOD/COD values vs. biodegradable organics coefficient (KBOD5), derived from multiple Monod kinetics for VF columns, and Monod kinetics for 80 HF systems in the UK (Sun and Saeed, 2009). It should be noted that among the 80 reed beds in this study, 22 have recorded values of BOD/ COD ratios. Fig. 10 indicates no clear correlation between BOD5 kinetic coefficients and influent BOD/COD ratios for VF and HF wetlands. Such observations are in agreement with the results observed from other HF wetlands (Caselles-Osorio and Garcia, 2006), despite of higher mean influent COD concentration across their systems. These findings indicate that the efficiency of organics removal in HF wetlands appears to be insensitive to the nature of organic matter, whether the organic matter is easily or slowly degradable.

NO3-N Removal Rates

600 400 200 0

gNO3-N/m2d

1.5

VF Columns

800 gNO3-N/m2d

NO3-N Removal Rates

1000

widely implemented rate constants might result incorrect bed configurations, enhancing clogging and wastewater shortcircuiting.

HF columns

1.0

0.5

0.0

0

20

40

Effluent COD/TN value

60

0

10

20

30

Effluent COD/TN value

Fig. 11 e Correlation of effluent COD/TN values vs. NO3-N removal rates in VF and HF wetland columns.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

4.5.

Effect of COD/N ratio on denitrification

Fig. 11 indicates plot of effluent COD/N values vs. denitrification coefficients (from alternative approach) in VF and HF columns. As observed in Fig. 11, higher denitrification kinetic coefficients coincided with greater effluent COD/TN values for VF columns. These results coincide with the findings illustrated by the Monod approach (Fig. 5), enlightening the dependency of denitrification on COD availability in VF wetlands. In contrast, the trend was completely opposite in HF columns, where higher effluent COD/TN values coincided with lower denitrification kinetic coefficients. These results suggest that, availability of NO3-N was the main limiting steps for denitrification in HF systems, as illustrated by positive correlation between input NO3-N loading and removal rates (Fig. 9).

5.

Conclusions

(1) Compared with simple first order kinetics, multiple Monod kinetics (when combined with CSTR flow pattern) allowed closer match between predicted and experimental data of NO3-N, COD and BOD5 removal in a vertical flow wetland. (2) The combination of Monod kinetics and CSTR flow pattern allows closer match for predicting NH4-N removal by nitrification in a vertical flow system. (3) Monod plug flow model matched inlet and outlet NO3-N, COD and BOD5 data closely (over first order kinetics) across lab-scale HF columns, and 80 full-scale horizontal flow wetlands, operated in the UK. (4) The statistical correlation showed better performances (as indicated by the alternative models) for predicting NO3-N and COD removal in experimental VF columns, compared to HF columns. Such results illustrate the contribution of different pollutant removal mechanisms over heterotrophic biodegradation, in the latter systems. (5) No clear correlation was observed between influent BOD/ COD values and kinetic coefficients of BOD5 in VF and HF wetlands. Such findings indicate that the removal of biodegradable organics was insensitive to the nature of organic matter, and was more dependent on the availability of electron acceptors (to foster heterotrophic degradation). (6) Higher effluent COD/TN values coincided with greater denitrification kinetic coefficients for VF columns, coinciding with the results illustrated by the alternative approach of the kinetic modelling. In contrast, the trend was completely opposite in HF columns, where higher effluent COD/TN values coincided with lower denitrification kinetic coefficients. (7) The investigation on the validity of the developed models, for predicting nitrogen and organics removal in wetland systems treating high strength wastewater are recommended for future studies. In addition, incorporating temperature coefficient in Monod and multiple Monod kinetics, to reflect the temperature effect on pollutant removal in these systems (when operated in different seasons) are also recommended.

3151

Acknowledgements The authors would like to thank Monash Research Graduate School and Department of Civil Engineering, Monash University, Australia for sponsoring Tanveer Saeed’s PhD studies.

references

Albuquerque, A., Oliveira, J., Semitela, S., Amaral, L., 2009. Influence of bed media characteristics on ammonia and nitrate removal in shallow horizontal subsurface flow constructed wetlands. Bioresource Technology 100, 6269e6277. Bachand, P.A.M., Horne, A.J., 2000. Denitrification in constructed free water surface wetlands: I. Very high nitrate removal rates in a macrocosm study. Ecological Engineering 14, 9e15. Bitton, G., 1994. Wastewater Microbiology. Wiley-Liss, Inc, New York, USA. Calherios, C.S.C., Rangel, A.O.S.S., Castro, P.M.L., 2009. Treatment of industrial wastewater with two-stage constructed wetlands planted with Typha latifolia and Phragmites australis. Bioresource Technology 100, 3205e3213. Caselles-Osorio, A., Garcia, J., 2006. Performance of experimental horizontal subsurface flow constructed wetlands fed with dissolved or particulate organic matter. Water Research 40, 3603e3611. Cerezo, R.G., Suarez, M.R., Vidal-Abarca, M.R., 2001. The performance of a multi stage system of constructed wetlands for urban wastewater treatment in a semiarid region of SE Spain. Ecological Engineering 16, 501e517. Coombes, C., Collett, R.J., 1995. Use of constructed wetland to protect bathing water quality. Water Science and Technology 32 (3), 149e158. Cooper, P.F., 1990. European Design and Operation Guidelines for Reed Bed Treatment Systems. Prepared by EC/EWPCA Emergent Hydrphyte Treatment Systems Expert Contact Group. Water Research Centre, Swindon, UK. Cooper, P.F., Job, G.D., Green, M.B., Shutes, R.B.E., 1996. Reed Beds and Constructed Wetlands for Wastewater Treatment. WRc Publications, UK. Cooper, P., 2001. Constructed wetlands and reed-beds: mature technology for the treatment of wastewater from small populations. Journal of the Chartered Institute of Water and Environment Management 15, 79e85. CWA, 2008. Constructed Wetlands Database Version 11.01. Constructed Wetland Association, Rugeley. Garcia, J., Aguirre, P., Mujeriego, R., Haung, Y., Ortiz, L., Bayona, J.M., 2004. Initial contaminant removal performance factors in horizontal flow reed beds used for treating urban wastewater. Water Research 38, 1669e1678. Garcia, J., Rousseau, D.P.L., Morato, J., Lesage, E., Matamoros, V., Bayona, J.M., 2010. Contaminant removal process in subsurface-flow constructed wetlands: a review. Critical Reviews in Environmental Science and Technology 40 (7), 561e661. Giraldi, D., de’Michieli Vitturi, M., Iannelli, R., 2010. FITOVERT: a dynamic numerical model of subsurface vertical flow constructed wetlands. Environmental Modelling & Software 25, 633e640. Grady, J.R.C.P.L., Daigger, G.T., Lim, H.C., 1999. Biological Wastewater Treatment, second ed,. Marcel Dekker, Inc. Green, M., Friedler, E., Ruskol, Y., Safrai, I., 1997. Investigation of alternative method for nitrification in constructed wetlands. Water Science & Technology 35 (5), 63e70.

3152

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 3 7 e3 1 5 2

Gujer, W., Henze, M., Mino, T., van Loodrecht, M., 1999. Activated sludge model no. 3. Water Science and Technology 39 (1), 183e193. Hamersley, M.R., Howes, B.L., 2002. Control of denitrification in a septage-treating artificial wetland: the dual role of particulate organic carbon. Water Research 36, 4415e4427. Hawkes, H.A., 1983. Activated sludge. In: Curds, C.R., Hawkes, H.A. (Eds.), Ecological Aspects of Used Water Treatment, vol. 1. Academic Press, London. Henze, M., Harremoes, P., Jansen, J.L.C., Arvin, E., 1995. Wastewater Treatment: Biological and Chemical Processes. Springer-Verlag, Berlin Heidelberg New York. Kadlec, R.H., 1994. Detention and mixing in free water wetlands. Ecological Engineering 3, 345e380. Kadlec, R.H., Knight, R.L., 1996. Treatment Wetlands. CRC Press, Boca Raton, USA.   Langergraber, G., Sim unek, J., 2005. Modeling variably saturated water flow and multi component reactive transport in constructed wetlands. Vadoze Zone J 4 (4), 924e938. Langergraber, G., Giraldi, D., Mena, J., Meyer, D., Pen˜a, M., Toscano, A., Brovelli, A., Korkusuz, E.A., 2009. Recent developments in numerical modelling of subsurface flow constructed wetlands. Science of the Total Environment 407 (13), 3931e3943. Lavrova, S., Koumanova, B., 2010. Influence of recirculation in a lab-scale vertical flow constructed wetland on the treatment efficiency of landfill leachate. Bioresource Technology 101, 1756e1761. Mayo, A.W., Bigambo, T., 2005. Nitrogen transformation in horizontal subsurface-flow constructed wetlands I: model development. Physics and Chemistry of the Earth 30, 658e667. Metcalf and Eddy, 2003. Wastewater Engineering Treatment and Reuse, fourth ed. McGraw-Hill, USA. Mitchell, C., McNevin, D., 2001. Alternative analysis of BOD removal in subsurface flow constructed wetlands employing Monod kinetics. Water Research 35 (5), 1295e1303. Reilly, J.F., Horne, A.J., Miller, C.D., 2000. Nitrate removal from a drinking water supply with large free surface constructed wetlands prior to groundwater recharge. Ecological Engineering 14, 33e47. Rousseau, D.P.L., Vanrolleghem, P.A., Pauw, N.D., 2004. Modelbased design of horizontal subsurface flow constructed treatment wetlands: a review. Water Research 38, 1484e1493. Saeed, T., Sun, G., 2011a. Enhanced denitrification and organics removal in hybrid wetland columns: comparative experiments. Bioresource Technology 102 (2), 967e974.

Saeed, T., Sun, G., 2011b. The removal of nitrogen and organics in vertical flow wetland reactors: predictive models. Bioresource Technology 102 (2), 1205e1213. Sklarz, M.Y., Gross, A., Soares, M.I.M., Yakirevich, A., 2010. Mathematical model for analysis of recirculating vertical flow constructed wetlands. Water Research 44, 2010e2020. Stenstrom, M.K., Song, S.S., 1991. Effect of oxygen transport limitation on nitrification in the activated sludge process. Research Journal of the Water Pollution Control Federation 63, 08e219. Sun, G., Saeed, T., 2009. Kinetic modelling of organic matter removal in 80 horizontal flow reed beds. Process Biochemistry 44 (7), 717e722. Tanner, C.C., 2001. Plants as ecosystem engineers in subsurface flow treatment wetlands. Water Science & Technology 44 (11), 9e17. Vaccari, D.A., Storm, P.F., Alleman, J.E., 2006. Environmental Biology for Engineers and Scientists. John Wiley & Sons, Inc., Hoboken, New Jersey. Verstraete, W., Vaerenbergh, E.V., 1986. Aerobic activated sludge. In: Rehm, H.J., Reed, G., Schonborn, W. (Eds.), Biotechnology. Aerobic Degradations, vol. 8. VCH, Germany, pp. 43e112. Vymazal, J., Greenway, M., Tonderski, K., Brix, H., Mander, U., 2006. Constructed wetlands for wastewater treatment. In: Verhoeven, J.T.A., Beltman, B., Bobbink, R., Whigham, D.F. (Eds.), Wetlands and Natural Resource Management, vol. 190, pp. 69e96. Wang, Y., Zhang, J., Kong, H., Inamori, Y., Xu, K., Inamori, R., Kondo, T., 2009. A simulation model of nitrogen transformation in reed constructed wetlands. Desalination 235 (1e3), 93e101. Wiesmann, U., 1994. Biological nitrogen removal from wastewater. In: Fletcher, A. (Ed.), Advances in Biochemical Engineering Biotechnology, vol. 51. Springer-Verlag, Berlin and Heidelberg, pp. 113e154. Wong, C.H., Barton, G.W., Barford, J.P., 2003. The nitrogen cycle and its application in wastewater treatment. In: Mara, Duncan, Horan, Nigel (Eds.), The Handbook of Water and Wastewater Microbiology. Elsevier Ltd. Wynn, T.M., Liehr, S.K., 2001. Development of a constructed subsurface-flow wetland simulation model. Ecological Engineering 16, 519e536.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Laboratory assessment of factors affecting soil clogging of soil aquifer treatment systems P. Pavelic a, P.J. Dillon b,*, M. Mucha c, T. Nakai b,d, K.E. Barry b, E. Bestland d a

International Water Management Institute, Hyderabad, India CSIRO Land and Water, Adelaide, Australia c Veolia Water (OEWA), Leipzig, Germany d Flinders University, School of Chemistry, Physics and Earth Sciences, Adelaide, Australia b

article info

abstract

Article history:

In this study the effect of soil type, level of pre-treatment, ponding depth, temperature and

Received 26 November 2010

sunlight on clogging of soil aquifer treatment (SAT) systems was evaluated over an eight

Received in revised form

week duration in constant temperature and glasshouse environments. Of the two soil types

8 March 2011

tested, the more permeable sand media clogged more than the loam, but still retained an

Accepted 15 March 2011

order of magnitude higher absolute permeability. A 6- to 8-fold difference in hydraulic

Available online 12 April 2011

loading rates was observed between the four source water types tested (one potable water and three recycled waters), with improved water quality resulting in significantly higher

Keywords:

infiltration. Infiltration rates for ponding depths of 30 cm and 50 cm were higher than 10 cm,

Clogging

although for 50 cm clogging rates were higher due to greater compaction of the clogging

Biofilm

layer. Overall, physical clogging was more significant than other forms of clogging. Microbial

Recycled water

clogging becomes increasingly important when the particulate concentrations in the source

SAT

waters are reduced through pre-treatment and for finer textured soils due to the higher specific surface area of the media. Clogging by gas binding took place in the glasshouse but not in the lab, and mechanical clogging associated with particle rearrangement was evident in the sand media but not in the loam. These results offer insight into the soil, water quality and operating conditions needed to achieve viable SAT systems. Crown Copyright ª 2011 Published by Elsevier Ltd. All rights reserved.

1.

Introduction

Soil Aquifer Treatment (SAT), the practice of intermittently ponding treated wastewater within shallow basins and recovering the recharged water from nearby production wells for non-potable uses, is becoming an increasingly important part of contemporary water management strategies. It enables water reuse thereby preserving valuable fresh water resources and avoids wastewater disposal problems. SAT is particularly suited to seasonally arid landscapes that feature permeable, free-draining soils (Fox et al., 2001; Bouwer, 2002; Toze et al., 2002; Dillon et al., 2006).

It is common practice to operate SAT infiltration basins under a cyclic wetting and drying regime so as to restore infiltration rates and assist in the removal of nitrogen, phosphorus, organic carbon and other contaminants present in the infiltrate (Lance et al., 1980; Kopchynski et al., 1996; Fox et al., 2001; Rauch-Williams and Drewes, 2006). This also enables the organic-rich surficial deposits to quickly desiccate for restoration of infiltration rates in subsequent wetting cycles. Movement of the infiltrated wastewater through the unsaturated zone is largely controlled by the formation of a low conductivity clogging layer within the upper layer of the soil (Rice, 1974; Pell and Nyberg, 1989). The clogging layer can

* Corresponding author. E-mail address: [email protected] (P.J. Dillon). 0043-1354/$ e see front matter Crown Copyright ª 2011 Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.027

3154

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

develop due to physical clogging caused by the filtration of suspended solids present in the recharge water. In addition, biological clogging caused by the accumulation of bacterial cells and the production of extracellular polysaccharide supported by bio-available nutrients is also important (Pell and Nyberg, 1989; Baveye et al., 1998). One of the major challenges in designing SAT schemes is to evaluate the level of source water pre-treatment required, particularly in terms of suspended solids and bio-available nutrient removal. Too little pre-treatment may result in inadequate rates of recharge but excessive pre-treatment may be economically prohibitive. Another design variable that impacts on infiltration rates is the ponding depth of water within the recharge basin. Conflicting observations have been reported between ponded depth and infiltration rate, with some studies proposing a positive correlation due to the enhanced hydraulic gradient across the upper soil profile and the increase in wetted surface area of the basin (Houston et al., 1999). Others have proposed a negative correlation due to enhanced compression of clogging layers (Rice, 1974; Bouwer and Rice, 1989). Due to the unique combination of source water quality, soil properties and climatic conditions at each site, it is difficult to predict a priori the level of pre-treatment and depth of ponding needed to achieve acceptable rates of infiltration. As such, these are generally deduced by trial and error through laboratory studies or in the field (Bouwer, 2002). Field scale experiments are often costly and must contend with environmental factors such as soil and climate variability. Columns studies, because of their greater ease and reduced cost, have been conducted in a range of settings, including at constant temperature (Taylor and Jaffe, 1990; Rinck-Pfeiffer et al., 2000), in uncontrolled laboratory environments (Lance, 1977) and in glasshouse environments (Rice, 1974). The extent to which results derived from controlled (constant temperature) laboratory conditions can be extrapolated to field conditions remains largely unknown. Most column studies compare different site selection and operational factors individually, such as soil type (Rice, 1974), or ponding depth (Bouwer and Rice, 1989), but little work has been performed to systematically test a range of factors on clogging of porous media. In this paper we present the findings of a detailed laboratory column experiment that examines clogging and infiltration rates as affected by different soil types, levels of pre-treatment, ponding depths and the role of climate variability, particularly the role of sunlight and temperature fluctuations. By establishing the relative significance of each factor, a firmer understanding of clogging processes can emerge that can improve the design and operation of SAT systems.

2.

Materials and methods

2.1.

Experimental design and operation

The experiment involved the operation of 34 columns in parallel in two different environments: 28 in 20
C constant temperature (CT) and 6 in glasshouse (GH) (Fig. 1). The majority of the treatments were performed at CT in the

absence of light. At CT, two soil types were subjected to four different qualities of water. Also at CT, the effect of three different ponding depths typical for SAT operations (10, 30 and 50 cm), was tested on both soil types with one water quality. A subset of columns, including two extended columns, were operated with one water type in the GH to more closely represent actual field conditions. Each treatment was conducted on duplicate columns (apart from the extended columns). Columns were manufactured from polycarbonate tubing with an internal diameter of 1.9 cm and supported at the base with stainless steel mesh (mesh size of 0.16 mm  0.1 mm for sand and 0.1 mm  0.065 mm for loam). The total length of the 32 short columns was 11 cm with a soil depth of 10 cm and of the remaining 2 extended columns was 71 cm with 70 cm soil depth. A saturated headspace of 1 cm was common to all columns to allow surface clogging to be visually observed. Flow was vertically downwards with free-drainage at the outlet end of the columns. The internal diameter was of the smallest feasible scale to minimize the volume requirements for feed water that in most cases had to be transported over long distances, but still meet a minimum column diameter to grain size ratio of 30 to minimize edge effects (Smith and Dillon, 1997). The 10 cm soil depth for the majority of the columns reflects the focus on shallow clogging, given that the severest clogging typically occurs within the first few centimeters of the soil surface and limits rates of groundwater recharge (Rice, 1974; Bouwer, 2002). This was subsequently verified by hydraulic head observations in the extended columns. All columns were operated for four cycles each consisting of 7 days of wetting followed by 7 days of drying. This cycling regime was selected because it has been commonly used in SAT operations (Rice and Bouwer, 1984; Bouwer, 2002) and column studies (Kopchynski et al., 1996). The experiment ran for a duration of eight weeks (10 October to 5 December 2003). After the bulk soils were homogenized, sieved and the gravel fraction >3 mm excluded, columns were dry packed to a uniform bulk density consistent with field measurements (Table 1) by adding small portions of soil and gently tapping the base to create as uniform a grain size distribution as possible. Next, the soil was saturated by slowly wetting from the bottom then allowed to stand overnight to enable entrapped air to dissolve and particles to settle. The initial saturated hydraulic conductivity (Ko) was then determined from a brief (5 min) infiltration test with potable quality water. Columns with distinctly outlying values of Ko were repacked and retested. The coefficient of variation in Ko for the columns used was <10% for sands and <20% for loams. Bulk water samples were stored in the dark at 4
C prior to use. Subsamples from bulk storage were recovered daily, allowed to reach thermal equilibrium then used to replenish each reservoir. During each drying cycle reservoirs were emptied, flushed and refilled. In the CT environment a temperature of 20  1
C was maintained throughout except for during the last cycle when the temperature rose up to 26
C due to mechanical breakdown of the temperature-controlling system. The experiment was conducted in the dark, apart from brief exposure to fluorescent light during sampling each day. In the GH

3155

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

Fig. 1 e Schematic illustration of the experimental setup in the constant temperature and glasshouse settings. Each column is identified alpha-numerically as indicated.

environment temperature ranged from 16 to 37
C. Pan evaporation rates ranged from 1.5 to 3.5 mm/day; two- to seven-fold higher than at CT. Only the soil surface was exposed, with the outer surface of the columns was wrapped in aluminium foil to prevent algal growth. The feed water was gravity-fed to the soil columns from a storage reservoir via a constant head tank of 10 cm above the soil surface as the base-case, assuring stable hydraulic conditions over the duration of the wetting cycles. Fully saturated conditions were consistently observed during the wetting periods. Daily measurements of infiltration rates, hydraulic heads, evaporation potential and temperature were made and samples of inflow and outflow water collected for analysis of EC, pH and turbidity. Where the volume of outflow allowed, analyses for a range of nutrient parameters were undertaken in the middle stage of each wetting cycle. At the end of the experiment the columns were sectioned into five intervals under sterile conditions and the physical and microbial properties of the soil determined.

2.2.

Analytical methods

Water samples were collected during each cycle and analysed by the Australian Water Quality Centre laboratory in Adelaide for suspended solids, nutrients and organic carbon. Turbidity was measured using a HACH turbidity analyser. EC and pH were determined using a standard laboratory analyser. Membrane Filtration Index (MFI) was determined according to the method described by Dillon et al (2001). At the end of the experiment bacterial biomass and polysaccharide concentrations were measured along the length of the column. Bacterial biomass in each of the soil increments was quantified with the MIDI HewlettePackard Microbial Identification System, based on extraction of fatty acids and conversion to fatty acid methyl esters from bacterial membranes (Haack et al., 1994). This method gives an accurate estimation of the viable biomass equivalent (biomass index) but not absolute biomass due to the variable fatty acid contribution from all bacterial cell components, including cell membranes.

Table 1 e Physicoechemical properties of the two study soils. Ka(m/day) d50b (mm) Porosity (%) Bulk density (g/cm3) Sand (%) Silt (%) Clay (%) TSAc (cm2) ESPd (%) OCe (%) CaCO3(%) Sand Loam a b c d e

23 0.10

0.68 0.15

54 44

1.60 1.35

K ¼ Hydraulic conductivity. d50 ¼ Median particle size. TSA ¼ Total surface area per column (from Mucha, 2004). ESP ¼ Exchangeable sodium percentage. OC ¼ Organic carbon.

97 70

2 17

1 13

1824 9392

«1.5 0.3

0.03 0.52

<1 <1

3156

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

Bacterial polysaccharide content was quantified using the phenol-sulphuric acid method (Dubois et al., 1956). Biomass index and polysaccharide are reported per unit weight of dry soil by determining the gravimetric water content of a subsample from each increment. Greater detail on the sampling and analysis methods is provided by Mucha (2004).

2.3.

Soil characterization

The two study soils were collected from the Arid Zone Research Institute (AZRI), located approximately 5 km south of Alice Springs in central Australia and the site of ongoing SAT investigations (Knapton et al., 2004; Dillon et al., 2006). The two soils are considered representative ‘end-members’ of the predominant soil types within the upper few metres of the profile at the AZRI site (Lennartz, 2003). One of the soils is a coarse textured palaeochannel sand, which for convenience, will simply be referred to as a ‘sand’; the other a finer textured floodplain silty sand, likewise referred to as a ‘loam’. Physicochemical property data for both soil types are given in Table 1. The most distinguishing feature of the two soils is a two order of magnitude difference in the initial Ko. Similar mineralogical make-up for both soil types, which is composed primarily of quartz with varying amounts of albite, orthoclase, mica/illite, amphibole and kaolin (Gates et al., 2009), suggests common parent material.

2.4.

Feed water characterization

The four qualities of water used in the study included three recycled waters and one potable water as follows:  Primary treated recycled water from the Alice Springs waste water treatment plant (WWTP) collected from the Blatherskite Park Irrigation Pump Station, Alice Springs (RW1)  Primary treated recycled water from the Alice Springs WWTP, followed by rock-bed filtration (RW2)  Advanced secondary treated recycled water from the Bolivar WWTP, Adelaide, South Australia (RW3)

 Potable quality groundwater from the Roe Creek borefield, Alice Springs (POT) Sufficient volumes of water for the experimental duration were collected in a single batch. The three waters from Alice Springs were collected on 30 September 2003 and transported via rail to Adelaide. The water from Adelaide was obtained on 3 October 2003 at the Dissolved Air Flotation Filtration (DAF/F) plant from a float tank before the final filtration step (to mimic the level of pre-treatment for SAT trials at Alice Springs). Water quality analyses of samples taken during the fourth SAT cycle, presented in Table 2, indicate the particulate and nutrient levels of the recycled waters cover wide ranges, depending on the level of pre-treatment. Water quality, in relation to nutrient and particulate concentrations, may be ranked from lowest to highest quality generally as follows: RW1 < RW2 < RW3 < POT. The two recycled waters from Alice Springs (RW1 and RW2) have significantly higher particulate concentrations as reflected in the levels of total algae, total suspended solids, turbidity and MFI. The percentage of particulate organic carbon (POC) ranges from 50% for RW1 and RW2 to below detection for POT. Water quality changes during the experimental period (for storage and experimental conditions combined) occurred due to microbial activity and were greatest for RW1 and RW2 waters. Some degree of change was evident for RW3 and POT was the most stable overall. These variations were insufficient to eclipse the strong distinctions between the water types. Over the course of the experiment turbidity of RW1 declined from around 40 to 15 NTU due to oxidation and flocculation/ settling of POC within the refrigerated bulk storages and the reservoir. DOC increased from 15 to 22 mg/L, presumably due to degradation of POC. Degradation of organic nitrogen was inferred from a steady increase in free ammonia that was accompanied by a steady decrease in TKN, with nitrate consistently low. Similar behaviour occurred for RW2, with an initial drop in turbidity over the first cycle of about 25 NTU, presumably due to settling out of particulate matter since neither Psol nor DOC increased. For the RW3 water in both environments, turbidity, Psol, pH and EC remained reasonably

Table 2 e Values for source water quality parameters as determined during the fourth wetting cycle. Parameter pH EC SAR Turbidity TSS Algaetot MFI TKN-N NH3eN NOxeN Psol Ptot TOC DOC

Units

RW1

RW2

RW3-CT

RW3-GH

POT

e mS/cm e NTU mg/L cells/mL s/L2 mg/L mg/L mg/L mg/L mg/L mg/L mg/L

7.88 1635 6.1 29.6 142 9.5  106 9600 13.6 4.3 <0.005 7.2 9.1 46.0 21.8

7.90 1583 6.5 17.1 49 1.0  107 5900 17.2 10.0 0.68 4.2 6.5 37.2 17.8

8.04 2190 8.0 0.7 2 2.7  102 44 2.4 1.1 1.1 4.2 5.0 11.7 11.1

8.62 2490 8.5 0.4 5 5.3  102 36 2.16 0.70 6.3 4.5 5.6 12.7 12.1

8.86 776 2.8 0.3 <1 0 22 0.25 <0.005 <0.005 0.009 0.02 1.1 1.2

CT ¼ constant temperature room environment; GH ¼ glasshouse environment.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

stable. TKN fell from 6.5 mg/L to 2 mg/L for the glasshouse and from 4 mg/L to 2.5 mg/L at CT. An increase in nitrate concentration of about 15 mg/L occurred after the first cycle at CT, whilst in the GH, nitrate concentrations showed no such shift over time and were high over the entire period. A sharp decrease in DOC from w30 to 12 mg/L occurred in the last cycle in both environments. Small declines in DOC and nitrate were observed for POT.

3.

Results and discussion

3.1.

Hydraulic behaviour

3.1.1.

Soil type

The average daily hydraulic rates for the sand and loam columns after four cycles for the various treatments are presented in Table 3. For 10 cm ponding depths in the sand columns the cumulative flux ranged from 2.0 m/day for RW2 water to 14.1 m/day for POT water, and for the loam columns from 0.06 m/day for RW1 to 0.34 m/day for the RW3 water at constant temperature (RW3-CT). The cumulative flux of RW1 water was about 34 times more for sands than for loams c.f. 230 for the Ko ratio. Corresponding ratio’s were 26 for RW2, 17 for RW3-CT, 51 for RW3-GH and 59 for POT. The duplicate columns behaved similarly except for the loam, RW2-fed columns (F2; Fig. 1) that was subsequently eliminated from consideration due to non-representative horizontal fissuring from excessive soil shrinkage/swelling. Examination of the time-series data (Mucha, 2004) reveal that quasi steady-state hydraulic conditions were achieved earlier for loams than for sands. Whilst average permeabilities and infiltration rates continued to decline over the first three cycles for sands, by the second cycle the loams had largely stabilised. The relative hydraulic conductivity (K/Ko) decline in sands was significantly larger than for loams, irrespective of water quality, with average K/Ko values in the last cycle ranging from 0.01e0.25 for sands compared with 0.2e2.5 for loams (Fig. 2). In absolute terms, rates of infiltration through sand remained greater than loam due to the higher values of K. In the last

cycle 4e46 times higher infiltration occurred for sands compared with the >200-fold difference in the initial infiltration rates between the two soils. Pronounced clogging was evident for even the POT water, with minimum K/Ko values of 0.02 for sand and 0.35 for loam at CT. Microbial clogging can reportedly occur with even high-quality potable water under laboratory conditions in the absence of light (Bouwer and Rice, 2001), and as will be shown later, these results would appear to be in agreement with the finding by Bouwer and Rice. In the GH environment infiltration rates exhibited up to three fold variations on a diurnal time-scale due to various biophysical processes that caused ebullition and entrainment of gas bubbles in soil pore-spaces (data given in Mucha, 2004). This was not observed at CT. Thus, tests conducted in the CT environment may overestimate infiltration rates as compared to the GH by up to 60% for the loam, with the difference apparently greater for finer-texted soils since gas binding effects are more pronounced. For sands, the GH produced 20% higher infiltration rates than for the equivalent ponding depth in CT.

3.1.2.

Water type

RW1 RW2 RW3-CT

RW3-GH POT

Ponding depth (cm)

10 10 10 30 50 10 10* 10

*70 cm columns (N ¼ 1).

Hydraulic loading Mean turbidity rates (m/day) (NTU) Sand

Loam

2.2  0.18 2.0  0.01 6.0  0.53 10.9  0.50 11.0  1.52 7.5  0.37 7.7 14.1  2.35

0.06  0.01 0.07  0.03 0.34  0.07 0.53  0.06 0.51  0.27 0.13  0.02 0.10 0.23  0.01

29.6 17.1 0.70 0.70 0.70 0.4 0.4 0.3

Ponding depth

The intermediate ponding depth of 30 cm, which increased the hydraulic gradient across the soil by w80% relative to the 10 cm ponding depth, enhanced the total infiltrated flux by approximately 80% for sands and 50% for loams. The highest ponding depth of 50 cm produced no greater mean flux than the 30 cm depth, although higher infiltration rates were observed in the initial cycles, but later the rates of decline in K/Ko at 50 cm were faster than for 30 cm for both soil types (Pavelic et al., 2006). Adjustment of the ponding depth affects two competing factors that control infiltration rates through a surficial clogged zone by: 1) increasing the hydraulic gradient and wetted surface area, and 2) reducing the permeability of the clogged zone by either greater compaction or penetration of clogging agents within the soil. Therefore, a limiting depth of ponding can be defined such that further increasing the water depth causes infiltration rates to be reduced rather than enhanced. The limiting depth determined from this study is in the order of 30 cm, as compared with around 60 cm or less for previous studies (Bouwer and Rice, 1989; Houston et al., 1999).

3.1.3. Table 3 e Mean hydraulic loading rates over four wetting cycles (mean ± half-difference between replicates given) and source water turbidity.

3157

Column length

Overall, the extended (70 cm) columns resulted in <5% higher flux for sand and 40% lower flux for loam as compared to the regular (10 cm) columns. Extending the soil depth reduced the hydraulic gradient from 2.0 to 1.4 implying an expected corresponding flux reduction compared to the shorter columns. The similar fluxes observed in the case of the sand must therefore be compensated by greater clogging in the shorter columns, presumably brought about by the higher hydraulic gradient. Only marginally greater clogging was observed for the loam columns (Fig. 2). Drying periods led to variable improvements in K/Ko. For the sands the general trend was for diminishing restoration in K/Ko with each successive cycle for poorer quality RW1 and RW2 waters and minimal change in K/Ko throughout for RW3 and POT waters. The trends for loams were weak and only marginally more pronounced for the two poorer water qualities. Higher restoration rates for sand is attributed to better

3158

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

RW1

1

C1 D1

0.8

3

C2 D2

2

0.6 0.4

E1 F1

3

E2 F2

2

0.6 0.4

1

1

0.2 0

0

0 20

30

40

RW3-CT 10cm

1 0.8

2

0.6 A7 B7

0.2

A8 B8

1

0 10

20

30

RW3-GH 10cm

1 0.8

40 A9 B9

20

30 POT

40 G1 H1

0.8

50 3

G2 H2

2

0.6 0.4

1

0.2

0 0

10

1

3

0.4

0 0

50

Sand K/Ko

10

Loam K/Ko

0

Loam K/Ko

0.2

Sand K/Ko

RW2

1 0.8

0

3

A10 B10

0 0

50

10

20

30

RW3-GH 70cm

1

40

S70

50 3

L70

0.8 2

0.6 0.4

1

2

0.6 0.4

1

0.2

0.2 0

0 0

10

20

30

40

0

0

50

0

10

Time (days)

20 30 Time (days)

40

50

Fig. 2 e K/Ko data for replicate sand and loam columns over four SAT cycles consisting of 7 days wetting followed by 7 days drying (data for sands is indicated by solid symbols and for loams by cross symbols).

drainage properties improving entry of atmospheric oxygen by diffusive/conductive processes and thereby allowing greater opportunity for oxidation of organic matter (Bancole et al., 2003).

3.1.4.

Water quality

For sands, the hydraulic loading rates were higher for water of higher quality, although the difference between RW1 and RW2 was only marginal (Table 3). For loams, the trends were slightly ambiguous, although RW3 and POT still out-performed RW1 and RW2. Sand and loam RW3-fed columns at CT experienced clear K/Ko declines in each wetting cycle but the restoration of the subsequent drying cycle overcompensated for this reduction through a net improvement in K/Ko over time that was consistent between replicates. K/Ko of sand and loam soils for RW3 in the GH environment showed no significant increase due to drying. A variation on the trends for RW3-CT and RW3-GH were evident for POT-fed loam columns. No clear explanation can be given for the increase over time, however the gradual displacement of residual air from within the pore space of the soil and variations in source water quality offer two possible causes.

3.2.

filtering-out of particles was more effective for loam than for sand due to the smaller pore sizes and filtration and for waters with higher particulate content. A proportion of the observed clogging must therefore be attributed to physical processes. These processes must have taken place within the profile rather than at the soil surface since a surficial clogged zone was not evident for either soil type. The trapped particles from treated wastewater are

Table 4 e Turbidity changes during soil passage through sand and loam columns during the fourth wetting cycle. Water type RW1 RW2 RW3-CT RW3-GH

Particulate deposition during soil passage

Differences between turbidity levels into and out of the column indicated highly variable rates of retention by the soil (Table 4). The amount of particulate matter removal determined from turbidity measurements during the final cycle was 21  21% for sand and 47  28% for loam. Generally, the

POT

Soil type

Inflow (NTU)

Outflow (NTU)

% Removal

Sand Loam Sand Loam Sand Loam Sand SandA Loam Loama Sand Loam

29.6

22.8e24.2 3.8e4.7 8.2e8.5 7.1 0.4e0.6 0.2e0.3 0.4e0.5 0.3 0.2e0.3 0.3 0.21e0.3 0.25e0.3

21 86 51 59 29 64 13b 25 38 25 15 8

17.1 0.7 0.4

0.3

a 70 cm columns. b Negative removal values may be due to minor leaching of fines from the column media.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

primarily of organic origin (known from the high proportion of volatile solids), and provide a source of substrate for microbial growth in addition to that from the soluble nutrients.

3159

Biomass and polysaccharide levels were higher near the inlet end of the column where substrate and electron acceptor availability were highest, but microbial growth had penetrated the entire length of the regular and extended columns (Mucha, 2004). Values varied by no more than a factor of three along the entire length. This is at least several-fold lower than observed in previous studies with water similar to RW3 under conditions of continuous infiltration (Rinck-Pfeiffer et al., 2000), and in urban stormwater by Wakelin et al (2010), and presumably reflects the smaller soil depth and possibly the inclusion of wetting and drying cycles in this study. Although the microbiological data provides only a representative ‘snap-shot’ of conditions at the end of the experiment, the hydraulic data suggests that quasi-steady-state growth conditions had been established. For example, K/Ko within different increments of the soil, for the three columns run with RW3 water that were fitted with manometers at depths of 1, 3 and 8 cm’s, was reasonably consistent across all intervals, which agrees with the soil microbial data (Mucha, 2004). This is also consistent with the lack of removal of phosphorus during soil passage in the latter cycles as compared to the earliest stages during the establishment of the biofilm. During steadystate, the additional biofilm added due to microbial growth is offset by an equivalent loss due to shear stresses that develop at the pore-scale, and thus only a finite mass of biological clogging agents accumulate within the soil. For sands, the levels of biomass index at the end of the experiment were similar (within limits of variability) for all waters except for RW1 where levels were 40% higher (Fig. 3). For loams, a clearer inverse trend between biomass levels and feed water nutrient concentrations was evident, with the overall difference between RW1 and POT being almost 50%. Polysaccharide concentrations show no correlation to nutrient levels in source waters and are consistently higher for loams than for sands (Fig. 4). Values for the loam soil fed by RW3 water at CT and GH were high, at around twice that of the

other waters, although concentrations in the sands were not elevated in the same way. The relationship between polysaccharide concentration and water quality, even considering differences in fluxes and total nutrient loads, were inconsistent. Previous column studies with water similar to RW3 demonstrated that polysaccharide production is the result of microbial activity within the media and not due to retention of polysaccharides present within the feed water (Rinck-Pfeiffer et al., 2000). Polysaccharide concentrations indicate a strong dependence on soil type. When the data is normalised to account for differences in specific surface area between sand and loam using the particle size distribution data (note the loam is higher in specific surface area by a factor of 5.2), polysaccharide concentrations per unit area of soil surface were similar across the four water types (Fig. 5). This suggests a physical constraint on the capacity of the soil to support polysaccharide production. Further evidence from a previous batch study with varying dilutions of water like RW3 and distilled water also showed polysaccharide production to be limited by soil surface area, and not by nutrient availability (Pavelic et al., 2007). Polysaccharide producing micoorganisms occupy the surfaces of soil grains and therefore greater surface area translates to higher microbial growth potential. Extracellular polymeric substances serve to trap and store particulates and nutrients for cell metabolism, and are generally thought to comprise the major component of the bacterial biofilm which is a major contributor to biological clogging (Ragusa et al., 1994; Rinck-Pfeiffer et al., 2000). In contrast, biomass, a measure of the microbial cells within a biofilm community, was independent of soil type but strongly dependent on water quality. Biomass levels were correlated to levels of N, P, OC and particulate matter in the feed water. Correlations were consistently higher for loam than for sand. Growth of the bacterial cells is dependent on physico-chemical factors associated with the soil and water and tends not to be limited by the availability of surfaces for bacterial attachment, as is the case for polysaccharide. Extending the drying phase of the RW3-CT columns by a further six weeks gave no indication of polysaccharide reductions, suggesting that bioclogging was stable over the extended rest period.

Fig. 3 e Column-averaged biomass index concentration for all soil types and water qualities for columns with ponding depth of 10 cm (N [ 2).

Fig. 4 e Column-averaged polysaccharide concentration for all soil types and water qualities for columns with ponding depth of 10 cm (N [ 2).

3.3.

Microbial growth

3160

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

Fig. 5 e Polysaccharide concentration per unit surface area for all soil types and water qualities (N [ 2 where variance bars indicated, otherwise N [ 1).

Numerous studies have shown that there can be a close correlation between the amount of soil polysaccharide and the observed reduction in hydraulic conductivity of soils (e.g. Ragusa et al., 1994; Rinck-Pfeiffer et al., 2000). The correlation of hydraulic conductivity and the corresponding average polysaccharide levels in the soil is poor (R20 s < 0.2 in this study) and counter to the anticipated trend for both soil types. The correlation between the turbidity of the feed water and the hydraulic conductivity were much better than for polysaccharide (R2 ¼ 0.77 for sand and R2 ¼ 0.39 for loam) and follows the anticipated trend in both cases. The poor correlation between polysaccharide levels and the magnitude of the K reductions can be explained by the K response being the combined effect of both physical and microbial clogging (and particle rearrangement at least in the case of the sand), with polysaccharides only constituting a proportion of the clogging agents within the soil. It was earlier shown that polysaccharide levels are relatively constant for the different water types in sand (Fig. 4), and are sufficiently high as to indicate some degree of microbial clogging (Rinck-Pfeiffer et al., 2000). It therefore follows that both microbial and physical processes influence clogging, but that the physical component is more direct than the biological one. Rice (1974), in his column study with secondary treated wastewaters, found that physical clogging was the predominant cause of infiltration rate reductions, but that biological clogging gained relative importance with decreasing concentrations of particulate matter in the source water.

3.4.

Chemical clogging

Chemical clogging due to clay dispersion would have had a negligible effect for either soil type given the low ESP of the soils (Table 1), even considering the low to moderate sodicity (SAR) (2.8e8.5; Table 2) and fresh to brackish EC (776e2490 mS/ cm) (Ragusa et al., 1994). The dominant clay mineral, kaolinite, is largely unreactive and the most reactive mineral, smectite, is not present in either soil.

3.5.

Mechanism for initial declines in K

Low values of K/Ko for the sands from the initial stages of the experiment were a pronounced feature of the hydraulic data.

Earliest measured values ranged from around 0.8 to 0.3 (Fig. 2); or expressed in other words, a 20%e70% reduction in hydraulic conductivity had occurred within the first day of the experiment. Declines were found to be quite rapid and largely independent of the setting and feed water quality, suggesting that some mechanism other than gas entrainment, particle filtration or microbial growth was responsible. In contrast, initial K/Ko values for loam columns were generally higher, and more variable (0.4e2.4). The cause of the rapid initial drop in hydraulic conductivity (K ) of the sand was investigated for a smaller set of column experiments focussing on the initial hydraulic response. High quality potable water dosed with hypochlorite solution was used to eliminate other potentially confounding forms of clogging. The sand media was selected for testing owing to the greater K reductions observed. Control columns (CN1eCN3) were only subjected to a brief (5 min) Ko test (as per the previous experiment) with no further testing thereafter, whilst treatment columns underwent tests of up to 24 h following the brief test. Experimental conditions were otherwise unchanged from the previous experiment. During the 24 h tests K declined rapidly and irreversibly to around 90% of the Ko value within the first 200e400 min of infiltration, followed by more stable conditions thereafter (Fig. 6). Interestingly, whilst the initial Ko values differed by almost 50% due to subtle differences in the packing of individual columns, K values appeared to converge towards a common value after 24 h. Note that the initial values in K were around double those of the previous experiment (Table 1), with numerous attempts to repack the columns to replicate the earlier value of K unsuccessful, thereby suggestive of differences between the two soil batches. Whilst both batches were obtained from the same bulk (20 L) source, fractionation effects may have still occurred. Ko values for control columns were stable over the brief test period (Fig. 6). Particle-sizes from sieve analyses on four equi-thick (2.5 cm) sections for control and treatment columns show a tri-modal frequency distribution of particle sizes for all depth intervals. The poor degree of sorting of the soil (not shown) is characteristic of a high energy transport and depositional environment. Variability in particle size within any column was substantially higher than between treatment and control columns. Longitudinal thin section images confirmed the occurrence of sub-column scale heterogeneities within the column (Nakai, 2006). They also reveal the presence of aggregates of fine particles within the control that are not present within the treatment, confirming that the fines had been washed-out of the treatment during extended infiltration (Fig. 7). The sieve analysis indicated that no more than 3% of the particle mass could have passed through the mesh, although the resultant effect would be expected to have increased rather than a decreased K. The rapid declines in K are thought to result from mobilization of particles within the column. Several hydrodynamic forces act upon the soil grains when water flow begins (particle sizes in the test soil are too large for other physicochemical forces to act), with the principle force involving the ‘rolling’ over surrounding larger grains of the smallest grains most affected by the shear stress induced by water flow by

3161

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

80

80

K (m/day)

60

60

40

CN1 CN2 CN3

K (m/day)

20

0

40

0

1

2

3

4

5

Time (minutes)

20 T1

T3

T2

T4

0 0

500

1000

1500

Time (minutes)

Fig. 6 e Decrease in hydraulic conductivity (K ) of duplicate columns during 24 h (T1eT2) and 8 h (T3eT4) infiltration tests due to passage of chlorinated potable water. Inset graph shows the behaviour for three control columns (CN1eCN3).

‘drag’ and ‘lift’ forces (Blatt et al., 1980). Fine particles mobilized by drag/lift forces are redeposited downstream, thereby destabilizing the larger particles that were initially bridged by the fines in the ‘wash-out’ zone and reducing the effective pore space and K by the accumulation of fines in the ‘wash-in’ zone (West et al., 1992). Since systematic trends between the sectioned intervals were not observed, the distance of migration must be less than the measurement scale (2.5 cm). The soil K stabilizes when the particles have settled down and particle mobilization no longer occurs. Temporary cessation of flow prior to stabilization alleviates the hydrodynamic forces and loosens grain contacts, thereby increasing K. This behaviour was observed in two of the columns that were allowed to rest overnight after w120 min and lead to some temporary recovery in K (expts. T3 and T4; Fig. 6). The consistently steady Ko values during the brief testing suggest

that the effect of particle mobilization had yet to appear over this time-scale, even though steady-state flow conditions had been established. A follow-up experiment with the loam soil using the same approach showed no change in K over 24 h, reaffirming that mobilization in the initial stage of the main experiment was limited to the sand soil only. Interestingly, field infiltration tests at the AZRI site with POT water and identical wet/dry cycle lengths revealed no short or long term change in soil permeability for either sand or loam basins (Knapton et al., 2004). Clearly the field soils, which are of a similar texture to the lab soils, have a more stable packing arrangement, and the process of reconstructing the soil within the column had a detrimental effect on soil stability. Differences in the pore system between natural soil and repacked columns have been reported as a key factor for

Fig. 7 e Thin section images under uncrossed polarised light: a) sand control sample showing heterogeneity of silty matrix between quartz sand grains; b) sand treatment sample showing lack of silty matrix. Size of individual grains range from 100 to 2000 microns.

3162

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

the divergent responses between soil column and corresponding field infiltration studies (Bouma, 1975). The absence of field clogging over the longer term could be due to other processes. For example, the maintenance of a chlorine residual in the water entering the basins may have inhibited microbial growth, whereas the water fed into the laboratory columns during the main experiment had no residual chlorine. In addition, the higher thermal mass associated with field conditions may have limited the temperature gradients between infiltrating water and soil water that induce gas entrainment.

4.

Conclusions

Infiltration rates were generally commensurate with the feed water quality for both soil types. The difference in hydraulic loading rates for four water types examined was 6-fold for sands and 8-fold for loams. Greater and more rapid clogging was observed for the sands, although the overall flux for sands still remained 17e59 times higher than for loams. The clogging observed during inundation of the sand and loam soils with recycled water and potable water is a response to several processes. Physical clogging would appear to be the most direct form of clogging, with its effect strongly determined by particulate concentrations in the source water, as determined by the level of treatment afforded. Microbial clogging, which develops gradually over time, is common to all cases and soil surface area availability is the limiting factor rather than nutrient status. Consequently, microbial clogging becomes relatively more significant as particulate levels in the source water are reduced. In the constant temperature room setting the diurnal gas clogging evident in the glasshouse setting was not observed; the latter is probably more representative of actual field processes. The depth of ponding above the soil surface also affects rates of infiltration. The highest ponding depth of 50 cm increased infiltration rates over the short-term but did not translate to higher rates over the longer-term. The greatest change in K consistently occurred within the first 24 h due to mechanical clogging caused by mobilization and small-scale transport of soil grains within the sand columns. This phenomenon is an artefact of repacking media composed of large particle sizes and low uniformity and was not directly evidenced in equivalent field studies with undisturbed soil, nor for the loam soil in the laboratory. The high dependence of hydraulic loading rates on the level of treatment given to the source water, and in particular, to the level of particulates present in the source water has been consistently observed. Ponding depth is also important, but to a lesser extent than water quality. These results provide insight on the water quality and operating conditions needed to achieve sustainable SAT systems.

Acknowledgements This work was financially supported by the CSIRO Water for a Healthy Country National Research Flagship Program and

Power Water Corporation (PWC). Anthony Knapton, Graham Ride and Peter Jolly (Northern Territory Department of Natural Resources, Environment and the Arts); Don Pidsley and Kelly Mashford (PWC); Dr. Simon Toze, Jon Hanna and Toney Hirnyk (CSIRO Land and Water); Danny Tintor (United Water); and Quenton Bedeau (Kingston Community School) contributed to logistical and analytical aspects of the study. The occupational traineeship of Mathias Mucha from the Technical University of Berlin at CSIRO Land and Water took place with support from Prof. Dr. Martin Jekel. Dr. Declan Page and Dr. Steven Wakelin of CSIRO Land and Water provided helpful comments on the manuscript.

references

Bancole, A., Brissaud, F., Gnahne, T., 2003. Oxidation processes and clogging in intermittent unsaturated infiltration. Water Science and Technology 48 (11e12), 139e146. Baveye, P., Vandevivere, P., Hoyle, B.L., DeLeo, P.C., Sanchez de Lozada, D., 1998. Environmental impact and mechanisms of the biological clogging of saturated soils and aquifer materials. Critical Reviews in Environmental Science and Technology 28 (2), 123e191. Blatt, H., Middleton, G., Murray, R., 1980. Origin of Sedimentary Rocks, second ed.. Prentice-Hall Inc., Englewood Cliffs, New Jersey. Bouma, J., 1975. Unsaturated flow during treatment of septic tank effluents. Journal of the Environmental Engineering Division 101 (6), 967e983. Bouwer, H., 2002. Artificial recharge of groundwater: hydrogeology and engineering. Hydrogeology Journal 10 (1), 121e142. Bouwer, H., Rice, R.C., 1989. Effect of water depth in groundwater recharge basins on infiltration. Journal of Irrigation and Drainage Engineering 115 (4), 556e567. Bouwer, H., Rice, R.C., 2001. Capturing flood waters for artificial recharge of groundwater. In: Proc. 10th Biennial Symposium on Artificial Recharge of Groundwater. Arizona Hydrological Society, Tucson, Arizona, pp. 99e106. Dillon, P., Pavelic, P., Massmann, G., Barry, K., Correll, R., 2001. Enhancement of the membrane filtration index (MFI) method for determining the clogging potential of turbid urban storm water and reclaimed water used for aquifer storage and recovery. Desalination 140 (2), 153e165. Dillon, P.J., Pavelic, P., Toze, S., Rinck-Pfeiffer, S.M., Martin, R., Knapton, A., Pidsley, D., 2006. Role of aquifer storage in water reuse. Desalination 188, 123e134. Dubois, M., Gilles, K.A., Hamilton, J.K., Rebers, P.A., Smith, F., 1956. Colorimetric method for determination of sugars and related substances. Analytical Chemistry 28, 350e356. Fox, P., Narayanaswamy, K., Genz, A., Drewes, J.E., 2001. Water quality transformations during soil aquifer treatment at the Mesa Northwest Water Reclamation Plant, USA. Water Science and Technology 43 (10), 343e350. Gates, W., Janik, L., Pavelic, P., Dillon, P., Barry, K., 2009. Characterisation of the physical and geochemical properties and infrared spectra of five soil cores at the AZRI site near Alice Springs, Northern Territory. CSIRO Water for a Healthy Country Flagship Report. Haack, S.K., Garshow, H., Odelson, D.A., Forney, L.J., Klug, M.J., 1994. Accuracy, reproducibility and interpretation of fatty acid methyl ester profiles of model bacteria communities. Applied and Environmental Microbiology 60, 2483e2493. Houston, S.L., Duryea, P.D., Hong, R., 1999. Infiltration considerations for groundwater recharge with waste effluent.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 5 3 e3 1 6 3

Journal of Irrigation and Drainage Engineering 125 (5), 264e272. Knapton, A., Jolly, P., Pavelic, P., Dillon, P., Barry, K., Mucha, M., Gates, W., 2004. Feasibility of a pilot 600 ML/yr soil aquifer treatment plant at the arid zone Research Institute. Department of Infrastructure, Planning and Environment, Alice Springs. Report No. 29/2004A. Kopchynski, T., Fox, P., Alsmadi, B., Berner, M., 1996. The effect of soil type and effluent pre-treatment on Soil Aquifer Treatment. Water Science and Technology 34 (11), 235e242. Lance, J.C., 1977. Phosphate removal from sewage water by soil columns. Journal of Environmental Quality 6 (3), 279e284. Lance, J.C., Rice, R.C., Gilbert, R.G., 1980. Renovation of wastewater by soil columns flooded with primary effluent. Journal of Water Pollution Control Fed. 52 (2), 381e387. Lennartz, R., 2003. Soils and geomorphology of “the outer Farm” area south of Heavitree Gap and north of Colonel rose Drive. Department of Infrastructure, Planning & Environment, Alice Springs. Report No. 25/2001A. Mucha, M., 2004. A laboratory column study of the effect of wastewater quality on clogging during soil aquifer treatment. Diploma Thesis, Technical University of Berlin, Germany. Nakai, T., 2006. A laboratory study on impacts of particle rearrangement on short-term hydraulic conductivity reductions. Honours Thesis, School of Chemistry, Physics and Earth Sciences, Flinders University of South Australia. Pavelic, P., Mucha, M., Dillon, P.J., Barry, K.E., 2006. Laboratory column study on the effect of ponding depth on infiltration rate during SAT. Berlin, Germany, 11e16 June 2005. In: Recharge Systems for Protecting and Enhancing Groundwater Resources. Proceedings of the 5th International Symposium on Management of Aquifer Recharge (ISMAR5). IHP-VI Series on Groundwater, 13, pp. 624e629. Pavelic, P., Dillon, P.J., Barry, K.E., Vanderzalm, J.L., Correll, R.L., Rinck-Pfeiffer, S.M., 2007. Water quality effects on clogging rates during reclaimed water ASR in a carbonate aquifer. Journal of Hydrology 334, 1e16.

3163

Pell, M., Nyberg, F., 1989. Infiltration of wastewater in a newly started pilot sand-filter system: II. Development and distribution of the bacterial populations. Journal of Environmental Quality 18, 457e462. Ragusa, S.R., de Zoysa, D.S., Rengasamy, P., 1994. The effect of microorganisms, salinity and turbidity on hydraulic conductivity of irrigation channel soil. Irrigation Science 15, 159e166. Rauch-Williams, T., Drewes, J.E., 2006. Using soil biomass as an indicator for the biological removal of effluent-derived organic carbon during soil infiltration. Water Research 40 (5), 961e968. Rice, R.C., 1974. Soil clogging during infiltration of secondary effluent. Journal of Water Pollution Control Fed. 46, 708e716. Rice, R.C., Bouwer, H., 1984. Soil-aquifer treatment using primary effluent. Journal of Water Pollution Control Fed. 56 (1), 84e88. Rinck-Pfeiffer, S.M., Ragusa, S.R., Sztajnbok, P., Vandevelde, T., 2000. Interrelationships between biological, chemical and physical processes as an analog to clogging in Aquifer Storage and Recovery (ASR) wells. Water Research 34 (7), 2110e2118. Smith, A., Dillon, P., 1997. Fluid flow and biogeochemical processes in porous media: a survey of laboratory columns designs. Centre for Groundwater Studies Report No. 77. Taylor, S.W., Jaffe, P.R., 1990. Biofilm growth and the related changes in the physical properties of a porous medium - 1. Experimental stage. Water Resources Research 26 (9), 2153e2159. Toze, S., Hanna, G., Smith, A., Hick, W., 2002. Halls Head Indirect Treated Wastewater Reuse Scheme. CSIRO Land and Water Report to Water Corporation, Western Australia. Wakelin, S.A., Page, D.W., Pavelic, P., Gregg, A.L. and Dillon, P.J. Rich communities of archaea, bacteria, and eukarya inhabit water treatment biofilters and are differentially affected by filter type and sampling depth. Water Science and Technology, accepted. West, L.T., Chiang, S.C., Norton, L.D., 1992. The morphology of soil crusts. In: Sumner, M.E., Stewart, B.A. (Eds.), Soil Crusting: Chemical and Physical Processes. Lewis Publishers, Boca Raton, Florida, pp. 55e92.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Chloramination of nitrogenous contaminants (pharmaceuticals and pesticides): NDMA and halogenated DBPs formation Julien Le Roux, Herve´ Gallard*, Jean-Philippe Croue´ 1 Laboratoire de Chimie et Microbiologie de l’Eau (CNRS UMR 6008), Universite´ de Poitiers e ENSIP, 40 Avenue du Recteur Pineau, 86022 Poitiers Cedex, France

article info

abstract

Article history:

Disinfection with chloramines is often used to reduce the production of regulated disin-

Received 21 December 2010

fection by-products (DBPs) such as trihalomethanes (THMs) and haloacetic acids (HAAs).

Received in revised form

However, chloramination can lead to the formation of N-nitrosamines, including

17 March 2011

N-nitrosodimethylamine (NDMA), a probable human carcinogen. Previous research used

Accepted 19 March 2011

dimethylamine (DMA) as a model precursor of NDMA, but certain widely used tertiary

Available online 26 March 2011

dimethylamines (e.g. the pharmaceutical ranitidine) show much higher conversion rates to NDMA than DMA. This study investigates the NDMA formation potential of several tertiary

Keywords:

amines including pharmaceuticals and herbicides. The reactivity of these molecules with

NDMA

monochloramine (NH2Cl) is studied through the formation of NDMA, and other haloge-

Nitrosamine

nated DBPs such as haloacetonitriles (HANs) and AOX (Adsorbable Organic Halides).

Chloramination

Several compounds investigated formed NDMA in greater amounts than DMA, revealing

Disinfection by-products

the importance of structural characteristics of tertiary amines for NDMA formation.

Ranitidine

Among these compounds, the pharmaceutical ranitidine showed the highest molar conversion to NDMA. The pH and dissolved oxygen content of the solution were found to play a major role for the formation of NDMA from ranitidine. NDMA was formed in higher amounts at pH around pH 8 and a lower concentration of dissolved oxygen dramatically decreased NDMA yields. These findings seem to indicate that dichloramine (NHCl2) is not the major oxidant involved in the formation of NDMA from ranitidine, results in contradiction with the reaction mechanisms proposed in the literature. Dissolved oxygen was also found to influence the formation of other oxygen-containing DBPs (i.e. trichloronitromethane and haloketones). The results of this study identify several anthropogenic precursors of NDMA, indicating that chloramination of waters impacted by these tertiary amines could lead to the formation of significant amounts of NDMA and other nonregulated DBPs of potential health concern (e.g. dichloroacetonitrile or trichloronitromethane). This could be of particular importance for the chloramination of wastewater effluents, especially during water reuse processes. ª 2011 Elsevier Ltd. All rights reserved.

* Corresponding author. Tel.: þ33 5 49 45 44 31; fax: þ33 5 49 45 37 68. E-mail addresses: [email protected] (J. Le Roux), [email protected] (H. Gallard), [email protected] (J.-P. Croue´). 1 Present address: King Abdullah University of Science and Technology (KAUST), Thuwal 23955-6900, Saudi Arabia. 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.035

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

1.

Introduction

A large diversity of disinfection by-products (DBPs) are formed during water treatment processes using chlorination, including trihalomethanes (THMs) and haloacetic acids (HAAs). Disinfection with chloramines is known to significantly reduce the formation of regulated DBPs as compared to chlorination. However, chloramination favors the formation of N-nitrosamines, including N-nitrosodimethylamine (NDMA). The US Environmental Protection Agency classifies NDMA as a probable human carcinogen, evaluating a 106 risk level of cancer from NDMA concentration at 0.7 ng/L in drinking water (U.S. Environmental Protection Agency, 1987). Over the last decade, interest has been growing about NDMA formation during water treatment process. Several studies examined the mechanisms explaining the formation of NDMA during chlorination and chloramination. In most studies, dimethylamine (DMA) served as the model NDMA precursor (Choi and Valentine, 2002a,b, 2003; Choi et al., 2002; Mitch and Sedlak, 2002; Schreiber and Mitch, 2005, 2006). However, some studies indicated that the amount of dimethylamine present in surface waters (Gerecke and Sedlak, 2003) or secondary municipal wastewaters (Mitch and Sedlak, 2004) are not sufficient to explain the amount of NDMA formed. The role of tertiary amines presenting dimethylamine functional groups has been pointed out (Mitch and Sedlak, 2004; Schmidt et al., 2006). Recent studies looked at diuron as a precursor of NDMA. Results showed that the molar conversion rate is relatively low (<1.5% of diuron forms NDMA) (Chen and Young, 2008, 2009). Another tertiary amine ranitidine, a histamine antagonist widely used for peptic ulcer treatment was found to be an important NDMA precursor (62.9% NDMA yield obtained by Schmidt et al., 2006 and 89.9% by Shen and Andrews, 2011). Other tertiary amines led to less or equal NDMA formation than DMA, revealing the importance of structural characteristics of tertiary amine compounds for NDMA formation (Schmidt et al., 2006). Shen and Andrews (2011) demonstrated that several tertiary amines including pharmaceuticals and personal care products are nitrosamine precursors during chloramines disinfection. According to these authors, the presence of electron donating group such as furan can increase the electron density on the nitrogen atom and then favors the reaction with chlorine leading to high NDMA yields observed with some pharmaceuticals (especially ranitidine). Ranitidine is sold worldwide as a gastrointestinal drug and has been detected at concentrations ranging from 70 ng/L to 540 ng/L in primary effluents of wastewater treatment plants (WWTP) in Spain (Radjenovic et al., 2009) and at w10 ng/L in several surface waters (Kolpin et al., 2002; Zuccato et al., 2000). Ranitidine and other pharmaceuticals are not well removed by biological treatments and can be found in river waters receiving the WWTP effluents (Castiglioni et al., 2006; Radjenovic et al., 2009). Chloramination of wastewaters (e.g. for wastewater reuse purposes) impacted by pharmaceuticals is of great concern because of the potential risk of NDMA formation. NDMA formation occurring during chloramination has previously been explained as a nucleophilic substitution reaction between monochloramine (NH2Cl) and dimethylamine

3165

(DMA) to form an unsymmetrical dimethylhydrazine intermediate (UDMH) (Choi and Valentine, 2002b; Choi et al., 2002; Mitch and Sedlak, 2002). UDMH is then rapidly oxidized by NH2Cl to NDMA at <3% yields. Over the past few years, studies have addressed the importance of chloramines speciation and dissolved oxygen (Schreiber and Mitch, 2006). Dichloramine (NHCl2) was found to contribute to the production of NDMA during chloramine disinfection, occurring through the formation of a chlorinated UDMH (UDMH-Cl) as an intermediate rather than UDMH. Enhancing the formation of NHCl2 by increasing the Cl:N ratio also lead to higher yields of NDMA during chloramination of tertiary amines (Shen and Andrews, 2011). Dissolved oxygen was also described as a critical parameter (Schreiber and Mitch, 2006). The authors proposed that the last step of the formation of NDMA consists in the incorporation of dissolved O2 into UDMH-Cl, which would then lead to NDMA. Degradation of tertiary amines may form other nitrogenous disinfection byproducts (N-DBPs) of potentially health concern, such as haloacetonitriles (HANs), halonitromethanes (HNMs), haloketones (HKs) or cyanogen chloride (CNCl). HANs have been proved to be more toxic than HAAs and other regulated DBPs (Muellner et al., 2007). Trichloronitromethane (TCNM), also known as chloropicrin, was the first of the HNMs to be identified as a DBP in drinking water (Hoigne and Bader, 1988; Thibaud et al., 1987). Potential health effects of HNMs have already been studied (National Cancer Institute, 1978; Schneider et al., 1999). They were found to be more mutagenic than the corresponding halomethanes, and TCNM has been demonstrated to be particularly genotoxic (Plewa et al., 2004). TCNM formation mechanisms have been proposed by chlorination and chloramination of monomethylamine and n-propylamine (Joo and Mitch, 2007). TCNM formation is expected to increase with pH during chlorination, and to be more important during chlorination than during chloramination. TCNM formation from chlorination of lake waters was 40 times lower than that of chloroform (Hoigne and Bader, 1988). Major haloketones (HKs) identified in chlorinated or chloraminated waters are 1,1-dichloro-2-propanone (1,1-DCP) and 1,1,1-trichloro-2-propanone (1,1,1-TCP). DCAN, 1,1-DCP and CNCl formation were found to decrease when increasing pH, with maximum yields around pH 5e6 (Yang et al., 2007). DCAN formation during chloramination was much lower than during chlorination, whereas CNCl and 1,1-DCP yields were higher in chloraminated water (Yang et al., 2007). The goal of this study was to investigate the reactivity of several nitrogen-containing organic compounds with monochloramine, through the formation of NDMA, HANs and AOX (Adsorbable Organic Halides). Model compounds investigated included three herbicides (diuron, isoproturon, trifluralin) and five pharmaceuticals: ranitidine (peptic ulcer treatment); doxepin and amitriptyline (tricyclic antidepressants); mifepristone (an abortifacient) and minocycline (an antibiotic used for acne treatment). All of them are tertiary amines presenting DMA functional groups. These anthropogenic compounds are likely to enter natural waters via wastewater discharges (i.e. pharmaceuticals) or agricultural runoff (i.e. herbicides). Because our objective was to study byproducts formation mechanisms, solutions of model compounds were prepared at

3166

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

concentrations that are significantly higher than what can be found in natural waters or wastewater effluents. As a result, DBPs were formed at relatively high concentrations that are not likely to be found in treated waters. The influence of several parameters (i.e. nitrites concentration, pH, chloramines speciation and dissolved oxygen concentration) was investigated in order to better understand the reaction mechanisms that lead to the formation of NDMA and some other DBPs (HANs, HKs, TCNM, AOX) during chloramination of tertiary amines.

2NH2 Cl þ Hþ ¼ NHCl2 þ NHþ 4

(1)

Free chlorine and total chlorine concentrations in the stock solutions of sodium hypochlorite were determined iodometrically with sodium thiosulfate 0.1 M (Prolabo, >99.9%). Initial NH2Cl and NHCl2 concentrations were determined by spectrophotometric measurement using their respective molar extinction coefficients at 245 nm and 295 nm and solving simultaneous equations (Schreiber and Mitch, 2005). Residual chloramines were analyzed iodometrically (Eaton et al., 1995).

2.

Materials and methods

2.3.

2.1.

Materials

All glassware used during these experiments was washed with deionized water and baked at 500  C for at least 5 h prior to use. Reactions were conducted in sealed 1 L amber glass bottles at 20  C in a temperature-controlled room, under dark conditions to avoid photolysis of NDMA. Chloramination experiments were conducted following the approach of Mitch et al. (Mitch et al., 2003), using high concentrations of NH2Cl (200e300 mg/L as Cl2) and a reaction time of 5 days for most of our experiments. NH2Cl remained in excess during all the reaction time. Solutions were prepared by dissolving a predetermined amount of compound in 1 L of 10 mM acetate, phosphate or carbonate buffer. 100 mL of preformed monochloramine was then added to the working solution. Chloramination experiments were conducted in triplicate. All series of experiments were completed with the chloramination of a corresponding blank solution. At given contact times, 350 mL of samples were transferred for residual chlorine, HANs and AOX analyses, and 750 mL were processed for nitrosamines analyses. Percent molar yields were calculated using the initial molar concentration of the studied compounds, following Equation (2).

All experiments were conducted using deionized water (MilliQ, Millipore) buffered with sodium acetate (pH ¼ 4.0e5.5), a mixture of sodium phosphate monobasic and sodium phosphate dibasic (pH ¼ 7.0e8.5), or sodium carbonate (pH ¼ 10). pH values were adjusted as needed using sodium hydroxide or sulfuric acid (0.1 N, Fisher Scientific). Fluka Analytical methyl tert-butyl ether (>99%), Fisher Scientific methylene chloride (GLC grade) and Carlo Erba methanol (>99.9%) were used without further purification. Amitriptyline (>98%), diuron (>98%), doxepin (>98%), isoproturon (99.8%), mifepristone (>98%), minocycline (92%, 8% water), ranitidine and trifluralin (>99%) were used without further purification and were supplied through SigmaeAldrich. Sodium hypochlorite (NaOCl, 13%, Acros Organics) and ammonium chloride (Fisher Scientific, 99.9%) were used to prepare chloramine reagents. Anhydrous sodium sulfite (Fisher Scientific) was used to quench residual chloramines. Isotopically labeled standards, [6-2H] N-nitrosodimethylamine (NDMA-d6, 98%, 1 mg mL1 in methylene chloride) and [14-2H] N-nitrosodin-propylamine (DPNA-d14, 98%, 1 mg mL1 in methylene chloride) were obtained from Cambridge Isotope Laboratories (Andover, MA, USA). A standard solution containing seven N-nitrosamines (2000 mg/mL each in methylene chloride) was purchased from Supelco (SigmaeAldrich). The SPE materials used to extract nitrosamines from aqueous solutions consisted in Supelclean prepacked coconut charcoal EPA 521 tubes, 2 g/6 ml, supplied from Supelco. A mixed standard containing haloacetonitriles (HANs), trichloronitromethane (TCNM) and haloketones (HKs) (EPA 551B Halogenated Volatiles Mix) and internal standard 1,2-dibromopropane were supplied from Supelco. All reagents not specified were obtained from Fisher Scientific.

2.2.

Preparation and analysis of chloramines

Monochloramine (NH2Cl) stock solutions were prepared daily by slowly adding sodium hypochlorite (NaOCl) into a rapidly stirred ammonium chloride (NH4Cl) solution adjusted to pH ¼ 8.5 with sodium hydroxide, and using a Cl:N M ratio of at least 1:1.2 to avoid breakpoint chlorination resulting from local excess of hypochlorite (Mitch and Sedlak, 2002). Adjusting the pH at 8.5 minimizes the disproportionation of NH2Cl to dichloramine (NHCl2), since NHCl2 forms at pH < 8 (U.S. Environmental Protection Agency, 1999) according to the equilibrium:

Chloramination experiments

DBP yield ð%Þ ¼

½DBP ðnMÞ  100 ½Organic compound0 ðnMÞ

(2)

AOX formation rates were calculated as follows:

AOX formation rate ðmol=molÞ ¼

½AOXðmg=L as ClÞ=35:5 ½Organic compound0 ðmMÞ (3)

2.4.

Influence of dissolved O2

Experiments were performed in saturated dissolved O2 solution and in absence of oxygen. The removal of oxygen was operated prior to chloramination by bubbling nitrogen gas through a Teflon line until dissolved O2 concentration was less than 0.3 mg O2/L. The dissolved O2 concentration was monitored using a WTW Oxi 330 oxygen meter. The samples were continuously bubbled until the end of the experiment (2 h contact time). Previous experiments were conducted with NDMA standard solutions in order to verify that nitrogen bubbling for 2 h did not lead to any significant NDMA or chlorinated DBPs stripping.

3167

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

Nitrosamines analysis

NDMA analysis was performed according to the US EPA method (U.S. Environmental Protection Agency, 2004), consisting in a solid-phase extraction (SPE) using coconut charcoal EPA 521 tubes followed by GC/MS analysis in EI mode. Analytical details are provided elsewhere (Le Roux et al., 2010) and summarized below. Chloramination reactions were quenched using 2.5 g sodium sulfite before SPE. Unlike previous studies (Chen and Young, 2009; Schreiber and Mitch, 2006), no ascorbic acid was used because it was found to degrade into furfural during SPE in coconut charcoal tubes, which led to poor NDMA recovery. Prior to the extraction, 200 ng of NDMA-d6 was added to each 1 L sample as an internal standard. Each sample was extracted at a continuous flow rate through the SPE tube. Analytes were eluted from the SPE bed with 15 mL of methylene chloride. Extracts were then filtered through 5 g anhydrous sodium sulfate column to remove residual water. Methylene chloride extracts were then concentrated down to 1 mL under a stream of N2, after addition of DPNA-d14 (200 ng) used as recovery standard. Samples extracts were analyzed immediately after SPE using a HP 6890 series gas chromatograph system coupled with a HP 5973 mass selective detector (MSD) in electron impact (EI) mode. Samples were injected in pulsed splitless mode using helium as the carrier gas. A Varian VF-5ms capillary column (30 m by 0.25 mm i.d. by 0.25 mm film thickness) supplied through Interchim was used to separate the analytes. Quantitative analyses were performed in selected-ion monitoring (SIM) mode. Full scan mode (40e240 m/z) analyses were also conducted for complementary spectral information. This method reached extraction efficiencies of approximately 85%. The method detection limit (MDL) for NDMA at the 99% confidence level was determined to be 33 ng/L.

2.5.2.

HANs, HKs, TCNM and AOX analysis

Chloramination reactions were stopped using 250 mg ascorbic acid prior to HAN and AOX analysis to avoid HANs degradation occurring in the presence of excess sodium sulfite (Croue and Reckhow, 1989). HANs, HKs and TCNM analysis was based on the US EPA 551.1 method (Munch and Hautman, 1995). 50 mL of samples were transferred to amber glass bottles and 1,2-dibromopropane (100 mg/L) was added as an internal standard. Samples were extracted by shaking for 4 min into 3 mL MTBE. Extracts were analyzed using GC/MS (same equipment as used for nitrosamines analysis), along with HANs, HKs and TCNM calibration standards. 1 mL was injected in pulsed splitless mode with an inlet temperature of 150  C. The column temperature program was 40  C held for 3 min, ramping to 55  C at 2  C/min and holding for 1 min, then a ramp of 5  C/min to 85  C, and a final ramp of 40  C/min to 200  C held for 1 min. The MDL for this method is about 0.1 mg/L. AOX were determined using a Dohrmann DX 20 analyzer after adsorption onto activated carbon (European Standard EN 1485, 1996). The detection limit for this method is about 20 mg as Cl/L.

3.

Results and discussion

Fig. 1 shows as example the kinetic results for AOX, DCAN and NDMA formation obtained with 3 mM ranitidine and 2.5 mM

NDMA yield (%)

2.5.1.

60

a

50

3.0 2.5

40

2.0

30

1.5

20

1.0

Cl(+I)

10

0.5

NDMA

0

Concentration (µg/L)

Analytical methods

b

200

Residual chlorine (mM)

2.5.

0.0

150 100 2

DCAN DCAN AOX

1 0

0

40

80 120 160 200 240 Time (hours)

Fig. 1 e NDMA formation from 3 mM ranitidine at pH 8.5 with 10 mM phosphate buffer, 2.5 mM monochloramine. Error bars represent one standard deviation (n [ 3). NDMA molar yields were calculated based upon the initial ranitidine concentration.

NH2Cl at pH 8.5. NH2Cl consumption over 120 h was always about 50% of the initial concentration. Same results were obtained for the other investigated compounds. Results from control samples exhibited similar chloramine decay. Kinetic modeling performed using Copasi software and Jafvert and Valentine’s model (Jafvert and Valentine, 1992) confirmed that monochloramine (NH2Cl) predominantly decays by selfdisproportionation under our experimental conditions (pH 8.5, 10 mM phosphate buffer). Hence, the consumption of NH2Cl by the model compounds investigated was insignificant and could not be quantified. AOX formation leveled off after only 2 h contact time, whereas NDMA and DCAN formation were slower and reached their maximum after 24 h. This observation is in accordance with results from the chlorination of proteins (one of the most important precursors of DCAN in drinking waters), that shows a two-step process (Reckhow et al., 2001). First, rapid reactions with reactive sites form THMs and Total Organic Halides (TOX) (Hureiki et al., 1994), then slow degradation of proteins leads to DCAN formation. A similar behavior for DBPs formation kinetics could occur during the chloramination of ranitidine. The formation of NDMA, HANs and AOX at pH 8.5 from selected compounds was monitored after 5 days of contact time (Table 1). Ranitidine exhibited the highest molar yield with 40.2% NDMA formed. Similar amounts of NDMA were produced after 5 days of contact time for initial monochloramine concentrations of 0.5 mMe2.5 mM and 100 nM ranitidine solutions i.e. for large excess of monochloramine. Yields for the other pharmaceuticals ranged from 0.4 to 8.2% and less than 0.4% for diuron and isoproturon. NDMA formation from DMA is known to be <3% molar conversion (Schmidt et al., 2006; Schreiber and Mitch, 2006) Minocycline and especially ranitidine exhibited higher molar yields than other tertiary amines or DMA.

3168

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

Table 1 e Nitrosamine and DCAN formation from compounds investigated at pH 8.5 (5 days contact time). Molar yieldb (%) (SDc)

AOX formation ratee (mol/mol) (SDc)

Compound concentration (nM)

NH2Cl concentrationa (mM)

14760

4.0

40.22 (1.43)

5.8 (0.2)

0.95 (0.18)

Minocycline

2820

2.5

8.21 (0.72)

1.5 (0.1)

8.98 (0.89)

Doxepine

1730

2.5

2.32 (0.01)

0.5 (0.1)

N.D.

Amitriptyline

3480

2.5

1.15 (0.04)

0.8 (0.4)

N.D.

Mifepristone

3170

2.5

0.39 (0.02)

0.2 (0.1)

N.D.

Isoproturon

5290

2.5

0.34 (0.02)

N.D.

N.D.

810

2.5

0.18 (0.01)

N.D.

N.D.

Compound investigated

Ranitidine

Trifluralin

Molecular structure

Nitrosamined DCAN

(continued on next page)

3169

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

Table 1 (continued ) Compound investigated

Diuron

Molecular structure

Compound concentration (nM)

NH2Cl concentrationa (mM)

16560

4.0

Molar yieldb (%) (SDc)

AOX formation ratee (mol/mol) (SDc)

Nitrosamined DCAN

0.15 (0.01)

N.D.

N.D.

N.D. ¼ Not Detected. a Initial NH2Cl concentration applied to a solution containing a compound investigated in deionized water with 10 mM phosphate buffer (pH 8.5). b Molar yields were calculated based upon the initial compound concentration. c SD ¼ Standard Deviation on 3 replicates. d Nitrosamine formed is NDMA except for trifluralin (DPNA). e AOX formation rates expressed as mol AOX as Cl/mol of initial compound.

Compounds presenting heterocyclic ring in their structure (e.g. furan in ranitidine) produced more NDMA than compounds with DMA functions near carbonyl groups (i.e. diuron and isoproturon) (Schmidt et al., 2006) and compounds with aromatic rings (e.g. minocycline or mifepristone). According to Shen and Andrews (2011), the higher yield observed for ranitidine would be explained by the electron-donating effect of furan group that increases electron density on the nitrogen atom and thus enhance electrophilic substitution of chlorine atom. This mechanism would involve the formation of dimethylchloramine (DMCA), DMA and then NDMA (Mitch and Sedlak, 2004). However, the presence of DMA as a key intermediate could not explain the high yield obtained with ranitidine because the NDMA yields from DMA are always < 3% in literature. An alternative mechanism would involve the nucleophilic substitution of NH2Cl on nitrogen atom instead of electrophilic substitution (i.e. chlorine transfer with formation of a DMCA group). Further research is needed to fully address the formation mechanism of NDMA from dimethylaminomethylfuran group. As suggested by (Shen and Andrews, 2011), ranitidine can be considered as a significant NDMA precursor because 6e39% of ranitidine is excreted as the parent form by human body (Jjemba, 2006) and its metabolites maintain the furan and DMA groups in their structures. Moreover, the removal of ranitidine through WWTP can be relatively low (Castiglioni et al., 2006). The presence of ranitidine and its metabolites in wastewaters could contribute significantly to the high NDMA formation potentials observed at several WWTP, which are much higher than concentrations predicted based upon DMA concentrations in raw waters and calculated following previously proposed formation mechanisms (Mitch and Sedlak, 2004; Mitch et al., 2003). Minocycline, the second highest NDMA precursor of the pool of compounds studied (8.2% NDMA molar conversion) contains two dimethylamine functional groups that probably partly explain the significant formation of NDMA. Amitriptyline and doxepin have similar molecular structures and formed 1.15 and 2.32% of NDMA, respectively. The three carbon atoms between the DMA group and the three rings could explain their lower reactivity compared to ranitidine (Shen and Andrews, 2011). The presence of the oxygen atom in

doxepin would explain the higher yield of NDMA for this molecule compared to amitriptyline. Chloramination of trifluralin led to the formation of 0.18% DPNA, half the formation of NDMA obtained from mifepristone that also incorporates an aromatic ring substituted with a dialkylamine group. The lower yield for trifluralin can be attributed to the electron withdrawing effect of the two nitro groups. The electron withdrawing effect of the carbonyl group would also explain the low formation yield of NDMA from isoproturon and diuron (Schmidt et al., 2006). In full scan mode, the GC/MS chromatogram of the extracts revealed the presence of dimethylformamide (DMF) and dimethylcyanamide (DMC) as by-products of the reaction of monochloramine with diuron or ranitidine. These compounds are known to be UDMH oxidation products, as well as NDMA (Mitch and Sedlak, 2002). However, formation mechanisms of DMC and DMF remain unclear. No other nitrosamine was detected during the experiments with compounds containing dimethylamine functional groups. Ranitidine formed about 10 times less DCAN than NDMA (Table 1). Minocycline was the second highest DCAN precursor (1.5% DCAN yield). For the other compounds studied, the amounts of DCAN formed were quite similar to those of NDMA (<1% yield). No TCAN formation was detected during these experiments. No correlation could be made between NDMA formation and DCAN formation but more DCAN was generally formed when NDMA was produced in higher amounts. Minocycline exhibited the highest AOX formation rate (8.98 mol/mol), which could be related to its highly aromatic and oxygen-containing structure. Ranitidine was the second AOX precursor with 0.95 mol/mol formation rate. The other compounds investigated did not lead to any significant AOX formation in our experimental conditions. These results indicate that compounds producing high amount of NDMA tend also to form more of other DBPs (AOX, and especially DCAN).

3.1.

Influence of nitrites

Previous research (Choi and Valentine, 2003) proposed an “enhanced nitrosation pathway” describing NDMA formation

3170

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

from the reaction of DMA with nitrite and hypochlorite. Nitrites were also found to enhance the formation of NDMA during the chlorination of diuron (Chen and Young, 2009). Because low amount of free chlorine may be present in monochloramine solution, nitrites could contribute to the formation of NDMA by chloramination. Experiments conducted with 1 mM amitriptyline and 1 mM mifepristone showed that NDMA formation was not significantly different in presence and in absence of 1 mM nitrites (Table 2). These results indicate that the formation of NDMA from tertiary amines during chloramination is not enhanced by any nitrosation mechanism, which could have occurred in presence of free chlorine and nitrites.

3.2.

Effect of pH

To assess the influence of pH on the formation of DBPs, NH2Cl (2.5 mM) was applied to ranitidine solutions (3 mM) in deionized water buffered at pH ranging from 4 to 10 (Table 3). NDMA, HANs, HKs, and TCNM were analyzed after a contact time of 5 days. NDMA formation from chloramination of ranitidine exhibited a maximum (59.6% yield) at pH 7.9, which is similar to 62.9% reported in Schmidt et al. (2006) for the same conditions. Amitriptyline and mifepristone followed similar trends, forming much less NDMA at pH 10 than at pH 8 (Table 2). Other studies showed that NDMA formation from chloramination of DMA or diuron varied with pH with a maximum formation rate between pH 7 and 9 (Chen and Young, 2008; Mitch and Sedlak, 2002; Schreiber and Mitch, 2006). Self-decomposition and hydrolysis of NH2Cl at pH < 8 are known to lead to the formation of NHCl2 (Valentine and Jafvert, 1988). Because NHCl2 is known to enhance NDMA formation (Schreiber and Mitch, 2006), then acid-catalyzed disproportionation of NH2Cl into NHCl2 could explain the higher formation of NDMA at pH 7.9 compared to pH > 8. However, kinetic modeling of NH2Cl decomposition indicates that NHCl2 is not present in important amounts at pH 7.9. Furthermore, NDMA formation from ranitidine at pH where NHCl2 is the major specie (i.e. pH w 4) was much lower than at pH 8, indicating that other factors than chloramines speciation may play a role in NDMA formation mechanisms. Thus, ranitidine acidebase equilibrium (pKa ¼ 8.2) could explain the

Table 3 e Effect of pH on NDMA and chlorinated DBPs formation from 3 mM ranitidine and 2.5 mM NH2Cl (5 days contact time). pH

AOX formation rate NDMA DCAN TCAN TCNM 1,1- 1,1,1(mol/mol) DCP TCP

4 5.5 7 7.9 8.5 10

Molar yield (%)

0.2 20.6 42.2 59.6 46.6 10.4

1.33 1.08 1.65 0.81 0.55 0.06

0.31 0.49 0.31 0.28 0.33 0.61

12.57 6.14 3.05 1.43 0.70 0.09

N.D. 1.06 1.51 0.24 0.12 N.D.

N.D. 0.38 1.37 0.03 N.D. 0.02

1.63 1.63 1.74 1.31 1.07 0.56

decrease of NDMA formation when the protonated form of ranitidine decreases at pH > 8 (Fig. 2). At pH < 8, NDMA formation seems to be strongly dependent on the NH2Cl concentration in the solution, and was not enhanced by the presence of NHCl2. As shown in Table 3, important amounts of trichloronitromethane (TCNM) were formed from ranitidine at acidic pH (12.57% at pH 4). The amounts of TCNM formed decreased as the pH was raised from pH 4 to pH 10, but were still higher than other chlorinated DBPs at neutral and basic pH. Whereas NDMA formation was maximum around pH 8, DCAN, 1,1-DCP and 1,1,1-TCP exhibited a maximum formation yield at pH 7. Moreover, TCAN formation from ranitidine was low and relatively constant when varying pH from 4 to 10. The lower concentrations of DCAN and 1,1-DCP at pH > 7 can be explained by basecatalyzed decomposition (Croue and Reckhow, 1989; Reckhow et al., 2001; Yang et al., 2007). AOX formation was constant from pH 4 to 7 and then decreased at alkaline pH (Table 3). As shown in Fig. 3, analyzed DBPs represent only a few percent of the AOX formed. TCNM accounts for 20% of the produced AOX at pH 4. However, the proportion of identified DBPs is decreasing when increasing pH.

3.3.

Influence of dichloramine

To evaluate the influence of NHCl2 on NDMA formation from ranitidine, preformed NHCl2 or NH2Cl (1 mM) were applied to

Table 2 e Effect of pH and NO2L on NDMA formation from amitriptyline and mifepristone over 5 days with 10 mM buffer (phosphate for pH 8.5 and carbonate for pH 10). Expt

Compound investigated

1

Amitriptyline Mifepristone

2

Amitriptyline Amitriptyline þ 1 mM NO 2 Mifepristone Mifepristone þ 1 mM NO 2

Compound concentration (mM)

NH2Cl concentration (mM)

NDMA yielda (%) (SDb)

pH

0.38 0.38 0.35 0.35 1 1 1 1

3.8 3.8 3.8 3.8 3.4 3.4 3.4 3.4

2.37 (0.34) 0.08 (0.01) 1.00 (0.30) 0.04 (0.01) 1.93 (0.15) 1.72 (0.15) 0.89 (0.09) 0.97 (0.09)

8.5 10 8.5 10 8.5 8.5 8.5 8.5

a Molar yields were calculated based upon the initial compound concentration. b SD ¼ Standard Deviation on 3 replicates.

3171

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

NHCl2 þ H2 O/NH2 Cl þ HOCl

(4)

NHCl2 þ NH3 þ Hþ /2 NH2 Cl þ Hþ

(5)

The use of the model showed that the residual chlorine concentrations of 0.3 mM analyzed after 24 h of contact time could be explained by the formation of NH2Cl from NHCl2 decomposition, which is almost complete after 24 h. In this condition, the simulated NH2Cl exposure (i.e. the C.t value) represents about 38% of the NH2Cl exposure from direct NH2Cl addition. Thus, NH2Cl formed from the decomposition of NHCl2 could explain the amounts of NDMA formed during the chloramination of ranitidine using dichloramine. These results seem to indicate that dichloramine would not be involved into the formation of NDMA from ranitidine. No significant differences were observed for DCAN formation after the application of either NH2Cl or NHCl2 to 100 nM ranitidine at pH 8 (Fig. 4).

100 Ranitidine H+ NDMA Yield

80

100

Percentage of AOX (%)

ranitidine solutions. Previous research indicated that NDMA formation from DMA and NHCl2 was much higher than in the presence of NH2Cl (Schreiber and Mitch, 2006). Our results showed that NDMA formation from 100 nM ranitidine buffered at pH 8 and after 24 h was significantly lower with NHCl2 than with NH2Cl (46.8% and 80.2% molar yields respectively, Fig. 4). Total chlorine decay during our experiments with NHCl2 was about 85% after 24 h, while it was only 25% with NH2Cl. Thus, NHCl2 decomposition is more rapid than NH2Cl at pH around pH 8, which could explain why less NDMA was formed in presence of NHCl2. The autodecomposition of NHCl2 in our experiments could be well simulated by the kinetic model of Jafvert and Valentine (1992). According to this model, the hydrolysis of dichloramine (Equation (4)) and inverse dismutation (Equation (5)) lead to the formation of significant amounts of monochloramine.

25

Other compounds HK HAN TCNM

20 15 10 5 0

4

5.5

7

7.9

8.5

10

pH Fig. 3 e AOX repartition between trichloronitromethane (TCNM), haloacetonitriles (HAN: sum of DCAN and TCAN) and haloketones (HK: sum of 1,1-DCP and 1,1,1-TCP) at different pH from 3 mM ranitidine and 2.5 mM NH2Cl. Note the scale break.

3.4.

Influence of dissolved oxygen

It has been demonstrated that dissolved oxygen concentration plays a major role in the formation of NDMA by chloramination of DMA (Schreiber and Mitch, 2006). Moreover, a recent study showed that the formation of NDMA from DMA could be catalyzed by activated carbon, and that the presence of oxygen was a critical factor in this mechanism (Padhye et al., 2010). In order to assess whether or not dissolved oxygen would influence the formation of NDMA from the chloramination of other model compounds, 2.7 mM NH2Cl was applied to 3 mM ranitidine during 2 h in presence and in absence of dissolved O2. NDMA formation was significantly inhibited for low oxygen concentration (w 0.2 mg O2/L) compared to ambient O2 concentration (w 9 mg O2/L) (molar yields of 4.01% and 54% respectively, Fig. 5a). Dissolved O2 concentration did not affect AOX formation as much as NDMA formation (Fig. 5b). Moreover, DCAN

100

%

60

NDMA

Simulated NH2Cl residual

80

20 0 3

4

5

6

7 pH

8

9

10

Fig. 2 e Effect of pH on NDMA formation from 3 mM ranitidine and 2.5 mM monochloramine over 5 days with 10 mM buffer (acetate for pH 4.0e5.5, phosphate for pH 7.0e8.5 and carbonate for pH 10); and NH2Cl residuals calculated using Jafvert and Valentine model (1992). NDMA yields were calculated based on the initial ranitidine concentration; percentages of NH2Cl residuals were calculated based on the initial NH2Cl concentration.

Yields (%)

40

DCAN

60

40

20

0

NH2Cl

NHCl2

Fig. 4 e NDMA and DCAN formation after 24 h following the application of 1 mM monochloramine or dichloramine to 100 nM ranitidine buffered at pH 8.

3172

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

10 0.2 mg O2 /L 8.6 mg O2 /L

Yields (%)

formation was not influenced by dissolved O2 concentration while the formation of other halogenated DBPs containing oxygen atoms (nitro or ketones functional groups, i.e. TCNM, 1,1-DCP and 1,1,1-TCP) was approximately an order of magnitude lower in the presence of 0.2 mg O2/L (Fig. 6). This indicates that dissolved oxygen could be incorporated into those DBPs, i.e. an oxygen atom of dissolved oxygen could serve as a source for the oxygen atom in nitroso or ketone functions of DBPs. Further research with model compounds and using inhibitors of oxygen species need to be done to elucidate the mechanisms involved into the formation of NDMA and to understand the role of dissolved oxygen.

1

0.1

0.01 DCAN

4.

Conclusion

 Even if concentrations of compounds used for our study were relatively high and are not likely to be found in natural waters, we observed that several nitrogenous anthropic compounds can lead to important concentrations of N-DBPs including NDMA, DCAN, 1,1-DCP, or TCNM. From the seven compounds investigated in our study, four compounds contain dimethylamine functional groups and exhibited yields higher than 1.15% (ranitidine, minocycline, doxepin, amitriptyline). Especially, the pharmaceutical ranitidine is

NDMA yield (%)

80

a

0.2 mg O2/L 8.6 mg O2/L

60

40

20

0

mol AOX as Cl / mol ranitidine

2.0

8.5

pH

b

9.5

0.2 mg O2/L 8.6 mg O2/L

1.6

1,1-DCP

TCNM

1,1,1-TCP

Fig. 6 e Effect of dissolved oxygen on DCAN, 1,1-DCP, TCNM and 1,1,1-TCP formation from 3 mM ranitidine and 2.7 mM NH2Cl over 2 h at pH 8.5 with 10 mM phosphate buffer.

of great concern regarding its high molar yield into NDMA (w60% at pH 7.9), as shown in earlier studies.  Such differences in NDMA formation can not be explained by the release of DMA and the reactions of DMA with chloramines. More simple compounds than those described in the present work need to be studied to improve our understanding of molecular structure influence on the formation of NDMA.  Our results demonstrate that the reaction of NHCl2 with ranitidine would not form more NDMA than NH2Cl. However, we confirmed the implication of dissolved oxygen in NDMA formation mechanisms. Dissolved oxygen was found to play a role into the formation of other oxygencontaining DBPs (TCNM, 1,1-DCP and 1,1,1-TCP) but did not influence DCAN formation. These results need further investigation to better understand the incorporation of dissolved oxygen into DBPs.  Considering the high conversion of ranitidine to NDMA, the use of chloramination as a disinfection for wastewaters containing ranitidine can lead to the formation of important amounts of NDMA. This could explain the high NDMA formation potentials observed at several WWTP, which are much higher than concentrations predicted based upon DMA concentrations in raw waters.

1.2

Acknowledgment

0.8

We would like to thank the French Ministry of Higher Education and Research (Ministe`re de l’Enseignement Supe´rieur et de la Recherche) for its financial support.

0.4 0.0 8.5

9.5 pH

Fig. 5 e Effect of dissolved oxygen and pH on (a) NDMA and (b) AOX formation from 3 mM ranitidine and 2.7 mM NH2Cl, over 2 h with 10 mM buffer. NDMA molar yields were calculated based upon the initial ranitidine concentration.

references

Castiglioni, S., Bagnati, R., Fanelli, R., Pomati, F., Calamari, D., Zuccato, E., 2006. Removal of pharmaceuticals in sewage treatment plants in Italy. Environ. Sci. Technol. 40 (1), 357e363.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

Chen, W., Young, T.M., 2008. NDMA formation during chlorination and chloramination of aqueous diuron solutions. Environ. Sci. Technol. 42 (4), 1072e1077. Chen, W., Young, T.M., 2009. Influence of nitrogen source on NDMA formation during chlorination of diuron. Water Res. 43 (12), 3047e3056. Choi, J., Duirk, S.E., Valentine, R.L., 2002. Mechanistic studies of N-nitrosodimethylamine-(NDMA) formation in chlorinated drinking water. J. Environ. Monit. 4 (2), 249e252. Choi, J., Valentine, R.L., 2002a. A kinetic model of N-nitrosodimethylamine (NDMA) formation during water chlorination/chloramination. Water Sci. Technol. 46 (3), 65e71. Choi, J., Valentine, R.L., 2002b. Formation of Nnitrosodimethylamine (NDMA) from reaction of monochloramine: a new disinfection by-product. Water Res. 36 (4), 817e824. Choi, J., Valentine, R.L., 2003. N-Nitrosodimethylamine formation by free-chlorine-enhanced nitrosation of dimethylamine. Environ. Sci. Technol. 37 (21), 4871e4876. Croue, J.P., Reckhow, D.A., 1989. Destruction of chlorination byproducts with sulfite. ESandT Contents 23 (11), 1412e1419. Eaton, A.D., Clesceri, L.S., Greenberg, A.E. (Eds.), 1995. Standard Methods for the Examination of Water and Wastewater, nineteenth ed. American Public Health Association/American Water Works Association/Water Environment Federation Publishers, Washington D.C., USA. European Standard EN 1485, 1996. Water Quality Determination of Adsorbable Organically Bound Halogens (AOX). Gerecke, A.C., Sedlak, D.L., 2003. Precursors of N-nitrosodimethylamine in natural waters. Environ. Sci. Technol. 37 (7), 1331e1336. Hoigne, J., Bader, H., 1988. The formation of trichloronitromethane (chloropicrin) and chloroform in a combined ozonation/chlorination treatment of drinking water. Water Res. 22 (3), 313e319. Hureiki, L., Croue´, J., Legube, B., 1994. Chlorination studies of free and combined amino acids. Water Res. 28 (12), 2521e2531. Jafvert, C.T., Valentine, R.L., 1992. Reaction scheme for the chlorination of ammoniacal water. Environ. Sci. Technol. 26 (3), 577e786. Jjemba, P.K., 2006. Excretion and ecotoxicity of pharmaceutical and personal care products in the environment. Ecotoxicol. Environ. Saf. 63 (1), 113e130. Joo, S.H., Mitch, W.A., 2007. Nitrile, aldehyde, and halonitroalkane formation during chlorination/chloramination of primary amines. Environ. Sci. Technol. 41 (4), 1288e1296. Kolpin, D.W., Furlong, E.T., Meyer, M.T., Thurman, E.M., Zaugg, S. D., Barber, L.B., Buxton, H.T., 2002. Pharmaceuticals, hormones, and other organic wastewater contaminants in U. S. streams, 1999e2000: a national reconnaissance. Environ. Sci. Technol. 36 (6), 1202e1211. Le Roux, J., Gallard, H., Croue, J.P., 2010. Formation of NDMA by chloramination of nitrogenous pharmaceuticals. Water Pract. Technol. 5 (4). Mitch, W.A., Gerecke, A.C., Sedlak, D.L., 2003. A N-Nitrosodimethylamine (NDMA) precursor analysis for chlorination of water and wastewater. Water Res. 37 (15), 3733e3741. Mitch, W.A., Sedlak, D.L., 2002. Formation of Nnitrosodimethylamine (NDMA) from dimethylamine during chlorination. Environ. Sci. Technol. 36 (4), 588e595. Mitch, W.A., Sedlak, D.L., 2004. Characterization and fate of N-Nitrosodimethylamine precursors in municipal wastewater treatment plants. Environ. Sci. Technol. 38 (5), 1445e1454. Muellner, M.G., Wagner, E.D., Mccalla, K., Richardson, S.D., Woo, Y., Plewa, M.J., 2007. Haloacetonitriles vs. regulated

3173

haloacetic acids: are nitrogen-containing DBPs more toxic? Environ. Sci. Technol. 41 (2), 645e651. Munch, D.J., Hautman, D.P., 1995. EPA Method 551.1 Determination of Chlorination Disinfection Byproducts, Chlorinated Solvents, and Halogenated Pesticides/Herbicides in Drinking Water by Liquid-Liquid Extraction and Gas Chromatography with Electron-Capture Detection, Revision 1. 0. Available online from. http://www.epa.gov/sam/pdfs/EPA551.1.pdf (accessed 22.04.09.). National Cancer Institute, 1978. Bioassay of Chloropicrin for Possible Carcinogenicity (CAS No. 76-06-2). National Toxicology Program Carcinogenesis Technical Report Series. U.S. Department of Health and Human. Padhye, L., Wang, P., Karanfil, T., Huang, C., 2010. Unexpected role of activated carbon in promoting transformation of secondary amines to N-nitrosamines. Environ. Sci. Technol. 44 (11), 4161e4168. Plewa, M.J., Wagner, E.D., Jazwierska, P., Richardson, S.D., Chen, P.H., McKague, A.B., 2004. Halonitromethane drinking water disinfection byproducts: chemical characterization and mammalian cell cytotoxicity and genotoxicity. Environ. Sci. Technol. 38 (1), 62e68. Radjenovic, J., Petrovic, M., Barcelo´, D., 2009. Fate and distribution of pharmaceuticals in wastewater and sewage sludge of the conventional activated sludge (CAS) and advanced membrane bioreactor (MBR) treatment. Water Res. 43 (3), 831e841. Reckhow, D.A., MacNeill, A.L., Platt, T.L., MacNeill, A.L., McClellan, J.N., 2001. Formation and degradation of dichloroacetonitrile in drinking waters. J. Water Supply Res. Technol. AQUA 50 (1), 1e13. Schmidt, C.K., Sacher, F., Brauch, H., 2006. Strategies for minimizing formation of NDMA and other nitrosamines during disinfection of drinking water. In: Proceedings of the American Water Works Association Water Quality Technology Conference. Denver, CO. Schneider, M., Quistad, G.B., Casida, J.E., 1999. Glutathione activation of chloropicrin in the salmonella mutagenicity test. Mutat. Res./Genet. Toxicol. Environ. Muta. 439 (2), 233e238. Schreiber, I.M., Mitch, W.A., 2005. Influence of the order of reagent addition on NDMA formation during chloramination. Environ. Sci. Technol. 39 (10), 3811e3818. Schreiber, I.M., Mitch, W.A., 2006. Nitrosamine formation pathway revisited: the importance of chloramine speciation and dissolved oxygen. Environ. Sci. Technol. 40 (19), 6007e6014. Shen, R., Andrews, S., 2011. Demonstration of 20 pharmaceuticals and personal care products (PPCPs) as nitrosamine precursors during chloramine disinfection. Water Res. 45 (2), 944e952. Thibaud, H., de Laat, J., Merlet, N., Dore´, M., 1987. Chloropicrin formation in aqueous solution: effect of nitrites on precursors formation during the oxidation of organic compounds (Formation de chloropicrine en milieu aqueux: influence des nitrites sur la formation de precurseurs par oxydation de composes organiques). Water Res. 21 (7), 813e821. U.S. Environmental Protection Agency, 1987. Integrated Risk Information System (IRIS), N-nitrosodimethylamine. Available online from. Office of Research and Development (ORD), National Center for Environmental Assessment. www.epa. gov/iris/subst/0045.htm (accessed 20.10.08.). U.S. Environmental Protection Agency, 1999. Alternative Disinfectants and Oxidants. Available online from. EPA Guidance Manual. http://www.epa.gov/OGWDW/mdbp/ alternative_disinfectants_guidance.pdf (accessed 10.04.09.). U.S. Environmental Protection Agency, 2004. Method 521: Determination of Nitrosamines in Drinking Water by Solid Phase

3174

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 6 4 e3 1 7 4

Extraction and Capillary Column Gas Chromatography with Large Volume Injection and Chemical Ionization Tandem Mass Spectrometry (MS/MS). Available online from. EPA/600/R-05/054. http://www.epa.gov/nerlcwww/m_521.pdf (accessed 20.10.08.). Valentine, R.L., Jafvert, C.T., 1988. General acid catalysis of monochloramine disproportionation. Environ. Sci. Technol. 22 (6), 691e696.

Yang, X., Shang, C., Westerhoff, P., 2007. Factors affecting formation of haloacetonitriles, haloketones, chloropicrin and cyanogen halides during chloramination. Water Res. 41 (6), 1193e1200. Zuccato, E., Calamari, D., Natangelo, M., Fanelli, R., 2000. Presence of therapeutic drugs in the environment. The Lancet 355 (9217), 1789e1790.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

A real-time RT-PCR method to detect viable Giardia lamblia cysts in environmental waters Robert H. Baque a, Amy O. Gilliam a,*, Liza D. Robles a,1, Walter Jakubowski b, Theresa R. Slifko c a

Orange County Utilities, 9124 Curry Ford Rd, Orlando, FL 32825, USA WaltJay Consulting, Spokane, WA 99223, USA c County Sanitation Districts of Los Angeles County, Whittier, CA 90601, USA b

article info

abstract

Article history:

Currently, USEPA Method 1623 is the standard assay used for simultaneous detection of

Received 10 June 2010

Giardia cysts and Cryptosporidium oocysts in various water matrices. However, the method

Received in revised form

is unable to distinguish between species, genotype, or to assess viability. Therefore, the

16 March 2011

objective of the present study was to address the shortcomings of USEPA Method 1623 by

Accepted 16 March 2011

developing a novel molecular-based method that can assess viability of Giardia cysts in

Available online 23 March 2011

environmental waters and identify genotypes that pose a human health threat (assemblage groups A and B). Primers and TaqMan
probes were designed to target the beta-

Keywords:

giardin gene in order to discriminate among species and assemblages. Viability was

Giardia

determined by detection of de-novo mRNA synthesis after heat induction. The beta-giardin

Viability

primer/probe sets were able to detect and differentiate between Giardia lamblia assem-

Assemblage

blages A and B, and did not detect Giardia muris (mouse species) or G. lamblia assemblages

Beta-giardin

C, D, E and F (non-human), with the exception of Probe A which did detect G. lamblia

Reclaimed water

assemblage F DNA. Additionally, DNA or cDNA of other waterborne organisms were not detected, suggesting that the method is specific to Giardia assemblages. Assay applicability was demonstrated by detection of viable G. lamblia cysts in spiked (assemblage B) and unspiked (assemblage A and B) reclaimed water samples. Published by Elsevier Ltd.

1.

Introduction

Giardia is a ubiquitous flagellated protozoan parasite that can cause diarrhea and gastrointestinal disease in humans (Adam, 2001). It is found in a wide variety of wild and domestic animals and is transmitted by the fecal-oral route. The cyst is considered to be the infectious stage of the organism and can be transmitted by ingesting contaminated food, by personto-person contact or by ingestion of contaminated water (Barbosa et al., 2008). There are several Giardia species with only Giardia lamblia (synonyms Giardia duodenalis, Giardia

intestinalis) being infectious to humans. G. lamblia is known to have at least seven genotypes (assemblages) but only two of these, designated assemblages A and B, cause infections in humans (Read et al., 2004). Various methods have been developed to discern cyst viability or genotype, however none exist that are capable of both. USEPA Method 1623 is a performance-based method applicable to the determination of Cryptosporidium and Giardia in aqueous matrices (USEPA, 2005). The three general components of Method 1623 include water concentration by filtration, immunomagnetic separation (IMS) of the oocysts

* Corresponding author. Tel.: þ1 407 254 9551; fax: þ1 407 254 9558. E-mail address: [email protected] (A.O. Gilliam). 1 Present address: Institute of Translational Research, Florida Hospital, Orlando, FL 32804, USA. 0043-1354/$ e see front matter Published by Elsevier Ltd. doi:10.1016/j.watres.2011.03.032

3176

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

and cysts from the material captured, and immunofluorescent assay (IFA) for determination of the oocyst and cyst concentrations. Although it is a standard method utilized by water utilities, USEPA Method 1623 is limited in the information it provides. The most pronounced deficiency is that it is unable to definitively characterize cysts as viable or to determine Giardia species or genotype, which is pertinent for the determination of public health significance. The method relies upon antibody staining for confirming the presence of Giardia cysts, while “potential viability” is determined through both DAPI staining and internal feature characterization. Additionally, the microscopic examination of stained internal features can be subjective and labor intensive (Lemos et al., 2005). More importantly, it cannot evaluate if intact cysts exposed to UV radiation and other disinfectants are alive or dead. Molecular methods have been developed for clinical diagnostic purposes as well as genotyping (Singh, 1997). Some of these earlier methods used radiolabelled and non-radiolabelled probes that hybridized to various parasitic target DNA sites. For Giardia, the small-subunit ribosomal RNA (ssu-rRNA) gene was one of the first targets for cDNA probing; however this assay was only genus specific (Abbaszadegan et al., 1991). Other studies investigated options for detecting Giardia species. The beta-giardin gene is a commonly investigated gene believed to be specific to Giardia spp. and was used as the target for a PCR assay that was able to differentiate between species of Giardia, but was not capable of quantifying cysts in environmental waters (Mahbubani et al., 1992). Another report described a real-time quantitative PCR (qPCR) assay capable of detecting DNA corresponding to the theoretical amount of a single cyst (Guy et al., 2003). The work produced by Guy et al. (2003) revealed the potential to quantify the levels of Giardia found in environmental waters. These advancements in methodology increased sensitivity and genotyping capabilities; however, cyst viability was not addressed. The ability to discriminate between viable and non-viable organisms is vital to the accurate assessment of water quality, since only viable Giardia cysts are infectious. Even though in vivo methods, such as animal infectivity models, can be used to detect viable cysts, the procedures have many limitations. These methods do not address human health risks because they are unable to differentiate between the currently recognized assemblages that affect humans. Previous work by Abbaszadegan et al. (1997) demonstrated the ability to determine viability of Giardia by heat induction of the hsp70 gene. Likewise, Mahbubani et al. (1991) suggested that the beta-giardin gene present in G. lamblia is heat inducible. These findings served as the basis for the current project. In theory, exposure of cysts to constant heat should cause de novo mRNA production from viable organisms that can be detected by real-time reverse-transcription PCR (qRTPCR). If the cysts are viable, they can be detected using qRTPCR directed to the beta-giardin gene. Recently, work by Bertrand et al. (2009) demonstrated the ability to assess Giardia spp. viability by measuring an increase in transcriptional activity but does not address assemblage specificity. The present report describes the species-specific G. lamblia primers and TaqMan
assemblage-specific probes that were developed, optimized, and evaluated for a qRT-PCR assay capable of determining the presence/absence of viable cysts

and differentiating between G. lamblia genotypes (assemblages A and B) gDNA.

2.

Materials and methods

2.1.

Organisms

Frozen aliquots of G. lamblia assemblage A (strain WB) and assemblage B (strain H3) trophozoites were obtained from Waterborne, Inc. (New Orleans, LA). These aliquots were thawed and maintained as axenic cultures in TYI-S33 media. Live G. lamblia (strain H3) and Giardia muris cysts were obtained from Waterborne, Inc. (New Orleans, LA). Waterborne, Inc. confirm genus and species by direct immunofluorescence microscopy using genus-specific monoclonal antibodies. Cysts were stored in phosphate buffered saline (PBS) containing penicillin, streptomycin, gentamycin and 0.01% Tween 20. Cysts were kept at 4  C until time of use. Live Cryptosporidium parvum oocysts and Entamoeba intestinalis cysts were obtained from Waterborne, Inc. (New Orleans, LA). C. parvum oocysts were stored in PBS containing antibiotics and kept at 4  C. E. intestinalis cysts were stored in PBS with penicillin, streptomycin, and gentamycin. A fecal sample from a Cyclospora cayetanensis infected patient, stored in 5% potassium dichromate was provided by J.H. Cross, Uniformed Services University of the Health Sciences (Bethesda, MD). Non-viable Toxoplasma gondii (strain RH1) tachyzoites were obtained from BIODESIGN International/OEM Concepts (Saco, ME) and stored in PBS at 4  C until time of use. Live Ascaris suum ova were obtained from Excelsior Sentinel Inc. (Trumansburg, NY). The following organisms were obtained from ATCCª (Manassas, VA): Klebsiella (Raoultella) terrigena (#33257), Escherichia coli (#11775), and poliovirus type 2 (PV-2) (#VR-301). G. lamblia genomic DNA (gDNA), assemblages C and E, were provided by Dr. Lihua Xiao, Centers for Disease Control and Prevention (Atlanta, GA). Partial beta-giardin gene clones from G. lamblia (assemblages D and F) were provided by Dr. Simone Caccio, Istituto Superiore di Sanita (Rome, Italy).

2.2.

Nucleic acid extraction/isolation

DNA was extracted from organisms using DNAzol reagent (Invitrogen, Carlsbad, CA). Organisms were resuspended in 1 mL of DNAzol reagent and placed through 8 cycles of both freezing for 1 min in liquid nitrogen and boiling for 1 min in water. This process was followed by proteinase K (10 mg) and poly-A (10 mg) incubation for 1 h at 56  C. Following the 1 h incubation, 500 mL of chilled (w4  C) 100% ethanol was added. Samples were centrifuged at 4000  g for 4 min at room temperature. The supernatant was discarded and the DNA pellet was washed in 800 mL of 75% chilled ethanol, followed by centrifugation at 4000  g for 4 min at room temperature. The supernatant was discarded and the DNA pellet was resuspended in 50 mL of nuclease-free 8 mM sodium hydroxide. DNA extraction from the C. cayetanensis oocysts was performed using a QIAamp
DNA Stool Mini Kit (Qiagen, Valencia, CA). Poliovirus type 2 RNA was extracted using a QIAamp
Viral RNA Mini Kit (Qiagen, Valencia, CA) after propagation using BGMK cells and then reverse transcribed into cDNA as

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

described by Reynolds et al. (1996). The genomic DNA or cDNA, in the case of poliovirus type 2, (w50 ng for each organism) was used for subsequent qPCR experiments. Quantification of gDNA was performed by spectrophotometer analysis. Nucleic acid absorbance of ultraviolet light (l ¼ 260 nm) was measured using an Eppendorf BioPhotometer (Brinkmann Instruments Inc., Westbury, NY).

2.3.

mRNA induction and extraction

A maximum volume of 100 mL containing 1000 Giardia cysts were heat induced by placing cysts into 1.5 mL microcentrifuge tubes and incubated in a dry-bath block at 42  C for 40 min. Extraction of mRNA was performed immediately after induction using the Dynabeads
mRNA Direct Micro Kit (Invitrogen-Dynal, Carlsbad, CA). One-hundred microliters of lysis-binding buffer (100 mM Tris, [pH 7.5], 500 mM LiCl, 10 mM EDTA [pH 8.0], 1% lithium dodecyl sulfate, 5 mM dithiothreitol) was added to each sample in order to disrupt the cysts. Samples were then mixed by pipetting and allowed to stand at room temperature for 10 min. During this 10 min incubation step, the oligo(dT)25 magnetic beads (Invitrogen-Dynal, Carlsbad, CA) were prepared following the Dynabeads
mRNA Direct Micro Kit’s specifications. After the 10 min incubation, the samples were mixed by pipetting and added to the prepared beads. Hybridization was performed by placing the microcentrifuge tubes in a rotator, set at 35 rpm for 5 min. Using a magnetic particle concentrator, the mRNA was captured and washed following the Dynabeads
mRNA Direct Micro Kit specifications. The mRNA was eluted using 11 mL of 10 mM TriseHCl and placed in a waterbath set at 65  C for 2 min. After the 2 min incubation, the samples were placed in a magnetic particle concentrator and the eluate was transferred to a new microcentrifuge tube. Eluted mRNA was stored at 4  C until DNase treatment. Following manufacturer’s specifications, Ambion Turbo DNA-free Kit (Austin, TX) was used to degrade contaminating DNA.

2.4.

Oligonucleotides

In order to identify conserved regions of the beta-giardin gene among Giardia spp., a representative number of available sequences from G. lamblia assemblages A to F and G. muris were used for sequence alignment analysis using ClustalW software, (DNASTAR, Madison, WI). The following sequences from environmental isolates with GenBank accession numbers were used: G. lamblia assemblage A [M36728 (human), AY258617 (human), X07919 (human), AY072723 (human), AY072724 (human), AY545643 (human), AY545644 (human), AY545645 (human), AY655702 (calf), AY545649 (calf), AY545642 (dog)]; G. lamblia assemblage B [AY258616 (human), AY072725 (human), AY072726 (human), AY072727 (human), AY072728 (human), AY647265 (calf), AY647266 (calf)]; G. lamblia assemblage C [AY545646 (dog), FJ009206 (dog), HM171701 (human)]; G. lamblia assemblage D [AY545647 (dog), AY545648 (dog)]; G. lamblia assemblage E [AY655703 (calf), AY545650 (calf), AY653159 (calf), AY072729 (pig)], G. lamblia assemblage F [AY647264 (cat), GU574802 (rat) ], G. muris [AY258618 (mouse)]. The beta-giardin forward and reverse primers were designed to detect G. lamblia by recognizing a region of the

3177

Portland-1 strain beta-giardin gene (GenBank Accession #M36728). This primer set was based on the region encompassing nucleotides 437e579. Two assemblage-specific TaqMan
probes, Probe A and Probe B, were designed to recognize assemblages A and B, respectively. Probe A was based on the sequence of the Portland-1 strain of G. lamblia. Probe B was based on the sequence of the BAH8 strain of G. lamblia (GenBank Accession # AY072727). The beta-giardin primer set and TaqMan
probes were designed using Primer Express software (Applied Biosystems, Foster City, CA). Specificity of the primer set and probes were initially screened by performing a nBLAST homology search (http://blast.ncbi. nlm.nih.gov/Blast.cgi). The primer set was synthesized by MWG-biotech (High Point, NC). Probes A and B were synthesized by Applied Biosystems (Foster City, CA) labeled at the 50 end with 6-carboxyfluorescein (6-FAM) and 30 end labeled with a minor groove binding protein (MGB; non-fluorescent quencher).

2.5.

cDNA generation

Isolated mRNA was reverse transcribed using a 9700 GeneAmp PCR System (Applied Biosystems, Foster City, CA). Ten microliters of isolated mRNA was used as template. A final volume of 25 mL was used for each reaction. The following reagents and buffers (Applied Biosystems, Foster City, CA) were utilized at the specified final concentrations: 1 RT Buffer without MgCl2, 1.5 mM MgCl2, 1.0 mM dNTP mix, 500 nM downstream primer, Multiscribe Reverse Transcriptase (12.5 U/rxn), and RNAse Inhibitor (5.0 U/rxn). For environmental samples, non-acetylated BSA was used (400 ng/ml). The following temperature conditions were used: incubation at 24  C for 10 min; incubation at 44  C for 60 min; incubation at 99  C for 5 min followed by a 4  C hold.

2.6.

qPCR

Following cDNA generation, qPCR was performed using the TaqMan
Universal PCR Master Mix Kit (Applied Biosystems, Foster City, CA). Eleven microliters of isolated cDNA was used as template. A final volume of 50 mL was used for each reaction with the following concentrations of reagents and buffers (Applied Biosystems, Foster City, CA): 1 TaqMan
Universal PCR Master Mix, 600 nM of each forward and reverse betagiardin primer, 150 nM of either TaqMan
Probe A or B. In order to determine if conditions in each reaction were suitable for successful qPCR detection, each one included 1 Exogenous Internal Positive Control (IPC) DNA and 1 Exogenous IPC Mix (which contains IPC primers). Non-amplification control samples included exogenous IPC DNA, IPC mix and 1 IPC Blocking reagent. IPC DNA, IPC mix and 1 IPC Blocking reagents were obtained from an Exogenous Internal Positive Control Kit (Applied Biosystems, Foster City, CA). Amplifications were performed in a 7300 Real Time PCR System (Applied Biosystems, Foster City, CA). Cycling conditions consisted of a 2 min incubation at 50  C, followed by one 10 min incubation at 95  C, followed by 40 cycles of alternating temperatures of 95  C for 15 s and 60  C for 1 min. For the unspiked environmental samples, 45 cycles were utilized. At the end of each cycle, normalized fluorescence was recorded by the 7300 Real

3178

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

Time PCR system (Applied Biosystems, Foster City, CA). For quality control purposes a no-template control and nonamplification control was included with every batch of realtime PCR assays. The non-amplification control serves as a negative control for the IPC.

2.7.

3.

Results and discussion

3.1.

Assay overview

de novo mRNA transcription assessment

Seven separate lots of live G. lamblia (assemblage B) cysts were obtained over an 18 month time period. Each lot of cysts was used for experimentation within two weeks of receipt from Waterborne, Inc. (New Orleans, LA). Each trial consisted of four aliquots of approximately 1000 or 100 cysts, each representing the four treatment groups used for this series of experiments: 1) live, induced; 2) live, non-induced; 3) killed, induced and 4) killed, non-induced. For both live and killed cysts, transcription activity was measured by using non-induced sample results as the baseline and compared to their respective induced sample results. Cysts were killed by being placed in boiling water for 1 min or by exposure to chlorine concentrations of 2e2.5 mg/L for 1 h at 25  C and pH 7.0.

2.8.

induced samples for each set of probes were analyzed by qRTPCR. Each aliquot represented 4 L of the original sample collected.

Environmental water samples

Water samples were obtained from a reclamation facility located in Orlando, Florida. One liter of final effluent (post tertiary treated and disinfected) reclaimed water was collected. The sample was divided into two 500 mL aliquots and placed into 500 mL conical tubes. One tube was spiked with w5000 live G. lamblia assemblage B (H3) stock cysts. The other tube was spiked with w5000 killed (boiled for 1 min) G. lamblia assemblage B (H3) cysts. Tubes were then centrifuged at 1000  g for 15 min at 4  C. The supernatant was aspirated until 5 mL remained. The final pellet was resuspended and equally aliquoted into five microcentrifuge tubes along with two wash steps using PBS. Microcentrifuge tubes were then centrifuged at 800  g for 5 min. The supernatant was aspirated leaving w100 mL in each microcentrifuge tube and w1000 cysts. Two of the five aliquots were assayed for viability. Six 20 L unspiked (post tertiary treated and disinfected) reclaimed water samples were collected from the same reclamation facility using Envirochek HV sampling capsules. Samples were filtered on-site following USEPA Method 1623 guidelines (USEPA, 2005), transported to the laboratory on ice and stored at 4  C. Filters were eluted with a 5% w/v sodium hexametaphosphate solution and collected into 500 mL conical tubes. Tubes were then centrifuged at 1000  g for 15 min at 4  C. The supernatant was aspirated until 5 mL remained. The final pellet was resuspended and equally aliquoted into five microcentrifuge tubes along with two wash steps using PBS. Microcentrifuge tubes were then centrifuged at 800  g for 5 min. The supernatant was aspirated leaving w100 mL in each microcentrifuge tube. Two of the aliquots were screened using Probe A and two aliquots were screened using Probe B for assemblage A and B cysts, respectively. For each pair of aliquots, one aliquot was subjected to heat induction and the other was left as the non-induced control. One aliquot was not analyzed. Both the induced and non-

The result of this work was the development of a qRT-PCR assay, referred to as Orange County Utilities qRT-PCR (OCU qRT-PCR), designed to detect viable G. lamblia cysts, which are potentially infectious to humans. The assay can generally be divided into four stages: (1) heat induction; (2) mRNA isolation; (3) cDNA generation; and (4) qPCR analysis as shown in Fig. 1.

3.2.

PCR primer and probe design

The development of a single primer set and two distinct probes served as the foundation for developing an assay capable of discriminating between G. lamblia genotypes. The target gene for this primer set and probe duo was beta-giardin, which is believed to be unique to Giardia spp. (Palm et al., 2002) therefore reducing the probability of cross reactivity with genetic material from other organisms. After sequence alignment analysis, a span of conserved nucleotide bases was found flanking a region containing assemblage-specific variability. This region was found among all assemblage A sequences analyzed in this study and, as a result, Portland-1 strain beta-giardin gene (GenBank Accession #M36728) was chosen as a representative sequence for assemblage A (Holberton et al., 1988). Likewise, a region containing assemblage B specific variability was found among all assemblage B sequences analyzed in this study, and the BAH8 strain of G. lamblia (GenBank Accession #AY072727) was chosen as a representative sequence for assemblage B (Caccio et al., 2002). The A29 strain of G. lamblia (GenBank Accession #AY545646) was chosen as the representative sample for assemblage C and strain A21 (GenBank Accession #AY545647) was chosen for assemblage D (Lalle et al., 2005). As with

Fig. 1 e Simplified schematic overview of developed Orange County real-time RT-PCR Giardia Viability Assay (OCU qRT-PCR).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

assemblages A through D, a region containing assemblage E specific variability was found among all assemblage E sequences and the Unk2 strain (GenBank Accession #AY655703) was used to represent assemblage E (Trout et al., 2004). Two strains of assemblage F, A101 (GenBank Accession #AY647264) and BR1 (GenBank Accession # GU574802) were screened independently to determine primer and probe specificity. Using Portland-1 strain beta-giardin gene sequences as a nucleotide positioning reference point, OCU primer 36 (forward), at position 437e455 and OCU primer 39 (reverse), at position 561e579, were designed to be species specific. These primers amplify a target region found within G. lamblia assemblage groups AeD. Assemblage specificity was incorporated into the assay by the development of TaqMan
probes. As previously mentioned only assemblage A and assemblage B are believed to have the ability to infect humans. Subsequently, probes

targeting these assemblages were designed. Beta-giardin gene probes found at positions 463e475 and 461e473, were designed to target assemblages A and B, respectively (Table 2). A single primer set helped to eliminate complications that may arise when multiplexing in a PCR.

3.3.

qPCR sensitivity assessment

In order to quantify method sensitivity, serial dilutions of genomic DNA (gDNA) were initially used. According to previously published data (Guy et al., 2003), 195 fg (fg) correspond to the theoretical estimate of gDNA found within a single Giardia cyst containing four nuclei. Using this information, it was calculated that w12 fg correspond to a single gene copy and this was used to determine primer/probe sensitivity. The 10-fold serial dilutions of gDNA from axenic cultures of G. lamblia assemblage A (strain WB) and assemblage B (strain H3) trophozoites were analyzed by qPCR. The following gDNA amounts were tested: 250 ng, 25 ng, 2.5 ng, 2.5  101 ng, 2.5  102 ng, 2.5  103 ng, 2.5  104 ng, 1.5  104 ng. Real-time PCR assays using primer set OCU 36/39 with probe A (Fig. 2A) and primer set OCU 36/39 with probe B (Fig. 2B) gave Ct (cycle threshold) values which were linear over a 6 log10 concentration range. In addition to the 10-fold serial dilutions, a separate dilution containing 150 fg of gDNA was used in order to assess if the OCU qRT-PCR’s primer/probe sets could detect the amount of gDNA expected to be found from a single cyst. Taking into account that 195 fg of gDNA is the theoretical amount found in a single cyst, detection of 150 fg dilutions by the OCU qRT-PCR assay indicated the ability to detect a single organism.

3.4.

Fig. 2 e Amplification sensitivity and linearity of the Orange County Utilities Real-time RT-PCR Giardia viability assay (OCU qRT-PCR). Standard curve analyses of the betagiardin qPCR assay using gDNA from a 10-fold dilution series starting with 250 ng to 250 fg and an additional sample having 150 fg. Serial dilutions were made from extracted gDNA. A) Probe A targeting G. lamblia assemblage A (strain WB). B) Probe B targeting G. lamblia assemblage B (strain H3). Each sample was run in triplicate and average Ct values were plotted with standard error bars. All samples were detected with the following exceptions: one sample of the 150 fg dilution was not detected with probe A, one sample of the 150 fg dilution and 250 fg dilution were not detected with probe B.

3179

Specificity of primer and probe sets

In order to address the possibility of cross reactivity with other waterborne organisms, the OCU qRT-PCR primer and probe sets were used to screen some of these organisms’ gDNA or cDNA in the case of poliovirus type 2. Approximately 50 ng of gDNA or cDNA was used for each experiment. The specificity of the primer and probe sets was assessed using gDNA extracted from various G. lamblia assemblages and other waterborne organisms commonly found in wastewater (Table 3). Probe A detected gDNA from G. lamblia assemblages A and F, while Probe B detected only G. lamblia assemblage B gDNA. Neither probe detected the following organisms: G. muris, E. intestinalis, C. cayetanensis, A. suum, T. gondii, K. terrigena, poliovirus type 2 (PV-2), E. coli, and C. parvum. It is important to note that two G. lamblia assemblage F sequences were aligned to the area encompassing the OCU primers. While one strain (GenBank Accession #AY647264) matched to both forward and reverse primers, another strain (GenBank Accession #GU574802) matched to the OUC 39 reverse primer, but contained two mismatched base pairs with the OUC 36 forward primer (Table 1). Due to the betagiardin gene sequence variability in assemblage F strains, the OCU qRT-PCR primer set may not be useful in detecting the presence or absence of assemblage F in environmental samples. In our study, the OCU primer set amplified assemblage F and although probe A was capable of detecting it, assemblage F is generally associated with cat giardiasis (Adam, 2001) and not with human infections. It has been

3180

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

Table 1 e ClustalW alignment used for primer design.a Organism Giardia Giardia Giardia Giardia Giardia Giardia Giardia

lamblia duodenalis duodenalis duodenalis duodenalis duodenalis duodenalis

Organism Giardia Giardia Giardia Giardia Giardia Giardia Giardia

lamblia duodenalis duodenalis duodenalis duodenalis duodenalis duodenalis

Strain Host-assemblage P1 BAH8 A29 A21 Unk2 A101 BR1

Human-A1 Human-B3 Dog-C Dog-D Calf-E Cat-F Rat-F

Forward primer (Sequence 5’e3’) G e e e e e e

G e e e e e e

C e e e e e e

C e e e e e e

Strain Host-assemblage P1 BAH8 A29 A21 Unk2 A101 BR1

Human-A1 Human-B3 Dog-C Dog-D Calf-E Cat-F Rat-F

C e e e e e e

T e e e e e e

C e e e e e T

A e e e e e e

A e e e e e e

G e e e e e e

A e e e e e e

G e e e e e e

C e e e e e e

C e e e e e e

T e e e e e e

Nucleotide Position G e e e e e e

A e e e e e e

A e e e e e e

C e e e e e T

Reverse primer (Sequence 5’e3’) G e e e e e e

A e e e e e e

G e e e e e e

A e e e e e e

A e e e e e e

G e e e e e e

G e e e e e e

A e e e e e e

G e e e e e e

A e e e e e e

C e e e e e e

G e e e e e e

A e e e e e e

T e e e e e e

C e e e e e e

437e455

Nucleotide Position G e e e e e e

C e e e e e e

C e e e T e e

C e e e e e e

561e579

a Multiple nucleic acid sequence alignment of beta-giardin from G. lamblia and G. duodenalis environmental isolates. The nucleotide position for each probe is indicated on the right. Nucleic acid bases identical to the reference sequence (G. lamblia -p1 GenBank accession no. M36728) are shown as "e" and any variations are shown as their nucleic acid abbreviations. Sequences displayed were used as the representative sequence for assemblages A-F.

recently reported, however, that assemblage F was recovered from a human giardiasis patient using restriction fragment length polymorphism (RFLP) analysis, but the finding could not be confirmed using sequencing analysis (Gelanew et al., 2007). Due to the popularity of cats as domestic pets it can be expected that assemblage F cysts be found in wastewater and possibly could be detected by Probe A. An additional analysis using PCR-RFLP can be used for confirmation of assemblage A or F in samples that are detected by Probe A. PCR-RFLP leads to distinct detectable banding patterns corresponding to specific assemblages (Lalle et al., 2005). This will allow for the differentiation between assemblages A or F being present in a water sample. It is important to note that assemblages E and F share the same nucleotide sequence spanning Probe A, however

assemblage E was not detected. There is a single mismatch between the OCU 39 reverse primer and assemblage E at position 578. With a relatively long oligonucleotide it is not expected that a single mismatch will inhibit PCR amplification. However, it may be possible that amplification of assemblage E gDNA was inhibited due to this mismatch, which is located in the 30 end of the reverse primer, and subsequently detection by qPCR was unsuccessful.

3.5.

qRT-PCR viability assessment

Work performed by Mahbubani et al. (1991) has implicated the beta-giardin gene as being heat inducible. Since only viable organisms are capable of de novo mRNA transcription, this finding revealed a convenient parameter to differentiate

Table 2 e ClustalW alignment used for probe design.a Organism Giardia Giardia Giardia Giardia Giardia Giardia

lamblia duodenalis duodenalis duodenalis duodenalis duodenalis

Organism Giardia Giardia Giardia Giardia Giardia Giardia

duodenalis lamblia lamblia lamblia lamblia lamblia

Strain

Host-assemblage

P1 BAH8 A29 A21 Unk2 A101

Human-A1 Human-B3 Dog-C Dog-D Calf-E Cat-F

Strain

Host-assemblage

BAH8 P1 A29 A21 Unk2 A101

Human-B3 Human-A1 Dog-C Dog-D Calf-E Cat-F

Probe A (sequence 5’e3’) A e e e e e

G e e e e e

A e e e e e

C e e e e e

G e C e e e

G A e e e e

G e e e e e

C e e e e e

A e e e e e

Nucleotide position T e e e e e

T C C e C C

G e e e e e

C e e e e e

Probe B (sequence 5’e3’) C e e T e e

G e e e e e

A e e e e e

G e e e e e

A e e e e e

C e e e e e

A G C C G G

G e e e e e

G e e e e e

463e475

Nucleotide position C e e e e e

A e e e e e

T e e e e e

C T e e e e

461e473

a Multiple nucleic acid sequence alignment of beta-giardin from G. lamblia and G. duodenalis environmental isolates. The nucleotide position for each probe is indicated on the right. Nucleic acid bases identical to the reference sequence (G. lamblia -p1 GenBank accession no. M36728) are shown as "e" and any variations are shown as their nucleic acid abbreviations. Sequences displayed were used as the representative sequence for assemblages AeF.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

3181

between living and non-living Giardia cysts by measuring the increase in transcriptional activity after heat induction. This discovery was the basis for determining viability with the OCU qRT-PCR assay. In order to determine if the assay was able to differentiate between living and non-living cysts, replicate trials (N ¼ 20) of OCU qRT-PCR analysis were performed using w1000 assemblage B cysts.

Each trial consisted of four treatment groups prepared in the following manner: 1) live, induced; 2) live, non-induced; 3) killed, induced and 4) killed, non-induced. For live cysts, in all twenty trials it was observed that a significant increase (>90fold) in mRNA transcription occurs after induction (Fig. 3). For 16 of 20 trials, the killed cysts demonstrated zero increase in transcription. Gene activity for killed cysts was detected in 4 of 20 trials, with a 2e4-fold increase in transcription. These results suggest only viable cysts demonstrate a marked increase in transcriptional activity after heat induction. The transcriptional response to heat induction and corresponding comparison of non-induced control cysts, as measured by qRT-PCR, was able to differentiate between viable and killed cysts. The variance of the increase in gene activity after heat induction for live cysts was determined by calculating the coefficient of variation (CV). The CV for viability trials (N ¼ 20) using 1000 assemblage B cysts is 0.97. Assay reproducibility and sensitivity was evaluated by an alternative analyst who performed replicate trials (N ¼ 5) of approximately 100 G. lamblia (assemblage B) cysts. Four treatment groups were assessed for transcriptional activity in each trial as described above. A >30-fold increase in mRNA transcription was observed for live cysts in all five of the trials after induction (Fig. 4). In 3 of the 5 trials, the killed cysts displayed no increase in transcription. In 2 trials, killed cysts demonstrated a 1 and 3-fold increase in gene activity. These results show that at lower cyst concentrations, live Giardia demonstrate a considerable increase in mRNA transcription when induced by heat, thereby allowing differentiation from non-living cysts using this method.

Fig. 3 e Bar graph derived from viability experiments. Each trial consisted of four 1000 cyst aliquots each treated in one of four ways. Treatments: 1) live, induced; 2) live, noninduced; 3) killed, induced and 4) killed, non-induced. For live and killed cysts the gene activity observed for their respective non-induced aliquot served as the baseline measure. Dark bars represent the percent change in betagiardin gene transcription for live cysts, after heat induction. Lighter bars represent the change in betagiardin gene transcription for killed cysts, after heat induction. Cysts were used within two weeks of receipt from Waterborne, Inc. (New Orleans, LA). “*” represents killed cyst trials that did not demonstrate an increase in beta-giardin gene activity.

Fig. 4 e Bar graph derived from viability experiments. Each trial consisted of four 100 cyst aliquots each treated in one of four ways. Treatments: 1) live, induced; 2) live, noninduced; 3) killed, induced and 4) killed, non-induced. For live and killed cysts the gene activity observed for their respective non-induced aliquot served as the baseline measure. Dark bars represent the percent change in betagiardin gene transcription for live cysts, after heat induction. Lighter bars represent the change in betagiardin gene transcription for killed cysts, after heat induction. Cysts were used within two weeks of receipt from Waterborne, Inc. (New Orleans, LA). “*” represents killed cyst trials that did not demonstrate an increase in beta-giardin gene activity.

Table 3 e qPCR specificity screening experiments.a Giardia lamblia/duodenalis assemblages Giardia Giardia Giardia Giardia Giardia Giardia

lamblia Assemblage A lamblia Assemblage B duodenalis Assemblage duodenalis Assemblage duodenalis Assemblage duodenalis Assemblage

Waterborne organisms Giardia muris Encephalitozoon intestinalis Cyclospora cayetanensis Ascaris suum Toxoplasma gondii Klebsiella terrigena poliovirus type-2 (PV-2) cDNA Escherichia coli Cryptosporidium parvum

C D E F

Probe A

Probe B

þ e e e e þ

e þ e e e e

e e e e e e e e e

e e e e e e e e e

a gDNA from all organisms was extracted and used for screening with the exception of PV-2. PV-2 RNA was converted to cDNA and then used for screening. Any sample that resulted in a detectable Ct value was recorded as (þ).

3182

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

Fig. 5 e Bar graph derived from viability experiments using chlorine disinfection as an alternative means to evaluate killed cysts. Each trial consisted of four 1000 cyst aliquots each treated in one of four ways. Treatments: 1) live, induced; 2) live, non-induced; 3) killed (chlorination), induced and 4) killed (chlorination), non-induced. For live and killed cysts the gene activity observed for their respective non-induced aliquot served as the baseline measure. Dark bars represent the percent change in beta-giardin gene transcription for live cysts, after heat induction. Lighter bars represent the change in beta-giardin gene transcription for killed cysts, after heat induction. Cysts were used within two weeks of receipt from Waterborne, Inc. (New Orleans, LA). “*” represents killed cyst trials that did not demonstrate an increase in beta-giardin gene activity. Further experiments evaluating assay reproducibility include investigating chlorine disinfection as an alternative to boiling of control cysts. For these experiments (N ¼ 15) 1000 assemblage B live and chlorine killed cysts were assayed. The fold increase in gene activity for live cysts ranged from 12 to >700 (Fig. 5). In 12 of the 15 trials, the chlorine killed cysts displayed no increase in transcription. In 3 of the 15 trials, chlorine killed cysts demonstrated a 2e3-fold increase in gene activity. These results show that chlorine disinfection is an effective alternative for evaluating killed cysts. In addition, live cyst average fold increase in transcription levels of >250 are comparable to previous studies using 1000 cysts. The CV for these experiments (N ¼ 15) using 1000 assemblage B cysts is 0.77.

3.6.

been possible that heating the cysts further compromised the cyst wall, increasing lysing efficiency and release of greater amounts of gDNA. This may have contaminated the qPCR assays. Additionally, the increased gDNA may have overwhelmed the DNAse treatment, thereby leading to the reduction of the ability to truly discriminate among live and dead cysts. In order to support the claim that the developed assay detected induced mRNA and not increased amounts of contamination by gDNA, samples were treated with RNase A (SigmaeAldrich St. Louis, MO) prior to and after reverse transcription (Table 4). When mRNA from w1000 live, induced G. lamblia (H3-assemblage B) cysts were treated with RNase A prior to reverse transcription, the phase where only mRNA should be present, no signal was detected. Furthermore, when the RNase A treatment was performed after the reverse transcription was completed, when mRNA is present along with the newly generated cDNA, a signal was detected. These results show that if mRNA is targeted for degradation prior to cDNA generation there is no detection by qPCR. These findings indicate that the detection is, in fact, due to de novo mRNA transcription. Also, the control sample lacking any RNase A treatment delivered Ct values comparable to the sample treated after cDNA generation, indicating that the additional RNase A treatment did not adversely affect the OCU qRT-PCR assay.

3.7.

Environmental water sample analysis

Demonstration of assay applicability was accomplished by spiking samples of finished reclaimed water with w5000 live Giardia assemblage B cysts and observing beta-giardin gene activity. Water samples were taken from a central Florida reclamation facility and spiked with either live or killed cysts. Spiked samples were split into five aliquots, leaving w1000

Role of de novo mRNA

After induction, an increase in beta-giardin transcription was detected by the OCU qRT-PCR assay. However, it could have

Table 4 e Comparison of OCU qRT-PCR RNAse A experiments.a Treatment designations RNAseA before reverse transcription RNAse A after reverse transcription No RNAse A treatment

Ct value ND;ND 28.84;29.09 29.40;29.51

a Samples were spiked with w1,000 live Giardia lamblia (H3) cysts. "ND" ¼ Not Detected.

Fig. 6 e Bar graph derived from environmental water experiment. Each trial consisted of four 1000 cyst aliquots each treated in one of four ways. Treatments: 1) live, induced; 2) live, non-induced; 3) killed, induced and 4) killed, non-induced. For live and killed cysts the gene activity observed for their respective non-induced aliquot served as the baseline measure. Dark bars represent the percent change in beta-giardin gene transcription for live cysts, after heat induction. Cysts were used within two weeks of receipt from Waterborne, Inc. (New Orleans, LA). “*” represents killed cyst trials that did not demonstrate an increase in beta-giardin gene activity.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

cysts in each. Two aliquots for each matrix were assayed using the OCU qRT-PCR assay. Heat induced mRNA was recovered from each aliquot of live cysts analyzed. All live cyst trials (N ¼ 3) demonstrated a significant increase in gene transcription. The fold increase in gene activity for live cysts ranged from w30 to >4000-fold (Fig. 6). There was no increase in gene activity observed from killed cyst samples. The findings demonstrate that the assay was able to detect live G. lamblia (assemblage B) cysts recovered from finished reclaimed water. Following the successful demonstration of live cyst detection in spiked reclaimed water samples, the OCU qRT-PCR viability method was evaluated for its’ ability to detect potentially viable Giardia in unspiked environmental samples from the same reclamation facility. Current cyst enumeration data collected by our laboratory from this facility using USEPA Method 1623 show Giardia concentrations typically above 2000 cysts/100 L. Three replicate 20 L reclaimed water samples were collected from two post-chlorination treatment locations within the facility, resulting in 6 total samples evaluated. One out of the six samples analyzed by Probe A was positive for assemblage A cysts. Additionally, one of the six samples analyzed by Probe B was positive for assemblage B cysts. The Ct values for the two positive samples evaluated by Probe A (40.04) and Probe B (40.00), independently, are consistent with OCU qRT-PCR sensitivity data of <10 viable G. lamblia cysts. There was no increase in gene activity observed from killed cyst controls. These results validate the applicability of this assay on unspiked environmental samples.

4.

Conclusion

The developed assay has demonstrated significant advantages over other published qPCR methods, particularly since, it has the ability to discriminate between Giardia assemblage groups and concurrently assess viability, which specifically addresses the shortcomings of existing methods used to screen environmental waters. The organisms present in a reclaimed water sample will span the microbiological spectrum and potentially include bacteria, viruses, and protozoa (Leclerc et al., 2002). The OCU-qRT-PCR assay does not cross react with many of these waterborne organisms. This assay has shown great potential in detecting viable G. lamblia cysts of human health significance, specifically assemblages A and B. The work performed has confirmed that ability to detect assemblage B cysts; unfortunately, assemblage A cysts could not be obtained in a timely manner and subsequently were not tested. It would be important to also screen Giardia assemblage G cysts in order to discern the extent of primer/probe set specificity. The OCU qRT-PCR method has demonstrated the ability to detect live spiked G. lamblia (assemblage B) cysts as well as low concentrations of live G. lamblia (assemblage A and B) cysts from unspiked finished reclaimed water. Further studies will include continued analysis of large volume environmental water matrices using the combined methodology of the OCU qRT-PCR assay with USEPA Method 1623 for sample collection. The combined method will be evaluated for sensitivity, potential inhibition and optimization. Ultimately, this method would serve as a robust alternative for determining viability of Giardia

3183

cysts in environmental waters and identification of genotypes that pose a human health threat (assemblage groups A and B). This study found that killed cysts (boiled or chlorine disinfected) were not able to initiate a significant mRNA transcriptional response after heat induction using this assay, supporting the claim that differentiating between viable and non-viable cysts is possible. It is important to note that disinfection methods employed by reclamation facilities do not boil cysts and may utilize disinfectants other than chlorine. The effects of other common disinfection techniques, such as UV radiation, will be evaluated as they may influence mRNA transcription or pose as PCR inhibitors. After these factors have been evaluated, various environmental waters will be assayed and levels of viable G. lamblia cysts (assemblage A and B) will be assessed.

Acknowledgments This work was supported by funding from Orange County Utilities (OCU), Orlando, FL. We would like to especially thank Kimberly Kunihiro (OCU) and George DiGiovanni (Texas A&M University) for advice and technical assistance. We would also like to thank Dr. Lihua Xiao (CDC) for providing extracted gDNA from Giardia lamblia (assemblages C and E), Dr. Simone Caccio (Istituto Superiore di Sanita) for providing partially cloned beta-giardin gene plasmids corresponding to Giardia lamblia (assemblages D and F) and Dr. Shawn Thompson and Dr. Ryan Reinke (Sanitation Districts of Los Angeles County) for critical review of the manuscript.

references

Abbaszadegan, M., Gerba, C.P., Rose, J.B., 1991. Detection of Giardia cysts with a cDNA probe and application to water samples. Applied Environmental Microbiology 57, 927e931. Abbaszadegan, M., Huber, M.S., Gerba, C.P., Pepper, I.L., 1997. Detection of viable Giardia cysts by amplification of heat shock induced mRNA. Applied Environmental Microbiology 63, 324e328. Adam, R.D., 2001. Biology of Giardia lamblia. Clinical Microbiology Reviews 14, 447e475. Barbosa, J., Costa-de-Oliveira, S., Rodrigues, A.G., Pina-Vaz, C., 2008. Optimization of a flow cytometry protocol for detection and viability assessment of Giardia lamblia. Travel Medicine and Infectious Disease 6, 234e239. Bertrand, I., Maux, M., Helmi, K., Hoffman, L., Schwartzbrod, J., Cauchie, H.M., 2009. Quantification of Giardia transcripts during in vitro excystation: interest for estimation of cyst viability. Water Research 43, 2728e2738. Caccio, S.M., De Giacomo, M., Pozio, E., 2002. Sequence analysis of the beta-giardin gene and development of a polymerase chain reaction-restriction fragment length polymorphism assay to genotype Giardia duodenalis cysts from human fecal samples. International Journal for Parasitology 32, 1023e1030. Gelanew, T., Lalle, M., Hailu, A., Pozio, E., Caccio, S., 2007. Molecular characterization of human isolates of Giardia duodenalis from Ethiopia. Acta Tropica 102, 92e99. Guy, R.A., Payment, P., Krull, U.J., Horgen, P.A., 2003. Real-time PCR for quantification of Giardia and Cryptosporidium in environmental water samples and sewage. Applied Environmental Microbiology 69, 5178e5185.

3184

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 7 5 e3 1 8 4

Holberton, D., Baker, D.A., Marshal, J., 1988. Segmented alphahelical coiled-coil structure of the protein giardin from the Giardia cytoskeleton. Journal of Molecular Biology 204, 789e795. Lalle, M., Pozio, E., Capelli, G., Bruschi, F., Crotti, D., Caccio, S., 2005. Genetic heterogeneity at the B-giardin locus among human and animal isolates of Giardia duodenalis and identification of potentially zoonotic subgenotypes. International Journal for Parasitology 35, 207e213. Leclerc, H., Schwartzbrod, L., Dei-Cas, E., 2002. Microbial agents associated with waterborne diseases. Critical Reviews in Microbiology 28, 371e409. Lemos, V., Graczyk, T., Alves, M., Lobo, M.L., Sousa, M.C., Antunes, F., Matos, O., 2005. Identification and determination of the viability of Giardia lambila cysts and Cryptosporidium parvum and Cryptosporidium hominis oocysts in human fecal and water supply samples by fluorescent in situ hybridization (FISH) and monoclonal antibodies. Parasitology Research 98, 48e53. Mahbubani, M.H., Bej, A.K., Perlin, M., Schaefer, F.W., Jakubowski, W., Atlas, R.M., 1991. Detection of Giardia cysts by using the polymerase chain reaction and distinguishing live from dead cysts. Applied Environmental Microbiology 57, 3456e3461. Mahbubani, M.H., Bej, A.K., Perlin, M., Schaefer, F.W., Jakubowski, W., Atlas, R.M., 1992. Differentiation of Giardia

duodenalis from other Giardia spp. by using polymerase chain reaction and gene probes. Journal of Clinical Microbiology 30, 74e78. Palm, D., Weiland, M., McArthur, A.G., Winiecka-Krusnell, J., Cipriano, M.J., Birkeland, S.R., Pacocha, S.E., Davids, B., Gillin, F., Linder, E., Svard, S., 2002. Development changes in the adhesive disk during Giardia differentiation. Molecular and Biochemical Parasitology 141, 199e207. Read, C.M., Monis, P.T., Thompson, R.C., 2004. Discrimination of all genotypes of Giardia duodenalis at the glutamate dehydrogenase locus using PCR-RFLP. Infection Genetics and Evolution 4, 125e130. Reynolds, K.A., Gerba, C.P., Pepper, I.L., 1996. Detection of infectious enteroviruses by an integrated cell culture-PCR procedure. Applied Environmental Microbiology 62, 1424e1427. Singh, B., 1997. Molecular methods for diagnosis and epidemiological studies of parasitic infections. International Journal for Parasitology 27, 1135e1145. Trout, J.M., Santin, M., Greiner, E., Fayer, R., 2004. Prevalence of Giardia duodenalis genotypes in pre-weaned dairy calves. Veterinary Parasitology 124, 179e186. USEPA, 2005. USEPA Method 1623: Cryptosporidium and Giardia in Water by Filtration/IMF/FA. Office of Water, Washington, D. C. EPA 815-R-05.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Effect of ciprofloxacin on microbiological development in wetland mesocosms Kela P. Weber a,b, Michael R. Mitzel c, Robin M. Slawson c, Raymond L. Legge a,* a

Department of Chemical Engineering, University of Waterloo, 200 University Avenue W., Waterloo, ON N2L 3G1, Canada Department of Civil and Environmental Engineering, University of Western Ontario, London, Ontario N6A 3K7, Canada c Department of Biology, Wilfrid Laurier University, Waterloo, ON N2L 3C5, Canada b

article info

abstract

Article history:

An understanding of how antibiotics and other “emerging contaminants” affect both water

Received 27 October 2010

treatment systems and natural environments is of growing interest. Ciprofloxacin is

Received in revised form

a broad-spectrum antibiotic active against both Gram-positive and Gram-negative bacteria

18 March 2011

and has been extensively used over the past 20 years. The objective of this research was to

Accepted 21 March 2011

study the effect of an antibiotic such as ciprofloxacin on the development, function and

Available online 29 March 2011

stability of bacterial communities in wetland systems. Four mesocosm wetlands planted with Phragmites australis were initially seeded with activated sludge from a waste water treatment plant and allowed to develop for a 1 week period, after which 2 of the 4 meso-

Keywords: Emerging contaminants

cosms were exposed to a ciprofloxacin concentration of 2 mg/mL for a 5 day period. The 4

Antibiotics

mesocosms were then monitored for several microbiological and hydrological parameters

Wetland

over the course of 22 weeks. The bacterial community species distribution and catabolic

Mesocosms

capabilities were assessed via denaturing gradient gel electrophoresis (DGGE) and

Denaturing gradient gel

community level physiological profiling (CLPP), based on carbon utilization. DGGE results

electrophoresis (DGGE)

indicated that the ciprofloxacin decreased the total number of PCR-amplifiable DNA

Community profiling (CLPP)

level

physiological

(bacteria) and the overall diversity of bacterial operational taxonomic units (OTUs). Through CLPP it was shown that the interstitial microbiological community was initially adversely affected by the ciprofloxacin, creating a temporary decrease in the activity and overall catabolic capabilities of the inherent wetland bacterial communities. However, after a 2e5 week recovery period the activities and catabolic capabilities of the bacterial communities exposed to ciprofloxacin returned to levels comparable to those found for bacterial communities not exposed to the ciprofloxacin. These findings suggest that ciprofloxacin exposure may have an adverse effect on the inherent bacterial communities in wetland systems initially reducing their ability to assimilate anthropogenic carbonbased compounds; however, normal functionality may resume after a 2e5 week period. It was also observed that plants in the ciprofloxacin-treated mesocosms did not adapt to the antibiotic presence, instead showing initial browning of above ground parts and eventual die-off. Reduced porosity, evapotranspiration, and overall hydrological mixing in the ciprofloxacin-treated mesocosms was observed. ª 2011 Elsevier Ltd. All rights reserved.

* Corresponding author. Tel.: þ1 519 888 4567x36728; fax: þ1 519 746 4979. E-mail address: [email protected] (R.L. Legge). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.042

3186

1.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

Introduction

Emerging contaminants are defined as chemicals or microbiological constituents previously undefined or not recognized as being of concern to human or environmental health (Petrovic et al., 2004). Of particular interest are pharmaceuticals, antibiotics, and antibiotic resistant pathogenic microorganisms, all of which can be exceedingly challenging to identify, remove or inactivate in water. Although this field of research is still quite new, a growing effort into this area has been initiated worldwide (Hotchkiss et al., 2008). Both natural and constructed wetlands contain a microbiological regime consisting of a complex, dynamic and mixed species microbiological community associated with the wetland substrate which plays a major role in ecosystem health, cycling of nutrients and in the degradation of contaminants (Parkinson and Coleman, 1991; Aelion and Bradley, 1991; Wynn and Liehr, 2001; Truu et al., 2009; Faulwetter et al., 2009; Kadlec and Wallace, 2009; Weber and Legge, 2008). The role of the microbiological regime and the related mechanisms associated with nutrient cycling and contaminant treatment in wetland treatment systems is, however, not fully understood. Research into understanding microbial population density and diversity, both spatially and temporally, will help in understanding natural wetland ecosystem health, stability and robustness, in addition to the ongoing optimization and future design of constructed treatment wetlands (Faulwetter et al., 2009; Weber and Legge, 2009). There are a number of bacterial community profiling techniques currently used to characterize wetlands. Some of the more popular molecular methods include denaturing gradient gel electrophoresis (DGGE), terminal restriction fragment length polymorphism (TRFLP), and fluorescent in situ hybridization (FISH) (see Malik et al., 2008 for a review). Non-molecular techniques include microscopy-based identification, fatty acid methyl ester (FAME) analysis and phospholipid fatty acid (PLFA) analysis, culture-based identification, and community level physiological profiling (CLPP) using BIOLOG plates. Two recent studies focused on temporal shifts and changes in the catabolic capabilities of bacterial communities in wetland mesocosms. Weber et al. (2008) studied the temporal response of the resident bacteria in wetland mesocosms in response to a controlled perturbation, whereas Weber and Legge (2010a) studied the temporal development of wetland resident bacteria in mesocosm systems during a developmental period. Ciprofloxacin is a contaminant of concern as it carries increasing environmental risk with the current pattern of use (Halling-Sørensen et al., 2000). Ciprofloxacin is a broad-spectrum antibiotic active against both Gram-positive and Gramnegative bacteria and has seen extensive use due to its potency (Hooper, 1998). Ciprofloxacin functions by inhibiting DNA gyrase, some topoisomerases and, resultantly, cell division (Drlica and Zhao, 1997). The effect which ciprofloxacin and other antibiotics have on the bacterial communities, in the natural environment and microbiologically mediated water treatment systems, are unclear. Resultant changes in the function of bacterial communities in waterways and water saturated areas such as wetlands are of concern, as bacterial communities play a large role in both nutrient and carbon

cycling in the natural environment and in the interactions with macrophytes. The microbial ecology and function of resident treatment-oriented bacterial communities in both conventional and alternative water treatment systems is also of concern, as this could affect the treatment efficiency and stability of such systems reducing effluent water quality. The objective of this study was to investigate the effect ciprofloxacin has on the development, function and stability of bacterial communities in wetlands. Four mesocosm wetlands planted with Phragmites australis were seeded with activated sludge from a waste water treatment plant and allowed to develop for a 1 week period after which 2 of the 4 systems were exposed to a ciprofloxacin. The 4 mesocosms were then monitored for several microbiological and hydrological parameters over a 22 week period to assess changes in the microbiological development and stability, in addition to overall mesocosm hydrological characteristics. The bacterial community was assessed using an overall microbial activity measure, PCR/DGGE and CLPP. Hydrological parameters assessed included porosity, evapotranspiration, and intrinsic mixing measured as the dispersion coefficient associated with a 1D advection-dispersion model.

2.

Material and methods

2.1.

Experimental design and setup

Several mesocosm studies have recently been used in undertaking a quantitative approach to the study of constructed wetland (CW) systems (Kappelmeyer et al., 2001; Stein et al., 2006; Werker et al., 2007; Weber et al., 2008, 2010). The mesocosm approach has been shown to suitably represent interactions between microorganisms, differing substrates, and contaminants within a complex rhizosphere system (Kappelmeyer et al., 2001; Stein et al., 2006; Werker et al., 2007; Weber et al., 2008; Weissner et al., 2008). Mesocosms cannot be said to completely represent full-scale wetlands, as full-scale wetlands can contain many different hydrological, biological and geochemical sub-environments. The mesocosm is taken to represent a single unit in a fullscale CW or natural wetland system where the geochemical, hydrological, and microbiological characteristics are relatively constant. By controlling different characteristics such as aspect ratios, flow rate, granular media etc., it is possible to create different mesocosms with relatively well controlled hydrological, biological and geochemical environments and used to better understand full-scale CW functionality or natural wetland robustness. Experiments were started with a 1 week operational period after which 2 of the 4 systems were exposed to ciprofloxacin at a concentration of 2 mg/mL for a 5 d period after which the ciprofloxacin was discontinued. Two mesocosms that were not exposed to ciprofloxacin are identified as “no cipro” and considered as controls, the 2 mesocosms exposed to the ciprofloxacin are identified as “cipro”. During and after the initial exposure period the mesocosms were monitored for changes in several microbiological and hydrological characteristics.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

The mesocosms used here were of the same design as those used previously by Weber and Legge (2010c) and Weber et al. (2010). Water entered approximately 65 cm from the bottom of the mesocosm, resulting in an oxidative reduction potential (ORP) of w125 mV at the bottom. Each mesocosm was constructed of schedule 80 (wall thickness of 1.5 cm), polyvinylchloride (PVC) columns (90 cm by 25 cm diameter) filled to w 80 cm with pre-sorted pea gravel (mean equivalent spherical diameter of 20 mm, with an approximate D10 of 17 mm, and a D60 of 22 mm, 80% limestone) and operated to 70 cm with water. Starting void volume was w12 L for all mesocosms, however as biomass accumulated the void volume changed. Water was circulated with a 1/200 HP, 3200 rpm, March (Glenview, Illinois) series 1 (1A-MD 1/2) centrifugal pump. The water inlet was positioned 5 cm below the water level. Bacterial community seeding was completed by adding fresh limestone gravel alternating with w160 mL of inoculum at depths of 10 cm, w40 cm and w65 cm. Plants (P. australis) were collected from a local marsh, cultured in pots with peat moss and transferred to the mesocosms with a small amount of peat moss in the top section. Three small plants (w30 cm high) were used initially for each mesocosm. Initial root depth was w20 cm in all cases. The P. australis used was of the North American native non-invasive type. The mesocosms were maintained under laboratory conditions with a relative humidity and temperature of 40e60% and 24e28  C, respectively. All mesocosms were exposed to artificial illumination (14,000 lumens) with a 15 h photoperiod. Plants were sprayed daily with tap water to reduce drying. The mesocosms were completely drained once per week. After draining, the mesocosms were then refilled with a simulated wastewater solution based on the descriptions of Droste (1996) and solutions used by Kargi and Karapinar (1995) and Wang et al. (2008). As a plant supplement, a nutrient solution according to Hoagland and Arnon (1938) was also added to the simulated wastewater. The nutrient solution was mixed in regular tap water and fed to the wetlands resulting in interstitial concentrations of 28.75 mg/L NH4H2PO4; 151.5 mg/L KNO3; 236 mg/L Ca(NO3)2$4H2O; 123.25 mg/L MgSO4$7H2O; 9.175 mg/L FeNaEDTA; 0.715 mg/L H3BO3; 0.4525 mg/L MnCl2$4H2O; 0.055 mg/L ZnSO4$7H2O; 0.0125 mg/L CuSO4 and 0.005 mg/L (NH4)6Mo7O24$4H2O. The simulated wastewater consisted of w1 g/L molasses, 0.049 g/L urea, 0.0185 g/L NH4H2PO4, yielding a glucose concentration of w0.588 g/L from the molasses, a COD of w500 mg/L and a COD:N:P ratio of w100:5:1. The mesocosms were operated in a constant recycle mode where water was circulated at approximately 2.4 L/min, resulting in an average cyclic hydraulic retention time of approximately 4e5 min. Each week the mesocosms were filled with the simulated wastewater and operated under constant recycle for 1 week which is similar to hydraulic retention times reported for full-scale CW systems. After 1 week the mesocosms were completely drained before being refilled with new simulated wastewater for the following week’s operation. This design and operation allowed the determination of true evapotranspiration and drainable porosity values without using estimation methods. Interstitial bacterial communities were assessed in this study. To study attached biofilm bound bacterial communities

3187

the mesocosms would need to be disassembled. Due to the interest in the temporal nature of the response, disassembling and thus discontinuing mesocosm development and operation was not an option. Due to the design, sloughed biofilm with this mesocosm set-up is retained, so assessment of the interstitial water communities appears to provide a reasonable representation of the outermost attached biofilm communities over time.

2.2.

Community level physiological profiling

CLPP gives information regarding the carbon-based substrates that can be utilized by a bacterial community; therefore, this method is often referred to as a “functional characterization” and performed as previously described (Weber et al., 2007; Weber and Legge, 2010b,c). Samples were gathered from the interstitial water of all mesocosms on days 1, 7, 14, 29, 49 and 73. BIOLOG ECO plate wells were inoculated with 150 mL of mesocosm interstitial water and incubated at room temperature. Inoculants were diluted to an optical density (OD) of w0.19. The BIOLOG ECO plates were read at an absorbance of 590 nm after an 84 h incubation period. In addition, the original activated sludge inoculum was also characterized using the same CLPP protocol; however, the results were analogous to the initial mesocosm interstitial water characterizations and therefore are not presented. Analysis of the CLPP data was performed as previously described (Weber et al., 2007; Weber and Legge, 2010b,c). Absorbance readings (590 nm) at 84 h were identified as the metric for further CLPP data analysis. BIOLOG ECO plates (Biolog Inc., Hayward CA., USA) consist of 96 wells. The wells contain 31 different carbon sources (Insam, 1997), and a blank in triplicate. One plate was used for each sampling time point (6 sampling time points in all) for each mesocosm. For each plate 3 replicate carbon source utilization patterns (CSUPs) were collected giving a total of 72 objects (data sets). Each data set represents 31 variables apiece giving a total of 6912 data points for analysis. The absorbance values from 84 h were initially standardized by first correcting by the corresponding blank value and then dividing by the average well colour development (AWCD) for that time point. Assessment of normality, homoscedasticity and linear correlations in the entire data set according to Weber et al. (2007) and Weber and Legge (2010b), yielded a recommended natural logarithmic transformation for subsequent principle component analysis (PCA). Negative well responses were coded as zeros during data treatment. PCA was completed using Statistica 8.1 (StatSoft, Tulsa, OK). The AWCD as described here can be thought of as an activity measure for the community being assessed. For general background and a description of PCA see Legendre and Legendre (1998).

2.2.1.

Functional community divergence measure

The community divergence measure was calculated according to Weber and Legge (2009). The Euclidean distance measure was used in this study as a measure of dissimilarity of the CSUPs gathered for any one mesocosm at any point, in comparison to the original day 0 CSUP for that mesocosm. The Euclidean distance can be calculated in n dimensions, where

3188

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

in this study n ¼ 31 (31 different carbon source utilization responses). Given the two points:

the activity of each well response ( pi) and the number of responding wells (S ) into a single ecological metric.


P ¼ p1 ; p2 ; .; pn

(1)

2.3.


 Q ¼ q1 ; q2 ; .; qn

(2)

DGGE is a technique used to separate DNA sequences based on the melting point of the DNA fragments extracted from the bacterial community. Sequence variants of similar-sized fragments stop migrating at different positions during DGGE, resulting in a unique banding pattern seen in DGGE gels. The use of primers targeting 16S rDNA in prokaryotes, originally proposed by Muyzer et al. (1993), allows for the formation of a structural bacterial community fingerprint based on genetic diversity within a given environment, as these genes are ubiquitous and contain variable and conserved regions that differ phylogenetically (Woese, 1987). Bands can be excised and sequenced to provide information for phylogenetic trees; however, the banding pattern of the structural fingerprint can itself be used as a source of high-dimensional data. Recent wetland studies utilizing DGGE include those of Buesing et al. (2009) and Calheiros et al. (2009). DGGE fingerprints were initially extracted using Gel Compar II (Applied Maths, Texas, USA). Both band movement (taken as number of pixels from the top of the gel), and relative intensities (i.e. % intensity of each band compared to total intensity of entire lane) were extracted.

The Euclidean distance can be calculated as: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
2 2
p1  q1 þ p2  q2 þ/ þ pn  qn

(3)

The reference point used for all Euclidean distance calculations was the original CSUP of any single mesocosm gathered at day 0; therefore, 4 separate Euclidean distance trends (one for each mesocosm) were calculated to fully represent the divergence of the interstitial bacterial communities in the mesocosms.

2.2.2.

Substrate related diversity indices

As first proposed by Zak et al. (1994), BIOLOG plates can also be used in a more traditional ecological sense, to calculate diversity indices based on carbon source utilization patterns (CSUPs). The Shannon index which is a measure of what is often called “diversity” is a common ecological metric used to track and understand shifts in communities over space and time. Using the CSUP gathered from a single BIOLOG plate, substrate diversity (H) can be calculated as: HCLPP ¼ 

X


 pi ln pi

(4)

where: HCLPP e substrate diversity; pi e ratio of the activity of a particular substrate to the sums of activities of all substrates; activity e chosen metric for analysis (absorbance value (590 nm) at 84 h). Another parameter associated with substrate diversity which can be calculated using CSUPs is substrate richness (S ). Substrate richness is a measure of the number of different substrates utilized by a microbial population and is calculated as the number of wells with a corrected absorbance greater than 0.25. Recent examples of studies utilizing the Shannon index or richness values calculated from CSUPs gathered using BIOLOG plates include He et al. (2008), Farnet et al. (2008), Weber et al. (2008), and Weber and Legge (2010c). In addition to the Shannon Index a modified Hurlbert Index was also used to characterize the substrate diversity found using CLPP. In a similar fashion to the way the Shannon index was adapted to give a substrate-related diversity index based on CLPP data, the Hurlbert index was calculated here as:  PIECLPP ¼

 X i S h 1 p2i S1

(5)

where: pi e ratio of the activity of a particular substrate to the sums of activities of all substrates; S e richness (number of wells with a corrected absorbance greater than 0.25). The use of the probability of interspecific encounters (PIE) here has a different meaning than what was first proposed by Hurlbert (1971). Where Hurlbert (1971) offered the PIE index in an effort to give scientists an index to describe the probability that successive encounters with individuals of a population will be with differing species, the context of the index here is changed. Much like the Shannon index the PIE index as presented here is simply a population index which weights both

2.3.1.

Denaturing gradient gel electrophoresis

DNA extraction

Five mL of interstitial water or original sludge inoculum was filtered using a 22 mm 0.22 mm polycarbonate filter (Millipore, Fisher Scientific) within 3 hours of sampling. Each filter was placed into a PowerSoil DNA isolation kit (Mo Bio Laboratories Inc., CA) bead tube and DNA was extracted following the protocol supplied by the manufacturer.

2.3.2.

PCR

PCR was performed with 5 mL of template DNA using the primers 357f (50 -CCTACGGGAGGCAGCAG-30 ) with a GC-clamp (50 -CGCCCGCCGCGCCCCGCGCCCGTCCCGCCGCCCCCGCCCG30 ) added to the 50 end, and 518r (50 -ATTACCGCGGCTGCTGG30 ), targeting the V3 region of bacterial 16S rDNA (Muyzer et al., 1993). PCR mastermix for this universal primer set was prepared such that each 50 mL reaction contained 1 Go-Taq Flexi (Promega, Fischer Science) Green PCR Buffer, 0.5 mM of each primer, 1.5 mM MgCl, 1.5U Go-Taq Flexi (Promega), 200 mM dNTP and 21.3 mL of Milli-Q water for each 50 mL reaction. PCR was performed using a BioRad I-cycler iQ PCR machine. Touch-down PCR conditions consisted of an initial denaturation step of 94  C for 5 min, followed by 20 cycles of 94  C, 65  C and 72  C for 1 min each, in which the annealing temperature of 65  C was decreased by 1  C every 2 cycles to a temperature of 56  C on the 20th cycle. Ten additional cycles of 94  C, 55  C and 72  C for 1 min each followed. PCR concluded with a 7 min, 72  C extension step and was held at 4  C until storage at 10  C. PCR reaction success was measured by loading 10 mL PCR product into 1.5% agarose gel in 1 TAE buffer. Gels were run for 60 min at 100 V, stained with ethidium bromide solution for 15 min and visualized using BioRad Universal Hood II to confirm the presence of only a 233 bp band in amplified samples and absence of

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

product in the blank, which consisted of the same mastermix with Milli-Q water instead of template DNA.

2.3.3.

Running conditions for DGGE

DGGE was performed with 8% (wt/vol) acrylamide gels containing a linear denaturant gradient ranging from 40 to 65%, with 100% denaturant defined as a solution of 7 M urea and 40% formamide. Gels were run for 17 h at 75 V (1190 V h) using a CBS Scientific DGGE-2401 machine set to a constant temperature of 60  C. 15 mL of PCR product was added to each lane and 10 mL of ladder was added to a central lane and both outside lanes. Gels were stained with 1 SYBR Gold solution for 1 h and subsequently placed on a UV-transilluminator (VWR International) flooded with deionized water before being photographed with an Olympus C-5060 5.1 megapixel camera with SYBR Gold filter and the image captured after being exposed for 3 s. Individual image contrasts were set to a maximized logarithmic scale using Alpha Digi Doc RT software (Alpha Innotech, CA). Gel images were normalized using GelCompar II software (Applied Maths, Inc., TX) and a report exported for further analysis.

2.3.4.

Genetic-related diversity indices

Genetic diversity indices were calculated based on both the Shannon index (H ) and the probability of interspecific encounters (PIE): HDGGE ¼ 

X


 pi ln pi

(6)

where HDGGE e genetic diversity; pi - relative band intensity.  PIEDGGE ¼

 X i S h 1 p2i S1

(7)

where pi e relative band intensity; S - richness (number of identifiable bands in sample lane). The use of the PIE index here has a similar meaning as to what was first proposed by Hurlbert (1971). Species richness was calculated from the number of different bands observed in a sample lane, representing the number of different species (also known as operational taxonomic units) in a sample.

2.3.5.

Principal component analysis (PCA)

For PCA sample band movement was normalised to the movement of the DNA ladders for each gel, with the ladder movement set to lie between 1 and 1000. For each gel, movement normalisation was completed separately for the right side and left side of the gel. In all cases the ladder was run on the furthermost left lane, the centre lane, and the furthermost right lane. For movement normalisation of the sample bands on the left side of the gel the average movement of both the centre ladder and the leftmost ladder positions were used to set the 1 to 1000 movement scale. The process was also completed for the right side of the gel. Some sample fragments were found to lie outside the ladder movement range, both below and above the ladder on the gel. Therefore a correction factor of 1000 was added to all movement values allowing for only positive integers in the data set. The final movement values were found to lie between 0 and 2500. Each band was then classified into different movement groupings in blocks of 50 (i.e. 1e50, 51e100, 101e150 etc.) giving a total of 50 groupings. Each of the 50

3189

groupings was then treated as a variable, with the sample site treated as an object. For the non-quantitative analysis presence/absence of each band in each group was coded as 1 or 0. For what is being referred to here as the quantitative analysis, relative band intensity was used as the metric. Following the analysis methods outlined by Weber et al. (2007), a natural logarithmic transform and Taylor power law transform were examined and the resulting data sets assessed for normality and homoscedasticity. The Taylor power law transform (b ¼ 0.608) was found to best improve the normality and homoscedasticity of the quantitative data sets and was therefore applied and used for all subsequent PCAs. PCs were extracted from the covariance matrix of the data using Statistica 8.1.

2.3.6.

Species divergence

Similar to the divergence measure described in section 2.1.2 a community divergence measure was also calculated using the DGGE data. The Euclidean distance measure was again used to calculate the similarity or (distances) between the different samples based on the untransformed data set created for PCA.

2.4.

Evapotranspiration

Evapotranspiration reflects the sum of evaporation and plant transpiration. Measurements were taken as frequently as possible given time constraints; representative values were gathered wevery 1e2 weeks. Water loss was measured by topping up the mesocosms until the overflow point was reached. Water loss was equated to be entirely due to evapotranspiration, as no other outputs existed.

2.5.

Dispersion coefficient

The internal hydraulic performance of aquatic systems can be quantified using inert, soluble chemical tracers (Dierberg et al., 2005; Kadlec, 1994). NaBr tracer tests were conducted on the 4 mesocosms over the 22 wk period. Two mL aliquots of a 200 g/L NaBr stock solution were injected into the mesocosms through the inlet port, a handheld conductivity probe (VWR conductivity/temperature meter, catalogue number 21800-012) was then inserted into the same port, measuring the conductivity of the circulating water. Conductivity readings were taken until stable values were reached (typically 20e45 min). Data were then fit to a 1D advection dispersion equation (Equation (8)) using Aquasim v.1.0.0.1 (Eawag Institute, Switzerland, 1995). 1D advection dispersion equation: vC vC v2 C ¼ v þD 2 vt vx vx

(8)

where C ¼ concentration (mg/mL); t ¼ time (min); v ¼ velocity (cm/min); x ¼ distance (cm); D ¼ Dispersion coefficient (cm2/min). Average flow rate and apparent cross sectional area were manually entered based on porosity and flow rates measured before starting the tracer test. The dispersion coefficient (D) was then determined using the parameter estimation

3190

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

function via both the simplex (Nelder and Mead, 1965) and secant methods (Ralston and Jennrich, 1978).

2.6.

Porosity

Overall mesocosm porosity was calculated by determining the total volume of material (Vm) and the non-solid volume (Vp). Vm was calculated as the cylindrical volume of the mesocosm from the bottom to the overflow point. Vp was calculated by completely draining and then measuring the volume of water required to refill the column to the overflow point. This is essentially a measurement of the drainable porosity, as not all water will be drained due to adherence to the greater bed media and filling of smaller pore spaces. Hereafter, the term porosity will be used to represent what is more accurately described as “drainable porosity”. Differences in the starting porosities of each mesocosm occurred due to inconsistencies in the limestone gravel bed media, therefore porosity data is presented as normalised to the starting porosity (day 1) for each individual mesocosm (i.e. porosity/initial porosity).

2.7.

Microbiological activity

The overall microbiological activity (MA) of the mesocosms was assessed using an indirect measurement based on the enzymatic activity associated with the fixed microbiological regime and has been previously used for systems of this type (McHenry and Werker, 2005). Schnurer and Rosswall (1982) demonstrated that the overall activity of the biomass in a complex sample can be correlated with the presence of esterases, lipases, and proteases, which can catalyze the transformation of fluorescein-diacetate (FDA) to fluorescein (FL). This assay was used to assess the total microbiological

activity in the mesocosm. One mL of a 5 mM FDA (Aldrich Chemical, Milwaukee, WI) stock solution in acetone was added to the mesocosm wetland. Two mL interstitial water samples were then drawn for a 1 h period every 30 s for the first 10 min, once every min for the following 20 min and every 2 min for the final 30 min. Fluorescein was then measured using a handheld fluorometer (Turner Designs, PicofluorTM, 490 nm excitation, 520 nm emission). Readings were normalized with respect to the maximum observable FL concentration based upon the FDA aliquot and mesocosm volume. The FDA utilization rate was determined for all time increments between 8 and 15 min. The final FDA utilization rate was calculated as the average of these 10 incremental slopes. Averaging of incremental utilization rates was performed to reduce error due to possible fluctuations in fluorescence readings. For the purposes of this study this final value was multiplied by 1000 and then used as a comparative microbiological activity (MA) metric.

2.8.

Time course data

Some of the data collected for this study is represented as time course data. The average of the 2 mesocosms exposed to ciprofloxacin for 5 days (cipro) and the average of the 2 mesocosms not exposed to ciprofloxacin for 5 days (no cipro) are shown in the different plots. Error bars representing 1 standard deviation are included in all figures where appropriate.

3.

Results and discussion

Fig. 1A summarizes the overall community size based on the total intensity of the stained DNA in any one sample lane on

Fig. 1 e Summary of bacterial community structure data based on DGGE analysis. A) Community size based on total amplifiable DNA; B) substrate richness based on the number of identifiable bands in a sample; C) Shannon index; D) Hurlbert (PIE) index results for the control “no cipro”, and “cipro” treated mesocosms. Five day ciprofloxacin exposure period is represented by the grey bar. Error bars represent 1 standard deviation.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

the gel normalised to the total intensity found in the ladder lane. This can also be described as the total amplifiable DNA segments based on the universal primers used. The cipro mesocosms showed decreased numbers of total bacteria after day 8 in comparison to the control mesocosms. Ciprofloxacin is a broad-spectrum antibiotic affecting both Gram positive and Gram negative bacteria and likely was able to either hinder or kill-off specific species of bacteria. The richness results for the mesocosms over the 10 wk period are presented in Fig. 1B. Although there are not statistical differences initially, the number of different bacterial species in the interstitial water of the mesocosms subjected to the ciprofloxacin appeared to be greater after day 15. This may be the result of an increased number of species either not attaching to the bed media or detaching from the biofilm on the bed media and/or the roots. Species diversity can be described as a mix of species richness and evenness. Different diversity indices weight either evenness or richness to greater degrees. The Shannon index is a commonly used indice for representing species distribution data (Zak et al., 1994; Weber and Legge, 2010c); however, depending on the situation and the question being asked, other species diversity indices may be more appropriate or show differing trends (Hurlbert, 1971). From Fig. 1C it can be seen that the species diversity based on the Shannon index was approximately the same for all mesocosms, up until day 70 at which point the cipro mesocosms showed a sharp decrease in species diversity. Fig. 1D shows a similar trend with species diversity for the cipro mesocosms being less than that for the control mesocosms directly following ciprofloxacin exposure (day 8). It can be concluded that exposure to ciprofloxacin decreased the total number of bacteria, while increasing the number of different species in

3191

the interstitial space, as well as decreasing the overall species diversity of the bacteria found. AWCD can be thought of as an overall activity and catabolic capability measure of the microbiological community being assessed. In general the trend for both control and cipro mesocosms are similar, with larger fluctuations seen for the mesocosms exposed to the ciprofloxacin (Fig. 2). As seen from Fig. 2A for the cipro mesocosms the AWCD was lower than the AWCDs for the control mesocosms at approximately day 30. This result agrees with the total community results (Fig. 1A), showing that with a smaller number of bacteria present, lower activity was also observed; however, after wday 75 no discernable difference between the AWCDs of the control and treated mesocosms can be seen. The ciprofloxacin had an initial effect of reducing the overall microbiological activity seen in the interstitial water of the wetland mesocosms but activities recovered to values analogous to those found for the control mesocosms. A community which is able to utilize a greater number of substrates can be said to have a greater catabolic capability. Fig. 2B shows a similar trend to that of Fig. 2A. Fluctuations in substrate richness for the mesocosms exposed to ciprofloxacin are greater than that of the control mesocosms, with the reduction in richness at day 30 being greater for the cipro mesocosms. Similar to the results seen for AWCD the substrate richness values, the cipro mesocosms recover to levels similar to the control mesocosms over the monitoring period. Fig. 2C and D reveals very similar trends for substrate diversity based on the Shannon and Hurlbert indices. The substrate diversity trends were seen to be analogous to both the AWCD and richness trends with a larger decrease seen at day 30 for the cipro mesocosms followed by a recovery near

Fig. 2 e Summary of bacterial community function data collected via CLPP utilizing BIOLOG ECO plates over the monitoring period. A) average well colour development (AWCD); B) substrate richness; C) Shannon index; D) Hurlbert (PIE) index results for the control “no cipro”, and “cipro” treated mesocosms. The five day ciprofloxacin exposure period is represented by the grey bar. Error bars represent 1 standard deviation.

3192

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

the end of the monitoring period. Based on the data in Fig. 2, it can be said that the interstitial microbiological community was initially adversely affected by the ciprofloxacin. It is likely that specific species were catabolically hindered or simply eliminated due to the antibiotic. This resulted in a temporary decrease in the activity and overall catabolic capabilities of the bacterial communities. In a 2e5 wk recovery period the activities and catabolic capabilities of the bacterial communities exposed to ciprofloxacin returned to levels analogous to those found for the control mesocosms. These results have implications for natural and treatment wetlands. The presence of an antibiotic such as ciprofloxacin can have an adverse effect on the bacterial communities suggesting that natural systems will have a reduced ability to assimilate anthropogenic carbon-based compounds resulting in reduced treatment and increased transport of these compounds through the wetland system reaching other water systems. For treatment wetlands exposure to antimicrobials may result in a reduction in the treatment ability both in terms of overall activity and in terms of the range of carbon based compounds that can be treated, thus reducing the overall treatment effectiveness or efficiency of the system. This adverse effect on the treatment-oriented bacteria, however, was short term, with a recovery of the bacterial communities seen after approximately 2-5 weeks. Therefore, although ciprofloxacin was shown to have a negative effect on the functionality of the bacterial population, the same communities were shown to have resilience and recovered to normal operating functionality. Attention was also given to the overall divergence and shifts seen in the bacterial communities during the study period. The bacterial communities in the mesocosms of control and ciprofloxacin exposed mesocosms were characterized in terms of the species distribution and the overall catabolic or “functional” characteristics. The divergence or change in the species distribution and functional characteristics was measured using a one dimensional divergence metric (Weber and Legge, 2009). The divergence trends (Fig. 3A and B) are quite similar; the major difference being a large increase in functional divergence for the mesocosms exposed to the antibiotic at day 30. This large increase in divergence is likely due to the reduction in activities, reduced substrate richness, and reduced substrate diversities.

The change in the bacterial communities in terms of the species distribution (DGGE) and the overall catabolic or “functional” characteristics can also be viewed as ordination plots resulting from PCA of the data (Weber and Legge, 2010c). On these plots similarity between objects (bacterial community samples in this case) is measured as distance on the plot, with similar samples being found closer on the plot in comparison to dissimilar samples. Fig. 4 presents PCA ordinations for the DGGE species distribution data and the CLPP functional characteristics data for the interstitial bacterial communities in the mesocosms. Fig. 4A presents all of the DGGE data on a single ordination plot. A large grouping consisting of the day 0, 7 and 14 samples for all mesocosms is observed, with the day 49 and day 73 samples found to be dispersed outside of this grouping. For an ordination of just the day 0, 7 and 14 DGGE samples (Fig. 4B) it can be seen that samples are quite similar on day 0 at which point they shift in a similar manner at day 7 and then again shift all in a similar manner on day 14. No discernable groupings of the control or cipro mesocosms could be seen in either plot. For CLPP data on a single ordination plot (Fig. 4C) a large grouping consisting of the day 0, 7 and 14 samples for all mesocosms is observed with the day 29, 49 and day 73 samples found outside of this grouping. For an ordination of the day 0, 7 and 14 samples (Fig. 4D) it can be seen that samples are quite similar on day 0 at which point they shift in a similar manner on day 7 and then again shift all in a similar manner on day 14. Again, no discernable groupings of control or cipro mesocosms could be seen in either plot. The trends and groupings seen here for the DGGE and CLPP data are similar and represent similar trends. From Fig. 4 it can be concluded that the bacterial species and functional distributions were comparatively similar for days 0e14 at which point a shift in both the species and catabolic capability distributions occurred. The divergence trends observed after day 14 could be said to be mesocosm specific, whereby critical populations involved in biofilm establishment may have been variably affected by the antibiotic exposure resulting in a delayed shift in the microbial profile, with no generalised groupings apparent. This study was performed during the start-up phase for the mesocosms, a period where the bacterial community is adjusting to the new environment of the mesocosm and

Fig. 3 e Community divergence calculated as the Euclidean distance of A) each respective monitoring day DGGE banding pattern with respect to the day 0 banding pattern, and B) each respective monitoring day CSUP with respect to the day 0 CSUP for control “no cipro” and “cipro” treated mesocosms. The five day ciprofloxacin exposure period is represented by the grey bar. Error bars represent 1 standard deviation.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

3193

Fig. 4 e PCA ordinations based on the Taylor transformed (b [ 0.608) DGGE banding patterns for A) all days and B) days 0, 7 and 14; PCA ordinations based on CSUPs collected via BIOLOG ECO plates for C) all days and D) days 0, 7 and 14 for control (3,4) and cipro treated (7,8) mesocosms. Output generated using Statistica 8.1.

establishing specific sub-environments around the bed media allowing for attachment (Weber and Legge, 2010c) therefore, a general divergence trend is expected. The fact that there was no observable difference in the initial divergence trend between control and cipro mesocosms suggests that the respective bacterial community’s ability to respond to the new environment had a stronger influence on the divergence trends than did the effect of ciprofloxacin exposure. A difference in mesocosm bacterial communities development during start-up based on plant presence/absence has been previously observed (Weber and Legge, 2010c) although in this case all mesocosms were planted. The results presented here suggest that although ciprofloxacin had an effect on the bacterial species and functional diversities, the overall divergence trend seen through PCA was attributed more so to the overall adaptation to the new environment rather than in response to the ciprofloxacin. Hydrological and microbial activity characteristics are summarized in Fig. 5. Porosities in all mesocosms decreased over the study period likely due to biofilm build-up and organic matter deposition (Engesgaard et al., 2002; Nivala and Rousseau, 2009). Porosity in the cipro mesocosms did not decrease as quickly initially as those of the control mesocosms (Fig. 5A). This is likely due to the decreased number of bacteria observed following ciprofloxacin exposure. As the ciprofloxacin reduced the number of bacteria in the mesocosm at day 8, the overall ability for the bacterial community to attach to the mesocosm bed media was also reduced, thus reducing the build-up of biofilm in the bed media. After day 100 it can be seen that this initial trend is reversed, with the cipro mesocosms showing lower normalised porosities

compared to the control mesocosms. Weber and Legge (2010c) saw that different bacterial communities developed in mesocosms introduced to different inocula and that this had an effect on the observed porosities over the start-up period. In this case the starting inoculum was the same for all mesocosms. This result may be partly due to a different bacterial species distribution found in the cipro mesocosms. The lower porosities found for the cipro mesocosms after 100 days were likely due to the health of the plants in the different mesocosms. The plants in the cipro mesocosms did not develop over the start-up period, but rather went through a steady phase of increasing chlorosis and declining health. The P. australis in the control mesocosms exhibited some chlorotic symptoms during the first 100 days, however later had new growth. Weber and Legge (2010c) observed that planted mesocosms have increased porosities in comparison to unplanted mesocosms and that this difference was only apparent after approximately 100 days following start-up. The results here are in agreement with these observations. The plants in the cipro mesocosms did not adapt well to the start-up period displaying reduced growth and exhibiting chlorotic symptoms. The lack of plant growth and hence root development likely accounts for the reduced porosity values after day 100 that is seen for the cipro mesocosms. Reed et al. (2005) showed that certain “plant growth promoting” bacteria are able to facilitate the growth of P. australis. It is proposed that the decrease in plant growth and vigor observed for the cipro mesocosms could in part be due to death of specific bacteria that may have been beneficial with regards to P. australis growth. After approximately day 30 the evapotranspiration rates of the control mesocosms were greater than those which were

3194

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

Fig. 5 e Summary of hydrological and overall microbial activity parameters over the monitoring period. A) normalized porosity; B) evapotranspiration data (L/day); C) dispersion coefficient (cm2/min); D) microbial activity data for control ”no cipro”, and “cipro” treated mesocosms. The five day ciprofloxacin exposure period is represented by the grey bar. Error bars represent 1 standard deviation.

exposed to ciprofloxacin (Fig. 5B). This can again be attributed to the plants which in the control mesocosms developed and grew whereas the plants in the cipro mesocosms deteriorated over the same time period. The differences in the dispersion coefficients (Fig. 5C) can also be attributed to plant health in the different mesocosms. Increased dispersion coefficient values have been observed for planted mesocosms when compared to unplanted mesocosms (Weber and Legge, 2010c). It is likely that the difference in underground root mass due to the increased activity and health of the mesocosms accounted for the differences in the hydrological mixing properties observed for the systems. It should be noted though that the differences in the dispersion coefficient values here are not very large and that this difference would not likely be noticeable from a hydrological perspective in large scale systems. The microbial activity measurements (Fig. 5D) show that there were no major differences between the mesocosms. This result highlights the resilience and ability for wetland systems to adapt to disturbances and perturbations such as the ciprofloxacin administered here.

4.

Conclusions

The results of this study have implications for both natural wetland systems and treatment wetlands. The presence of ciprofloxacin will have an adverse effect on the inherent bacterial communities in wetland systems resulting in a reduced ability to assimilate anthropogenic carbon-based compounds, and reducing their ability to handle as diverse a range of carbonbased pollutants. This effect though may be transitory, as the bacterial communities were shown here to return to normal functionality in approximately 2-5 weeks. It was also shown that

the plants in the mesocosms exposed to the ciprofloxacin did not favourably adapt, which resulted in reduced porosity, evapotranspiration, and overall hydrological mixing. This study focused on the bacterial community development and stability in the face of the antibiotic ciprofloxacin. It is suggested that similar studies be completed looking at the effect of ciprofloxacin on already operational/developed wetland systems. In addition, further scientific understanding of the effects emerging contaminants may have on wetland systems could be discerned through direct observation and characterization of biofilm materials. Although this would require disassembly of experimental systems or disruption/ disturbing healthy or operational wetland systems, this knowledge would help both wetland scientists and water treatment engineers gain a greater understanding for the effects antibiotics and other emerging contaminants have on wetland systems.

Acknowledgements Support from NSERC in the form of a Discovery Grant to RLL and from ORF in the way of funding to the Centre for Control of Emerging Contaminants (CCEC) to RMS & RLL is gratefully acknowledged.

references

Aelion, C.M., Bradley, P.M., 1991. Aerobic biodegradation potential of subsurface microorganisms from a jet fuel-contaminated aquifer. Appl. Environ. Microbiol. 57, 57e63.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

Buesing, N., Filippini, M., Burgmeann, H., Gessner, M.O., 2009. Microbial communities in contrasting freshwater marsh microhabitats. FEMS Microbiol. Ecol. 69, 84e97. Calheiros, C.S.C., Duque, A.F., Moura, A., Henriques, A.M., Correia, A., Rangel, A.O.S.S., Castro, P.M.L., 2009. Changes in the bacteria community structure in two-stage constructed wetlands with different plants for industrial wastewater treatment. Bioresour. Technol. 100, 3228e3235. Dierberg, F.E., Juston, J.J., DeBuska, T.A., Pietro, K., Gub, B., 2005. Relationship between hydraulic efficiency and phosphorus removal in a submerged aquatic vegetation-dominated treatment wetland. Ecol. Eng. 25, 9e23. Drlica, K., Zhao, X.K., 1997. DNA gyrase, topoisomerase IV, and the 4-quinolones. Microbiol. Mol. Biol. Rev. 61, 377e392. Droste, R.L., 1996. Theory and Practice of Water and Wastewater Treatment. John Wiley and Sons, New Jersey, USA. Engesgaard, P., Holm, J., Jensen, K.H., Henze, M., Albrechtsen, H.J., 2002. Development of preferential flow in bioclogging of porous media. Develop. Water Sci. 47, 803e810. Farnet, A.M., Prudent, P., Cigna, M., Gros, R., 2008. Soil microbial activities in a constructed soil reed-bed under cheese-dairy farm effluents. Bioresour. Technol. 99, 6198e6206. Faulwetter, J.L., Gagnon, V., Sundberg, C., Chazarenc, F., Burr, M. D., Brisson, J., Camper, A.K., Stein, O.R., 2009. Microbial processes influencing performance of treatment wetlands: a review. Ecol. Eng. 35, 987e1004. He, X.Y., Wang, K., Zhang, W., Chen, Z., Zhu, Y., Chen, H., 2008. Positive correlation between soil bacterial metabolic and plant species diversity and bacterial and fungal diversity in a vegetation succession on Karst. Plant Soil 307, 123e134. Halling-Sørensen, B., Holten Lu¨tzhoft, H.C., Anderson, H.R., Ingerslev, F., 2000. Environmental risk assessment of antibiotics: a comparison of mecillinam, trimethoprim and ciprofloxacin. J. Antimicrob. Chemother. 46 (S1), 53e58. Hoagland, D.R., Arnon, D.I., 1938. The Water Culture Method for Growing Plants Without Soil. University of California College of Agriculture Agricultural Experiment Station, Berkeley, CA. Circular 347. Hooper, D.C., 1998. Expanding uses of fluoroquinolones: opportunities and challenges. Ann. Int. Med. 129, 908e910. Hotchkiss, A.K., Rider, C.V., Blystone, C.R., Wilson, V.S., Hartig, P.C., Ankley, G.T., Foster, P.M., Gray, C.L., Gray, L.E., 2008. Fifteen Years after “Wingspread”denvironmental endocrine disrupters and human and wildlife health: where we are today and where we need to go. Toxicol. Sci. 105 (2), 235e259. Hurlbert, S.H., 1971. The nonconcept of species diversity: a critique and alternative parameters. Ecology 52, 577e586. Insam, H., 1997. A new set of substrates proposed for community characterization in environmental samples. In: Insam, H., Rangger, A. (Eds.), Microbial Communities: Functional Versus Structural Approaches. Springer-Verlag, Berlin. Kadlec, R.H., 1994. Detention and mixing in free water wetlands. Ecol. Eng. 3, 345e380. Kadlec, R.H., Wallace, S., 2009. Treatment Wetlands, second ed. CRC Press, Boca Raton, FL, USA. Kappelmeyer, U., Wiessner, A., Kuschk, P., Kastner, M., 2001. Planted fixed bed reactor (PFR) e Eine universelle testenheit fur bewachsene bodenfilter. Chem. Ingenieur Tech. 73, 1472e1477. Kargi, F., Karapinar, I., 1995. Kinetics of COD removal from a synthetic wastewater in a continuous biological fluidized bed. Bioproc. Eng. 13, 265e270. Legendre, P., Legendre, L., 1998. Numerical Ecology, second ed. Elsevier B.V., Amsterdam, Netherlands. Malik, S., Beer, M., Megharaj, M., Naidu, R., 2008. The use of molecular techniques to characterize the microbial communities in contaminated soil and water. Environ. Int. 34, 265e276.

3195

McHenry, J., Werker, A., 2005. In-situ monitoring of microbial biomass in wetland mesocosms. Water Sci. Technol. 51, 233e241. Muyzer, G., de Waal, E.C., Uitterlinden, A.G., 1993. Profiling of complex microbial population by DGGE analysis of polymerase chain reaction amplified genes encoding for 16S rRNA. Appl. Environ. Microbiol. 62, 2676e2680. Nelder, J., Mead, R., 1965. A simplex method for function minimization. Comp. J. 7, 308e313. Nivala, J., Rousseau, D.P.L., 2009. Reversing clogging in subsurface-flow constructed wetlands by hydrogen peroxide treatment: two case studies. Water Sci. Technol. 59 (10), 2037e2046. Parkinson, D., Coleman, D.C., 1991. Microbial communities, activity, and biomass. Agric. Ecosyst. Environ. 34, 3e33. Petrovic, M., Eljarrat, E., Lopez de Alda, M.J., Barcelo, D., 2004. Endocrine disrupting compounds and other emerging contaminants in the environment: a survey on new monitoring strategies and occurrence data. Analy. Bioanaly. Chem. 378, 549e562. Ralston, M., Jennrich, R., 1978. DUD e a derivative free algorithm for non linear least squares. Technometrics 20, 7e14. Reed, M.L.E., Warner, B.G., Glick, B.R., 2005. Plant growthepromoting bacteria facilitate the growth of the common reed Phragmites australis in the presence of copper or polycyclic aromatic hydrocarbons. Curr. Microbiol. 51, 425e429. Schnurer, J., Rosswall, T., 1982. Fluorescein diacetate hydrolysis as a measure of total microbial activity in soil and litter. Appl. Environ. Microbiol. 43, 1256e1261. Stein, O.R., Biederman, J.A., Hook, P.B., Allen, W.C., 2006. Plant species and temperature effects on the k-C* first order model for COD removal in batch-loaded SSF wetlands. Ecol. Eng. 26, 100e112. Truu, M., Juhanson, J., Truu, J., 2009. Microbial biomass, activity and community composition in constructed wetlands. Sci. Total Environ. 407, 3958e3971. Wang, Y., Inamori, R., Kong, H., Xu, K., Inamori, Y., Kondo, T., Zhang, J., 2008. Influence of plant species and wastewater strength on constructed wetland methane emissions and associated microbial populations. Ecol. Eng. 32, 22e29. Weber, K.P., Legge, R.L., 2010a. Dynamics in the bacterial community level physiological profiles and hydrological characteristics of constructed wetland mesocosms during start-up. Ecol. Eng. 1661, 1e12. Weber, K.P., Legge, R.L., 2010b. Community level physiological profiling. In: Cummings, S.P. (Ed.), Bioremediation: Methods in Molecular Biology 599. Humana Press, New York, NY, pp. 263e281. Weber, K.P., Legge, R.L., 2010c. Method for the detachment of culturable bacteria from wetland gravel. J. Microbiol. Methods 80, 242e250. Weber, K.P., Werker, A., Gehder, M., Senger, T., Legge, R.L., 2010. Influence of the microbial community in the treatment of acidic iron-rich water in aerobic wetland mesocosms. Biorem. J. 14, 28e37. Weber, K.P., Legge, R.L., 2009. One dimensional metric for tracking bacterial community divergence using sole carbon source utilization patterns. J. Microbiol. Methods 79 (1), 55e61. Weber, K.P., Gehder, M., Legge, R.L., 2008. Assessment of changes in the microbial community of constructed wetland mesocosms in response to acid mine drainage exposure. Water Res. 42, 180e188. Weber, K.P., Legge, R.L., 2008. Pathogen removal in constructed wetlands. In: Russo, R.E. (Ed.), Wetlands: Ecology, Conservation and Restoration. Nova Publishers, New York, pp. 176e211. Weber, K.P., Grove, J.A., Gehder, M., Anderson, W.A., Legge, R.L., 2007. Data transformations in the analysis of community-level substrate utilisation data from microplates. J. Microbiol. Methods 69, 461e469.

3196

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 8 5 e3 1 9 6

Weissner, A., Kuschk, P., Jechorek, H., Seidel, H., Ka¨stner, M., 2008. Sulphur transformation and deposition in the rhizosphere of Juncus effusus in a laboratory-scale constructed wetland. Environ. Pollut. 155, 125e131. Werker, A.G., Van Loon, W., Legge, R.L., 2007. Tracers for investigating pathogen fate and removal mechanisms in mesocosms. Sci. Total Environ. 380, 188e195.

Woese, C.R., 1987. Bacterial evolution. Microbiol. Rev. 51, 221e271. Wynn, T.M., Liehr, S.K., 2001. Development of a constructed subsurface-flow wetland simulation model. Ecol. Eng. 16, 519e536. Zak, J.C., Willig, M.R., Moorhead, D.L., Wildman, H.G., 1994. Functional diversity of microbial communities: a quantitative approach. Soil Biol. Biochem. 26, 1101e1108.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 9 7 e3 2 0 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Comparison of the performance of UF membranes in olive mill wastewaters treatment A. Cassano*, C. Conidi, E. Drioli Institute on Membrane Technology, ITM-CNR, University of Calabria, via P. Bucci, 17/C, I-87030 Rende (CS), Italy

article info

abstract

Article history:

Olives are known to contain an appreciate amount of phenols with good antioxidant

Received 11 February 2011

properties which are lost in large part in olive mill wastewaters (OMWs) during olive oil

Received in revised form

production.

19 March 2011

Membrane technology offers several advantages (low energy consumption, no additive

Accepted 21 March 2011

requirements, no phase change) over traditional techniques to recover phenolic

Available online 29 March 2011

compounds from OMWs.

Keywords:

treatment of OMWs finalized to the recovery of polyphenols. For this purpose, OMWs were

Olive mill wastewater (OMW)

processed, in selected operating conditions, with four flat-sheet UF membranes charac-

Ultrafiltration (UF)

terized by different molecular weight cut-off (MWCO) (4, 5 and 10 kDa) and polymeric

Polyphenols

material (regenerated cellulose and polyethersulphone). Permeate fluxes, fouling index and

Total antioxidant activity (TAA)

retention coefficients towards phenolic compounds, total antioxidant activity (TAA) and

The aim of this work was to evaluate the performance of different UF membranes in the

total organic carbon (TOC) were evaluated. Regenerated cellulose membranes exhibited lower rejections towards phenolic compounds, higher permeate fluxes and lower fouling index if compared with PES membranes. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Mediterranean countries such as Italy, Spain, Greece and Turkey are responsible for 95% of the world-wide olive oil production. They produce about 11 million tons of olives per year from which about 1.7 million tons of olive oil are extracted. Olive mill wastewaters (OMWs) are by-products of the olive oil extraction whose production is estimated to 1.1e1.5 times the weight of milled olives (considering the three-phase centrifugal olive oil extraction) (Tsonis et al., 1989). OMWs are liquids with high electric conductivity constituted in average by 83.4% of water, 1.8% of inorganic salts and 14.8% of organic compounds. Their composition, however, is affected by different parameters such as the oil extraction

technology, the kind of olives and ripeness. The pH of OMWs is in the range 4.9e5.3 due to the presence of organic acids such as acetic, malic, fumaric, lactic, malonic, citric, tartaric, ossalic and succinic. Organic substances, mainly in solution, and in lower quantities in suspension or in emulsion (oil), determine the high polluting load of these wastewaters. Biochemical oxygen demand (BOD5) and Chemical Oxygen Demand (COD) may be as high as 100 and 200 g/L, respectively. The organic fraction contains sugars, tannins, polyphenols, polyalcohols, pectins, lipids, proteins and organic acids (Fiestas Ros de Ursinos, 1981; Jaouani et al., 2003). In recent years various management options have been proposed for the treatment and valorisation of OMWs. They include evaporation (Jarboui et al., 2010), direct application on soil (Paredes et al., 1999), physico-chemical treatments (Aktas

* Corresponding author. Tel.: þ39 0984 492067; fax: þ39 0984 402103. E-mail address: [email protected] (A. Cassano). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.041

3198

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 9 7 e3 2 0 4

et al., 2001; Azbar et al., 2004), and biotechnological transformations such as microbiological treatments (RamosCormenzana et al., 1996; Piperidou et al., 2000), composting (Tomati et al., 1995) and anaerobic treatments (Ammary, 2005; Hamdi, 1996; Marques, 2001; Martinez Nieto et al., 1993). Membrane operations, including microfiltration (MF), ultrafiltration (UF), nanofiltration (NF) and reverse osmosis (RO), and integrated membrane processes have been also proposed to obtain effluent streams from OMWs of acceptable quality for safe disposal to the environment (surface water or soil), for irrigation or even for recycling and use in the olive mill (Akdemir and Ozer, 2009; Borsani and Ferrando, 1996; Coskun et al., 2010; Paraskeva et al., 2007; Stoller, 2008; Turano et al., 2002). Independently of the efficiency of physical, chemical and biological treatments the disposal of great amounts of sludge remains the main environmental problem of OMWs. However, in the last years OMWs have received a great attention for the presence of high added value compounds such as antioxidant substances and phenols, which can be used in pharmaceutical, nourishment and cosmetical applications. In particular, European researchers consider OMWs as a potential source for the recovery of antioxidant, antiatherogenic and anti-inflammatory biophenols rather than a by-product (Obied et al., 2005). The biophenolic content of OMWs and their biological activity has been extensively studied by Visioli et al. (1999) and Obied et al. (2007). The release of polyphenols from OMW by using fungi (such as Phanerochaete chrysosporium, Aspergillus niger, Aspergillus terreus and Geotrichum candidum) and fungal enzymes has been studied by Garcia Garcia et al. (2000) and Bouzid et al. (2005), respectively. Solvent extraction techniques have been also studied on small scale (Lesage-Meessen et al., 2001; Allouche et al., 2004). However, these methods are more expensive than aqueous processing; the presence of residual solvents in the extract and the co-extraction of undesirable compounds are additional drawbacks. Recently, supercritical fluids, particularly supercritical CO2, have been also used as alternative systems to organic solvents (Crea and Creagri Inc., 2004; Takac and Karakaya, 2009). However, the extraction with supercritical fluids has the disadvantage of the requirement of expensive high pressure equipment. Membrane operations can be considered a valid approach for the selective removal of polyphenols from OMWs. They offer several advantages over traditional technologies mainly in terms of low energy consumption, no additive requirements and no phase change. Many studies indicate that the future direction of the processes for the recovery of antioxidants from OMWs is presumably towards the utilization of membranes in a sequential design (Takac and Karakaya, 2009). Integrated membrane processes for the selective fractionation, recovery and concentration of polyphenols from OMWs have been recently proposed (Garcia-Castello et al., 2010; Paraskeva et al., 2007; Russo, 2007) and patented (Pizzichini et al., 2005; Sniderman Zachary et al., 2006). In particular, Paraskeva et al. (2007) found that the OMW may be treated efficiently by using UF, NF and/or RO to obtain a permeate fraction which can be discharged in aquatic

systems according to national or EU regulations or to be used for irrigation. In this case NF was employed for the separation of the most part of phenols present. Russo (2007) proposed a preliminary MF of the OMW, followed by two UF steps realised by using 6 kDa and 1 kDa membranes, respectively, and a final RO treatment. The RO retentate, containing enriched and purified low molecular weight polyphenols, was proposed for food, pharmaceutical or cosmetic industries, while MF and UF retentates can be used as fertilizers or in the production of biogas in anaerobic reactors. In the integrated process proposed by Garcia-Castello et al. (2010) the vegetation water was pre-treated by MF without preliminary centrifugation; this step allowed to achieve a 91% and 26% reduction of suspended solids and total organic carbon (TOC), respectively. The MF permeate, containing 78% of the initial content of polyphenols was submitted to an NF treatment. In this step almost all polyphenols were recovered in the permeate stream, while TOC was reduced from 15 g/L to 5.6 g/ L. A concentrated solution containing 0.5 g/L of free low molecular weight polyphenols was obtained by treating the NF permeate by osmotic distillation (OD). Performances of UF and NF membranes are remarkably affected by fouling phenomena which reduce the permeate flux and determine both efficiency decrease and variation of membrane selectivity; fouling also makes the process highly expensive owing to repeated plant shut-down for cleaning and washing the membranes. This study aims to evaluate the performance of different UF membranes in the treatment of OMWs in terms of: i) permeate flux; ii) fouling index; iii) selectivity towards phenolic compounds, total organic carbon (TOC) and total antioxidant activity (TAA). The membrane filtration of pre-treated OMWs was investigated by using four membranes characterized by different MWCO (between 4 and 10 kDa) and membrane material (regenerated cellulose and polyethersulphone).

2.

Materials and methods

2.1.

Solutions and reactants

2.1.1.

Feed solution

Olive mill wastewaters, produced according to a 3-phase centrifugation process, were supplied by APOR (Gioia Tauro, Reggio Calabria, Italy). Before use, liquors were filtered through a cotton fabric filter to remove the coarse material (leaves, etc.) and then submitted to an MF treatment (by using a 0.2 mm polypropylene tubular membrane module). The suspended solids content of the original solution was 5.4% (w/ w): it was reduced to 3.8% (w/w) after the filtration with the cotton fabric filter. Suspended solids were completely removed in the MF step and TOC was reduced to 6106 mg/L (from 9968 mg/L).

2.1.2.

Reactants

2,20 -azinobis(3-ethylbenzothiazoline-6-sulfonic acid) diammonium salt (ABTS), 6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid (trolox) and potassium persulfate were obtained from SigmaeAldrich (Milan, Italy); sodium hydrogen

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 9 7 e3 2 0 4

and dihydrogen phosphate were from Carlo Erba (Milan, Italy). FolineCiocalteu reagent were from SigmaeAldrich (Milan, Italy). Hydroxytyrosol (MW 154.17 g/mol), protocatechuic acid (MW 154.12 g/mol) and tyrosol (MW 138.17 g/mol) were from Extrasynthese (Genay, France). Caffeic acid (MW 284.30 g/mol) and p-cumaric acid (MW 164.16 g/mol) were from SigmaeAldrich (Milan, Italy).

2.2.

Equipments

2.2.1.

Pre-treatment of OMWs

OMWs were pre-treated by MF in order to remove suspended solids and reduce fouling phenomena in the subsequent UF treatment. MF was performed by using a laboratory pilot unit supplied by Verind SpA (Milano, Italy) equipped with an MD 020 TP 2N tubular membrane module (membrane material polypropylene, nominal pore diameter 0.2 mm, open porosity 40e55%, membrane surface 0.036 m2) supplied by Microdyn (Wuppertal, Germany). The MF system was operated at a transmembrane pressure (TMP) of 0.5 bar and at a temperature of 25  0.1  C according to the batch concentration mode (recycling the retentate stream and collecting separately the permeate).

2.2.2.

UF equipment and procedures

UF experiments were performed in a bench-scale cross-flow membrane system (Fig. 1). The plant consisted of a feed tank with a capacity of 3 L, a magnetic drive gear pump (Micropump Mod. GC-M25 JF5 SA), a stainless steel cross-flow cell able to accommodate a flat-sheet membrane with a diameter of 7 cm (effective membrane surface area 38.465 cm2) and a permeate tank. Two manometers before and after the cell were used to measure the inlet and the outlet pressure and, consequently, the TMP. The feed flow rate and the TMP value were regulated by a pressure control valve, on the retentate side, and by regulating the velocity of the gear pump. A cooling system, fed with tap water, was used to keep the temperature of the feed solution at 25  1  C.

3199

Four commercial polymeric flat-sheet membranes, all supplied by Microdyn-Nadir (Wiesbaden, Germany), were investigated. Their characteristics are presented in Table 1. Microfiltered OMWs were submitted to the UF process in selected operating conditions. In particular, experimental runs with membranes having NMWCO lower than 10,000 Da were performed at a TMP of 5 bar and a temperature of 25  C; UF membranes with NMWCO of 10,000 Da were tested at 2 bar and 25  C. UF was operated according the batch concentration configuration up to reach a weight concentration ratio (WCR, defined as the ratio between the initial feed weight and the weight of the resulting retentate) of 3. During the filtration experiments the permeate flux was continuously monitored by measuring the permeate weight collected in a certain time according to the following equation: Jp ¼

Wp A$t

(1)

where Jp is the permeate flux (kg/m2h), Wp the permeate weight (kg) at time t (h) and A the membrane surface area (m2).

2.2.3.

Hydraulic permeability and membrane cleaning

The water permeability (Lp) of each membrane was determined by the slope of the straight line obtained plotting the water flux values, measured in fixed conditions of temperature (25  C), versus the applied TMP. The fouling index (If), expressed as a percentage drop in the water permeability, was estimated by measuring the water permeability before and after the treatment of OMWs, according to the following equation (Ma¨ntta¨ri and Nystro¨m, 2007): lf ¼

  Lp1 $100 1 Lp0

(2)

where Lp0 and Lp1 are the water permeabilities measured before and after the treatment of OMWs, respectively. After the treatment with OMWs UF membranes were cleaned in two steps. The first cleaning step was performed recirculating tap water for 30 min through the membranes in

Fig. 1 e Schematic flow diagram of the UF experimental set-up.

3200

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 9 7 e3 2 0 4

Table 1 e Characteristics of UF flat-sheet membranes. Type

UF-PES-004H

C005F

C010F

P010F

Membrane material

Permanently hydrophilic PES 4000

Regenerated cellulose 5000

Regenerated cellulose 10,000

Permanently hydrophilic PES 10,000

0e14 e 95 15.5

1e11 1e12 55 37.5

1e11 1e12 55 42.6

0e14 1e12 95 73.0

Nominal molecular weight cut-off (Da) pH operating range pH range in cleaning conditions Max. operating temperature ( C) Water permeabilitya (L/m2hbar) a Experimental values measured at 25  C.

order to remove the reversible polarized layer. In the second step the membrane module was submitted to a cleaning procedure by using 1 w/w% Ultrasil 50 (Henkel Chemicals Ltd., UK) at 40  C for 60 min. At the end of each cleaning procedure membranes were rinsed with tap water for 20 min and the hydraulic permeability was measured once again.

2.3.

Analytical methods

UF permeate and retentate samples were analysed for total phenols, free low molecular weight polyphenols, total antioxidant activity (TAA) and total organic carbon (TOC). The rejection (R) of UF membranes towards specific compounds was determined as: R¼

  cp $100 1 cf

(3)

where Cp and Cf are the concentration of a specific component in the permeate and feed, respectively.

2.3.1.

Total phenols

Total phenols were estimated colorimetrically by using the FolineCiocalteu method (Singleton et al., 1999). The method is based on the reduction of tungstate and/or molybdate in the FolineCiocalteu reagent by phenols in alkaline medium resulting in a blue coloured product (lmax 756 nm). The

80 C010F; 2 bar P010F; 2 bar UF-PES-004H; 5 bar C005F; 5 bar

70

Jp (kg/m2h)

60

estimation of total phenols was carried out in triplicate and results were expressed as mg/L gallic acid.

2.3.2.

Free low molecular weight polyphenols

Free low molecular weight polyphenols were determined according to the procedure reported in Obied et al. (2005) by using an HPLC system (Agilent Technologies, 1200 Series, USA) equipped with a UV detector. Chromatographic separation was performed by using a Luna C18 column (250  4.6 mm, 5 mm) (Phenomenex, Torrance, CA). Operating conditions were as follows: flux 1 ml/min, temperature 25  C, pressure 100 bar, wavelength 280 nm. The mobile phase was a mixture of 100:1 water/acetic acid (v/v) (solvent A) and a mixture of 90:10:1 methanol/acetonitrile/acetic acid (v/v/v) (solvent B). A sixthstep linear gradient analysis for a total run time of 60 min was used as follows: starting from 90% solvent A and 10% solvent B, increase to 30% solvent B over 10 min and then isocratic for 5 min, increase to 40% solvent B over 10 min, to 50% over 15 min and to 100% solvent B over 10 min, and finally isocratic for 10 min. The system was equilibrated between runs for 20 min using the starting mobile phase composition. Prior to HPLC analysis, all samples were filtered using cellulose acetate filters with 0.45 mm pore size and diluted 1:3 with pure water. The external standard method was applied. The concentration of free low molecular weight polyphenols was determined from experimental peak areas by analytical interpolation in a standard calibration curve. Each assay was performed in triplicate. The deviation of each measurement was of 2% from the average value.

2.3.3.

Total antioxidant activity (TAA)

TAA was determined by an improved version of the ABTS assay in which the radical cation is generated by reaction with

50 40 30

Table 2 e Hydraulic permeability of UF membranes before and after the treatment of OMWs and relative fouling index.

20 10

Membrane type

0 0

50

100

150

200

250

300

350

Lp0 (L/m2hbar)

LP1 (L/m2hbar)

If (%)

15.5 37.5 73.0 42.6

8.36 22.1 17.1 30.5

46.1 41.0 76.5 28.4

400

operating time (min)

Fig. 2 e Time course of permeate flux in the treatment of OMWs with UF membranes (T [ 25  C; WCR [ 3).

UF-PESe004H C005F P010F C010F

3201

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 9 7 e3 2 0 4

Table 3 e Analyses of polyphenols in samples of OMWs treated by UF membranes.

Table 4 e Analyses of TAA in samples of OMWs treated by UF membranes.

Membrane type

Membrane type

Polyphenols (mg/L gallic acid) Feed

UF-PESe 004H C005F C010F P010F

Permeate

RPolyphenols (%)

Feed

Retentate

704.5  31.2 363.9  10.3 1018.5  8.2

48.3

648.1  22.6 509.6  20.8 789.0  22.4 568.4  24.5 518.6  13.9 685.6  32.9 595.2  11.4 395.5  3.0 873.2  14.3

21.3 8.7 33.5

UF-PESe004H C005F C010F P010F

potassium persulfate before the addition of the antioxidant (decolourization assay) (Re et al., 1999). This method gives a measure of the antioxidant activity of pure substances and of mixtures by monitoring the reduction of the radical cation as the percentage inhibition of absorbance at 734 nm. Spectrophotometric measurements were performed by using a UVeVisible recording spectrophotometer (UV-160 A, Shimadzu Scientific Instruments, Inc., Japan) at 30  C. ABTS was dissolved in water at 2 mM concentration: ABTS radical cation was produced by reacting 10 ml of ABTS stock solution with 100 ml of 70 mM potassium persulfate solution (ABTS:K2S2O8 ¼ 1:0.35 M ratio) and allowing the mixture to stand in the dark at room temperature for 12e16 h before use. Work solution was prepared diluting 1 ml of the ABTS radical cation solution to 25 ml with PBS buffer (5 mM Na2HPO4, 5 mM NaH2PO4, NaCl 9 g/L, pH ¼ 6.8) to a final UV absorbance of 0.70  0.02 at 734 nm. After addition of 10 ml of sample to 10 ml of ABTS work solution, the absorbance at 734 nm was recorded every min for a total of 6 min. The value at 5 min was used to calculate the results reported as TAA, expressed in terms of mM trolox equivalent. Each determination was performed in triplicate. Results were expressed as mean  SD of three samples.

2.3.4.

TOC

Total carbon (TC) and inorganic carbon (IC) were analyzed by a TOC analyzer (TOC-V CSN, Shimadzu, Kyoto, Japan). TOC values were obtained by difference between TC and IC.

RTAA (%)

TAA (mM trolox)

9.3  9.1  8.0  10.1 

Permeate Retentate

0.7 0.2 0.6 0.6

3.1 6.8 7.7 5.7

   

0.3 0.6 1.2 0.2

10.7 11.4 8.7 12.9

   

1.4 0.5 1.5 0.4

3.

Results and discussion

3.1.

Permeate fluxes and membrane fouling

66.6 25.2 3.7 43.5

Fig. 2 shows productivities of UF membranes, in terms of kg of permeate produced per unit area and time (kg/m2h), in the treatment of microfiltered OMWs at a temperature of 25  C up to a WCR value of 3. For the C005F membranes, the permeate flux quickly decreased due to concentration polarization and membrane fouling, reaching a steady-state value of about 33 kg/m2h (corresponding to a 51.1% reduction of the initial value). The UF-PES-004H membrane showed a lower permeate flux. In particular, the initial permeate flux of about 25 kg/m2h was reduced at about 20 kg/m2h when the WCR was 3. These values can be compared with those observed by Russo (2007) in the treatment of microfiltered OMWs with PES spiralwound membranes having an MWCO of the same order (6 kDa). Membranes with NMWCO of 10 kDa were tested at 2 bar and 25  C. Also in this case the regenerated cellulose membrane C010F exhibited a higher permeate flux. The reduction of permeate flux in comparison to the initial value was quite the same for both membranes (32% for the P010F membrane and 29.5% for the C010F membrane). Table 2 shows the fouling index values for the selected membranes based on their water permeability before and after the treatment of microfiltered OMWs. PES membranes fouled more than cellulose regenerated and PVDF membranes. Similar results were obtained by Ulbricht et al.

Table 5 e Analyses of free low molecular weight polyphenols in samples of OMWs treated by UF membranes. Membrane type

Sample

UF-PES-004H

Feed Retentate Permeate Feed Retentate Permeate Feed Retentate Permeate Feed Retentate Permeate

C005F

C010F

P010F

Hydroxytyrosol (mg/L)

Protocatechuic acid (mg/L)

Tyrosol (mg/L)

Caffeic acid (mg/L)

p-Cumaric acid (mg/L)

Total phenols (mg/L)

7.5 6.73 7.32 8.5 9.5 8.0 5.88 6.5 5.5 6.52 6.7 6.4

7.57 8.0 7.0 15.0 14.8 13.7 14.0 13.5 13.0 13.8 14.4 13.5

43.3 48.0 34.4 36.56 36.0 35.7 34.0 35.0 33.0 33.7 35.0 29.0

4.0 4.8 3.2 7.14 7.46 6.32 3.43 4.0 3.11 3.24 3.35 3.06

1.12 1.96 0.72 1.08 1.35 0.8 0.77 0.8 0.67 0.80 0.78 0.65

63.49 69.49 52.64 68.28 69.11 64.52 58.08 59.80 55.28 58.06 60.23 52.61

RTotal (%)

phenols

17.0

5.5

4.8

9.3

3202

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 9 7 e3 2 0 4

Table 6 e Analyses of TOC, TC and TIC in samples of OMWs treated by UF membranes. Membrane type

Sample

TOC (mg/L)

TC (mg/L)

TIC (mg/L)

RTOC (%)

UF-PES-004H

Feed Retentate Permeate Feed Retentate Permeate Feed Retentate Permeate Feed Retentate Permeate

5718 9386 3032 5636 5952 3937 4853 5502 3574 4326 7965 2870

5725 9396 3036 5640 5960 3941 4857 5510 3578 4329 7970 2875

7.182 9.734 4.076 4.604 7.374 3.674 4.765 8.666 4.167 3.906 8.442 4.478

46.9

C005F

C010F

P010F

(2009) which found a much stronger adsorption of polyphenols in wine-like ethanol-containing buffer on PES MF membranes than with polypropylene membranes. Polar interactions, such as van der Waals and electron donoreacceptor interactions, and multiple hydrogen bonds towards the additive PVP in PES, seems to be main mechanisms involved in the adsorption of polyphenols to PES membranes. All UF membranes were regenerated with simple cleaning procedures by using 1 wt% Ultrasil 50 solution at 40  C, recovering totally the initial permeability.

3.2.

Analytical characterization

Table 3 shows results concerning the analytical evaluation of polyphenols in permeate, retentate and feed samples coming from the treatment of OMWs with different membranes. Regenerated cellulose membranes showed lower rejections towards polyphenols if compared with PES membranes. Within membranes of the same material (regenerated cellulose and PES) membranes with low NMWCO exhibited higher rejection towards polyphenols. Table 4 shows results referred to the evaluation of TAA. Also in this case regenerated cellulose membranes showed a lower rejection towards TAA, if compared with PES membranes. These results can be attributed to the

0,18 feed permeate

0,16

Polyphenols/TOC

0,14 0,12 0,10 0,08 0,06

30.1

26.3

33.6

contribution of polyphenols to the TAA of OMWs. In particular, recent studies have shown that total phenols determined by FolineCiocalteu method can be correlated to the antioxidant activity determined by different methods (ABTS and DPPH assays, for instance) (Roginsky and Lissi, 2005). For this reason, the method described by Singleton et al. (1999) has been proposed recently as a standardized method for use in the routine quality control and measurement of antioxidant capacity of food products and dietary supplements (Prior et al., 2005). Moreover, the novel designation “FC reagent reducing capacity” was suggested (Huang et al., 2005). Also in this case within membranes of the same material the rejection towards TAA was higher for membranes with lower NMWCO. Table 5 shows the analytical evaluation of free low molecular weight polyphenols in samples of OMWs treated by the selected UF membranes. Tyrosol is the most representative compound of the feed solution representing from 53.5 to 68.2% of the total polyphenols with a concentration of 33e43 mg/L. All membranes showed lower rejection towards these compounds in comparison with values observed for total polyphenols. This is in agreement with the molecular weight of the investigated phenols which is in the range 138e284 g/mol and hence lower than the MWCO of each UF membrane. According to the observed fouling index cellulose regenerated membranes showed lower rejection values than PES membranes. In Table 6 the evaluation of TOC, TC and TIC in samples of OMWs treated by the selected UF membranes is reported. The rejection of different membranes towards TOC is in agreement with results observed for polyphenols and TAA. For regenerated cellulose membranes an increasing of the polyphenols/TOC ratio was also observed in the permeate stream in comparison with the feed solution (Fig. 3). This is in agreement with an improved separation of phenolic compounds from other organic compounds of the OMWs by regenerated cellulose membranes.

0,04

4.

0,02 0,00 C010 F

C005 F

P010 F

UF PES 004H

Fig. 3 e Polyphenols/TOC ratio in feed and permeate streams of UF membranes.

Conclusions

The selection of UF membranes is a critical step of the process for the selective separation of polyphenols from OMWs. An important implication of this study is that polyphenols, probably aggregated with polysaccharides, have a higher

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 9 7 e3 2 0 4

affinity to the polar PES membranes leading to severe fouling by pore narrowing and blocking under UF conditions. Polyphenolic and antioxidant compounds were recovered in the permeate fractions of UF membranes: regenerated cellulose membranes offered best results if compared with PES membranes. For regenerated cellulose membranes an increasing of the polyphenols/TOC ratio was also observed in the permeate stream in comparison with the feed solution. On the other hand, PES membranes (and particularly the UF PES 004H membrane) exhibited higher rejection values towards phenolic compounds. Finally, in the selected operating conditions regenerated cellulose membranes showed average permeate fluxes higher than those observed for PES membranes and lower fouling index. Experimental results supply useful indications about the possibility to design integrated membrane systems for obtaining from OMWs high added value products of interest for food, cosmetic and pharmaceutical applications. Polyphenols yield in UF permeates could be increased working at higher values of WCR and with a diafiltration step. UF retentates, depleted of polyphenolic compounds, could be used as fertilizers or in the production of biogas in anaerobic reactors. Nanofiltration membranes can be utilized in order to concentrate specific phenol classes, with a simultaneous separation of dissolved minerals and low molecular organic matter, from the UF permeates. Finally, chromatographic techniques could be combined with membrane operations to isolate polyphenols in retentate fractions in function of the desired product.

Acknowledgements The authors gratefully acknowledge Calabria Region, Ministero dello Sviluppo Economico, Ministero dell’Istruzione, Universita` e Ricerca (Italy) and European Union which supported this work within the frame of “Calabria region contract APQ-olio”.

references

Allouche, N., Fki, I., Sayadi, S., 2004. Towards a high yield recovery of antioxidants and purified hydroxytyrosol from olive mill wastewaters. Journal of Agricultural and Food Chemistry 52, 267e273. Ammary, B.Y., 2005. Treatment of olive mill wastewater using an anaerobic sequencing batch reactor. Desalination 177, 157e165. Akdemir, E.O., Ozer, A., 2009. Investigation of two ultrafiltration membranes for treatment of olive mill wastewaters. Desalination 249, 660e666. Aktas, E.S., Imre, S., Ersoy, L., 2001. Characterization and lime treatment of olive mill wastewater. Water Research 35 (9), 2336e2340. Azbar, N., Bayram, A., Filibeli, A., Muezzinoglu, A., Sengul, F., Ozer, A., 2004. A review of waste management options in olive oil production. Critical Reviews in Environmental Science and Technology 34, 209e247.

3203

Borsani, R., Ferrando, B., 1996. Ultrafiltration plant for olive vegetation waters by polymeric membrane batteries. Desalination 108, 281e286. Bouzid, O., Navarro, D., Roche, M., Asther, M., Haon, M., Delattre, M., Lorquin, J., Labal, M., Asther, M., LesageMeessen, L., 2005. Fungal enzymes as a powerful tool to release simple phenolic compounds from olive oil byproducts. Process Biochemistry 40, 1855e1862. Coskun, T., Debik, E., Demir, N.M., 2010. Treatment of olive mill wastewaters by nanofiltration and reverse osmosis membranes. Desalination 259, 65e70. Crea, R., Creagri Inc., 2004. An hydroxytyrosol-rich composition from olive vegetation water and method of use thereof. Int. Patent WO/2004/005228. Fiestas Ros de Ursinos, J.A., 1981. Differentes utilisations des margines. In: Proc. of Seminaire International sur la valorisation des sous produits de l’olivier, FAO-UNAP, Monastir, Tunisie, pp. 93e110. Garcia-Castello, E., Cassano, A., Criscuoli, A., Conidi, C., Drioli, E., 2010. Recovery and concentration of polyphenols from olive mill wastewaters by integrated membrane system. Water Research 44, 3883e3892. Garcia Garcia, I., Jimenez Pen˜a, P.R., Bonilla Venceslada, J.L., Martin Martin, A., Martin Santos, M.A., Ramos Gomez, E., 2000. Removal of phenol compounds from olive mill wastewater using Phanerochaete chrysosporium, Aspergillus niger, Aspergillus terreus and Getrichum candidum. Process Biochemistry 35, 751e758. Hamdi, M., 1996. Anaerobic digestion of olive mill wastewaters. Process Biochemistry 31 (2), 105e110. Huang, D., Ou, B., Prior, R.L., 2005. The chemistry behind antioxidant capacity assays. Journal of Agricultural and Food Chemistry 53, 1841e1856. Jaouani, A., Sayadi, S., Vanthournhout, M., Penninckx, M., 2003. Potent fungi for decolourisation of olive oil mill wastewaters. Enzyme and Microbial Technology 33, 802e809. Jarboui, R., Chtourou, M., Azri, C., Gharsallah, N., Ammar, E., 2010. Time-dependent evolution of olive mill wastewater sludge organic and inorganic components and resident microbiota in multi-pond evaporation system. Bioresource Technology 101, 5749e5758. Lesage-Meessen, L., Navarro, D., Maunier, S., Sigoillot, J.C., Lorquin, J., Delattre, M., Simon, J.L., Asher, M., Labat, M., 2001. Simple phenolic content in olive oil residues as a function of extraction systems. Food Chemistry 75, 501e507. Ma¨ntta¨ri, M., Nystro¨m, M., 2007. Membrane filtration for tertiary treatment of biologically treated effluents from the pulp and paper industry. Water Science & Technology 55, 99e107. Marques, I.P., 2001. Anaerobic digestion treatment of olive mill wastewater for effluent re-use in irrigation. Desalination 137, 233e239. Martinez Nieto, L., Garrido Hoyos, S.E., Camacho Rubio, F., Garcia Pareja, M.P., Ramos Cormenzana, A., 1993. The biological purification of waste products from olive oil extraction. Bioresource Technology 43, 215e219. Obied, H.K., Allen, M.S., Bedgood, D.R., Prenzler, P.D., Robards, K., Stockmann, R., 2005. Bioactivity and analysis of biophenols recovered from olive mill waste. Journal of Agricultural and Food Chemistry 53, 823e837. Obied, H.K., Bedgood, D.R., Prenzler, P.D., Robards, K., 2007. Bioscreening of Australian olive mill waste extracts: biophenol content, antioxidant, antimicrobial and molluscicidal activities. Food and Chemical Toxicology 45, 1238e1248. Paraskeva, C.A., Papadakis, V.G., Tsarouchi, E., Kanellopoulou, D. G., Koutsoukos, P.G., 2007. Membrane processing for olive mill wastewater fractionation. Desalination 213, 218e229. Paredes, C., Cegarra, J., Roig, A., Sa´nchez-Monedero, M.A., Bernal, M.P., 1999. Characterization of olive-mill wastewater

3204

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 1 9 7 e3 2 0 4

(alpechı´n) and its sludge for agricultural purpose. Bioresource Technology 85, 1e8. Piperidou, C.I., Chaidou, C.I., Stalikas, D., Soulti, K., Pilidis, G.A., Balis, C., 2000. Bioremediation of olive mill wastewater: chemical alterations induced by Azotobacter vinelandii. Journal of Agricultural and Food Chemistry 48, 1941e1948. Pizzichini, M., Russo, C., Enea, Verdiana Srl, 2005. Process for recovering the components of olive mill wastewaters with membrane technologies. Int. Patent WO/2005/123603. Prior, R.L., Wu, X., Schaich, K., 2005. Standardized methods for the determination of antioxidant capacity and phenolics in foods and dietary supplements. Journal of Agricultural and Food Chemistry 53, 4290e4302. Ramos-Cormenzana, A., Juarez-Jimenez, B., Garcı´a-Pareja, M.P., 1996. Antimicrobial activity of olive mill wastewaters (alpechı´n) and biotransformed olive oil mill wastewater. International Biodeterioration and Biodegradation 38, 283e290. Re, R., Pellegrini, N., Proteggente, A., Pannala, A., Yang, M., RiceEvans, C., 1999. Antioxidant activity applying an improved ABTS radical cation decolorization assay. Free Radicals in Biology and Medicine 26, 1231e1237. Roginsky, V., Lissi, A.E., 2005. Review of methods to determine chainbreaking antioxidant activity in food. Food Chemistry 92, 235e254. Russo, C., 2007. A new membrane process for the selective fractionation and total recovery of polyphenols, water and organic substances from vegetation waters (VW). Journal of Membrane Science 288, 239e246. Singleton, V.L., Orthofer, R., Lamuela-Ravento´s, R.M., 1999. Analysis of total phenols and other oxidation substrates and

antioxidants by means of FolineCiocalteu reagent. Methods in Enzymology 299, 152e178. Sniderman Zachary, N., Ibarra, A., Natraceutical Industrial Slu, 2006. Olive polyphenols concentrate. Int. Patent WO/2006/005986. Stoller, M., 2008. Technical optimization of a dual ultrafiltration and nanofiltration pilot plant in batch operation by means of the critical flux theory: a case study. Chemical Engineering and Processing 47, 1165e1170. Takac, S., Karakaya, A., 2009. Recovery of phenolic antioxidants from olive mill wastewater. Recent Patents on Chemical Engineering 2, 230e237. Tomati, U., Galli, E., Pasetti, L., Volterra, E., 1995. Bioremediation of olive mill waste-waters by composting. Waste Management and Research 13, 509e518. Tsonis, S., Tsola, V.P., Grigoropoulos, S.G., 1989. Systematic characterization and chemical treatment of olive oil mill wastewater. Toxicological and Environmental Chemistry 20, 437e457. Turano, E., Curcio, S., De Paola, M.G., Calabro`, V., Iorio, G., 2002. An integrated centrifugation-ultrafiltration system in the treatment of olive mill wastewater. Journal of Membrane Science 209, 519e531. Ulbricht, M., Ansorge, W., Danielzik, I., Ko¨nig, M., Schuster, O., 2009. Fouling in microfiltration of wine: the influence of the membrane polymer on adsorption of polyphenols and polysaccharides. Separation and Purification Technology 68, 335e342. Visioli, F., Romani, A., Mulinacci, N., Zarini, S., Conte, D., Vincieri, F.F., Galli, C., 1999. Antioxidant and other biological activities of olive mill wastewaters. Journal of Agricultural and Food Chemistry 47, 3397e3401.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Electrochemical degradation of the b-blocker metoprolol by Ti/Ru0.7Ir0.3O2 and Ti/SnO2-Sb electrodes Jelena Radjenovic a,*, Beate I. Escher b, Korneel Rabaey a a b

The University of Queensland, Brisbane, QLD Advanced Water Management Centre, 4072, Australia The University of Queensland, Brisbane, National Research Centre for Environmental Toxicology (Entox), QLD 4108, Australia

article info

abstract

Article history:

Electrochemical oxidation has been proposed for the elimination of pesticides, pharma-

Received 7 February 2011

ceuticals and other organic micropollutants from complex waste streams. However, the

Received in revised form

detrimental effect of halide ion mediators and the generation of halogenated by-products

21 March 2011

in this process have largely been neglected thus far. In this study, we investigated the

Accepted 21 March 2011

electrochemical oxidation pathways of the b-blocker metoprolol in reverse osmosis

Available online 29 March 2011

concentrate (ROC) from a water reclamation plant using titanium anodes coated with Ru0.7Ir0.3O2 or SnO2-Sb metal oxide layers. The results of liquid chromatography-mass

Keywords:

spectrometry analysis indicated that irrespective of the electrode coating the same oxidant

Electrochemical oxidation

species participated in electrochemical transformation of metoprolol in ROC. Although

Bioassays

Ti/SnO2-Sb exhibited higher oxidizing power for the same applied specific electrical charge,

Halogenated by-product

the generation of large fractions of chloro-, chloro-bromo- and bromo derivatives was

Micropollutant

observed for both electrode coatings. However, degradation rates of metoprolol and its degradation products were generally higher for the Ti/SnO2-Sb anode. Chemical analyses of metoprolol and its by-products were complemented with bioanalytical tools in order to investigate their toxicity relative to the parent compound. Results of the bioluminescence inhibition test with Vibrio fischeri and the combined algae test with Pseudokirchneriella subcapitata indicated a substantial increase in non-specific toxicity of the reaction mixture due to the formed halogenated by-products, while the specific toxicity (inhibition of photosynthesis) remained unchanged. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

The unique ability of electrochemical processes to oxidize and/ or reduce organic compounds at controlled electrode potentials and using only electrons as reagents represents a great advantage over existing advanced oxidation processes (AOPs). Due to the generation of a variety of reactive oxygen species (ROS) at the anode (e.g. OH , H2O2, O2, O3), electrochemical oxidation is considered to be versatile and capable of oxidizing persistent organic micropollutants, nitrogen species and -

microorganisms (Comninellis et al., 2010). In recent years, there has been an increasing interest in applying electrochemical oxidation for the treatment of refractory waste streams such as landfill leachate and reverse osmosis concentrate (ROC) from water treatment processes (Anglada et al., 2009; Van Hege et al., 2004; Dialynas et al., 2008; Perez et al., 2010). Due to the high concentrations of organic contaminants, disposal of these streams represents a major problem as they require an on-site treatment. High concentrations of Cl
ions in landfill leachate and ROC not only promote indirect oxidation via reactive

* Corresponding author. Tel.: þ61 7 3346 3234; fax: þ61 7 3365 4726. E-mail address: [email protected] (J. Radjenovic). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.040

3206

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

halogen species (RHS) (e.g. Cl2/ClO
, ClO2), they also lower the ohmic resistance of the system making it more energy-efficient. However, the main issue raised among environmental electrochemists is their detrimental effect on the outcome of oxidation due to the generation of chlorinated by-products. The scenario with a plurality of different ROS generated during electrochemical oxidation is already very complex, and besides ROS it will involve a number of different radical and non-radical RHS in the Cl
containing solutions. Generally, it is considered that the formation and distribution of electrochemical oxidation by-products is also governed by the oxidizing power of an anode and applied potential. The oxidizing power is determined by the anode overpotential toward the oxygen evolution reaction and adsorption enthalpy of the electrogenerated OH (Comninellis et al., 2010). Anodes such as boron-doped diamond, Ti/SnO2-Sb, Ti/PbO2 are characterized by a high overpotential for the oxygen evolution reaction. They are capable of combusting organic compounds and are usually referred to as non-active anodes (i.e. with weak electrode-OH interaction). Anodes such as Ti/IrO2 and Ti/RuO2 have a lower oxidizing power and rather promote selective oxidation, i.e. conversion rather than degradation. These electrodes have a strong interaction with the generated OH and are called active anodes. Extensive investigation has been performed on the electrochemical oxidation of specific organic micropollutants in supporting electrolytes (e.g. Na2SO4, KH2PO4, NaCl) using active and nonactive electrodes. Recent studies have addressed the formation of low molecular weight halogenated by-products such as trihalomethanes and haloacetic acids during electrochemical oxidation of landfill leachate (Anglada et al., 2011) and ROC (Perez et al., 2010; Bagastyo et al., 2011). However, to the author’s knowledge, no investigation has been done on the oxidative transformations of organic micropollutants in real waste streams, and hence the effects of a complex matrix with high chloride content on the formation of degradation products have not been established. In this study, the electrochemical oxidation pathways of a model organic micropollutant, the b-blocker metoprolol, were investigated in a real ROC matrix. Metoprolol is used for the treatment of a range of cardiac conditions such as hypertension, angina and arrhythmias, and it is commonly found in wastewater effluents and surface water (Hernando et al., 2007). The experiments were performed using two electrodes coated with active and non-active layers of Ru0.7Ir0.3O2 and SnO2-Sb, respectively, at two current densities. The ability of a hybrid triple quadrupole-linear ion trap mass spectrometer to establish fragmentation pathways in MS3 sequential product ion spectra experiments enabled elucidation of electrochemical oxidation pathways and identification of metoprolol by-products. Given that not all by-products can be readily quantified by chemical analysis, we additionally evaluated the toxicity of the reaction mixture (ROC alone and ROC spiked with metoprolol) during electrochemical oxidation in bioluminescence inhibition tests (Microtox) using the marine bacterium Vibrio fischeri, and a combined algae test with Pseudokirchneriella subcapitata, respectively. The combined algae test was chosen as previous work indicated a high algal toxicity of b-blockers (Neuwoehner and Escher, 2011; Escher et al., 2006). -

-

-

2.

Materials and methods

2.1.

Chemicals, media and setup

Metoprolol tartarate (analytical grade, 99%) was purchased from Sigma-Aldrich (U.S.A.). All solvents (methanol, acetonitrile and water) as well as formic acid (98%) were HPLC-grade (Merck, Germany). The ROC used in the experiments was sampled at an advanced water treatment plant (AWTP) in Bundamba, 30 km west of Brisbane, Australia. This AWTP receives a mixture of secondary treated effluents from four wastewater treatment plants. After pretreatment of the secondary effluents (addition of iron coagulants, separation of solids in the clarifier), water is passed through microfiltration and reverse osmosis membranes before being treated with hydrogen peroxide/UV advanced oxidation and chemical stabilization and desinfection. The ROC was obtained straight after the reverse osmosis step. The concentrations of chloride and bromide ions in ROC were measured to be 1.65 g L
1 and 1.48 mg L
1, respectively. The electrochemical cell was constructed by assembling two equal rectangular polycarbonate frames with internal dimensions of 20  5  1.2 cm. The frames were bolted together between two rectangular polycarbonate plates. The anode cell was separated from the cathode cell by a cation exchange membrane (Ultrex CMI-7000, Membranes International, U.S.A.). Sealing was ensured by a rubber o-ring inserted between the two frames. The net volume for each compartment was 114 mL. As anodes, either a Ti/Ru0.7Ir0.3O2 electrode or a Ti/SnO2-Sb electrode was used, both with 12 g m
2 coating on the Ti mesh (dimensions: 4.8  5 cm; thickness: 1 mm; specific surface area: 1.0 m2 m
2), supplied by Magneto Special Anodes (The Netherlands). The cathode used was a woven stainless steel mesh (dimensions: 4.8  5 cm; 80 mm  0.050 mm wire diameter), both the anode and cathode had a projected electrode surface area of 24 cm2. For electrochemical control, a Wenking potentiostat/galvanostat (KP07, Bank Elektronik, GmbH, Pohlheim, Germany) was used. The anodic half-cell potentials were measured by placing an Ag/AgCl reference electrode (assumed þ0.197 V vs SHE) in the anode compartment. Data was recorded every 60 s using an Agilent 34970A data acquisition unit (Agilent Technologies, U.S.A.). For the anode, a recirculation was foreseen at 162 mL min
1, on a 1 or 10 L buffer vessel. Likewise, cathode medium (0.1 M HCl) was recirculated at 162 mL min
1 over a 1 or 10 L buffer vessel.

2.2.

Experimental approach

To investigate the electrochemical oxidation pathways of metoprolol (MTPL), 1 L of ROC collected prior to the nitrification stage was amended with the target compound dissolved in water to a final concentration of MTPL of 50 mM. The spiked ROC was then recirculated over the electrochemical cell over a period of 3 h, at current densities J ¼ 100 and 250 A m
2 using Ti/Ru0.7Ir0.3O2 or a Ti/SnO2-Sb electrode. The system was operated in batch mode, with 1 mL samples taken every 5e15 min during the 3 h experiment. To determine the toxicity of formed by-products relative to the parent compound, 10 L of the same ROC matrix containing

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

MTPL (50 mM) was oxidized in batch mode using the same reactor and the Ti/SnO2-Sb anode, at J ¼ 250 A m
2. 100 mL samples were taken after 2, 4, 6, 8, 10 and 24 h and extracted by a previously optimized solid phase extraction (SPE) method (Radjenovic et al., 2010). Analysis of sample extracts used in bioassays enabled assignment of the observed ecotoxic effect to a mixture of specific, newly identified by-products. All experiments were performed in duplicate under galvanostatic control, at room temperature (25  1  C).

2.3.

Sample preparation

As a variety of generated oxidants can prevent efficient sample stabilization, no quenching agent was added to the sample, and samples were immediately stored at
20  C. Quenching of free chlorine may lead to errors in analytical determination of trace organics due to the formation of the original compounds from their N-chloro analogs (WulfeckKleier et al., 2010). In the experiment performed to determine the toxicity of MTPL by-products, solid phase extraction (SPE) of electrochemically oxidized samples was performed. The sample pH was adjusted immediately after sampling to pH 7 by adding an appropriate amount of 0.1 M NaOH, and 100 mL samples were extracted on a Visiprep manifold system (Sigma-Aldrich, U.S.A.) using Oasis HLB cartridges (200 mg, 6 ml) from Waters Corporation (U.S.A.), previously conditioned with 10 mL of methanol and 10 mL of deionized water (HPLC-grade). Chloride and bromide ions were measured by ion chromatography (IC)-Dionex 2010i (Dionex, U.S.A.).

2.4.

3207

quantum yield of photosystem II using Imaging pulse-amplitude modulated fluorometry (2 h IPAM and 24 h IPAM) and the inhibition of growth rate (24 h growth rate). The effect concentrations causing 50% of maximum effect (EC50) were derived for all endpoints from full concentration-response curves and were expressed in units of relative enrichment factor (REF). The results of the Microtox and the endpoint of 24 h algal growth rate inhibition were additionally converted to baseline-toxicity equivalent concentration (baseline-TEQ) relative to a virtual reference compound with an EC50 of 12.2 mg L
1 in the Microtox and 18.75 mg/L in the algae test (Escher et al., 2008).

3.

Results and discussion

In the first part of this work, four different experiments were conducted using 1 L of ROC amended with MTPL and Ti/RuIrO2 and Ti/SnO2-Sb electrode operating at J ¼ 100 and 250 A m
2. Samples from these experiments were all scanned in full-scan mode and identification of unknown by-products was performed as described in Chemical analysis section. The measured anode potentials (EAN) of Ti/RuIrO2 and Ti/SnO2-Sb were 1.77  0.1 and 2.25  0.06 V vs Ag/AgCl, respectively. No electrode passivation was observed in any of the experiments conducted. The anolyte pH decreased from the initial pH 7.75  0.1 to pH 3.4  0.15 (J ¼ 100 A m
2) and 2.4  0.1 (J ¼ 250 A m
2) using Ti/RuIrO2 anode, and to pH 2.7  0.2 (J ¼ 100 A m
2) and 2.2  0.1 (J ¼ 250 A m
2) using Ti/SnO2eSb anode.

Chemical analysis

Liquid chromatography-mass spectrometry (LC-MS) analyses were performed using a Shimadzu Prominence ultra-fast liquid chromatography (UFLC) system (Shimadzu, Japan) coupled with a 4000 QTRAP quadrupole-linear ion trap mass spectrometer (QqLIT-MS) equipped with a Turbo Ion Spray source (Applied Biosystems-Sciex, U.S.A.). Chromatographic separation was achieved with an Alltima C18 Column (250  4.6 mm, particle size 5 mm) run at 40  C, supplied by Alltech Associates Inc (U.S.A.). The structure of electrochemical oxidation by-products of MTPL was elucidated by isolating the protonated molecular ions, collision induced dissociation (CID) MS2 and MS3 experiments in (þ)ESI mode and mass spectral comparison with the parent compound. Additional confirmation of the identity was obtained by the retention time (tR) and isotope abundance and distribution that enabled confirmation of halogenated intermediates. Detailed description of the recorded mass spectra is given in the Supporting Information.

2.5. Microtox and Pseudokirchneriella subcapitata bioassays The Microtox assay with Vibrio fischeri and the combined algae test with the green algae Pseudokirchneriella subcapita were performed as described previously (Escher et al., 2008), on the samples pre-treated by SPE. Three endpoints were assessed in the combined algae test: the direct inhibition of photosynthesis after 2 h and 24 h incubation via measurement of the

3.1. Structural elucidation of electrochemical oxidation products of MTPL In the full-scan experiments recorded in the range m/z 100e600 several new peaks appeared during the course of the oxidation experiment with retention time (tR) lower than the one observed for MTPL (tR ¼ 10.1 min). Four new peaks eluting at tR ¼ 2.8, 7.2, 7.9 and 8.7 min were assigned to molecular ions m/z 134 (P133), 254 (P253), 238 (P237) and 226 (P225), respectively (Table S1, Fig. 2). Products P225 (formed by N-dealkylation of MTPL) and P253 (formed by the cleavage of terminal methyl group) were previously reported as ozonation products of MTPL (Benner and Ternes, 2009), while P237 (formed by the cleavage of the terminal methoxy group) was one of the main products of e-beam and g-radiolysis of MTPL (Slegers et al., 2006). On the other hand, in CID experiments molecular ion m/z 134 was identified as a polar intermediate P133, formed by the cleavage of aromatic ether bond in the MTPL molecule (Figure S1). Next, the same nominal mass of two molecular ions at m/z 284 (mass shift of þ16 Da) and their elution at tR ¼ 6.1 and 6.8 min suggested the attachment of eOH group in different positions in the MTPL molecule, leading to different polarities of the formed intermediates. The mass spectra of these two products marked as P283-I and P283-II, respectively, are identical to the ones reported in Benner and Ternes (2009) and Slegers et al. (2006). Thus implies that the hydroxylation occurred at the benzene ring and at the b-carbon in the methoxyethyl side chain, respectively. Besides the P283-II, product P281-II (tR ¼ 6.95 min, m/z

3208

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

Fig. 1 e a) (D)ESI-QqLIT-MS2 spectrum of P281, protonated molecular ion, [M D H]D, at m/z 282; insert: (D)ESI-QqLIT-MS3 spectrum: m/z 282/m/z 208, b) (D)ESI-QqLIT-MS2 spectrum of P315; insert: (D)ESI-QqLIT-MS spectrum of P315, [M D H]D at m/z 316, [M D Na]D at m/z 338, and c) (D)ESI-QqLIT-MS2 spectrum of P359; insert: (D)ESI-QqLIT-MS spectrum of P359, [M D H]D at m/z 360. 282) formed by the further oxidation of the hydroxyl group to keto group was also detected. Further attack of RHS on the primary electrochemical oxidation products described led to the generation of their

halogenated derivatives. Thus, attachment of one chloride atom at the benzene rings of P237, P281-II and P283-II yielded products P271, P315-II and P317-II. The structures of these monochlorinated oxidation by-products was elucidated by the

3209

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

RHS P237

P271

ROS

ROS

RHS

RHS MTPL

P253

P345

RHS

RHS P293

P379

P301

P225

P335

P133 P279 (R: =O)

P281 (R: -OH) P281-II (R: =O)

P283-I

P283-II (R: -OH)

P315-II (R: =O) P317-II (R: -OH)

P357 (R: =O)

P359 (R: -OH)

RHS

RHS

P313 (R: =O)

P347 (R: =O)

P381 (R: =O)

P315 (R: -OH)

P349 (R: -OH)

P383 (R: -OH)

Fig. 2 e Proposed electrochemical oxidation pathways of metoprolol (MTPL) in ROC. ROS-reactive oxygen species, RHS-reactive halogen species.

fragment ions obtained in MS2 and MS3 scans that were shifted up for 34 Da relative to the ones observed for MTPL, and by isotope ratios and distribution in the spectra of molecular ions m/z 272, 316 and 318. Although the monochlorinated derivative of P225 was not found in the experiments, a strong signal appeared at tR ¼ 10.6 min with a molecular ion m/z 294, corresponding to the dichlorinated product P293. Concomitantly with the disappearance of MTPL four new peaks eluted at longer tR, exhibiting isotopic patterns characteristic for (poly)halogenated compounds. Their mass spectral comparison with MTPL revealed identical fragmentation patterns of monochlorinated (m/z 302, P301) and monobrominated MTPL (m/z 346, P345), shifted up for 34 and 78 Da, respectively. Also, CID fragmentation of the molecular ions m/ z 336 and 380 corroborated the halogenation at the aromatic ring for products P335 and P379, identified as dichlorinated and bromo-chloro MTPL. Another product marked as P281 (m/z 282) eluted at tR ¼ 15.2 min. The mass spectrum of P281 did not indicate the presence of halogens. Also, it did not coincide with the MS2 and MS3 fragmentation pattern of P281-II. Furthermore, appearance of product P279 (m/z 280, tR ¼ 16.4 min) with both eOH groups oxidized to eC¼O groups implied that they were located in the side chain(s) and not at the benzene ring. Loss of water (m/z 264), methanol (m/z 250) and the isopropyl moiety (m/z 208) in the (þ)ESI product ion mass spectra of P281 indicated that the methoxyethyl side chain remained unchanged,

as well as the terminal dimethylamino group (Fig. 1a). The recorded MS3 spectrum of the sequence m/z 282/208 showed an intense fragment ion m/z 163 formed by the loss of 45 Da (NH3þCO), and fragment ion m/z 145 formed by further cleavages and intramolecular cyclization (insert in Fig. 1a). Absence of fragment ion m/z 116, and a strong signal observed at m/z 130 corroborated the assumption that the eC¼O group was inserted in the a-position of ethoxy group in ethanolamide side chain. Furthermore, fragment ion m/z 130 was mutual for all halogenated derivatives of P281. Products P279 and P281 were further oxidized to their monochlorinated (P313 and P315), dichlorinated (P347 and P349) and brominated derivatives (P357 and P359). Additionally, low intensity peaks with molecular ions m/z 382 and 384 that appeared at longer oxidation time at the end of the experiment were tentatively assigned to trichlorinated by-products P381 and P383.

3.2.

Electrochemical oxidation pathway of MTPL in ROC

The identified products allowed us to propose several electrochemical oxidation pathways of MTPL in ROC, as depicted in Fig. 2. Cleavage of terminal methyl and methoxy group in MTPL methoxyethyl side chain and formation of products P253 and P237, respectively, was achieved by the H-abstraction from the terminal, a-C atom. Besides ROS such as OH and O2, Cl2
and other radical RHS can also perform H-abstraction, although they are generally less reactive than OH (Deborde and von -

-

-

3210

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

Gunten, 2008). Another radical cleavage of the aromatic ether bond generated the most polar among the detected by-products, amino-diol P133, previously reported to be a radiolytic product of MTPL (Song et al., 2008). Although they are usually assigned to OH attack, the hydroxylated products P283-I (eOH at the aromatic ring) and P283-II (eOH at the b-carbon in the methoxyethyl side chain) could have been formed by a number of ROS in the solution (e.g. HO2 , H2O2 and/or O3). Further attack of ROS on the hydroxyl group resulted in the formation of P281II. Most of the abovementioned products were previously identified as products of MTPL oxidation by OH formed in radiolytic oxidation (Slegers et al., 2006; Song et al., 2008) and ozonation (Benner and Ternes, 2009). Scission of the aromatic ether bond and oxidative attacks at the methoxyethyl side chain by OH were observed during electro-Fenton and photoelectron-Fenton treatment of MTPL (Isarain-Chavez et al., 2011). The most likely mechanisms of N-dealkylation (responsible for the formation of P225) in oxidation processes is by oxidative attacks at the a-C atom of the dimethylamino moiety, either by radical RHS or ROS through H-abstraction and consecutive cleavage of CeN bond, or by addition of O2 and release of aldehyde and dealkylated amine through a carbinolamine intermediate. Considering that the secondary amine group of MTPL (pKa ¼ 9.7) was protonated in the experimental pH range (pH 2.6e7.7), its reactivity toward both RHS and ROS can be expected to be low. Surprisingly, product P225 was very abundant and formed in slightly higher amounts on the Ti/RuIrO2 anode, which is expected to produce less OH and other radical species than Ti/SnO2-Sb. Previously, N-dealkylated product was reported by Nouri-Nigjeh et al. (2010) as a major product of direct electrochemical oxidation of lidocaine. They explained the electrochemical N-dealkylation by the direct electron transfer from nitrogen to the anode, formation of secondary amine radical cation to give iminium intermediate, which after hydrolysis and intramolecular rearrangement leads to the formation of dealkylated amine. The identified products P279 and P281 formed by the introduction of a carbonyl group on the a-C atom of ethoxy group were similar to a-keto-ester that was previously identified as a photocatalytic oxidation product of another b-blocker, atenolol, by accurate mass measurements using time of flight mass spectrometer (Q-ToF) (Radjenovic et al., 2009). Similar product was also observed during electrochemical oxidation of atenolol at boron-doped diamond anodes (Sires et al., 2010). These abundant products P279 and P281 were further transformed to their halogenated derivatives with eCl and eBr substituents attached at the aromatic ring. Although the exact position of eBr and eCl could not be determined, it is likely that it occurred in ortho position to the ether group which is ortho and para directing substituent. Therefore, halogenation of P279 and P281 yielded monochlorinated (P313 and P315), dichlorinated (P347 and P349), trichlorinated (P381 and P383) and monobrominated products (P357 and P359). The parent compound underwent similar transformation reactions, and MTPl-Cl (P301), MTPL-Cl2 (P335), MTPL-Br (P345) and MTPL-BrCl (P379) were rapidly formed at the beginning of the experiment. Electrochemical oxidation products P237, P281-II and P283-II were transformed into their monochlorinated derivatives P271, P315-II and P317-II, respectively, while for dealkylated secondary amine P225 only the dichlorinated product P293 was observed. -

-

-

-

-

Generally, the formation of chlorinated by-products in electrochemical oxidation is assigned to active chlorine (i.e. HClO/ ClO
) dissolved in the bulk liquid, although radical RHS formed

in reactions of Cl
and Br
ions with OH (e.g. Br , Br2 , Cl , Cl2 ,
and ClBr ) are likely play an important role (Deborde and von Gunten, 2008; De Laat and Le, 2006). Radical and non-radical RHS can be expected to be the dominant oxidizing agent under acidic conditions, decreasing the importance of OH and other oxidants (Grebel et al., 2010). Previously, an N-chlorinated derivative was reported as the main product of MTPL chlorination in wastewater at circumneutral pH (Bedner and MacCrehan, 2006), and in aqueous solution in the pH range 5e8 (Pinkston and Sedlak, 2004). Preferential chlorination of the benzene ring is likely a consequence of protonated amine group, which is less reactive toward active chloro-species. Nevertheless, differences between the electrochemical treatment in the presence of Cl
and aqueous chlorination could also be caused by surface electrochemical reactions, in particular adsorbed chloro and oxychloro radicals (Martinez-Huitle et al., 2005). When comparing chemical and electrochemical oxidation of phenol, Comninelis and Nerini (Comninellis and Nerini, 1995) reported that chlorinated by-products formed in the presence of hypochlorite (ClO
) are rapidly degraded at the anode surface in the case of electrochemical oxidation. On the other hand, in spite of occurring in much lower concentrations than Cl
, Br
is an important scavenger of OH (Grebel et al., 2010), and HOBr is usually more reactive than active chlorine, especially with phenolic compounds (Gallard et al., 2003). Indeed, Br-based oxidants led to the formation of brominated by-products (i.e. P345, P379, P357 and P359) in electrochemical oxidation of MTPL, despite a thousand-fold lower concentration of Br
(1.48 mg L
1) vs. 1.65 g L
1 of Cl
ions in ROC. These Br-derivatives were probably formed by HOBr (and/or bromamines) formed either at the anode surface or by indirect oxidation in the bulk (e.g. by HOCl/OCl
), although the role of Br-based radical reactions cannot be excluded (Deborde and von Gunten, 2008; Gallard et al., 2003). This is especially concerning considering that brominated by-products are suspected to be more toxic, carcinogenic and mutagenic to humans than their chlorinated analogs (Krasner et al., 2006). -

-

-

-

-

-

-

-

3.3. Effect of electrode coating and applied current density on the formation and distribution of by-products Two reaction zones of an anode can be distinguished: 1) electrochemical reaction zone (i.e. anodic surface and diffusion layer) where direct oxidation by electron transfer and/or OH occurs, and 2) chemical reaction zone (i.e. bulk liquid) where compounds are oxidized by electrogenerated oxidant species (i.e. indirect oxidation). It is very hard to distinguish between the contribution of direct and indirect oxidation as they will be influenced by multiple factors (e.g. anode coating, current density, flow dynamic regime, waste stream matrix, reactor design). It is generally assumed that the contribution of direct electron transfer between organic matter and anode is insignificant under the conditions of high current density (Ramalho et al., 2010; Comninellis, 1994). Moreover, MTPL and its oxidation by-products will need to compete with other organic matter and inorganic ions (e.g. halides) present in ROC for adsorption and oxidation at the anode surface. -

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

As postulated by Comninellis (Comninellis, 1994), the type of anode coating is expected to influence the electrochemical reaction pathways and by-products distribution due to different formation rates of non-selective OH formed by water discharge. When comparing the degradation of phenol on Ti/SnO2-Sb and Ti/RuO2 anodes, Li et al (Li et al., 2005) explained poor performance of the latter by the shortened lifetime of OH that hindered a more complete oxidation. On the other hand, Panizza and Cerisola (2009) demonstrated that in practice OH accumulate only in the vicinity of a SnO2-coated anode. Although a more significant contribution of OH was expected for the Ti/SnO2-Sb anode, near identical electrochemical oxidation pathways were observed for both anodes: monochlorinated P271 and dichlorinated P293 were observed only for the Ti/RuIrO2 anode (Figure S21a). The appearance of these additional chlorinated by-products on Ti/RuIrO2 appears rather a consequence of its lower oxidizing power (i.e. less oxidant species other than chlorine present in the bulk), than the enhanced evolution of Cl2 at this anode. Since at the Ti/ RuIrO2 anode the oxygen evolution reaction (OER) takes place at lower potential this reaction is highly competitive with Cl2 -

-

-

-

3211

evolution, while in the case of Ti/SnO2-Sb, the overpotential for OER is so high that the oxidation of Cl
is kinetically favored (Ferro et al., 2000). A detrimental effect of Cl
and Br
ions can be expected for both RuIrO2 and SnO2-Sb coatings. Fig. 3 illustrates qualitative profiles of MTPL and its halogenated derivatives, determined by the peak area in full-scan (þ)ESI experiments and normalized to the initial peak area of MTPL (t ¼ 0 min), plotted versus specific electrical charge consumed (Q, Ah L
1) (Anglada et al., 2009). When comparing Ti/RuIrO2 and Ti/SnO2-Sb anodes under the same conditions of flow rate, pH and temperature, the generation of both halogenated and non-halogenated by-products of MTPL was enhanced on the active, RuIrO2-coated electrodes. The rate of formation and disappearance of by-products was observed to be significantly faster for Ti/SnO2-Sb anode. Similar effect was observed with the increase in current density (J) for each anode, as illustrated for Ti/RuIrO2 in Figure S22, Supporting Information. In all cases, products with an inserted carbonyl group in the side chain were more abundant than their corresponding hydroxylated forms (Figures S20 b and S21). When using Ti/PbO2 anodes, Panizza et al (Panizza et al., 2008) observed faster formation of oxidation intermediates and in

Fig. 3 e The peak areas of MTPL and its halogenated derivatives P301, P335, P345 and P379, formed during electrochemical oxidation, normalized to the initial value of the peak area of MTPL (t [ 0) presented vs. Q (An h LL1): a) Ti/Ru0.3Ir0.7O2, J [ 250 A mL2, and b) Ti/SnO2-Sb2O5, J [ 250 A mL2.

3212

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

higher amounts, as well as their faster degradation with the increase in current density. In our study, at higher current density the accumulation of intermediates was lowered for both electrodes. The initial pathways of electrochemical oxidation elucidated for MTPL in complex matrix of ROC seem to be identical for both electrodes and current densities applied. Enhanced competition of electrogenerated oxidants with the OER on the Ti/RuIrO2 anode led to their lower concentration in the system and lower oxidation rates, which was especially evident in the case of more persistent halogenated derivatives that were very slowly degraded. On the other hand, higher amounts of oxidants promoted the oxidation steps on the Ti/SnO2-Sb anode, and formation and generation of all by-products was shifted to lower electrical charge compared to Ti/RuIrO2. However, it is unclear whether OH plays a significant role in a complex matrix such as ROC, with high Cl
concentrations. Several small molecular weight intermediates could be expected in OH mediated degradation of MTPL in a chloridefree electrolyte (Isarain-Chavez et al., 2011; Sires et al., 2010). However, these products were not observed in the conducted experiments. Although Ti/SnO2-Sb electrodes exhibit higher oxidizing power, electrogenerated OH will rapidly be consumed by the complex matrix of ROC, primarily NOM and RHS, leaving only a small proportion of radicals for reaction with MTPL. Thus, the generation of considerable amounts of free OH would be needed to outweigh the contribution of free chlorine and bromine, as well as various radical RHS generated (e.g. Br2
, Cl2
, Cl , HClOH /ClOH
). Moreover, OH will rapidly form a variety of other oxidants at the anode surface (e.g. O2, O3, H2O2), which are able to diffuse away into the bulk solution and react chemically with the organics. -

-

Fig. 4 e Increase in toxicity, expressed as baseline-TEQ of spiked ROC minus ROC, and sum of peak areas of the identified products (see Figure S20), scaled with the retention times in the HPLC to account for toxicity differences due to hydrophobicity as a function of the applied charge during the electrochemical oxidation experiment. , Microtox bioassay, > algae bioassay, Sum (A/A0,MTPL* tR/tR,MTPL).

-

-

-

-

-

-

-

-

3.4. Increase in toxicity determined in Vibrio fischeri and Pseudokirchneriella subcapitata bioassays As the Ti/SnO2 anode provided the fastest degradation, an additional experiment was performed with this anode to test the toxicity of the ROC after treatment. The toxicity increased substantially in both ROC alone and ROC spiked with 50 mM MTPL for all tested toxicity endpoints, apart from the direct inhibition of photosynthesis after 2 h (Text S2). The toxicity increase could be assigned to non-specific toxicity and was expressed as baseline-toxicity equivalent concentration. MTPL at 50 mM did not contribute to the mixture toxicity at all. Thus all increase in toxicity could be attributed to the transformation products, be it from MTPL degradation or from NOM degradation. Toxicity increased linearly with the applied charge for both the spiked and unspiked samples. Baseline-TEQ are concentration additive in mixtures, therefore it was possible to subtract the contribution of the products formed during the electrochemical oxidation from the high background of increasing toxicity in the unspiked ROC sample. Although the toxicity experiments were performed under slightly different experimental conditions than the experiments targeted for product identification, as illustrated in Figure S23, 17 out of the 25 identified products were retained and detected in the SPE sample extracts used for the bioassay analyses and all product concentrations increased during the

course of the experiment. Introduction of eCl and eBr substituents in the organic molecule are expected to augment significantly its biological activity and toxicological properties. While the transformation products were not quantified due to lack of standards, the sum of the peak areas weighted by their relative hydrophobicity relative to MTPL (tR/tR,MTPL) can be used as a proxy for the prediction of mixture toxicity of the identified transformation products as is explained in more detail in the Supporting Information, Text S2. This measure (sum(A/A0,MTPL*tR/tR,MTPL)) correlates linearly with applied charge (Fig. 4), and thus the formed by-products correlate linearly with the increase in mixture toxicity. Given the good agreement of the experimental and calculated increase in toxicity, we can safely conclude that the formed products are very problematic as they cause the toxicity to increase by a factor of 40e60 times depending on the toxicity endpoint.

4.

Conclusions

The elucidated electrochemical oxidation pathways indicate a very similar initial mechanism of degradation on the Ti/ RuIrO2 and Ti/SnO2-Sb anode, with a combined effect of ROS (e.g. OH , O2
, H2O2 and/or O3) and RHS (e.g. HOCl, HOBr, Br2
, Cl2 , Cl ) on the formation and degradation of the identified oxidation by-products, likely in the bulk solution. The main differences were observed in the oxidation rates, which were always higher for the Ti/SnO2-Sb anode. However, it is unclear whether OH are responsible for the higher oxidizing power of the SnO2-coated electrode when treating a complex ROC matrix with high concentrations of chloride. Further investigation is necessary to clarify the role of different oxidants produced at the anode surface and in the bulk, and establish their fate during oxidative treatment. Nevertheless, a detrimental effect of Cl
and Br
was observed for both electrode coatings, with chlorinated and brominated compounds being formed either by direct attacks -

-

-

-

-

-

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

of RHS on the MTPL molecule, or by further halogenation of oxidation by-products formed by ROS mediation. Furthermore, consecutive attacks of RHS at the aromatic ring yielded polychlorinated and chloro-bromo derivatives, which were even more persistent toward further oxidation and might require impractical reactor residence times to be further degraded.

Acknowledgments This research was supported by the Australian Research Council (grant LP0989159), Veolia Water Australia, Water Secure, Magneto Special Anodes and the Urban Water Security Research Alliance. The authors would like to thank Hanne Thoen and Miroslava Macova of Entox for performing the bioassays.

Appendix. Supplementary material Supplementary data associated with this article can be found in the online version, at doi:10.1016/j.watres.2011.03.040.

references

˜ , Urtiaga, A., Ortiz, I., 2009. Contributions of Anglada, A electrochemical oxidation to waste-water treatment: fundamentals and review of applications. J. Chem. Technol. Biotechnol. 84 (12), 1747e1755. ˜ , Urtiaga, A., Ortiz, I., Mantzavinos, D., Anglada, A Diamadopoulos, E., 2011. Boron-doped diamond anodic treatment of landfill leachate: evaluation of operating variables and formation of oxidation by-products. Water Res. 45 (2), 828e838. Bagastyo, A.Y., Radjenovic, J., Mu, Y., Rozendal, R.A., Batstone, D.J., Rabaey, K., 2011. Impact of anode material on the electrochemical oxidation of reverse osmosis membrane concentrates, Water Res., submitted. Bedner, M., MacCrehan, W.A., 2006. Reactions of the aminecontaining drugs fluoxetine and metoprolol during chlorination and dechlorination processes used in wastewater treatment. Chemosphere 65 (11), 2130e2137. Benner, J., Ternes, T.A., 2009. Ozonation of metoprolol: elucidation of oxidation pathways and major oxidation products. Environ. Sci. Technol. 43 (14), 5472e5480. Comninellis, C., 1994. Electrocatalysis in the electrochemical conversion/combustion of organic pollutants for waste water treatment. Electrochim. Acta 39 (11e12), 1857e1862. Comninellis, C., Nerini, A., 1995. Anodic oxidation of phenol in the presence of NaCl for wastewater treatment. J. Appl. Electrochem. 25 (1), 23e28. Comninellis, C., Chen, G., Kapa1ka, A., Fo´ti, G., 2010. Electrochemistry for the Environment. Springer, New York, pp. 1e23. Deborde, M., von Gunten, U., 2008. Reactions of chlorine with inorganic and organic compounds during water treatmentkinetics and mechanisms: a critical review. Water Res. 42 (1e2), 13e51. De Laat, J., Le, T.G., 2006. Effects of chloride ions on the iron(III)catalyzed decomposition of hydrogen peroxide and on the efficiency of the Fenton-like oxidation process. Appl. Catal. B. Environ. 66 (1e2), 137e146.

3213

Dialynas, E., Mantzavinos, D., Diamadopoulos, E., 2008. Advanced treatment of the reverse osmosis concentrate produced during reclamation of municipal wastewater. Water Res. 42 (18), 4603e4608. Escher, B.I., Bramaz, N., Richter, M., Lienert, J., 2006. Comparative ecotoxicological hazard assessment of beta-blockers and their human metabolites using a mode-of-action-based test battery and a QSAR approach. Environ. Sci. Technol. 40 (23), 7402e7408. Escher, B.I., Bramaz, N., Mueller, J.F., Quayle, P., Rutishauser, S., Vermeirssen, E.L.M., 2008. Toxic equivalent concentrations (TEQs) for baseline toxicity and specific modes of action as a tool to improve interpretation of ecotoxicity testing of environmental samples. J. Environ. Monit. 10 (5), 612e621. Ferro, S., De Battisti, A., Duo, I., Comninellis, C., Haenni, W., Perret, A., 2000. Chlorine evolution at highly boron-doped diamond electrodes. J. Electrochem. Soc. 147 (7), 2614e2619. Gallard, H., Pellizzari, F., Croue, J.P., Legube, B., 2003. Rate constants of reactions of bromine with phenols in aqueous solution. Water Res. 37 (12), 2883e2892. Grebel, J.E., Pignatello, J.J., Mitch, W.A., 2010. Effect of halide ions and carbonates on organic contaminant degradation by hydroxyl radical-based advanced oxidation processes in saline waters. Environ. Sci. Technol. 44 (17), 6822e6828. Hernando, M.D., Gomez, M.J., Aguera, A., Fernandez-Alba, A.R., 2007. LC-MS analysis of basic pharmaceuticals (beta-blockers and anti-ulcer agents) in wastewater and surface water. Trends Anal. Chem. 26 (6), 581e594. Isarain-Chavez, E., Garrido, J.A., Rodriguez, R.M., Centellas, F., Arias, C., Cabot, P.L., Brillas, E., 2011. Mineralization of metoprolol by electro-Fenton and photoelectro-Fenton processes. J. Phys. Chem. 115 (7), 1234e1242. Krasner, S.W., Weinberg, H.S., Richardson, S.D., Pastor, S.J., Chinn, R., Sclimenti, M.J., Onstad, G.D., Thruston Jr., A.D., 2006. Occurrence of a new generation of disinfection byproducts. Environ. Sci. Technol. 40 (23), 7175e7185. Li, X.Y., Cui, Y.H., Feng, Y.J., Xie, Z.M., Gu, J.D., 2005. Reaction pathways and mechanisms of the electrochemical degradation of phenol on different electrodes. Water Res. 39 (10), 1972e1981. Martinez-Huitle, C.A., Ferro, S., De Battisti, A., 2005. Electrochemical incineration in the presence of halides. Electrochem. Solid State Lett. 8 (11). Neuwoehner, J., Escher, B.I., 2011. The pH-dependent toxicity of basic pharmaceuticals in the green algae Scenedesmus vacuolatus can be explained with a toxicokinetic ion-trapping model. Aquat.Toxicol. 101 (1), 266e275. Nouri-Nigjeh, E., Permentier, H.P., Bischoff, R., Bruins, A.P., 2010. Lidocaine oxidation by electrogenerated reactive oxygen species in the light of oxidative drug metabolism. Anal. Chem. 82 (18), 7625e7633. Panizza, M., Cerisola, G., 2009. Direct and mediated anodic oxidation of organic pollutants. Chem. Rev. 109 (12), 6541e6569. Panizza, M., Sires, I., Cerisola, G., 2008. Anodic oxidation of mecoprop herbicide at lead dioxide. J. Appl. Electrochem 38 (7), 923e929. Perez, G., Fernandez-Alba, A.R., Urtiaga, A.M., Ortiz, I., 2010. Electro-oxidation of reverse osmosis concentrates generated in tertiary water treatment. Water Res. 44 (9), 2763e2772. Pinkston, K.E., Sedlak, D.L., 2004. Transformation of aromatic ether- and amine-containing pharmaceuticals during chlorine disinfection. Environ. Sci. Technol. 38 (14), 4019e4025. Radjenovic, J., Sirtori, C., Petrovic, M., Barcelo, D., Malato, S., 2009. Solar photocatalytic degradation of persistent pharmaceuticals at pilot-scale: kinetics and characterization of major intermediate products. App. Catal. B. Environ. 89 (1e2), 255e264. Radjenovic, J., Bagastyo, A., Rozendal, R.A., Mu, Y., Keller, J., Rabaey, K., 2010. Electrochemical oxidation of trace organic

3214

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 0 5 e3 2 1 4

contaminants in reverse osmosis concentrate using RuO2/ IrO2-coated titanium anodes. Water Res. 45 (4), 1579e1586. Ramalho, A.M.Z., Martı´nez-Huitle, C.A., Silva, D.R.d, 2010. Application of electrochemical technology for removing petroleum hydrocarbons from produced water using a DSA-type anode at different flow rates. Fuel 89 (2), 531e534. Sires, I., Oturan, N., Oturan, M.A., 2010. Electrochemical degradation of b-blockers. Studies on single and multicomponent synthetic aqueous solutions. Water Res. 44 (10), 3109e3120. Slegers, C., Maquille, A., Deridder, V., Sonveaux, E., Habib Jiwan, J.L., Tilquin, B., 2006. LC-MS analysis in the e-beam and gamma

radiolysis of metoprolol tartrate in aqueous solution: structure elucidation and formation mechanism of radiolytic products. Radiat. Phys. Chem. 75 (9), 977e989. Song, W., Cooper, W.J., Mezyk, S.P., Greaves, J., Peake, B.M., 2008. Free radical destruction of Iˆ2-blockers in aqueous solution. Environ. Sci. Technol. 42 (4), 1256e1261. Van Hege, K., Verhaege, M., Verstraete, W., 2004. Electro-oxidative abatement of low-salinity reverse osmosis membrane concentrates. Water Res. 38 (6), 1550e1558. Wulfeck-Kleier, K.A., Ybarra, M.D., Speth, T.F., Magnuson, M.L., 2010. Factors affecting atrazine concentration and quantitative determination in chlorinated water. J. Chromatogr. A 1217 (5), 676e682.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Distribution of Asellus aquaticus and microinvertebrates in a non-chlorinated drinking water supply system e Effects of pipe material and sedimentation Sarah C.B. Christensen a,*, Erling Nissen b, Erik Arvin a, Hans-Jørgen Albrechtsen a a b

DTU Environment, Technical University of Denmark, Bygningstorvet B115, 2800 Kgs. Lyngby, Denmark VCS Denmark, Vandværksvej 7, 5000 Odense C., Denmark

article info

abstract

Article history:

Danish drinking water supplies based on ground water without chlorination were investi-

Received 3 October 2010

gated for the presence of the water louse, Asellus aquaticus, microinvertebrates (<2 mm) and

Received in revised form

annelida. In total, 52 water samples were collected from fire hydrants at 31 locations, and two

10 March 2011

elevated tanks (6000 and 36,000 m3) as well as one clean water tank at a waterworks (700 m3)

Accepted 21 March 2011

were inspected. Several types of invertebrates from the phyla: arthropoda, annelida (worms),

Available online 29 March 2011

plathyhelminthes (flatworms) and mollusca (snails) were found. Invertebrates were found at 94% of the sampling sites in the piped system with A. aquaticus present at 55% of the sampling

Keywords:

sites. Populations of A. aquaticus were present in the two investigated elevated tanks but not

Invertebrates

in the clean water tank at a waterworks. Both adult and juvenile A. aquaticus (length of

Microbial quality

2e10 mm) were found in tanks as well as in pipes. A. aquaticus was found only in samples

Distribution system

collected from two of seven investigated distribution zones (zone 1 and 2), each supplied

Cast iron

directly by one of the two investigated elevated tanks containing A. aquaticus. Micro-

Water storage tank

invertebrates were distributed throughout all zones. The distribution pattern of A. aquaticus had not changed considerably over 20 years when compared to data from samples collected in 1988e89. Centrifugal pumps have separated the distribution zones during the whole period and may have functioned as physical barriers in the distribution systems, preventing large invertebrates such as A. aquaticus to pass alive. Another factor characterising zone 1 and 2 was the presence of cast iron pipes. The frequency of A. aquaticus was significantly higher in cast iron pipes than in plastic pipes. A. aquaticus caught from plastic pipes were mainly single living specimens or dead specimens, which may have been transported passively trough by the water flow, while cast iron pipes provided an environment suitable for relatively large populations of A. aquaticus. Sediment volume for each sample was measured and our study described for the first time a clear connection between sediment volume and living A. aquaticus since living A. aquaticus were nearly only found in samples with sediment contents higher than 100 ml/m3 sample. Presence of A. aquaticus was not correlated to turbidity of the water. Measurements by ATP, heterotrophic plate counting and Colilert
showed that the microbial quality of the water was high at all locations with or without animals. Four other large Danish drinking water supplies were additionally sampled (nine pipe samples and one elevated tank), and invertebrates were found in all systems, three of four containing A. aquaticus, indicating a nationwide occurrence. ª 2011 Elsevier Ltd. All rights reserved.

* Corresponding author. Tel.: þ45 45251600; fax: þ45 932850. E-mail address: [email protected] (S.C.B. Christensen). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.039

3216

1.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

Introduction

Invertebrate animals are present in drinking water distribution systems worldwide. In tropical and subtropical countries, some species of invertebrates can act as secondary hosts for parasites and thereby pose a serious health risk to consumers (Evins, 2004). In temperate areas, the presence of the animals is largely regarded as an aesthetic problem (van Lieverloo et al., 2002). However, previous studies have shown that invertebrates such as crustaceans and nematodes can harbour bacterial pathogens and potential pathogens e.g. Escherichia coli (indicator organism for faecal contamination) (Bichai et al., 2009; Levy et al., 1984), Salmonella wichita (Smerda et al., 1971) and Campylobacter jejuni (Schallenberg et al., 2005) and may play a role in the survival of these organisms in drinking water systems. The Danish water supply systems are based on ground water without chlorination, which may increase the risks of growth of bacteria and biofilm formation in the water pipes (Martiny et al., 2003) that may serve as a food supply for animals in the system. The absence of hygienic barriers between waterworks and consumers in terms of chlorination increases the focus on any potential carrier of pathogens such as e.g. invertebrates. The abundance of invertebrates in distributed drinking water is a source of consumer complaints and the supply companies highly desire to control the invertebrate abundance. Well established sampling methods have been developed in the Netherlands to assess the abundance of most invertebrate taxa in distribution systems, and a two-year survey has confirmed the wide abundance of invertebrates (van Lieverloo et al., 2004). However, studies on the controlling parameters for the distribution of invertebrates on full scale distribution systems are still lacking. In order to obtain and distribute biostable drinking water, biostable materials are needed. van der Kooij et al. (1999), and van Lieverloo et al. (2002) therefore suggested that pipe material may influence the occurrence of invertebrates. This hypothesis has not been tested on a full scale distribution system, nor has the correlation to sedimentation in the pipes and turbidity of the water. van Lieverloo et al. (2002) suggest that multiplication of invertebrates in distribution systems depends on the presence of biofilms and sediment and it is known that keeping the pipes clean by e.g. flushing reduces the amount of invertebrates in the system (Levy, 1990; van Lieverloo et al., 1998). The

risk of high sedimentation rates may be enhanced in water pipes constructed for higher flows than the actual flow due to e.g. reduction of water consumption. The water louse, Asellus aquaticus, is present in water distribution systems globally (Australian Government, 2004; Gauthier et al., 1999; Gray, 1999), which often causes consumer complaints (Walker, 1983 and unpublished results) due to its size, which makes it visible to the naked eye. Another nuisance is discolouration of the water by the faeces (pellets) of A. aquaticus. A survey from the Netherlands showed that though A. aquaticus was not the most abundant of invertebrates present in water distribution systems, most of the invertebrate biomass (86%) was formed by A. aquaticus (van Lieverloo et al., 1998). The aims of this study were, a) to implement methods to examine the distribution of invertebrates in a drinking water system with special emphasis on A. aquaticus, b) to investigate the spatial distribution of A. aquaticus in different pressure zones and c) to identify factors influencing or being influenced by the presence of A. aquaticus with special emphasis on pipe materials, sedimentation, turbidity and microbial water quality.

2.

Material and methods

2.1.

Locations

The investigated water supply system, VCS Denmark, in Odense, Denmark supplies approximately 150,000 people via a distribution system with 1000 km of pipes and a total pipe volume of 40,000 m3. The supply company distributes about 10 million m3 per year with an average flow velocity in the pipes of 0e0.5 m/s. Hence the average residence time is two days but varies from 1 to 14 days. The majority of the pipes are of PVC (polyvinyl chloride) (46%) or PE (polyethylene) (33%), while 20% of the pipes are concrete, asbestos cement or ductile iron pipes (Table 1). The remaining cast iron pipes (1%) are currently being replaced by plastic pipes. The supply system is divided into eleven pressure zones of which seven zones were sampled (Table 1). Although connected, the pressure varies between the different zones, which are separated by centrifugal pumps. The supply network is constructed after a finger principle, which means that it is branched and has a unidirectional flow, hence terminating at the consumers. The transmission network on the other hand is designed as a ring system in order to obtain security of supply. The raw

Table 1 e Characteristics and number of sampling sites in the various distribution zones. Zone

1 2 3 5 6 7 8 Total

Area [km2]

78 78 23 16 7 4 2 208

Pipes [km]

463 383 43 22 8 12 5 936

Resident population #

Revenue water [m3]

93,567 54,467 1624 1557 281 1805 208 153,509

5,971,911 2,871,174 83,474 79,535 11,040 84,525 9616 9,111,275

Pipe material [%]

Samples taken #

Plastic

Cast iron

Other

Plastic

Cast iron

74 81 99 96 93 100 100 79.2

2 1 0 0 0 0 0 1.4

24 18 1 4 7 0 0 19.4

11 5 1 1 1 1 1 21

8 2 0 0 0 0 0 10

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

Table 2 e Main water quality parameters of the supply system in Odense, Denmark. Water quality parameter

Measured values in Odense

Danish guideline values

Oxygen NVOC Temperature

9.0e9.3 mg/l 1.3e2.0 mg/l 5e16  C

Conductivity

57e79 mS/m

Total hardness

14e21 H degrees

pH Iron Manganese Ammonium

7.4e7.6 <0.01e0.02 mg/l <0.005 mg/l <0.01e0.06 mg/l

Min. 5 mg/l Max. 4 mg/l Max. 12  C (recommended) Min. 30 mS/m (recommended) 5e30 H degrees (recommended) 7.0e8.5 Max. 0.1 mg/l Max. 0.02 mg/l Max. 0.05 mg/l

water is ground water treated only by aeration/stripping and biological rapid sand filtration, and distributed without the use of chlorination. Main water quality parameters are presented in Table 2.

2.2.

Sampling from pipes and in clean water tanks

Water samples from pipes were collected by flushing from above ground fire hydrants. Before each sampling, the part of the hydrant above the water main was flushed with tap water to remove terrestrial animals living in the water free part of the hydrant. Clean water (10e20 l) was poured into the hydrant and pumped out through a drainage valve with a manual pump. For sampling, a flowmeter and a fire hose were attached to the hydrant and the water was flushed directly into transparent single-use plastic bags in 1 m3 containers. The flowmeter was cleansed after each sampling and a fresh pre-rinsed fire hose was used at each site. No water was discharged by pre-flushing in order to be able to detect invertebrates inhabiting dead ends by the hydrants. At each site, samples were obtained by flushing 1 m3, corresponding to approximately 5e300 m of pipes, at maximum obtainable flow (turbulent flow). In 80% of the samples, main diameters were between 80 and 110 mm, which corresponds to 100e200 m of flushed pipes. The sampled volume was measured per time unit in order to calculate the flow velocity, and the Reynolds numbers, expressing the turbulence, were reported. Samples were obtained from 31 locations. To avoid public interest, the samples were not filtered in the street at the sampling point but were transported to the waterworks and slowly filtered (5e10 l/min) successively through two nets with mesh sizes of 500 and 100 mm. To avoid contamination from one sample to another the nets were cleansed with tap water at high flow. Reproducibility was investigated at three locations where sampling was repeated one or two times with varying time intervals (Table 3). Three water tanks: one 700 m3 clean water tank of a waterworks and two elevated tanks (elevated tank 1 containing 2  18,000 m3 and elevated tank 2 containing 6000 m3) were emptied and the floors were carefully inspected. In the elevated tank l, 20 random samples (each covering 0.35 m2)

3217

were taken on the floor in half of the tank. In the other half of the tank the flush channel (30 m2) in the length of the tank was inspected and the animals were sampled by 10 ml pipettes. A. aquaticus was easily visible in the 500 mm net samples, while 1e5 ml sediment per sample from 100 mm net samples and samples from clean water tanks were examined by stereo microscopy with a Protec digital camera (16  11.3  0.63e4.0 magnification). Invertebrates were identified, counted and measured (head to tail, rounded to nearest millimetre). To investigate whether the occurrence of invertebrates in the drinking water supply was nationwide, additional samples were taken from four large Danish water supply systems during MarcheDecember 2009. Three times three samples were obtained from cast iron pipes (Aarhus Water Ltd., Aalborg Supply, Water Ltd. and TRE-FOR Water Ltd.) by flushing and one sample was collected by visual inspection in an empty elevated tank (Copenhagen Energy Ltd.).

2.3.

Validation of sampling from pipes

Prior to the main sampling rounds, sampling efficiency was studied at varying flow velocities and with swabbing applied. Samples were filtered with nets of various mesh sizes and pieces of pipes were cut out. Up to three samples were taken at low laminar flow (Reynolds numbers < 2100) as well as up to three samples at maximum obtainable flow (turbulent flow, Reynolds numbers > 2100) at each locality. After sampling, 150 m of plastic pipe were swabbed with a foam sponge and finally 2 m of pipe were cut out for visual inspection. Swabbing was not possible in cast iron pipes due to scaling but 2 m of pipe were cut out for visual inspection. Four mesh sizes were tested for filtration of the water samples (500, 100, 20 and 10 mm).

2.4.

Analyses

2.4.1.

Bacterial analyses

Biofilm samples were collected from the inner pipe surfaces of the cut out pipe pieces by scraping of biofilm from 10 cm2 with a cotton bud. Three scrapes were taken from the plastic pipe (one before and two after swabbing with a sponge). Three samples were taken from two pieces of 1 m cast iron pipes (one from the end, one from the middle and one from a vent). Each cotton bud was kept cold in 10 ml sterile water until 50 ml of the suspension was spread on each R2A agar plate (triplicates) and 1 ml was spread on each yeast extract agar plates (triplicates) within 24 h and incubated 14 days at 20  C and 22  C. Regular bacterial control measurements by HPC (heterotrophic plate counts) on yeast extract agar at 22  C and 37  C as well as Colilert
on the supply system were conducted by Eurofins Environment Ltd., Vejen, Denmark. Sediment samples from the 36,000 m3 elevated tank 1 were investigated for bacterial numbers by R2A colony count 20  C and yeast colony count 22  C and 1e5 A. aquaticus per sample at randomly chosen samplings were crushed with a mortar and analysed for E. coli and other coliform bacteria by Colilert
. ATP measurements on the sediment were conducted on an Advance Coupe (Celsis, Landgraaf, The Netherlands) with a Celsis kit.

3218

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

Table 3 e Asellus aquaticus and sediment volume at repeated samplings. Sample locations

1 9 15

2.4.2.

Dates

1st Sampling 3

A. aquaticus/m 07.01.08 þ 24.03.09 23.10.08 þ 15.12.08 þ 10.06.09 15.12.08 þ 16.03.09

0 9 9

2nd Sampling

Sediment A. aquaticus/m Sediment A. aquaticus/m3 Sediment vol. [ml/m3] vol. [ml/m3] vol. [ml/m3] 180 5 170

Iron and Manganese

Sediment from the elevated tank 1 was dried at 105  C (18 h) and analysed for content of iron and manganese on a Varian Vista MPX Axial View Inductively Coupled Plasma (ICP) OES after acid digestion with 7 M HNO3, after 80  C sand bath (4 h) and 0.45 mm filtration (Fe: 259,940 nm, Mn: 293,931 nm).

2.4.3.

Turbidity

After settling for a minimum of 2 h, 5 L of sample were transferred to a plastic container. Following 5 s of shaking, turbidity was measured in triplicates on a Hach 2100N Laboratory Turbidimeter. A second 5 L sample was taken from each 1 m3 container when only 200 L of water sample remained in the container. Turbidity readings on the initial water were in accordance with the repeated measurements.

2.4.4.

Sediment volume

Sediment remaining in the 100 and 500 mm filters and sediment scraped from the 1 m3 plastic bags were stored in graduated glass bottles. After sedimentation for a minimum of seven days, the total sediment volume of all three fractions was measured. Statistical analyses were performed using R software (R Development Core Team, 2010).

3.

Results and discussion

3.1.

Validation of sampling methodology

3rd Sampling

3

Sampling at different flow rates revealed that only microscopic invertebrates and oligochaete worms were flushed out at laminar flow (Reynolds numbers < 2100). Highly turbulent flow (Reynolds numbers > 25,000) was necessary to flush out A. aquaticus. When a pipe was swabbed with a sponge following sampling at highly turbulent flow both A. aquaticus and microscopic invertebrates could still be found in the pipes. Additional invertebrates were not found in the cut out piece of plastic pipe nor in the cast iron pipe but the animals may have escaped while the pipes were being cut. Complete removal of terrestrial animals from the fire hydrants could not be validated. However, gill-bearing animals such as A. aquaticus would not originate from the water free part of the hydrants. In a previous study with flushing at 1.0 m/s, the removal efficiencies of different invertebrate groups varied between 30% and 75% assuming a complete removal by extensive cleaning (high velocity flushing and swabbing with 3 consecutive swabs) after sampling. Mains couplings though,

2 16 3

200 60 300

e 5 e

e 20 e

proved to be hide-outs for A. aquaticus out of reach for practical sampling methods (van Lieverloo et al., 2004). In studies operating with fixed flows of typically 1.0 m/s (e.g. van Lieverloo et al., 2004), the sampling procedure is only applicable on pipes within a certain interval of pipe diameters since flow velocities depend on the main diameters. In the present study pipes with diameters from 63 to 500 mm were sampled. To apply the method to all pipe sizes, a novel approach using Reynolds numbers was adopted, which allows for expressing the actual turbulence exerted on the invertebrates while flushing. Reynolds numbers for cast iron pipes will always be theoretical though, since the actual inner pipe diameter or roughness is not known due to corrosion and scaling. The 10 mm mesh clogged instantly, and the 20 mm mesh clogged frequently and were only used in the methodology studies. van Lieverloo et al. (2004) found that 100 mm nets retained 53e100% of the taxa with copepod larvae and nematodes being the hardest to retain. A 20 mm mesh could be used to obtain greater accuracy on the quantification on microinvertebrates but for the purpose of our study, processing of more samples was favoured. After implementation of the methodology, all subsequent sampling was done at maximum obtainable flow. Sampling size of 1 m3 was chosen as the standard sample size due to prioritisation of the quantity of sampling localities, though this volume is most likely to be too small to identify all positive samples. This sampling volume is in accordance with a 2-year survey in the Netherlands, where a sample volume of 1 m3 was recommended due to applicability (van Lieverloo et al., 2004). The low filtration rate of 5e10 L/min minimised injuring the invertebrates but damage during sampling may have led to an underestimation of the number of samples containing living A. aquaticus. Random sampling in the first half of the 36,000 m3 elevated tank 1 yielded only one A. aquaticus in total from 20 random samples covering a total area of 7 m2. Obviously, A. aquaticus was not randomly distributed on the floor of the tank but gathered in remaining pools of water. In the second half of the tank >200 A. aquaticus were sampled from an area of 30 m2 in the flush channel with remaining water, cutting transversely through the tank. The optimal sampling method in tanks was inspection of the entire floor, which was done in the 700 m3 and the 6000 m3 tanks. When tank size does not allow this method, samples should be collected from flush channels and similar low lying areas with water remaining.

3.2.

Reproducibility of flushing pipes

Three locations were sampled two or three times (Table 3). At site 1, no A. aquaticus was found during the first sampling,

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

although 3 m3 were flushed out at highly turbulent flow (Reynolds number: 100,000, flow: 1.1 m/s). Microscopy of the flushed out sediment revealed a high number of A. aquaticus pellets. When sampling at the same site approximately one year later, two A. aquaticus were caught in 1 m3 of flushed out water. Hence, A. aquaticus was present or had been present recently at site 1 during the first sampling and the population size remained relatively low over time. At the sites 9 and 15, A. aquaticus were caught at all samplings at higher as well as lower numbers per m3 than at the previous sampling. At a sampling conducted less than two months after the first sampling at site 9 the caught number of A. aquaticus was raised from 9/m3 to 16/m3, hence there was no indication of A. aquaticus being removed from the location on a long term scale by sampling at maximum obtainable flow (Reynolds number of 84,000).

3.3. tanks

Occurrence of invertebrates in pipes and clean water

Invertebrates within the phyla: arthropoda, annelida (worms) and plathyhelminthes (flatworms) were found in the drinking water distribution system (Fig. 1). The observed invertebrates are all commonly found in drinking water distribution systems (Evins, 2004; van Lieverloo et al., 2002). A land slug was observed on the wall of a clean water tank. The water louse, A. aquaticus, was found at 55% of the investigated sampling points, while 94% of the samples contained animals when microscopic invertebrates (<2 mm) and annelida were included. The highest concentrations of microinvertebrates observed in this study were 9000 specimens/m3 sample with an average of 800 specimens/m3 sample. Levels of 0e959

3219

invertebrates/m3 in drinking water leaving the waterworks were measured in a German ground water based supply (DVGW, 1997), while samples from pipes contained a rough mean of 1000 specimens/m3 in a national Dutch survey in 1993e95 with maximum values reaching more than 10,000 specimens/m3 (van Lieverloo et al., 2002). The concentrations of A. aquaticus in the positive samples of our study varied between 1 and 14 specimens/m3 with an average of 4/m3. This is slightly higher than observed in the German survey with 1e10 A. aquaticus/m3 and an average of 2/m3 (DVGW, 1997). Compared to observations decades ago these concentrations are relatively low, e.g. another survey from Germany reports concentrations of A. aquaticus of 5e30 specimens/m3 (Schwarz et al., 1966), while data from 1948e96 compiled by van Lieverloo et al. (2002) reported of means from 10 to 100 A. aquaticus/m3 in flushing water. A. aquaticus varied in length from 2 to 10 mm (Fig. 2), which is small compared to A. aquaticus from fresh water ponds reaching up to 20 mm. The average size in pipe samples was 4.3 mm (standard deviation 1.4 mm) and in the elevated clean water tanks 6.3 mm (standard deviation 1.2 mm). The abundance of large animals in the tanks may be caused by the stable environment with sufficient bacteria-rich sediment in the elevated clean water tanks. Part of the observed size difference was presumably an effect of the different sampling techniques since the nets retained all sizes of A. aquaticus while small specimens were easily ignored when tanks were visually inspected. Besides the small size, transparency of the juvenile A. aquaticus made them difficult to observe in the tanks. A. aquaticus sampled in this study were brown (adults) with small eyes (Fig. 1). Characteristic A. aquaticus pellets

Fig. 1 e A) Adult and juvenile Asellus aquaticus (Malacostraca) B) Seed schrimp (Ostracoda) C) Flatworm (Turbellaria) D) Land slug from a clean water tank E) Cyclops sp. (Maxillopoda) F) Tubifex sp. (Clitellata) G) Springtail (Entognatha). Photos: S.C.B. Christensen.

3220

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

content of iron (220 mg/g dry weight), manganese (150 mg/g dry weight) and bacteria (76,000  2700 pg ATP/ml wet sediment and 140,000 CFU/ml wet sediment by HPC 22  C). ATP measurements of water leaving the two elevated tanks before and after the periods of sampling were low, varying between 1 and 6 pg ATP/ml (Corfitzen and Albrechtsen, 2010). Samples taken from four additional large Danish distribution companies, nationwide, showed the presence of invertebrates in all investigated systems. A. aquaticus was found in three of four systems.

3.4.

Fig. 2 e Size distribution of Asellus aquaticus from all pipe samples and from three sampling rounds in a 36,000 m3 elevated clean water tank.

(DVGW, 1997; Walker, 1983) were observed in many sediment samples (Fig. 3) and could be used as an indication of the presence of A. aquaticus populations. The highest occurrence of A. aquaticus in the clean water tanks was found in the 36,000 m3 elevated tank 1 with an average of 7/m2 in the flush channel. In the elevated tank 2 of 6000 m3, an equivalent of 0.1 A. aquaticus/m2 was found on the floor of the tank. A. aquaticus, annelida and microinvertebrates were found in both elevated tanks but not in the clean water tank of the waterworks. The water supply company has never observed A. aquaticus nor their trails (Fig. 3) during previous controls in clean water tanks of any waterworks. In a German drinking water supply system, partially supplied by ground water, A. aquaticus was also found in 50% of the samples from the distribution system, while no A. aquaticus could be found at the waterworks (DVGW, 1997). Both of the investigated elevated tanks contained a layer of fine grained sediment. There was no sediment in the 700 m3 clean water tank at the waterworks and the bacterial concentration in the water of this tank was 23 CFU/ml water (HPC 22  C). The sediment from the elevated tank 1 had a high

Distribution between pressure zones

Pressure zone 1 with the elevated tank 1 contained the majority of the caught A. aquaticus (68% positive samples in zone 1, Fig. 4), while microinvertebrates were present in all parts of the investigated distribution system (94% positive samples) (Fig. 5). Pressure zone 2 with the elevated tank 2 had a few A. aquaticus positive samples, with only one living A. aquaticus and only an average of 1 specimen per positive sample. No A. aquaticus were caught in the remaining zones; zone 3 e zone 8 (Fig. 4). Samples from 1988e89 from the same area showed a similar distribution pattern: 46% of the samples in zone 1 were positive of A. aquaticus while only 5% of the samples in zones 2e8 were positive and only containing dead A. aquaticus (Fig. 4). Hence, the distribution of A. aquaticus in the samples from 2008e09 was not different from the distribution in the samples from 1988e89 (p ¼ 1.000, Fisher’s exact probability test for 2  2 tables). This indicates that the populations are quite stable once established or that newly entered specimens have similar habitat preferences as prior populations. Previous studies conclude that the establishment of breeding populations is responsible for the greatest number of invertebrates in distribution systems (Evins, 2004). DVGW (1997) pointed at a pipe leakage 30 years prior to the investigations as the way of entry for A. aquaticus, and Smalls and Greaves (1968) identified species in several distribution systems in the 1960s that according to Evins (2004) had not been recorded from natural water since the 1920s. The repeated samplings (Table 3) showed that the season of the year did not affect the occurrence of A. aquaticus. In nature, A. aquaticus breed between February and October (Gledhill et al., 1993), while we found juvenile A. aquaticus all year round in the investigated drinking water distribution

Fig. 3 e Traces of Asellus aquaticus. A) Trails on sediment in empty elevated tank. B) Pellets (faeces). The characteristic transverse fissure is seen on some pellets. Photos: S.C.B. Christensen.

3221

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

Samples 2008 - 2009 No A. aquaticus Living A. aquaticus Dead A. aquaticus Elevated tank 1 Elevated tank 2 Clean water Tank

1

Zone 7

17

2 7 34 14

30

Samples 1988 - 1989 No A. aquaticus Living A. aquaticus Dead A. aquaticus

35

20

Zone 4

Zone 3

38

33

9

3

10

5

4

8

29 24

25

15

22

21

36

Zone 1

28 23

27

11

6

32

16

Zone 2 Zone 5 Zone 6 31

13

Zone 8 18

0 0,5 1

2

3

4

5

6

7

8

9

Kilometers 10

Fig. 4 e Distribution of Asellus aquaticus in pressure zones 1e8. The distribution of A. aquaticus in the samples from 2008e09 was not different from the distribution in samples from 1988e89. The elevated water tanks in zones 1 and 2 contained A. aquaticus, while none was observed in the clean water tank in zone 8. Living A. aquaticus were observed in zone 1 covering a wide area while living A. aquaticus in zone 2 was found at only one sampling location. No A. aquaticus was observed in zones 3e8. Numbers refer to sampling locations.

system. A. aquaticus is known to adapt to changing environments over a small spatiotemporal scale (Hargeby et al., 2004). Our observations showed that populations in the drinking water system were able to increase their life span since natural populations in northern Europe are recorded a life

Fig. 5 e Samples containing invertebrates in distribution zone 1 and distribution zones 2e8.

span of up to 1 year (Gledhill et al., 1993) while the A. aquaticus collected in our study survived in culture (10  C, darkness, on sediment collected from water pipes and on maple leaves) for up to 2½ years. Zone 1 contained above 70% of the cast iron pipes of the system (Table 2) and was furthermore the earliest constructed zone (starting in the 19th century), which would provide plenty of time for the populations of A. aquaticus to establish. Zone 2 contained the remaining cast iron pipes (Table 2). It is likely that A. aquaticus over time has entered the distribution system in other zones than zone 1 and 2 but have not been able to establish breeding populations. Since zone 1 hosted a larger percentage of both cast iron pipes and A. aquaticus than zone 2, pipe material may have the greatest impact on the distribution of A. aquaticus. Previous literature states that a species like A. aquaticus is recruited into the system infrequently and in small numbers but reach high numbers by successful establishment and breeding (Smalls and Greaves, 1968). Alternatively, the elevated tanks in zone 1 and 2 may have functioned as sources for A. aquaticus but since the 36,000 m3 elevated tank 1 has been emptied, chlorinated and hosed down one year prior to sampling breeding populations are also likely to exist in the pipes. The presence of both juvenile and adult A. aquaticus in tanks as well as in pipes (Fig. 2) supports the presence of breeding populations in both systems. Finally, a factor which could inhibit migration between zones was the centrifugal pumps, which separated the zones, and may have functioned as physical barriers by killing larger invertebrates with the fast rotating blades.

3222

3.5.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

Sedimentation

The availability of food plays a great part in the ability of A. aquaticus to survive and establish breeding populations. Sediments in pipes and clean water tanks contain e.g. bacteria and protozoa and may function as a food source for A. aquaticus (Gauthier et al., 1999). This study showed that the vast majority of samples with living A. aquaticus contained a substantial volume of sediment (more than 100 ml/m3 sample) (Fig. 6). In 53% of the samples containing >100 ml sediment/m3 sample, living A. aquaticus were observed, while this was significantly lower (10%) in samples containing < 100 ml sediment/m3 sample (Fisher’s exact probability test for 2  2 tables, p ¼ 0.008). However, the number of living A. aquaticus was not directly correlated to the sediment volume in the samples (Pearson’s test for correlation). Dead A. aquaticus were equally distributed in samples containing low and high sediment volumes. This may be because dead specimens lose their grip instantly and are easily transported to neighbouring parts of the system or because A. aquaticus living in areas with low sediment volumes are less fit and more easily killed during sampling. All samples were collected at highly turbulent flows (Reynolds numbers > 25,000). At these velocities, sediment volume was not correlated to flushing flow velocities or Reynolds numbers (R-values below 0.22), hence the relationship between sediment volume and A. aquaticus positive samples cannot be explained by higher catchment rates due to more efficient flushing. Regular flushing of pipe systems can reduce the occurrence of A. aquaticus (van Lieverloo et al., 1998) but, to our knowledge, no quantitative relationship has been shown before. Repeated sampling at three localities showed that sediment volume varied from sampling to sampling and neither the sediment nor A. aquaticus were eliminated by sampling at maximum obtainable flow (Table 3). Flushing larger water volumes than 1 m3 at maximum obtainable flow may reduce the sediment to values below the threshold of approximately 100 ml sediment/m3 sample, where living A. aquaticus was found to occur, and hence reduce their occurrence.

3.6.

Pipe materials

To investigate the importance of pipe materials we compared samples from cast iron and plastic pipes in zone 1. Although present in both pipe types, significantly more samples from cast iron pipes than from plastic pipes contained A. aquaticus (100% positive samples versus 45% positive samples) (p ¼ 0.018, Fisher’s exact probability test for 2  2 tables) (Fig. 7). Furthermore, the average concentration of A. aquaticus was higher in cast iron pipes (6 specimens/m3) than in plastic pipes (1.6 specimen/m3) (p ¼ 0.037, ManneWhitney U-test) (Fig. 7). Five of the samples were taken at localities within a 300 m radius with the same source of water supplying all five points. Three of the sampled pipes were plastic pipes and the remaining two were cast iron pipes. Only the cast iron pipes contained A. aquaticus, which indicates that cast iron pipes provide an environment suitable for populations of A. aquaticus. A. aquaticus caught from plastic pipes elsewhere in the system were mainly single living specimens or dead specimens, which may have been transported passively through by the water flow. The dead specimens could also be an indication of less fit A. aquaticus, which were easily killed during sampling. There was no difference between the median value nor the mean value of sediment/m3 sample in cast iron and plastic pipes on a 5% level of significance (ManneWhitney U-test and a t-test with log transformation of the data), hence the amount of sediment was similar in the two pipe types. High sediment volumes (>100 ml sediment/m3 sample) were obtained from plastic pipes in 45% of the samples but only 40% of the fraction with high sediment volumes contained A. aquaticus. Therefore the pipe type itself had a large influence on the occurrence of A. aquaticus, which was not just caused by one pipe type accumulating more sediment than the other. Several factors may be involved in making cast iron pipes a preferable habitat for A. aquaticus: They provided many hiding places due to corrosion and scaling, and more food, e.g. from iron-oxidising and nitrite-oxidising bacteria may be

Fig. 6 e Numbers of living Asellus aquaticus and the relation to sediment volume per sample. Pointed bars show values above 2500 ml sediment or above two A. aquaticus/m3 sample. The proportion of living A. aquaticus in samples containing >100 ml sediment/m3 sample (53%) was significantly higher than in samples containing < 100 ml sediment/m3 sample (10%). * shows repeated samplings.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

Fig. 7 e The distribution of samples with living Asellus aquaticus and dead A. aquaticus from 8 cast iron pipes and 11 plastic pipes from zone 1. A. aquaticus was present in a significantly higher number of samples from cast iron pipes than plastic pipes (100% positive samples versus 45% positive samples). There was a significantly higher concentration of A. aquaticus in cast iron pipes 6.0/m3 than in plastic pipes 1.6/m3. Replicate samplings are removed. Dead A. aquaticus may be present in samples with living A. aquaticus.

available in cast iron pipes (Martiny et al. 2005). Finally, the cast iron pipes were old pipes (up to 90 years) providing an undisturbed environment. Since all cast iron pipes were more than 62 years old at the time of sampling, there was no basis for studying the effects of pipe age of cast iron pipes. For plastic pipes, the samples taken in 2008e09 containing A. aquaticus were all but one from pipes older than 32 years. In the 1988e89 samples all A. aquaticus positive samples were from pipes, which were 17e19 years old at the time of sampling. The common characteristics of these positive samples were that the pipes originated from around 1970. Hence, it may merely be due to factors related to the specific period of the construction of the system in 1970 than the pipe age itself.

3.7.

Turbidity

The occurrence of A. aquaticus did not correlate with turbidity. This was probably because high turbidity values were often measured due to red iron or black manganese colloidal particles, which did not settle though given several days. Hence, since turbidity did not simply reflect the amount of sediment, turbidity could not be used for prediction of the presence of A. aquaticus.

3.8.

Microbial water quality

Over the two years of sampling heterotrophic plate counts (HPC 37  C) did not exceed 5 CFU/ml at any control measurement at

3223

the sampling points. Neither were any E. coli nor other coliform bacteria detected at any sampling location or in the analyses of crushed A. aquaticus. This is contrary to land slugs intruding clean water tanks, which have been observed to cause measurable concentrations of coliform bacteria (unpublished results). Scrapes from biofilm (not sediment) in the cut out pieces of pipes showed low levels of heterotrophic bacteria (below an average of 190 CFU/cm2, HPC 22  C) in cast iron as well as plastic pipes. At 80% of the sampling locations, bacterial numbers measured prior to and after sampling did not exceed 10 CFU/ml (HPC 22  C). The Danish guideline value of 200 CFU/ ml (HPC 22  C for water at the consumers tap) was only exceeded at two locations, both related to cutting out pipes and most likely generated by the pipe work. At these two sites, bacterial concentrations increased from 3 CFU/ml before sampling to 210 CFU/ml after sampling and from 4 CFU/ml before sampling to 220 CFU/ml after sampling. There was no correlation between the distribution of A. aquaticus and heterotrophic bacteria based on the regular control measurements, and the microbial quality of the water in the distribution system was good in the investigated zones over the two years of sampling, including locations where A. aquaticus were caught repeatedly.

4.

Conclusions

In conclusion, this first investigation of invertebrate occurrence in a Danish drinking water distribution system showed that:  Flushing at highly turbulent flow (Reynolds numbers > 25,000) and preferably swabbing was necessary to sample A. aquaticus from drinking water pipes  Juvenile and adult invertebrates (A. aquaticus, microinvertebrates or annelida) were present in 94% of the samples, both in the distribution system in pipes and in the elevated clean water tanks  Microinvertebrates were present in all parts of the distribution system, while the occurrence of A. aquaticus was influenced by the location in the distribution system (percentage of cast iron pipes, separation by centrifugal pumps)  Data from 1988 e 89 samples showed that the distribution pattern of A. aquaticus had not changed considerably over 20 years  Microinvertebrates were present in cast iron as well as plastic pipes  A. aquaticus was present mainly in cast iron pipes and in higher concentrations than in plastic pipes  The vast majority of samples with living A. aquaticus contained a substantial volume of sediment (more than 100 ml/ m3 sample) however, the number of living A. aquaticus in the samples was not directly correlated to sediment volume in the samples  The microbial quality of the investigated drinking water distribution system was high and without correlation to the presence of A. aquaticus

3224

5.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 1 5 e3 2 2 4

Perspective

Despite various attempts over time, total removal of invertebrates from drinking water supply systems have shown close to impossible. A great nuisance to consumers is caused by larger animals like A. aquaticus. The knowledge obtained from this study can be applied to control the presence of A. aquaticus by replacing cast iron pipes with plastic pipes in areas with high concentrations of A. aquaticus. Sediment threshold values in supply system can be used to determine a feasible level of cleaning of the pipes in order to control A. aquaticus populations.

Acknowledgements We greatly acknowledge VCS Denmark and the Urban Water Technology Graduate School for co-funding the project. Special thanks to all involved people at VCS Denmark for being part of carrying out the project. We greatly acknowledge two anonymous reviewers for constructive comments on the manuscript. Thanks to Arnaud Dechesne, Peter Wieberg Larsen and Henrik Spliid for sharing their knowledge and to Copenhagen Energy Ltd., Aarhus Water Ltd., Aalborg Supply, Water Ltd. and TRE-FOR Water Ltd. for allowing us to sample from their supply systems. Thanks to Walter Bru¨sch (GEUS) for interesting field trips. Lisbeth Brusendorff is acknowledged for her assistance on graphics. Thanks to Susanne Kruse and Mona Refstrup for help in the lab and finally thanks to Char´ luva K. Vang for fruitful discussions lotte B. Corfitzen and O and support in the lab.

Appendix. Supplementary data Supplementary data related to this article can be found online at doi:10.1016/j.watres.2011.03.039.

references

Australian Government, 2004. Australian Drinking Water Guidelines 6, National Water Quality Management Strategy chap. 5. Bichai, F., Barbeau, B., Payment, P., 2009. Protection against UV disinfection of E. coli bacteria and B. subtilis spores ingested by C. elegans nematodes. Water Res. 43, 3397e3406. Corfitzen, C.B., Albrechtsen, H.J., 2010. Microbiological Investigations of the Effects of UV Treatment in the Supply System of Water Center South, Phase 2 (In Danish: Mikrobiologiske undersøgelser af effekten af UV-belysning i VandCenter Syds ledningsnet, fase 2). DTU Environment, Technical University of Denmark. DVGW, 1997. Animals in Water Supply Systems (In German: Tierische Organismen in Wasserversorgungsanlagen). DVGW Regelwerk, Technische Mitteilung, Hinweis W 271. Evins, C., 2004. Small animals in drinking water distribution systems. In: Safe Piped Water: Managing Microbial Water

Quality in Piped Distribution Systems. World Health Organization, IWA Publishing, London, pp. 101e120. Gauthier, V., Ge´rard, B., Portal, J.M., Block, J.C., Gatel, D., 1999. Organic matter as loose deposits in a drinking water distribution system. Water Res. 33, 1014e1026. Gledhill, T., Sutcliffe, D.W., Williams, W.D., 1993. British Freshwater Crustacea Malacostraca: A Key with Ecological Notes, vol. 52. Freshwater Biological Association, Scientific Publication no. 173. Gray, N.F., 1999. Water Technology an Introduction for Scientists and Engineers. Elsevier, pp. 548. Hargeby, A., Johansson, J., Ahnesjo¨, J., 2004. Habitat specific pigmentation in a freshwater isopod e adaptive evolution over a small spatiotemporal scale. Evolution 58, 81e94. Levy, R.V., Cheetham, R.D., Davis, J., Winer, G., Hart, F.L., 1984. Novel method for studying the public health significance of macroinvertebrates occurring in potable water. Appl. Environ. Microbiol. 47, 889e894. Levy, R.V., 1990. Invertebrates and associated bacteria in drinking water distribution lines. In: McFeters, G.A. (Ed.), Drinking Water Microbiology. Springer-Verlag, pp. 224e238. Martiny, A.C., Albrechtsen, H.J., Arvin, E., Molin, S., 2005. Identification of bacteria in biofilm and bulk water samples from a non-chlorinated model drinking water distribution system: detection of a large nitrite-oxidizing population associated with Nitrospira spp. Appl. Environ. Microbiol. 71, 8611e8617. Martiny, A.C., Jørgensen, T.M., Albrechtsen, H.J., Arvin, E., Molin, S., 2003. Long-term succession of structure and diversity of a biofilm formed in a model drinking water distribution system. Appl. Environ. Microbiol. 69 (11), 6899e6907. R Development Core Team, 2010. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, ISBN 3-900051-07-0. http://www. R-project.org/. Schallenberg, M., Bremer, P.J., Henkel, S., Launhardt, A., Burns, C.W., 2005. Survival of Campylobacter jejuni in water: effect of grazing by the freshwater crustacean Daphnia carinata (Cladocera). Appl. Environ. Microbiol. 71 (9), 5085e5088. Schwarz, H., Kucera, R., Klapper, H., Kramer, W., Schuster, W., 1966. Erfahrungen bei der Beka¨mpfung von Asellus aquaticus in den Wasserversorgungsanlagen der Stadt Magdeburg. Fortschritte der Wasserchemie und ihrer Grenzgebiete 4, 96e127 (in German). Smalls, I.C., Greaves, G.F., 1968. A survey of animals in distribution systems. Water Treat. Exam. 17, 150e186. Smerda, S.M., Jensen, H.J., Anderson, A.W., 1971. Escape of Salmonellae from Chlorination during Ingestion by Pristionchus Lheritieri (Nematoda: Diplogasterinae). van der Kooij, D., van Lieverloo, J.H.M., Schellart, J.A., Hiemstra, P., 1999. Distributing drinking water without disinfectant: highest achievement or height of folly? J. Water SRT Aqua 48, 31e37. van Lieverloo, J.H.M., Bosboom, D.W., Bakker, G.L., Brouwer, A.J., Voogt, R., De Roos, J.E.M., 2004. Sampling and quantifying invertebrates from drinking water distribution mains. Water Res. 38, 1101e1112. van Lieverloo, J.H.M., van der Kooij, D., Hoogenboezem, W., 2002. Invertebrates and protozoans (free-living) in drinking water distribution systems. In: Bitton, G. (Ed.), Encyclopedia of Environmental Microbiology. Wiley, New York, pp. 1718e1733. van Lieverloo, J.H.M., van Buuren, R., Veenendaal, G., van der Kooij, D., 1998. Controlling invertebrates in distribution systems with zero or low disinfectant residual. Water Supply 16, 199e204. Walker, A.P., 1983. The microscopy of consumer complaints. J. Inst. Water Eng. Sci. 37, 200e214.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Semi-quantitative evaluation of fecal contamination potential by human and ruminant sources using multiple lines of evidence Donald M. Stoeckel a,*, Erin A. Stelzer a, Robert W. Stogner b, David P. Mau b a b

USGS Ohio Water Science Center, Columbus, OH 43229, United States USGS Colorado Water Science Center, Southeast Colorado Office, Pueblo, CO 81003, United States

article info

abstract

Article history:

Protocols for microbial source tracking of fecal contamination generally are able to identify

Received 7 August 2010

when a source of contamination is present, but thus far have been unable to evaluate what

Received in revised form

portion of fecal-indicator bacteria (FIB) came from various sources. A mathematical approach

18 February 2011

to estimate relative amounts of FIB, such as Escherichia coli, from various sources based on the

Accepted 21 March 2011

concentration and distribution of microbial source tracking markers in feces was developed.

Available online 29 March 2011

The approach was tested using dilute fecal suspensions, then applied as part of an analytical suite to a contaminated headwater stream in the Rocky Mountains (Upper Fountain Creek,

Keywords:

Colorado). In one single-source fecal suspension, a source that was not present could not be

Microbial source tracking

excluded because of incomplete marker specificity; however, human and ruminant sources

Bacteroidales

were detected whenever they were present. In the mixed-feces suspension (pet and human),

Wastewater organic chemicals

the minority contributor (human) was detected at a concentration low enough to preclude

Nutrients

human contamination as the dominant source of E. coli to the sample. Without the semi-

Escherichia coli

quantitative approach described, simple detects of human-associated marker in stream

Quantitative PCR

samples would have provided inaccurate evidence that human contamination was a major source of E. coli to the stream. In samples from Upper Fountain Creek the pattern of E. coli, general and host-associated microbial source tracking markers, nutrients, and wastewaterassociated chemical detectionsdaugmented with local observations and land-use patternsdindicated that, contrary to expectations, birds rather than humans or ruminants were the predominant source of fecal contamination to Upper Fountain Creek. This new approach to E. coli allocation, validated by a controlled study and tested by application in a relatively simple setting, represents a widely applicable step forward in the field of microbial source tracking of fecal contamination. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

1.1.

Background

Fountain Creek is a high-gradient stream on the Front Range of the Rocky Mountains in Colorado. The headwaters of Fountain

Creek drain Pikes Peak, a major landmark and destination for tourism. In addition, Fountain Creek is a drinking water source for the City of Colorado Springs, Colorado, and is used for irrigation, recreation, and other purposes between Colorado Springs and the confluence with the Arkansas River at Pueblo, Colorado. In 2008, Fountain Creek was placed on the Colorado

* Corresponding author. Present address: Battelle Memorial Institute, 505 King Avenue, Columbus, OH 43201, United States. Tel.: þ1 614 670 9302. E-mail addresses: [email protected] (D.M. Stoeckel), [email protected] (E.A. Stelzer), [email protected] (R.W. Stogner), dpmau@usgs. gov (D.P. Mau). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.037

3226

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

303(d) list of impaired streams because of Escherichia coli (E. coli) contamination (CDPHE, 2008). The level of impairment is evident in historical data collected by the U.S. Geological Survey (USGS; available at http://waterdata.usgs.gov/nwis). Fecal coliform densities measured by USGS on 241 dates between 1980 and 2004 at Fountain Creek near Colorado Springs, Colorado (USGS site ID 07103700, site 12 in Fig. 1) ranged from 2 to 150,000 colony-forming units per 100 millilters (CFU/100 mL) with a median value of 200 CFU/100 mL. Similarly, E. coli densities measured by USGS on 95 dates between 2000 and 2005 ranged from 2 to 130,000 CFU/100 mL with a median value of 140 most probably number per 100 mL (MPN/100 mL). Fountain Creek is representative of a large number of contaminated streams that require remedial action by U.S. law and analagous regulations in other countries. With regards to Fountain Creek, Colorado historically has used a Class 1E (existing recreational usage) fecal coliform 30-day geometric mean standard of 200 per 100 mL or an E. coli geometric mean standard of 126 per 100 mL (CDPHE, 2005). By comparison to this and similar standards, over 15% of evaluated stream miles in the U.S. are not suitable for their designated use because of excessive fecal-indicator bacteria (FIB), making “pathogens” the largest listed cause of stream and river impairment (USEPA, 2010). This level of water-quality degradation remains despite the fact that nearly 9000 total maximum daily load (TMDL) assessments for pathogens have been done since 2002 (USEPA, 2010). Better and more effective tools clearly are called for to make more effective use of these efforts. The USEPA Protocol for Developing Pathogen TMDLs (USEPA, 2001) provides general guidance for development of management plans for waters impaired by fecal contamination, but no specific guidance on partitioning the relative contribution of contamination from, for instance, human sources as opposed to domestic animals, birds, or wildlife. Professionals in the field of microbial source tracking (MST) recognized this challenge to water management and proposed that MST tools might be used to partition sources of fecal contamination in support of TMDL development (USEPA, 2005). Existing MST tools have limitations, however. Although several reports over the past decade purport to partition fecal contamination loads according to source by use of MST tools (Booth et al., 2003; Samadpour et al., 2002; Shanks et al., 2006; Vogel et al., 2007), none has succeeded in controlled tests using artificial “aqueous test samples” of fecal contamination (Griffith et al., 2003; Vogel et al., 2007). Thus, no prior application of MST has the proven ability to measure the amount of fecal contamination coming from a particular source, as needed to effectively generate a TMDL plan. For this reason MST is characterized as an emerging field of research, typically with the intent to inform management decisions in recreational, fisheries, or drinking source waters (Reischer et al., 2008; Boehm et al., 2003; Gentry et al., 2007). Various researchers use different approaches to apply MST for sanitary water-quality managementda typical design uses a tiered strategy in which non-problem areas or areas with obvious sources of fecal contamination are identified and excluded during a sanitary survey and coarse-level monitoring (Boehm et al., 2003; Noble et al., 2006). Elimination of both non-problem areas and areas with obvious fecal

contamination sources allows efficient use of resources in later, more-intensive tiers of research. Field application of MST, particularly for watershed management systems such as the TMDL program in the U.S., is hindered by several factors. First, host-associated markers do not have absolute host specificity (Stoeckel and Harwood, 2007) and chemical tracers such as pharmaceuticals and caffeine can have non-fecal sources. This adds Type 1 (false positive) error to presenceeabsence evaluations of fecal sources. The second issue, which also can lead to Type 1 error, is that both MST markers (Bower et al., 2005) and chemical tracers (Haack et al., 2009) can be detected in water that meets FIB-based water-quality standards. Although it remains to be proven that this error is relevant to true public-health risk, at this time regulations meant to protect human health are driven by FIB concentration. This opens the possibility that human-origin fecal contamination (contributing FIB at levels lower than the relevant standard) can be detected and wrongly implicated as the major source of impairment in samples that are, in fact, grossly contaminated by a different source. These issues underscore the importance of quantification, which hitherto has not been possible, in MST studies.

1.2.

Theoretical approach

Early protocols for detection of host-associated markers for MST (reviewed in Stoeckel and Harwood, 2007) provided data in presence or absence format. More recent protocols (such as Reischer et al., 2007; Savichtcheva et al., 2007; Seurinck et al., 2005; Shanks et al., 2008, 2009) allow measurement of MST markers by use of quantitative polymerase chain reaction (qPCR). Furthermore, limited surveys of MST markers in human (Savichtcheva et al., 2007; Shanks et al., 2009) and ruminant (Shanks et al., 2008) fecal material indicate that the concentration ranges of both FIB and MST markers, while broad, have limits. Given knowledge of concentration distributions for FIB and MST markers in feces, we propose that it is mathematically possible to estimate the upper limit of hostorigin FIB in a sample based upon the measured concentration of host-associated MST markers. In cases where the calculated maximum level of FIB from a source is less than the measured FIB in water, then specific sources may be excluded as likely sources of fecal contamination. This report describes development and application of a semi-quantitative approach to fecal source tracking in validation samples and in a relatively simple example study area. A small portion of Fountain Creek, Colorado, was selected for initial evaluation of the approach because accuracy in larger, more complex, study areas would be more difficult to evaluate. Streams such as the one described in this study, on the other hand, drain areas of manageable size, potential sources of fecal contamination are relatively few and well characterized, and flow is unidirectional. This exemplar study area serves to demonstrate this semi-quantitative approach to MST, which is broadly applicable to water bodies in many geographic regions.

1.3.

Specific objectives

The long-term objective for development of quantitative MST data analysis is to inform management decisions that are

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

designed to reduce FIB concentrations and, thus, to meet locally applicable standards. In this study, our objective was to make innovative use of information including MST marker concentration measurements in water and feces, along with a suite of traditional environmental measurements, to evaluate potential contributions of FIB from various sources. Knowledge of sources can allow evaluation of public-health risk based on contamination levels from various sources (human, ruminants) as recommended for risk-based assessments by the World Health Organization (2003). A further objective of monitoring activities in this research was to provide detailed information about where and when fecal contamination entered Fountain Creek, specifically. These objectives represent a crucial first step in evaluating this approach, and thus to validate its application in similar study areas as well as more complex study areas.

2.

Materials and methods

2.1.

Study design

The research described in this report was designed in three steps: 1) collection and analysis of fecal samples from various sources 2) evaluation of quality-control fecal slurries and 3) application to a relatively simple watershed. The watershed study was a thorough investigation accomplished by collection of data related to FIB density, hydrology, chemical constituents, and MST marker concentration. The first tier of data collection was a sanitary survey, which included traditional measures such as monitoring FIB densities under different hydrologic conditions at 16 sites, collection of land-use and National Pollutant Discharge Elimination System (NPDES) discharge information (USEPA, 2009), and a visual sanitary survey. During the visual survey, scientists walked the watershed looking for leaky pipes and other apparent sources of fecal contamination and tested these sources for E. coli. On the basis of sanitary-survey results, 5 sites were selected for intensive monitoring in the second tier of the study.

2.1.1.

Known-source fecal sample collection

Known-source reference feces were collected from human sources (both septic systems and consolidated sewer flow), pets (cats and dogs), ruminants (range-fed beef cattle, elk, big horn sheep), horses, and birds (duck) (Table 2). Reference feces were composited by species; each sample composite represented 8e10 individuals of the species. Several species (human, dog, cat, cattle, horse) were sampled on two occasions (August and October 2008) and the remainder were sampled only in August. Human-source septic dips were collected from seven residences and human sewage flow was collected by dip sampling from main collector sewerlines. non-human feces samples were collected by taking equal-sized portions (approximately 2 g wet wt) from fresh feces and compositing them, by species, in a sterile container. Combined reference feces from each species were manually homogenized in plastic bags and subsamples of the homogenized samples were removed for subsequent analysis. Percent dry weight was measured by evaporating a subsample of each fecal reference sample at 105  C to constant weight.

2.1.2.

3227

Preparation and treatment of QC blind samples

One set of known-source reference feces was used to create positive-control QC blind samples. The slurries created for E. coli analysis were used as starting material, and dilutions were made in phosphate-buffered saline. Because these samples were created in the analytical laboratory, a technician not associated with the project was assigned to create QC blind samples to enhance independence: 1) all human, 2) all ruminant, 3) a single source neither human nor ruminant, and 4) a combination of human and a second source. The order, composition, and strength of each sample were unknown to those involved in the project and remained unknown until after the data were interpreted. Samples were processed in the same way as all other samples for FIB and MST markers.

2.1.3.

Sample and field data collection

Stream-water samples were collected by the hand-dip method according to standard methodology (USGS, 2006). The hand-dip method was used instead of depth-width integrated sampling because the stream is high gradient and expected to be well mixed. Streamflow was measured at each sampling site by use of Price AA or Pygmy current meters (Rickly Hydrological Company, Columbus, Ohio), depending on depth and velocity (Rantz et al., 1982). In addition to collecting water samples, field properties (pH, specific conductance, and temperature) were measured using a Beckman f240 m (Beckman Coulter, Fullerton, Calif., for pH) and an Orion model 128 m (Thermo Fisher Scientific, Pittsburgh, Pa., for temperature and specific conductance). Samples were preserved, as appropriate, by filtration, addition of acid, and (or) chilling until analysis (Wilde et al., 2004). Sixteen sites were sampled seasonally during the sanitary survey. For the year-long investigation, 5 sites were sampled monthly between October 2007 and May 2008 and semi-monthly between June and September 2008.

2.2.

Study area description

The study area is described in detail to enable full evaluation of experimental results. The upper Fountain Creek watershed extends approximately from Woodland Park, Colorado, to the confluence of Monument Creek with Fountain Creek at Colorado Springs, Colorado (Fig. 1). A continuously monitored USGS streamgage, Fountain Creek near Colorado Springs (station 07103700; sampling site 12 in Fig. 1), is located downstream from Manitou Springs, approximately 3 miles upstream from the confluence with Monument Creek. The drainage area at the streamgage is 103 sq mi (Bossong, 2001), and from 1958 to 2008 the mean daily streamflow was 16.0 cubic feet per second (CFS). The overall drainage of the study area is 118 sq mi. Elevation in the main channel drops from approximately 9250 feet at Woodland Park to approximately 5950 feet at the confluence of Monument Creek (PPACG, 2003). Streamflow in the Fountain Creek watershed has been described as seasonally variable with three distinct flow regimes: base flow, snowmelt, and summer flow (Stogner, 2000). Generally uniform base flow begins in late September or early October and extends until the following April. Snowmelt occurs from about mid-April to about mid-June with a peak in early to mid-May. Typically, more variable,

3228

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Fig. 1 e Map showing locations of sampling sites in the Fountain Creek and Ruxton Creek watersheds, Colorado, 2007e2008. Various municipalities are shown as shaded areas. Sample sites are indicated with numerals (1e15). Numerals that are bold underlined indicate intensive-monitoring sites at which a full suite of analytes was evaluated.

rainfall-driven summer streamflow follows the snowmelt period beginning about mid-June and extending through September (Stogner, 2000). Land use in upper Fountain Creek is primarily undeveloped except for areas associated with Manitou Springs and Colorado Springs (Table 1). Residential uses increase from about 2% in the headwaters to approaching 8% as Fountain Creek enters Colorado Springs. Land areas designated for agricultural and industrial uses are relatively minor (less than 3% agricultural, less than 0.5% industrial). Land designated for commercial use increases from less than 1% in the headwaters to approaching 5% at Colorado Springs. The land area designated as streets and easements is between 1% and 2% throughout the watershed. Thus, this is characterized as a generally rural, undeveloped watershed with major use as undeveloped woodland. Based on sanitary-survey results at 15 sites distributed throughout the study area (Fig. 1), 5 sites were sampled during the one-year intensive study period. These represented drainage of the area upstream from Manitou Springs (site 2, which frequently meets state standards for recreational water quality), drainage within Manitou Springs (sites 7.9, 9, and 10, which bracket the area where Fountain Creek frequently changes from meeting to exceeding recreational waterquality standards), and drainage between Manitou Springs and Colorado Springs (site 15, which frequently exceeds state standards for recreational water quality). Upstream from and surrounding Manitou Springs, most of the study area is undeveloped (Table 1). Tourism results in dramatic population fluctuations seasonally and, seasonal increases in weekend populations. There were no NPDES-permitted discharges to

Fountain Creek and tributary streams upstream from Manitou Springs (Rich Muzzy, PPACG, written commun., October 2007, USEPA, 2009). Within Manitou Springs are a significant tributary stream (Ruxton Creek) and the Pikes Peak Cog Railroad. Manitou Springs is heavily affected by seasonal tourism. Between Manitou Springs and Colorado Springs are a substantial number of NPDES discharge points located on tributary streams and the main stem, but few of the discharge-point descriptions (USEPA, 2009) indicate that they would carry fecal contamination. The main stem of Fountain Creek becomes a straightened urban/industrial reach beginning upstream from the USGS streamgage (site 12) and extending to the confluence with Monument Creek. The City of Colorado Springs intermittently diverts flow from Fountain Creek for use as drinking source water at a location downstream from the USGS streamgage. Likely controllable sources of fecal contamination, identified by local water managers, include horses, grazing cattle, wildlife, humans, and domestic pets. Horse access to Fountain Creek and its tributaries is distributed throughout the watershed, especially in the headwater areas and around the Garden of the Gods parkland in the lower reach. Some reaches in the headwaters, near Green Mountain Falls, run through horse paddocks. A small number of grazing beef cattle have access to Fountain Creek and its tributaries in the upper reaches. Pet and human access to the waterways generally is coincident, with many recreational trails running alongside waterways. Water could be contaminated by human feces directly (hikers in recreational areas, vagrants and the homeless population in peri-urban areas, and partial-immersion

Table 1 e Study site ID (as in Fig. 1), site description and USGS NWISWeb identifier, stream kilometer relative to the confluence of Monument Creek, drainage area contributing to the reach, and percent of area classified according to various land uses. Site ID

Site name and USGS NWISWeb identifier

Ruxton Creek sites 8 1 mi above mouth (site 385111104560801) 9 At mouth (site 385134104550901)

Reach area (km2)

22.4 21.2 17.5 15.6 15.0 12.9 10.5 8.9 8.5 6.9 5.0 4.2 2.7 0.5

9.9 8.9

Land use Agriculture

Commercial

Industrial

Residential

Streets and easements

Undefined/undeveloped

42.99 90.38 109.92 136.38 137.54 168.10 176.27 180.25 235.07 237.49 266.77 295.56 300.24 305.62

NAa 0.9% 2.5% 2.1% 2.0% 1.7% 1.6% 1.6% 1.3% 1.3% 1.2% 1.4% 1.4% 1.4%

NA 0.8% 0.9% 0.8% 0.8% 2.4% 3.4% 3.8% 3.5% 3.8% 3.6% 4.6% 4.6% 4.7%

NA <0.1% <0.1% <0.1% <0.1% <0.1% <0.1% <0.1% 0.1% 0.2% 0.2% 0.2% 0.3% 0.4%

NA 2.3% 5.0% 4.7% 4.9% 4.4% 4.3% 4.5% 4.4% 4.7% 8.0% 7.7% 8.1% 8.8%

NA 0.9% 1.0% 1.9% 1.9% 1.6% 1.5% 1.4% 1.1% 1.2% 1.1% 1.1% 1.1% 1.1%

NA 95% 91% 90% 90% 90% 89% 89% 90% 89% 87% 85% 84% 85%

42.80 45.70

0.1% 0.1%

0.2% 0.6%

<0.1% <0.1%

0.2% 2.1%

<0.1% <0.1%

99% 97%

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Fountain Creek sites 1 At Green Mountain Falls (site 07099990) 2 Below Crystal Creek (site 385550105003401) 3 Below Wellington Gulch (site 385438104583401) 4 Below Cascade Creek (site 385340104581001) 5 Above French Creek (site 385329104575801) 6 Below French Creek (site 385254104565901) 7 Above Cavern Gulch (site 385223104554201) 7.9 Above Ruxton Creek (site 385137104551001) 10 Below Ruxton Creek (site 385126104545101) 11 At Schryver park (site 385127104535201) 12 Near Colorado Springs (site 07103700) 13 Below Camp Creek (site 385102104521101) 14 Above 21st Street (site 385029104513001) 15 Below 8th Street (site 07103707)

Stream kilometer

a NA, not available, data were not available for land use in this reach.

3229

3230

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

bathing; recreational boating is not practical on these streams because of shallow depth), leachate from outdated on-site wastewater systems, leaky underground wastewater pipes, or accidental wastewater discharges directly to the streams. There are several campgrounds in the area, including those near 8th Street, 36th Street, and Garden of the Gods (Colorado Springs), near Schryver Park (Manitou Springs), and near Pikes Peak Highway.

2.3.

Sample analysis

2.3.1.

Fecal-indicator bacteria

E. coli density was measured in each sample by use of Colilert reagents and Quantitray 2000 most-probable number (MPN) trays (Idexx, Westbrook, Maine). Stream-water samples were processed within 6 h of collection. Dilutions (1:10 or 1:100) were made as necessary using 90 or 99 mL of sterile distilled water buffered with the Colilert reagent. E. coli density in reference fecal material was measured by first diluting a measured mass (approximately 1 g wet weight) in 100 mL of phosphate-buffered water (USEPA, 2000). Dilutions up to the final dilution were made in buffered water; the final dilution was made with sterile distilled water buffered with the Colilert reagents. Samples were incubated at 35  C in Quantitray 2000 trays for 24e28 h, and a manufacturerprovided MPN table (Idexx, Westbrook, Maine) was used to generate E. coli density estimates based on the proportion of positive reactions in each tray.

2.3.2.

Microbial source tracking analysis

Water samples were concentrated for MST analysis, prior to DNA extraction, by use of 47-mm diameter, 0.45-mm pore size acetate plus filters (GE Osmonics). Reusable polycarbonate filter funnels (Nalge Nunc International, Rochester, N.Y.) were cleaned and autoclaved, residual DNA was removed by use of DNA-Zap reagent (Applied Biosystems/Ambion, Austin, Tex.), and filters were loaded with 100 mL of sample (straight or diluted, as processed to obtain E. coli measurements within the range of detection). Vacuum was applied to dryness and loaded filters were removed, folded, and sliced with a nonoiled razor blade into 2-mm strips directly into lysis buffer (MoBio Power Soil, Carlsbad, Calif.). For reference feces samples, approximately 0.1 g of wet composited fecal sample was weighed and added directly to lysis buffer. Once in lysis buffer, samples were held frozen at 80  C for up to 6 months before extraction. Extraction was completed by lysing the sample in a BeadBeater (BioSpec Products, Bartlesville, OK) for 30 s on high setting followed by extraction with the DNA EZ kit (GeneRite, North Brunswick, N.J.) for all subsequent steps. Extracted and purified DNA was eluted from the column by use of two aliquots of 100 mL elution solution that were combined for analysis. Samples were stored at 20  C following extraction. Losses of DNA during the processes of concentration, holding, extraction, and subsequent storage were measured and corrected for by use of an exogenous spike and recovery control plasmid containing the dsRed2 sequence (Stoeckel et al., 2009). Purified, extracted DNA was analyzed for various microbial source tracking markers by use of previously published protocols. In brief, the AllBac marker (Layton et al., 2006) was

used to estimate total Bacteroidales; qHF183 (Seurinck et al., 2005) and BacHum (Kildare et al., 2007) were used to measure human-associated Bacteroidales; and BoBac (Layton et al., 2006) was used to measure ruminant-associated Bacteroidales in each sample. All qPCR analyses were done using 5 mL of DNA extract and 20 mL of master mix in an Applied Biosystems 7500 (Foster City, Calif.) thermal cycler. Applied Biosystems 2x master mix (universal TaqMan or Sybr Green formulation, as applicable) with uracil-N-glycosylase (UNG, to degrade qPCR product and avoid cross contamination) was used for each analysis. Quality-control samples were incorporated into the analysis at various stages (results summarized in Section 3.1). A processing blank (phosphate-buffered water processed in the same manner as a water sample) was run with water samples each day. An extraction blank (sterile distilled water processed in the same way as a DNA-containing sample) was done with each batch of DNA extracts. No-template controls and plasmid-based standard curves were included in all qPCR runs. Each quantitative PCR analysis was done in duplicate. Matrix inhibition was tested by addition of approximately 1000 copies per reaction volume of the qHF183 or BoBac standard-curve plasmid (respectively for Sybr and TaqManbased protocols) to the master mix (Stoeckel et al., 2009). Samples in which matrix inhibition was detected (threshold cycle of the sample more than 1 cycle higher than the threshold cycle of the spiked no-template control) were diluted 1:5, 1:10, 1:20, or 1:100. The lowest sample dilution at which matrix inhibition was no longer detected was used to generate qPCR data for that sample.

2.3.3.

Chemical and nutrient analyses

Water-quality data were collected and processed for wastewater compounds and nutrients using standard USGS techniques and procedures (USGS, 1977, 2006; Sylvester et al., 1990; Horowitz et al., 1994). All water-quality samples collected during this study were chilled and shipped by overnight express courier and analyzed by the USGS National WaterQuality Laboratory (NWQL), Lakewood, Colorado. Waterquality data reported in this report, along with associated quality-control information, are available through the USGS National Water Information System (NWISWeb, available at http://waterdata.usgs.gov/nwis; NWISWeb identifiers are given in Table 1).

2.4.

Calculations and statistical treatment of data

2.4.1.

Calculation of MST marker concentration

Samples were analyzed by qPCR over the course of a year. A composite standard curve was used to relate observed threshold cycle to marker concentration in each sample. For all runs, acceptance criteria for standard-curve coefficient of determination was 0.99 and amplification efficiency was between 0.8 and 1.1. If acceptance criteria were not met, the run was rejected and the analysis was redone. Each composite standard curve included results from 7 to 12 individual runs. In order to calculate a concentration per 100 mL of water sample from observed marker concentration per qPCR reaction, factors were applied to adjust for known volumetric

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

losses in the DNA extraction process and volumes conversions according to the following equation:

3231

obtained in 24 blank samples analyzed. The average relative percent difference for 29 duplicate sample analyses was 16%

Observed Concentrationðcopies=100 mLÞ ¼ Concentrationðcopies=reactionÞ  ðVolume extract=Volume reactionÞ  ð100=Volume filtered ðmLÞÞ

The recovery efficiency of the exogenous DNA control was calculated as the ratio of marker recovered to marker added (Stoeckel et al., 2009). The calculated marker concentration for each sample was then adjusted to the spike and recovery control according to the following equation:

(range 0e59%). For MST marker analysis, the 95th percentile value of 80e130 blank analyses was used to set detection limits that represented 45, 4, 10, and 8 copies/reaction for AllBac, qHF183, BacHum, and BoBac, respectively. The average relative percent differences among 6 duplicate measurements

Adjusted Concentration ¼ ðObserved Concentration=Recovery EfficiencyÞ

2.4.2.

Calculation of upper threshold levels

The upper credible limit for E. coli from a particular source was estimated using the range of MST marker concentrations and E. coli densities among the collection of reference materials. The 95% confidence interval was calculated for log-transformed E. coli density and log-transformed concentration of each MST marker in fecal reference materials. These values were used to estimate a maximum number of E. coli represented by one copy of each MST marker, for each source, according to the following equation.

Ratiomax;source ðMPN=copyÞ ¼

3.2. MST marker distribution in reference feces and upper thresholds of E. coli per source Reference feces samples were collected and analyzed for E. coli and MST marker concentration as described. Sample data for 5 human-source sewage samples, 4 ruminant-source fecal

Upper limit E:coli ðMPN=g dry weightÞ Lower limit marker ðcopy=g dry weightÞ

The value of Ratiomax, source was then multiplied against the concentration of MST marker detected in the water sample to estimate the maximum concentration of E. coli in the water sample that could have come from that source (Estmax,source). For evaluation of human sources other sources which were found to carry more than one marker, the lower value of Estmax,source between the markers present in that source was taken as a conservative estimate of the maximum number of E. coli that could have originated from the source, as indicated by measured MST marker concentration.

2.4.3.

(base 10 values calculated from log(concentration) data) ranged from 8% to 70% for the four MST markers evaluated (excluding samples in which the target was not detected).

Statistical regressions and means comparisons

Relations among variables were investigated by the traditional method of linear regression of untransformed or transformed data, as appropriate. The coefficient of determination (R2) was used to evaluate strength of the various relations. Microsoft Excel software was used to create these plots and generate regression equations. Means were evaluated by use of the t-test function in Microsoft Excel. The probability of equal means ( p value) was used to evaluate differences between means.

composite samples, 4 fecal composites from domestic pets, and 3 samples from other sources are compiled in Table 2. In general, the markers showed complete sensitivitydeach marker was detected in each sample from the expected hosts. Specificity, however, was in general poor. One or both of the human-associated markers was detected, albeit at low concentration, in 9 of 13 non-human samples. Similarly, the ruminant-associated marker, BoBac, was detected in 10 of 13 non-ruminant samples. Despite the lack of specificity observed for the host-associated markers in reference feces, each marker showed strong differential distribution. For example, although the humanassociated marker qHF183 was detected in 9 of 13 non-human feces samples, the detected concentrations were at least two orders of magnitude higher in human sewage than they were in non-human feces. This differential distribution was combined with the observed E. coli densities in the reference feces to calculate an upper threshold of E. coli that could be expected to originate with the represented source categories.

3.3. Test of approach against known-composition fecal suspensions

3.

Results

3.1.

Quality control

Quality-control blanks and replicates were done for each analysis. For E. coli analysis, no false-positive results were

To test whether the proposed approach could overcome limitations imposed by incomplete specificity of these markers, a series of known-source suspensions (prepared by non-project personnel) was analyzed as blind samples. Results for E. coli density observed in the test sample, upper

3232

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Table 2 e Characteristics of reference feces collected in the study area. Measured concentrations of E. coli were normalized to water content (percent dry weight) of reference samples (most-probable number per gram dry weight, MPN/g dry). Measured concentrations of fecal source tracking markers were normalized to water content and adjusted to account for recovery efficiency (copies per gram dry weight, copy/g dry). Shaded areas indicate tests that should result in no detection. All samples were composites of at least 10 individuals except Big Horn Sheep and Bear, for which only one individual scat was sampled. Source

Season

Percent dry weight

General

Human

E. coli (MPN/g dry)

AllBac (copy/g dry)

qHF183 (copy/g dry)

Ruminant

BacHum (copy/g dry)

BoBac (copy/g dry)

Human-source fecal reference materials Septic Warm Sewer 1 Cool Sewer 1 Warm Sewer 2 Cool Sewer 2 Warm

0.067% 0.060% 0.054% 0.066% 0.041%

2.5  106 9.1  107 1.6  107 2.0  108 6.6  107

1.9  1012 3.5  1012 8.2  1011 8.8  1012 1.1  1012

Ruminant-source fecal reference materials Cow Cool Cow Warm Cow Warm Big Horn Sheep (n ¼ 1) Cool Elk Warm

16% 16% 15% 35% 23%

1.0  107 3.9  106 8.3  106 6.0  105 1.0  107

4.3  1012 3.9  1012 3.3  1012 2.6  1012 9.6  1011

9.8  106 6.9  106 <1.2  107 E4.9  106 2.5  106

3.2  107 <4.9  106 <3.9  105 1.1  107 <2.6  106

2.2  1011 1.2  1012 8.2  1011 7.4  1010 1.9  1011

Alternate sources in domestic contact with humans Cat Cool 49% Cat Warm 49% Dog Cool 30% Dog Warm 33%

5.4  107 4.9  105 1.3  107 1.7  107

1.8  1012 2.4  1011 7.8  1012 1.7  1012

1.1  107 4.3  107 7.1  107 2.2  106

5.8  107 1.2  108 6.4  108 2.8  107

7.9  109 6.5  109 8.1  1010 1.1  1011

Alternate sources not in close contact with humans Horse Warm 28% Duck Warm 28% Pigeon Warm 36% Bear (n ¼ 1) Warm 21%

2.1  106 2.8  107 3.4  108 7.4  107

6.0  1011 4.6  107 8.0  106 1.0  109

<2.0  106 <2.6  105 <3.7  105 2.6  106

<5.4  105 <6.9  105 <6.5  105 5.5  107

3.5  109 6.8  1010 3.2  1010 2.0  1011 3.0  1010

1.2  1010 2.4  1011 7.0  1010 7.3  1011 7.6  1010

1.3  108 6.3  109 1.1  109 1.1  1010 1.1  109

<4.9  104 <2.8  106 <7.5  105 E1.0  107

E, estimated value, value reported is below the lowest standard in the standard curve. <, less than, value reported is calculated from the analytical detection limit.

threshold E. coli from each source calculated based on MST marker concentration, and the expected density of E. coli added to the test mixture (based on the mass of fecal material added) are compiled in Table 3. In all four cases, the source added was detected by the analysis and the calculated upper limit of E. coli from that source greatly exceeded the E. coli added. In some cases, false-positive indications of fecal

contribution were derived (such as 67,000 MPN/100 mL calculated upper limit of E. coli from ruminants in QC blind 1, and calculated upper limit densities of E. coli from pets in QC blind 2, 3, and 4 that were higher than the observed densities of E. coli despite absence of these sources in the respective samples). Thus, the approach did not allow discrimination between pet contamination and ruminant or human

Table 3 e Results for four quality-control (QC) samples. Observed E. coli densities (MPN/100 mL) are presented for each sample in italics followed by calculated upper limit (potential) E. coli contribution per fecal contamination category (Human, Ruminant, Pet, in MPN/100 mL). Sources for which host-associated MST markers were measured (Human, Ruminant) are denoted in italic when the detection was above the limit of quantification. E. coli added to the QC blind test mixture from each source (sewage for human, cattle for ruminant, cat for pets, horse for other) are presented last. QC blind 1

QC blind 2

QC blind 3

QC blind 4

Observed

E. coli

>24,000

24,000

830

930

Calculated upper limit

Human Ruminant Pets

62,000 67,000 1,300,000

ND 350,000 200,000

ND ND 4700

7900 ND 3500

Added to test mixture

Human Ruminant Pets Other

810 0 620,000 0

0 42,000 0 0

0 0 0 710

500 0 0 0

ND, not detected, analyte below limit of detection.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

3233

contamination because pets tended to shed qHF183, BacHum, and BoBac marker with their feces (Table 2) and no pet-associated MST marker was used. It was, however, possible to distinguish human from ruminant contamination. In no case was a false-negative result deriveddwhen a fecal source was present, it always was detected.

3.4. Application to the upper fountain creek test watershed 3.4.1.

Sanitary survey

Fifteen sites in the study area were sampled three or more times in 2007 (Fig. 1). One sample set was collected at high streamflow during snowmelt with no precipitation (May and early June), the second sample set was collected at low streamflow during the summer with no precipitation (July), and the third sample set was collected at medium streamflow during the winter with no precipitation (October). Based on early sanitary-survey results, a sixteenth site (designated 7.9) was added. The sanitary survey also included 5 partial sample sets, for which 5 of the 16 sites were sampled approximately monthly for 4 months. An additional late July “walking survey” was done of reaches where increases in fecal contamination previously were observed in early results. During the walking survey, samples were collected from all discharging pipes and tributaries in the segment, as well as from the main stem at locations between established sample locations. Three pipe leaks were identified during and after the walking survey. One cracked pipe released a trickle of flow containing 200,000 MPN/100 mL E. coli. Mass balance comparison of this discharge and the associated in-streamflow indicated that the cracked pipe contributed some, but not most, of the E. coli observed in Fountain Creek at the time of sampling (data not shown). One leaking pipe was on private property and the third leaking pipe was identified below ground during street repairs, shortly after the sanitary surveydneither was sampled for E. coli. All three leaking pipes were repaired before the one-year study was initiated. None of eight discharging pipes or tributaries (with the exception of Ruxton Creek) that were sampled during the sanitary survey carried higher E. coli density relative to the receiving water. Each sample was analyzed for E. coli (Fig. 2).

3.4.2.

Fig. 2 e E. coli density measurements collected during the sanitary survey, MayeOctober 2007. Site identifiers can be cross-referenced to Fig. 1 and Table 1; samples collected between established sites are plotted at corresponding locations in the plot. The upper detection limit for the method utilized (without dilution) is 2400 MPN/100 mL. The Colorado 30-day geometric mean standard of 126 MPN/100 mL is shown for reference.

after snowmelt and was defined as June 15eSeptember 15. Geometric mean E. coli density was higher in the warm season than the cool season at all five sites (Fig. 3; t-test p < 0.05). The standard generally was met in samples collected during cool season at all sites (Fig. 3)dwhen evident, exceedances were at the mouth of Ruxton Creek (site 9) and Fountain Creek below Ruxton Creek and below 8th stream (sites 10 and 15) rather than at upstream sites (sites 2 and 7.9). In contrast, the standard was exceeded more frequently during the warm season e routinely at the sites upstream from Ruxton Creek (sites 2 and 7.9) and always at Ruxton Creek (site 9) and sites downstream from Ruxton Creek (sites 10 and 15). In both seasons, upward trends in E. coli density were noted in an upstream-todownstream direction (linear regression of log E. coli with river mile, slope ¼ 0.089, R2 ¼ 0.25 (cool season); slope ¼ 0.104, R2 ¼ 0.46 (warm season), data not shown). The lower coefficient of determination in the cool-season regression appears

Year-long survey

During the year-long survey of Fountain Creek, the median flow (24 CFS) was higher than the long-term median for the entire period of record extending to 1959 (16 CFS). The maximum flow during the study period was, however, substantially lower than the maximum over the period of record (92 CFS compared to 813 CFS). Though typical streamflow was similar to or higher than normal during this investigation, there were no intense precipitation events. None of the samples collected was substantially influenced by precipitation runoff, though some of the samples (early May) were influenced by snowmelt (Fig. 2).

3.4.2.1. E. coli patterns. For the purposes of this study, samples were divided into cool-weather months (cool season) and warm-weather months (warm season) based on data patterns, hydrology, and tourism. The warm season began

Fig. 3 e E. coli density measurements during the year-long intensive monitoring, August 2007eSeptember 2008. Cool season is from September 16 to June 14; warm season is June 15eSeptember 15. Warm season and cool-season values are associated with the same sites, but are offset from the true site identifier to enhance visual clarity.

3234

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

largely to be caused by the presence of a step increase in E. coli density between Fountain Creek above Ruxton Creek (site 7.9) and Fountain Creek below Ruxton Creek at Manitou Springs (site 10) rather than the more consistent trend observed in the warm-season regression (Fig. 3).

3.4.2.2. Microbial source tracking marker patterns in stream water. The longitudinal and seasonal trends observed for the various MST markers generally were distinct from the trends observed for E. coli (Fig. 4). Like E. coli, AllBac had an upward trend and stepwise increases in concentration upstream to downstream, though site-to-site pairwise differences were significant only during warm season ( p < 0.05). In contrast, the concentrations of human-associated markers (qHF183 and BacHum) showed significant stepwise increases only at sites 7.9 and 10, in Manitou Springs, and only during the cold season (Fig. 4). The ruminant-associated marker (BoBac) showed no concentration differences longitudinally in either season ( p < 0.05 for all pairwise significant differences). Curiously, host-associated markers of fecal contamination showed seasonal differences that were inverted relative to the trends observed for E. coli (Fig. 4BeD compared to Fig. 3). Unlike E. coli, for which densities always were higher in warmseason samples compared with cool-season samples, average human-associated marker concentrations were not significantly different between seasons at the Fountain Creek sites. Human-associated markers were higher in cool-season samples than in warm-season samples at Ruxton Creek (site 9; p < 0.05). The BoBac marker of ruminant contamination tended to be higher in cool-season samples than warm-season samples at all sample sites (Fig. 4D; p < 0.05). Analysis of MST marker concentration against E. coli density by linear regression indicated that the AllBac marker of general fecal contamination was correlated with E. coli density in cool (R2 ¼ 0.36; p ¼ 0.0035) and to some extent warm (R2 ¼ 0.21; p ¼ 0.17) seasons (Fig. 5A). The qHF183 and BacHum human-associated markers were more strongly correlated with E. coli in cool (R2 ¼ 0.84; p ¼ 0.024) season than warm (R2 < 0.1; p > 0.1) season (Fig. 5B). Similarly, the BoBac ruminant-associated marker was more strongly correlated with E. coli density in cool (R2 ¼ 0.33; p ¼ 0.0011) than warm (R2 < 0.1; p > 0.1) season samples (Fig. 5C). Regression lines and equations are shown on the plots of Fig. 5AeC only for those relationships in which the coefficient of determination (R2) was greater than 0.1. MST samples from three cool-season dates were analyzed to obtain characteristic conditions when E. coli densities typically did not exceed the water-quality standard. Upper thresholds of E. coli calculated based on MST marker concentrations did not exclude any of the sources tested as a major or minor source of contamination to the samples (Table 4) except that ruminant and pet sources likely were not the major sources of 110 MPN/100 mL E. coli to Ruxton Creek (site 9) on February 20, 2008. In the two instances that E. coli density exceeded 126 MPN/100 mL (both collected on May 1, 2008), data did not exclude any of the sources tested as a potential primary contributor of E. coli to the samples. High upper thresholds for the bird category were caused by lack of a birdassociated marker (only the general marker was used to create

Fig. 4 e Microbial source tracking marker concentration measurements for (A) AllBac, (B) qHF183, (C) BacHum and (D) BoBac during the year-long intensive monitoring, August 2007eSeptember 2008. Cool season is from September 16 to June 14; warm season is June 15eSeptember 15. Warm season and cool-season values are associated with the same sites, but are offset from the true site identifier to enhance visual clarity.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

3235

calculated upper limit of E. coli density from these sources usually was lower than the E. coli densities observed in the sample; in other words, the low MST marker concentration indicated that most of the E. coli present were not from human or ruminant sources. The major exception was on August 12, 2008, when the MST marker distribution indicated that human sources may have been responsible for high E. coli densities in Fountain Creek above and below Ruxton Creek (sites 7.9 and 10) but not below 8th street (site 15). Pets frequently were excluded as major sources of E. coli based on absence of cross-carried human- and ruminant-associated MST markers in the samples. Birds, however, could not be excluded as the source of fecal contamination to these samples, because no bird-associated marker was used and because of the low concentrations of MST markers detected in the bird reference samples. In other words, bird was the only source category tested that had high E. coli but low AllBac general MST marker concentration in their feces, similar to the pattern in samples collected during warm season. Other non-human, non-ruminant sources that were not sampled may have a similar pattern, but avian is the common factor among sources with high E. coli and low MST markers (Table 2 and unpublished data representing avian (chicken, Canada goose, duck, pigeon) and nonavian (raccoon, horse, bear, cat, dog) sources).

3.4.2.3. Additional lines of evidence. Evaluation of MST

Fig. 5 e MST marker concentrations plotted against E. coli density measurements for (A) AllBac, (B) qHF183 and BacHum, and (C) BoBac during the year-long intensive monitoring, August 2007eSeptember 2008. Cool season is from September 16 to June 14; warm season is June 15eSeptember 15. The regressions for the two humanassociated markers were nearly identical; therefore, both are plotted in (B) and only the regression for BacHum is shown on the chart.

this estimate) and scarcity of data (only two composite bird samples were analyzed). All MST samples from three dates and several MST samples from four other dates during warm season, when the E. coli standard typically was exceeded, were characterized (Table 5). In most of these cases, the human- and ruminantassociated markers were not detected. When markers were detected, the marker concentration was so low that the

markers indicated that, in general, humans, ruminants, and (or) pets could have contributed E. coli at the levels observed during the cool season but not during the warm season (as defined in this study). To support this preliminary hypothesis, and to provide further support for the proposed approach, alternate lines of evidence (physical and chemical characteristics, nutrient concentrations, presence of wastewater organic chemicals) were evaluated. Analysis of physical and chemical characteristics revealed little novel information. Temperature was significantly higher in warm-season samples at all sites (Table 6), and correlation between temperature and E. coli density was observed (global R2 ¼ 0.35). Although specific conductance may be expected to vary with water source (base flow dominated as opposed to snowmelt or rainfall dominated), the few seasonal differences in specific conductance ( p < 0.05) did not follow any interpretable pattern. Specific conductance, in general, was lower in Ruxton Creek compared with Fountain Creek because of geology (Ruxton Creek drains a granitic watershed, while the headwaters of Fountain Creek are dominated by sandstone). The increase in specific conductance at the bottom of the watershed (site 15) may be caused by more urban influence, but specific conductance did not vary seasonally as did E. coli density. Observations of pH and streamflow did not vary with season. Streamflow in Fountain Creek increased in an upstream-to-downstream direction except for the reach between Manitou Springs and Colorado Springs (site 10 to site 15), likely because of withdrawals by Colorado Springs as a source for drinking water. Increased E. coli density was expected to coincide with elevated nutrient concentrations, because fecal material carries high levels of each. In agreement with this expectation, total nitrogen was significantly higher in the warm

3236

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Table 4 e Results of routine field sampling during cool season (September 16eJune 14) in Fountain Creek and Ruxton Creek. Ruxton Creek (site 9) data are presented between bracketing main stem sites but shaded to offset from main stem data. Observed E. coli densities (MPN/100 mL) are presented for each site in italics. Calculated upper limit (potential) E. coli contribution per fecal contamination category (Human, Ruminant, Pet, Bird, in MPN/100 mL) follow. Sources for which host-associated MST markers were measured (Human, Ruminant) are separated from other estimates and denoted in italic when the detection was above the limit of quantification. Site 2 Wednesday, February 20, 2008 Observed E. coli

3

Site 7.9

Site 9

Site 10

Site 15

5

110

28

12

84 100

350 4

790 130

310 520

1100 560,000

86 95,000

1700 2,800,000

2700 2,000,000

MST marker

Human Ruminant

38 24

No MST marker

Pets Bird

240 250,000

E. coli

12

11

58

29

11

MST marker

Human Ruminant

260 85

140 120

930 67

1700 270

160 150

No MST marker

Pets Bird

1500 1,700,000

1600 900,000

630 1,400,000

3100 5,900,000

3000 1,000,000

E. coli

11

80

380

61

2400

MST marker

Human Ruminant

200 72

310 280

8500 490

1200 290

9000 10,000

No MST marker

Pets Bird

1300 1,300,000

2100 2,000,000

8800 11,000,000

2200 6,300,000

51,000 60,000,000

Wednesday, April 23, 2008 Observed

Thursday, May 01, 2008 Observed

season compared to the cool season at three of four Fountain Creek sites (all except site 15; Table 7) but not at Ruxton Creek (site 9). When observed, increases in nitrogen were attributable to the nitrate-plus-nitrite component rather than organic nitrogen or ammonia nitrogen (Table 7). Unlike nitrogen, phosphorus tended to be enriched in warm season compared with cool season at downstream sites (sites 10 and 15) but not at upstream sites. A large portion of phosphorus enrichment was caused by orthophosphate concentration (Table 7). Although seasonal nutrient enrichment coincided with elevated E. coli density, nutrient concentrations were not correlated with E. coli concentrations in any combination of site or season (data not shown) except Fountain Creek at Manitou Springs (site 10). Increased E. coli density also was expected to coincide with detection of wastewater-associated chemicals at sites where the E. coli originated with human sewage. Various wastewater organic chemicals were detected more frequently in the urbanized sections of Fountain Creek than in the undeveloped sections (Tables 8 and 9) although some chemicals (particularly industrial and asphalt-associated chemicals such as isophorone, camphor, and naphthalene) were detected throughout the watershed. Many chemicals, including human-wasteassociated chemicals, were never detected in samples (Table 8).

4.

Discussion

A new conceptual approach to analysis of microbial source tracking data was developed and data were collected from

reference fecal material to test its applicability. The absolute distribution of markers in reference fecal material collected from the study area underscored the well-characterized risk of false-positive results when using presenceeabsence data to indicate fecal contamination sources (Shanks et al., 2010; Stoeckel and Harwood, 2007). Host-associated markers commonly were found in non-target reference materials; i.e., the lack of specificity was clear, particularly for the ruminantassociated marker BoBac. The markers were, however, more abundant in the respective associated hosts than they were in non-target hosts. The differential distribution of markers indicated that the proposed approach, using ratios of marker concentrations and E. coli densities to place an upper bound on the E. coli that could have come from a particular host, might provide useful information. The proposed approach was tested against fecal suspensions prepared in a similar way to previous studies (Griffith et al., 2003; Wang et al., 2010). The analytical results identified both strengths and limitations of the proposed approach. In the positives column, there were no false-negative results, the minority contributor was correctly discriminated from the majority contributor in the mixed sample, and the practical calculation steps were quite simple. These findings are in contrast with previously described false-negative results and inability to discriminate minority from majority contributors (Griffith et al., 2003) and the need for complex calculation steps that may be beyond commonly accessible capabilities (Wang et al., 2010). Although the ability to quantify contributions from fecal contamination sources under constrained conditions was reported by the latter researchers (Wang et al., 2010)

3237

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Table 5 e Results of routine field sampling during warm season (June 15eSeptember 15). Observed E. coli densities (MPN/ 100 mL) are presented for each site in italics. Calculated upper limit (potential) E. coli contribution per fecal contamination category (Human, Ruminant, Pet, Bird, in MPN/100 mL) follow. Sources for which host-associated MST markers were measured (Human, Ruminant) are separated from other estimates and denoted in italic when the detection was above the limit of quantification. Ruxton Creek (site 9) data are presented between bracketing main stem sites but shaded to offset from main stem data. Warm season calculated upper limit E. coli contributions in MPN/100 ml. Site 2

Site 7.9

Site 9

E. coli

140

100

280

170

8200

MST marker

Human Ruminant

e e

e e

e e

e e

ND ND

No MST marker

Pets Bird

e e

e e

e e

e e

800 890,000

E. coli

55

99

460

310

2600

MST marker

Human Ruminant

e e

e e

ND ND

ND 27

ND ND

No MST marker

Pets Bird

e e

e e

120 130,000

530 590,000

160 180,000

E. coli

140

170

5500

2000

4600

MST marker

Human Ruminant

ND ND

ND ND

ND ND

ND ND

ND ND

No MST marker

Pets Bird

190 210,000

130 142,000

150 170,000

480 540,000

250 280,000

E. coli

340

170

860

630

2000

MST marker

Human Ruminant

ND ND

ND ND

e e

ND ND

ND ND

No MST marker

Pets Bird

380 510,000

320 360,000

e e

210 230,000

760 850,000

E. coli

<100

460

2300

740

61,000

MST marker

Human Ruminant

330 ND

3600 ND

ND ND

14,000 91

290 20

No MST marker

Pets Bird

400 440,000

950 1,100,000

560 620,000

1800 2,000,000

390 430,000

E. coli

62

170

860

730

>2400

MST marker

Human Ruminant

e e

260 ND

ND ND

ND ND

810 430

No MST marker

Pets Bird

e e

320 350,000

270 300,000

320 350,000

8400 5,300,000

41

200

240

380

160

Tuesday, June 17, 2008 Observed

Monday, June 30, 2008 Observed

Tuesday, July 15, 2008 Observed

Tuesday, July 29, 2008 Observed

Tuesday, August 12, 2008 Observed

Tuesday, August 26, 2008 Observed

Tuesday, September 09, 2008 Observed E. coli

Site 10

Site 15

MST marker

Human Ruminant

e e

ND ND

e e

ND ND

e e

No MST marker

Pets Bird

e e

1000 1,100,000

e e

620 690,000

e e

e, analysis not done. ND, not detected, analyte below the limit of detection.

3238

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Table 6 e Physical and chemical characteristic averages, 95% confidence intervals about the means (CI), and numbers of observations (n) measured at study sites between May 2007 and September 2008. Data are divided into warm season (June 15eSeptember 15), roughly corresponding to summer vacation season, and cool season (September 16eJune 14), roughly corresponding to the off season. Ruxton Creek (site 9) data are presented between bracketing main stem sites but shaded to offset from main stem data. Season Fountain Creek below Fountain Creek upstream Crystal Creek (site 2) from Ruxton Creek (site 7.9) Average (CI, n)

Ruxton Creek Mouth (site 9)

Fountain Creek below Fountain Creek below Ruxton Creek (site 10) 8th Street (site 15)

Average (CI, n)

Average (CI, n)

Average (CI, n)

Average (CI, n)

Temperature ( C) Cool 5.0 (2.9e7.1, 12) Warm 12.2 (11.4e13.0, 11)

5.9 (3.7e8.2, 12) 13.9 (12.4e15.3, 10)

6.0 (4.0e8.0, 14) 13.4 (12.7e14.1, 18)

7.0 (4.4e9.6, 12) 14.3 (13.8e14.8, 16)

10.3 (7.0e13.5, 12) 19.1 (18.3e19.9, 17)

Specific conductance (mS/cm) Cool 322 (303e341, 12) Warm 321 (292e351, 11)

321 (281e361, 12) 298 (233e363, 10)

122 (97e146, 14) 176 (141e211, 19)

317 (267e367, 12) 326 (272e380, 18)

507 (401e613, 12) 829 (470e1190, 17)

pH (unitless) Cool 7.97 (7.72e8.22, 12) Warm 8.06 (7.92e8.20, 10)

7.30 (7.19e7.41, 12) 7.36 (7.21e7.52, 8)

7.66 (7.43e7.88, 14) 7.63 (7.50e7.77, 16)

7.48 (7.38e7.57, 12) 7.51 (7.41e7.62, 14)

8.06 (7.90e8.21, 12) 8.14 (7.99e8.29, 15)

Streamflow (CFS) Cool 4.78 (3.77e5.80, 12) Warm 4.72 (3.37e6.07, 9)

8.94 (6.61e11.3, 12) 10.4 (5.38e15.4, 7)

3.50 (1.83e5.17, 13) 2.21 (0.83e3.60, 14)

14.3 (8.87e19.7, 12) 12.9 (7.84e18.0, 13)

12.3 (6.05e18.6, 12) 14.4 (7.63e21.2, 14)



the samples in that study were not analyzed blind. The current research is the first report of correct semi-quantitative results from analysis of fecal suspensions that were analyzed blind. In the negatives column, incomplete specificity of the hostassociated markers resulted in common false-positive results, particularly for classes such as domestic pets that carried both human- and ruminant-associated markers (this report, Shanks et al., 2010; Stoeckel and Harwood, 2007). These falsepositive results were pronounced for sources such as pets that were incidental carriers, and not purposefully targeted with a host-associated marker. Further development of markers with better specificity and inclusion of additional host-associated markers, such as the dog-associated marker (Kildare et al., 2007; Wang et al., 2010) and bird-associated markers that under evaluation (such as Lu et al., 2008; Weidhaas et al., 2010) may at least partially alleviate this problem. In sum, results for the fecal suspension samples indicated that the proposed approach could be used reliably to exclude sources of fecal contamination, though not to directly implicate sources of fecal contamination. It remained to be determined whether these limitations to the approach would render it ineffective in environmental application, or whether practical value would be obtained despite the limitations imposed by incomplete host specificity. A thorough field study was done to provide a practical test of the approach. The sanitary survey allowed effective allocation of available resources, with a sample design that emphasized differences between a long cool season and a short warm season, and sites that bracketed a focus point in the watershed at which stream-water quality (as indicated by E. coli density) tended to breach the local water-quality standard of 126 MPN/100 mL. Furthermore, the sanitary survey uncovered unexpected sources of human wastewater to the stream that were repaired and (or) controlled before the yearlong intensive monitoring began. Despite these infrastructure repairs, data from the year of intensive monitoring (monthly during cool season and

biweekly during warm season) showed the same trends indicated by the sanitary survey. Combined evaluation of E. coli, MST marker, nutrient, and wastewater organic chemical data clearly indicated that human sources of fecal contamination typically were not the major cause of E. coli standard exceedances in Fountain and Ruxton Creeks during the summer of 2008. Though E. coli densities increased to more than 100 times the recreational water standard, host-associated markers for human and ruminant fecal contamination stayed the same or decreased during warm season (Fig. 4), nutrient concentrations remained steady (Table 7), and human- and wastewaterassociated organic chemicals rarely were detected (Tables 8 and 9). Though human sources of fecal contamination did not appear to cause chronic warm-season water-quality standard exceedances, instances of human-source contamination were apparent. On occasion, increased E. coli density coincided with increased MST markers. Further evaluation of the integrated data set was done to test the validity of field study results and to emphasize the value of the proposed upper limit calculations. In one example, Fountain Creek below 8th street (site 15) had relatively high E. coli and MST markers on May 1, 2008 (Table 4), indicating that human, ruminant, or pet sources could have caused the observed E. coli contamination level. In support of this hypothesis, ammonium was measured at 0.56 ppm (compared with a 90th percentile value of 0.08 ppm among all samples; Table 7). Ammonium concentrations in wastewater are estimated to be 12e50 ppm (as N; Tchobanoglous and Burton, 1991) so this ammonium concentration would be expected in 1:100 diluted wastewater. In addition, 3 of 4 wastewater-associated chemicals and 5 of 6 human-associated chemicals detected in this study were detected in the May 1st sample (see Tables 4 and 5). In this instance, presence of human-origin fecal contamination was supported by E. coli, MST marker, nutrient, and wastewater organic chemical analytical data although presence of fecal contamination from other sources could not be excluded.

Table 7 e Nutrient concentration averages, 95% confidence intervals about the means (CI), and numbers of observations (n) measured at study sites between May 2007 and September 2008. Data are divided into warm season (June 15eSeptember 15), roughly corresponding to summer vacation season, and cool season (September 16eJune 14), roughly corresponding to the off season. Ruxton Creek (site 9) data are presented between bracketing main stem sites but shaded to offset from main stem data. Season

Fountain Creek upstream from Ruxton Creek (site 7.9)

Ruxton Creek Mouth (site 9)

Fountain Creek downstream from Ruxton Creek (site 10)

Fountain Creek below 8th Street (site 15)

Average (CI, n)

Average (CI, n)

Average (CI, n)

Average (CI, n)

Average (CI, n)

Total nitrogen (parts per million as N) Cool 1.58 (1.47e1.69, 10) Warm 1.76 (1.69e1.83, 8)

1.05 (0.95e1.14, 10) 1.15 (1.12e1.17, 8)

0.29 (0.19e0.40, 10) 0.29 (0.22e0.37, 10)

0.95 (0.83e1.07, 10) 1.03 (1.01e1.06, 10)

1.10 (0.75e1.46, 10) 1.09 (0.95e1.23, 10)

Ammonium (parts per million as N) Cool 0.018 (0.015e0.020, 10) Warm 0.021 (0.018e0.024, 8)

0.023 (0.015e0.031, 10) 0.018 (0.015e0.021, 8)

0.031 (0.008e0.053, 10) <0.02 (no detect, 10)

0.026 (0.007e0.044, 10) 0.018 (0.016e0.021, 10)

0.075 (e0.031e0.182, 10) 0.017 (0.013e0.020, 10)

Nitrate-plus-nitrite (parts per million as N) Cool 1.41 (1.30e1.51, 10) Warm 1.64 (1.56e1.72, 8)

0.905 (0.791e1.02, 10) 0.989 (0.941e1.04, 8)

0.210 (0.170e0.250, 10) 0.199 (0.183e0.215, 10)

0.784 (0.683e0.885, 10) 0.914 (0.881e0.947, 10)

0.797 (0.737e0.857, 10) 0.796 (0.671e0.921, 10)

Total phosphorus (parts per million as P) Cool 0.037 (0.031e0.043, 10) Warm 0.047 (0.036e0.058, 8)

0.012 (0.010e0.015, 10) 0.038 (0.021e0.056, 8)

0.017 (0.004e0.029, 10) 0.029 (0.004e0.053, 10)

0.012 (0.010e0.015, 10) 0.039 (0.014e0.065, 10)

0.016 (0.002e0.030, 10) 0.033 (0.023e0.043, 10)

Orthophosphate (parts per million as P) Cool 0.018 (0.017e0.020, 9) Warm 0.024 (0.020e0.027, 8)

0.005 (0.004e0.006, 9) 0.009 (0.006e0.012, 8)

0.007 (0.002e0.013, 9) 0.009 (0.008e0.010, 10)

0.005 (0.004e0.006, 9) 0.012 (0.007e0.017, 10)

0.005 (0.004e0.006, 9) 0.013 (0.011e0.015, 10)

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Fountain Creek below Crystal Creek (site 2)

3239

3240

Table 8 e Wastewater organic chemicals detected fewer than three times in the study area. Chemicals are divided according to type and/or origin. Ruxton Creek (site 9) data are presented between bracketing main stem sites but shaded to offset from main stem data. Site 2 Wastewater chemicals

None

Site 7.9 None

Site 9 (Ruxton Creek) Triethyl citrate Hexahydrohexamethyl cyclopentabenzopyran

Site 10 None

Site 15 Triethyl citrate

Never detected Benzophenone Triclosan Acetyl hexamethyl tetrahydro naphthalene

None

None

DEET Triphenyl phosphate

DEET Menthol

DEET Cotinine Triphenyl phosphate Skatolea

3-tert-Butyl-4-hydroxyanisole D-Limonene 5-Methyl-1H-benzotriazole

Sterols and stanols

None

None

None

None

None

Beta-sitosterol Cholesterol Beta-stigmastanol 3-Beta-coprostanol

Pesticides/Herbicides

Metolachlor (M)

None

None

Bromacil Metolachlor

Prometon Tributyl phosphate

Chlorpyrifos Diazinon 1,4 Dichlorobenzene Metalaxyl

Synthesis, solvents, plasticizers

Asphalt

None

None

None

1-Methylnaphthalene 2-Methylnaphthalene

a This chemical classified in more than one category.

Phenol

None

Phenol

1-Methylnaphthalene 2-Methylnaphthalene

Tris(2-butoxyethyl) phosphate Phenol Acetophenone Isopropylbenzene

4-tert-Octylphenol diethoxylate

1-Methylnaphthalene 2-Methylnaphthalene Skatolea

Anthracene Benzo[a]pyrene 2,6-Dimethylnaphthalene

4-tert-Octtylphenol monoethoxylate 4-Nonylphenol diethoxylate 4-Nonylphenol 4-n-Octylphenol 4-tert-Octylphenol Indole Tribromomethane Isoquinoline Isoborneol Bisphenol A 4-Cumylphenol

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Human associated

3241

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Table 9 e Wastewater organic chemicals quantified and/or detected three or more times. Chemicals are divided according to type and/or origin. Values represent the number of samples in which the chemical was quantified and/or detected. Ruxton Creek (site 9) data and overall sums are shaded to offset from main stem data. Site 2

Site 7.9

Site 9

Site 10

Site 15

Overall

13

12

13

13

13

64

0 0

0 0

1 0

0 0

3 2

4 2

0 0

1 0

1 0

0 0

3 2

5 2

1 0

1 0

2 1

1 0

4 1

9 2

0 0

3 1

5 2

5 1

9 4

22 8

1 0

2 1

3 0

4 0

4 1

14 2

1 0

0 0

0 0

0 0

11 2

12 2

0 0

0 0

1 0

3 0

4 1

8 1

0 0

0 0

0 0

0 0

3 2

3 2

0 0

1 0

0 0

0 0

3 2

4 2

3 0

4 1

4 1

3 1

3 1

17 4

0 0

0 0

1 0

1 0

8 2

10 2

1 0

1 0

0 0

2 0

1 0

5 0

0 0

0 0

1 0

0 0

6 1

7 1

0 0

1 0

2 0

2 0

4 0

9 0

Total Wastewater chemicals

Human associated

Synthesis, solvents, plasticizers

Asphalt

Tris(2-cholorethyl) phosphate Detected Quantified Tris(dicholoroisopropyl) phosphate Detected Quantified Methyl salicylate Detected Quantified Caffeine Detected Quantified Isophorone Detected Quantified Tetrachloroethene Detected Quantified Phenanthrene Detected Quantified Carbazole Detected Quantified 9,10-Anthraquinone Detected Quantified Camphor Detected Quantified Fluoranthene Detected Quantified Naphthalene Detected Quantified Pyrene Detected Quantified p-Cresol Detected Quantified

On the same date, Ruxton Creek (site 9) had high E. coli and human-associated MST markers (Table 4), indicating that human or pet (but less likely ruminant) sources may have caused the observed fecal contamination. Nutrient concentrations measured in this sample were high (ammonium at 0.13 ppm, total nitrogen and total phosphorus higher than any other date during the study at this site). In addition, 3 of 4 wastewater-associated chemicals and 5 of 6 human-associated chemicals detected in this study were detected in the sample. The upper threshold limit approach was again supported by these additional lines of evidence in the Ruxton Creek sample. In other cases, the inverse happened e host-associated MST markers did not increase when E. coli densities

increased, indicating a source that was not detectable by the MST markers applied. On these dates, nutrients were present at baseline levels and wastewater- and humanassociated chemicals were rarely detected. As an example, Fountain Creek below 8th street (site 15) had relatively high E. coli densities in four samples between June 17 and July 29, 2008 (Table 5), without an associated increase in ruminant or human-associated MST markers. In all four cases, nutrient concentrations were similar to those in other samples. Wastewater compounds were analyzed for 3 of the 4 samples e two human-associated chemicals (caffeine and skatole) and two wastewater chemicals (chloroethyl phosphate and chloroisopropyl phosphate) were detected, albeit

3242

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

at relatively low concentrations (below the median of all observed and estimated concentrations). Similarly, Ruxton Creek had unusually high E. coli density on July 15 (Table 5) without increased host-associated MST marker, nutrient, or human- or wastewater-associated organic chemical concentrations. In this report, a novel approach that numerically relates E. coli density and MST marker concentration in reference feces was used to estimate the upper in-stream threshold of E. coli that one could reasonably expect from a source. In-depth evaluation of the integrated data set was done as a final evaluation of the upper threshold limit approach to exclude potential sources of FIB to stream water. Overall, though human and ruminant sources could explain observed levels of FIB primarily during the cool season an unexpected source was indicated during the warm season, when E. coli densities are chronically elevated. Similar observations of seasonality to fecal contamination sources have been made in other studies of small and medium-sized streams by use of other semi-quantitative MST methods (Vogel et al., 2007; Wijesinghe et al., 2009). By a process of elimination, the unexpected source of warm-season fecal contamination to the Fountain Creek watershed could be detected. Bird (two samples of composted fecal material, Table 2) was not excluded as the source of E. coli to the stream, based on absence of host-associated marker and relatively low observed concentrations of general marker relative to E. coli. Prior studies also have shown that bird feces has a low ratio of MST marker per E. coli (Haack et al., 2003, Stoeckel unpublished data) though no prior study indicated that birds may be a primary source of fecal contamination against land-use patterns that initially pointed to other sources. In this study, no substantial roosting by migratory birds was observed in the watershed and the only birds observed in substantial density were pigeons, associated with bridges and overpasses at Manitou Springs. Limited available information (Al-Harbi, 2003; Haack et al., 2003) indicates that the density of E. coli in pigeon feces in the summer time is between 106 and 1010 MPN/g wet weight. In this study, E. coli in pigeon feces was measured to be 3.4  108 MPN/g dry wt so a representative range was taken to be 108 to 109 MPN/g dry wt. During the warm season, typical streamflow in Manitou Springs was 13 CFS. If a pigeon defecates at a rate of 25e50 g wet wt fecal material per day, at 30e40% dry weight (typical of bird fecal material; Table 2) from a roost directly over water (in other words, 100% transfer efficiency) then the combined defecation of between 16 and 420 pigeons, distributed over 24 hours, would generate a sustained in-stream E. coli density increase of 1000 MPN/100 mL. It is, thereby, feasible that pigeons contribute significantly to warm season increases in FIB density in the study area.

5.

Conclusions

1. Host-associated markers were not specific to host a. Concentrations of human-associated marker were highest in human-source waste, but frequently were present in ruminant and pet feces as well as other sources.

2.

3.

4.

5.

6.

b. Concentration of ruminant-associated marker was highest in ruminant-source waste, but consistently was present in pets and human sources. The ruminant-associated marker was present at low level in duck and bear feces, but not detected in horse feces. c. The general marker (AllBac) was differentially distributed among hosts, with the lowest concentrations in a bird feces sample. Use of calculated E. coli upper threshold limits, the maximum concentration of E. coli expected from a source based on the pattern of MST marker concentrations, was validated through the use of blind-analyzed positivecontrol suspensions of reference feces and supported by multiple lines of evidence collected in this study. In the study area, E. coli densities were lower during cool season (September 16eJune 14) than warm season (June 15eSeptember 15). Only minor differences were observed in physical and chemical characteristics of the stream, except for a seasonal difference in temperature. Levels of E. coli detected during cool season, which rarely exceeded the criterion of 126 MPN/100 mL, consistently could be explained by upper threshold limit E. coli contributions based on measured concentrations of human- and ruminant-associated MST markers. Because of incomplete specificity, pets and birds could not be excluded as alternate contamination sources during this season. Levels of E. coli detected during warm season, which routinely exceeded the criterion of 126 MPN/100 mL at all but the uppermost site on Fountain Creek, did not fall within the human or ruminant-origin upper threshold limits. Because of observed cross-carriage of human- and ruminant-associated markers, pets also were excluded as sources of observed FIB in most samples during warm season. On-site observations and calculations using literaturederived data indicated that birds, particularly pigeons, could not be excluded as sources of chronic FIB contamination. This source would not have been identified except through application of the novel upper threshold of E. coli contamination calculation, and would have been less credible without the supporting lines of evidence collected.

Acknowledgements This work was supported, in part, by the Colorado Department of Public Health and the Environment, Pikes Peak Area Council of Governments, Colorado Springs Utilities, Colorado Springs Engineering, and the USGS Cooperative Water Program. Interpretations would not have been possible without valuable information from the aforementioned collaborators, the cities of Manitou Springs, Green Mountain Falls, and Woodland Park, and El Paso County. Special thanks to Dr. Mark Wilson and Colorado College for help and support during the sanitary survey. Study design, sample collection, and final data interpretation were done solely by the authors. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

references

Al-Harbi, A.H., 2003. Faecal coliforms in pond water, sediments and hybrid tilapia Oreochromis niloticus  Oreochromis aureus in Saudi Arabia. Aquacult. Res. 34, 517e524. Boehm, A.B., Fuhrman, J.A., Mrse, R.D., Grant, S.B., 2003. Tiered approach for identification of a human fecal pollution source at a recreational beach: case study at Avalon Bay, Catalina Island, California. Environ. Sci. Technol. 37 (4), 673e680. Booth, A.M., Hagedorn, C., Graves, A.K., Hagedorn, S.C., Mentz, K. H., 2003. Sources of fecal pollution in Virginia’s Blackwater River. J. Environ. Eng. 129, 547e552. Bossong, C.R., 2001. Summary of water-quality data, October 1987 through September 1997, for Fountain and Monument Creeks, El Paso and Pueblo Counties, Colorado. USGS Water-Resources Investigations Report WRIR00-4263. Bower, P.A., Scopel, C.O., Jensen, E.T., Depas, M.M., McLellan, S.L., 2005. Detection of genetic markers of fecal indicator bacteria in Lake Michigan and determination of their relationship to Escherichia coli densities using standard microbiological methods. Appl. Environ. Microbiol. 71 (12), 8305e8313. CDPHE, 2005. Colorado Department of Public Health and the Environment, Water Control Commission, Regulation Number 93 Section 303(d) list water-quality-limited segments requiring TMDLs (5 CCR 1002e93). Available at: http://www.cdphe.state. co.us/regulations/wqccregs/100293wqlimitedsegtmdls.pdf (accessed 27.05.09). CDPHE, 2008. Colorado Department of Public Health and the Environment, Water Control Commission, Regulation Number 31 The Basic Standards and Methodologies for Surface Water (5 CCR 1002-31). Available at: http://pubweb.epa.gov/ waterscience/standards/wqslibrary/co/reg31-20051231.pdf (accessed 27.05.09). Gentry, R.W., Layton, A.C., McKay, L.D., McCarthy, J.F., Williams, D.E., Koirala, S.R., Sayler, G.S., 2007. Efficacy of Bacteroides measurements for reducing the statistical uncertainty associated with hydrologic flow and fecal loads in a mixed use watershed. J. Environ. Qual. 36 (5), 1324e1330. Griffith, J.F., Weisberg, S.B., McGee, C.D., 2003. Evaluation of microbial source tracking methods using mixed fecal sources in aqueous test samples. J. Water Health 1 (4), 141e151. Haack, S.K., Duris, J.W., Fogarty, L.R., Kolpin, D.W., Focazio, M.J., Furlong, E.T., Meyer, M.T., 2009. Comparing wastewater chemicals, indicator bacteria concentrations, and bacterial pathogen genes as fecal pollution indicators. J. Environ. Qual. 38 (1), 248e258. Haack, S.H., Fogarty, L.R., Wright, C., 2003. Escherichia coli and enterococci at beaches in the Grand Traverse Bay, Lake Michigan: sources, characteristics, and environmental pathways. Environ. Sci. Technol. 37, 3275e3282. Horowitz, A.J., Demas, C.R., Fitzgerald, K.K., Miller, T.L., Rickert, D. A., 1994. U.S. Geological Survey protocol for the collection and processing of surface-water samples for the subsequent determination of inorganic constituents in filtered water. USGS Open File Report 93-656, 28 p.. Kildare, B.J., Leutenegger, C.M., McSwain, B.S., Bambic, D.G., Rajal, V. B., Wuertz, S., 2007. 16S rRNA-based assays for quantitative detection of universal, human-, cow-, and dog-specific fecal Bacteroidales: a Bayesian approach. Water Res. 41, 3701e3715. Layton, A., McKay, L., Williams, D., Garrett, V., Gentry, R., Sayler, G., 2006. Development of Bacteroides 16S rRNA gene TaqMan-based real-time PCR assays for estimation of total, human, and bovine fecal pollution in water. Appl. Environ. Microbiol. 72 (6), 4214e4224. Lu, J., Santo Domingo, J.W., Lamendella, R., Edge, T., Hill, S., 2008. Phylogenetic diversity and molecular detection of bacteria in gull feces. Appl. Environ. Microbiol. 74 (13), 3969e3976.

3243

Noble, R.T., Blackwood, A.D., Gregory, J.B., Griffith, J.F., Schiff, K., Fuhrman, J.A., Hernandez, X., Liang, X., Bera, A.A., 2006. Multitiered approach using quantitative PCR to track sources of fecal pollution affecting Santa Monica Bay, California. Appl. Environ. Microbiol. 72 (2), 1604e1612. PPACG (Pikes Peak Area Council of Governments), 2003. Fountain Creek Watershed Plan. http://www.fountain-crk.org/Plan% 20Documents/fc_watershedplan.html. Rantz, S.E., et al., 1982. Measurement and computation of streamflow. Measurement of Stage and Discharge, vol. 1. U.S. Geological Survey Water-Supply Paper 2175, 297 p. Reischer, G.H., Haider, J.M., Sommer, R., Stadler, H., Keiblinger, K. M., Hornek, R., Serobin, W., Mach, R.L., Farnleitner, A.H., 2008. Quantitative microbial faecal source tracking with sampling guided by hydrological catchment dynamics. Environ. Microbiol. 10 (10), 2598e2608. Reischer, G.H., Kasper, D.C., Farnleitner, A.H., Mach, R.L., 2007. A quantitative real-time PCR assay for the highly sensitive and specific detection of human faecal influence in spring water from a large alpine catchment area. Lett. Appl. Microbiol. 44, 351e356. Samadpour, M., Kitts, C., Moline, M.A., Schaffner, A., McNeill, K., Duffield, S., O’Connell, L., 2002. Identifying the sources of Escherichia coli contamination to the shellfish growing areas of the Morrow Bay estuary. Full report also available through the Institute for Environmental Health at:. In: Magoon, O.T., Converse, H., Baird, B., Jines, B., Miller-Henson, M. (Eds.), Proceedings of California and the World Ocean’02, October 27e30, 2002, Santa Barbara, California. American Society of Civil Engineers, Reston, Virginia http://www. microbialsourcetracking.com/PDF/Morro%20Bay.pdf. Savichtcheva, O., Okayama, N., Okabe, S., 2007. Relationships between Bacteroides 16S rRNA genetic markers and presence of bacterial enteric pathogens and conventional fecal indicators. Water Res. 41, 3615e3628. Seurinck, S., Defoirdt, T., Verstraete, W., Siciliano, S.D., 2005. Detection and quantification of the human-specific HF183 Bacteroides 16S rRNA genetic marker with real-time PCR for assessment of human faecal pollution in freshwater. Environ. Microbiol. 7 (2), 249e259. Shanks, O.C., Aticovic, E., Blackwood, A.D., Lu, J., Noble, R., Santo Domingo, J., Seifring, S., Sivaganesan, M., Haugland, R.A., 2008. Quantitative PCR for detection and enumeration of genetic markers of bovine fecal pollution. Appl. Environ. Microbiol. 74 (3), 745e752. Shanks, O.C., Kelty, C.A., Sivaganesan, M., Varma, M., Haugland, R.A., 2009. Quantitative PCR for genetic markers of human fecal pollution. Appl. Environ. Microbiol. 75 (7), 5507e5513. Shanks, O.C., Nietch, C., Simonich, M., Younger, M., Reynolds, D., Field, K.G., 2006. Basin-wide analysis of the dynamics of fecal contamination and fecal source identification in Tillamook Bay, Oregon. Appl. Environ. Microbiol. 72 (8), 5537e5546. Shanks, O.C., White, K., Kelty, C.A., Hayes, S., Sivaganesan, M., Jenkins, M., Varma, M., Haugland, R.A., 2010. Performance assessment PCR-based assays targeting Bacteroidales genetic markers of bovine fecal pollution. Appl. Environ. Microbiol. 76 (5), 1359e1366. Stoeckel, D.M., Harwood, V.J., 2007. Performance, design, and analysis in microbial source tracking studies. Appl. Environ. Microbiol. 73 (8), 2405e2415. Stoeckel, D.M., Stelzer, E.A., Dick, L.K., 2009. Evaluation of two spike-and-recovery controls for assessment of extraction efficiency in microbial source tracking studies. Water Res. 43 (19), 4820e4827. Stogner, R.A., 2000. Trends in precipitation and streamflow in the Fountain Creek watershed, southeastern Colorado, 1977e99. USGS Fact Sheet 136-00, pp. 1e5.

3244

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 2 5 e3 2 4 4

Sylvester, M.A., Kister, L.R., Garrett, W.B. (Eds.), 1990. Guidelines for the Collection, Treatment, and Analysis of Water Samples, U.S. Geological Survey Western Region Field Manual: Unpublished Report on File in the Pueblo, Colo.. Water Resources Discipline Office of the U.S. Geological Survey, 144 p. Tchobanoglous, G., Burton, F.L., 1991. Wastewater Engineering: Treatment, Disposal, and Reuse, third ed. Metcalf&Eddy, Inc and McGraw Hill, Inc., New York. USEPA (U.S. Environmental Protection Agency), 2000. Improved enumeration methods for the recreational water quality indicators: Enterococci and Escherichia coli. EPA 821-R-004, pp. 1e49. USEPA (U.S. Environmental Protection Agency), 2001. Protocol for developing pathogen TMDLs. EPA 841-R-00-002, pp. 1e174. USEPA (U.S. Environmental Protection Agency), 2005. Microbial source tracking guide document. EPA 800-R-05e064, pp. 1e150. USEPA (U.S. Environmental Protection Agency), 2009. Water Discharge Permits Information available at: http://oaspub.epa. gov/enviro/ef_home2.water search parameter El Paso County, CO. USEPA (U.S. Environmental Protection Agency), 2010. National Summary of Impaired Waters and TMDL Information. Available at: http://aspub.epa.gov/waters10/attains_nation_ cy.control?p_report_type=T#causes_303d (accessed 15.04.11). USGS (U.S. Geological Survey), 1977. National Handbook of Recommended Methods for Water-data Acquisition. U.S. Geological Survey Monograph, Office of Water Data Coordination, Reston, Va. various pagination. USGS (U.S. Geological Survey), 2006. Collection of Water Samples (ver. 2.0). U.S. Geological Survey Techniques of Water-Resources

Investigations. Available at: http://pubs.water.usgs.gov/ twri9A4/ Book 9, Chapter A4, September (accessed 19.01.10). Vogel, J.R., Stoeckel, D.M., Lamendella, R., Zeldt, R.B., Santo Domingo, J.W., Walker, S.R., Oerther, D.B., 2007. Fecal-source identification using multiple source-tracking tools in a selected subcatchment. J. Environ. Qual. 36, 718e729. Wang, D., Silkie, S.S., Nelson, K.L., Wuertz, S., 2010. Estimating true human and animal host source contribution in quantitative microbial source tracking using the Monte Carlo method. Water Res. 44, 4760e4775. Weidhaas, J.L., Macbeth, T.W., Olsen, R.L., Sadowski, M.J., Norat, D., Harwood, V.J., 2010. Identification of a Brevibacterium marker gene specific to poultry litter and development of a quantitative PCR assay. J. Appl. Microbiol. 109 (1), 334e347. Wijesinghe, R.U., Feng, Y., Wood, C.W., Stoeckel, D.M., Shaw, J.N., 2009. Population dynamics and genetic variability of Escherichia coli in the Catoma Creek watershed. J. Water Health 7, 484e496. Wilde, F.D., Radtke, D.B., Gibbs, J., Iwatsubo, R.T. (Eds.), 2004. Processing of Water Samples (Version 2.1). U.S. Geological Survey Techniques of Water-Resources Investigations Available at: http://pubs.water.usgs.gov/twri9A5/ Book 9, Chapter A5, April (accessed 19.01.10). World Health Organization, 2003. Faecal pollution and water quality. In: Guidelines for Safe Recreational Water Environments. Coastal and Fresh Waters, vol. 1. WHO, Geneva, Switzerland Available online at: www.whqlibdoc. who.int/publications/2003/9241545801.pdf (Chapter 4).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Roof selection for rainwater harvesting: Quantity and quality assessments in Spain Ramon Farreny a,b,*, Tito Morales-Pinzo´n a,c, Albert Guisasola d, Carlota Taya` d, Joan Rieradevall a,d, Xavier Gabarrell a,d a

SosteniPrA (ICTA-IRTA-Ine`dit), Institute of Environmental Science and Technology (ICTA), Universitat Auto`noma de Barcelona (UAB), 08193 Bellaterra, Barcelona, Spain b Ine`dit Innovacio´ SL, UAB Research Park, Carretera de Cabrils, km 2, IRTA, 08348 Cabrils, Barcelona, Spain c Facultad de Ciencias Ambientales, Universidad Tecnolo´gica de Pereira, AA 97, Pereira, Colombia d Department of Chemical Engineering, XRB, Universitat Auto`noma de Barcelona (UAB), 08193 Bellaterra, Barcelona, Spain

article info

abstract

Article history:

Roofs are the first candidates for rainwater harvesting in urban areas. This research inte-

Received 30 September 2010

grates quantitative and qualitative data of rooftop stormwater runoff in an urban Medi-

Received in revised form

terranean-weather environment. The objective of this paper is to provide criteria for the

24 February 2011

roof selection in order to maximise the availability and quality of rainwater. Four roofs

Accepted 21 March 2011

have been selected and monitored over a period of 2 years (2008e2010): three sloping

Available online 29 March 2011

roofs e clay tiles, metal sheet and polycarbonate plastic e and one flat gravel roof. The authors offer a model for the estimation of the runoff volume and the initial abstraction of

Keywords:

each roof, and assess the physicochemical contamination of roof runoff. Great differences

City ecodesign

in the runoff coefficient (RC) are observed, depending mostly on the slope and the

Runoff coefficient

roughness of the roof. Thus, sloping smooth roofs (RC > 0.90) may harvest up to about 50%

Sustainable urbanism

more rainwater than flat rough roofs (RC ¼ 0.62). Physicochemical runoff quality appears to

Stormwater runoff

be generally better than the average quality found in the literature review (conductivity:

Urban environment

85.0  10.0 mS/cm,

Water management

11.6  1.7 mg/L, pH: 7.59  0.07 upH). However, statistically significant differences are found

total

suspended

solids:

5.98  0.95 mg/L,

total

organic

carbon:

between sloping and flat rough roofs for some parameters (conductivity, total organic carbon, total carbonates system and ammonium), with the former presenting better quality in all parameters (except for ammonium). The results have an important significance for local governments and urban planners in the (re)design of buildings and cities from the perspective of sustainable rainwater management. The inclusion of criteria related to the roof’s slope and roughness in city planning may be useful to promote rainwater as an alternative water supply while preventing flooding and water scarcity. ª 2011 Elsevier Ltd. All rights reserved.

* Corresponding author. SosteniPrA (ICTA-IRTA-Ine`dit), Institute of Environmental Science and Technology (ICTA), Universitat Auto`noma de Barcelona (UAB), 08193 Bellaterra, Barcelona, Spain. Tel./fax: þ34 937532915. E-mail address: [email protected] (R. Farreny). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.036

3246

1.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

Table 1 e Review of runoff coefficient (RC) estimates.

Introduction

Roof Rainwater harvesting (RWH) in urban areas is a strategy that brings many benefits and may serve to cope with current water shortages, urban stream degradation and flooding (Fletcher et al., 2008; van Roon, 2007; Zhu et al., 2004). In this context, the assessment of the quantitative potential of RWH and the quality of stormwater runoff from several types of roofs is essential in order to set up criteria for the (re)design of cities from the perspective of sustainable rainwater management. Both aspects (quantity and quality) are necessary in order to select the most adequate roof for RWH. Since roofs represent approximately half of the total sealed surface in cities they contribute to the most important urban stormwater runoff flow. As a consequence, they offer a significant possibility for RWH (Villarreal and Dixon, 2005), which makes it relevant to have criteria for roof selection at one’s disposal. The RWH potential (in L/year) of a roof can be estimated based on the local precipitations (P, in mm/year), the catchment area (A, in m2) and the runoff coefficient (RC, nondimensional), as shown in Eq. (1): RWH potential ¼ P$A$RC

(1)

Eq. (1) draws inspiration from the rational method, which has traditionally been used in order to estimate the peak runoff rate of any watershed (McCuen, 2004; Viessman and Lewis, 2003). The RC is a dimensionless value that estimates the portion of rainfall that becomes runoff, taking into account losses due to spillage, leakage, catchment surface wetting and evaporation (Singh, 1992). Thus, the RC is useful for predicting the potential water running off a surface, which can be conveyed to a rainwater storage system. Since water shortage is recognised as an emerging problem which is becoming the number one problem in the world today (Sazakli et al., 2007), and many cities are facing water restrictions due to an increasing pressure on water resources (Fletcher et al., 2008), it is essential to consider the RC in the selection of roofs in order to maximise their RWH potential. The value of the RC has usually been selected from generic lists based on the degree of imperviousness and infiltration capacity of the drainage surface. Estimates so far consider that roof RCs are within the range of 0.7e0.95 for relatively frequent storms (see Table 1 for details). This broad range is the result of the interaction of many factors, both climatic (size and intensity of the rain event, antecedent moisture, prevailing winds) and architectural (slope, roof material, surface depressions, leaks/ infiltration, roughness). For this reason, it seems urgent to solve the lack of specific RC for different roof types under diverse environmental climatic conditions in the context of RWH. On the other hand, an increased interest in the monitoring of roof runoff quality has been observed recently (Skaryska et al., 2007). Roofs are the first candidates for RWH systems because their runoff is often regarded to be unpolluted (Fo¨rster, 1999) or, at least, it presents relatively good quality standards compared to the rainwater from surface catchment areas (Go¨bel et al., 2007). Despite this, there is still some disagreement about the quality of roof runoff water: the assessment or rooftop runoff quality ranges from good or acceptable (for example, Adeniyi and Olabanji, 2005; Melidis et al., 2007; Uba

RC

Roofs (in general)

Reference

0.7e0.9 Pacey and Cullis (1989) 0.75e0.95 ASCE (1969), McCuen (2004), Singh (1992), TxDOT (2009), Viessman and Lewis (2003) 0.85 McCuen (2004), Rahman et al. (2010) 0.8e0.9 Fewkes (2000) 0.8 Ghisi et al. (2009) 0.8e0.95 Lancaster (2006)

Sloping roofs Concrete/ 0.9 asphalt Metal 0.95 0.81e0.84 Aluminium 0.7

Lancaster (2006) Lancaster (2006) Liaw and Tsai (2004)) Ward et al. (2010)

Flat roofs Bituminous 0.7 Ward et al. (2010) Gravel 0.8e0.85 Lancaster (2006) Level 0.81 Liaw and Tsai (2004)) cement

and Aghogho, 2000) to severely polluted (for example, Chang et al., 2004; Gromaire et al., 2001; Simmons et al., 2001). Rooftop runoff quality is dependent on both the roof type and the environmental conditions (not only the local climate but also the atmosphere pollution). Most research on the quality of rainwater roof runoff has been carried out in East Asia (for example: Appan, 2000; Kim et al., 2005a; Kim et al., 2005b), in Central, Eastern and Northern Europe (for example: Albrechtsen, 2002; Fo¨rster, 1996, 1999; Gromaire et al., 2001; Moilleron et al., 2002; Polkowska et al., 2002; Ward et al., 2010; Zobrist et al., 2000), in the United States (for example: Chang et al., 2004; Van Metre and Mahler, 2003) and in Oceania (for example: Evans et al., 2006; Kus et al., 2010; Magyar et al., 2007; Simmons et al., 2001). However, there is scarce data from Southern Europe, in particular from Spain. This research integrates quantitative and qualitative data of rooftop stormwater runoff in an urban Mediterraneanweather environment in order to select the best roof type for RWH. The eventual purpose of this is to maximise the rainwater harvesting potential as a measure of adaptation to water scarcity and of climate change effects mitigation. The main objective of this paper is to provide criteria for roof selection in order to maximise the availability and quality of rainwater harvesting supplies. The specific objectives are to (1) develop a model for the estimation of the runoff and the initial abstraction of each roof; (2) estimate the global RC for the different roofs in Mediterranean-weather conditions; (3) estimate the physicochemical contamination of the roofs; (4) determine the degree of association between water quality parameters and storm characteristics and between water quality parameters themselves; and, (5) assess the differences in water quality between the different roofs.

2.

Materials and methods

2.1.

Study area

Four different roofs have been selected in the UAB University Campus, in Cerdanyola del Valle`s (metropolitan region of

3247

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

Barcelona, NE Spain). The climate in the area can be characterised as semi-wet Mediterranean and the average annual temperature and rainfall are 15.5 oC and 568 mm, respectively. The selection of roofs, the main characteristics of which are shown in Table 2, includes: clay tiles (CT), metal sheet (M), polycarbonate plastic (P) and flat gravel (FG). CT roof is a wellknown trademark of the constructed Mediterranean landscape, whereas M and P roofs are increasing in presence in the region. Flat roofs are the most common type of roofs in the driest regions within traditional Mediterranean architecture, being FG roofs increasingly popular in large buildings. This set of roofs presents two extreme positions regarding slope and roughness: the M and P roofs present high slope (30 ) and smooth surfaces, whereas the FG roof is flat and presents a rough surface. The CT roof, presents characteristics similar to the M and P roofs but with a slightly rougher surface. Our research hypothesis is that these two parameters (slope and roughness) are fundamental for the assessment of the quantity and quality of roof runoff. All roofs are located between 0.5 and 1 km from a motorway with dense traffic, and at less than 4 km from some industrial states (light industry), both of which are located to the E, NE or SE of the campus. Predominant winds in this area come from westerly directions.

2.2.

Experimental design

A one-way design with 4 levels e where the factor was the type of roof and the levels were CT, M, P and FG e was applied. The assessed variables were the runoff and the physicalechemical parameters; and rainfall height, predominant wind orientation and antecedent dry weather period (ADWP) were included as covariates. A rainwater conveyance and storage system was installed in each roof. The experimental design consisted of connecting the building’s gutters and downpipes to one or more polyethylene rainwater tanks with a capacity of 1 m3 (Fig. 1), without first flush diversion system. No special maintenance of the roofs was carried out.

Data for the calculation of RC and/or water samples for the quality analysis were collected after several rainfall events during the experimental campaign between June 2008 and September 2010. A rain event was defined as a rainfall of a total height of at least 1 mm and separated one from another by an ADWP of at least 2 h. The amounts of rainfall were monitored using a rainwater gauge on each roof.

2.3.

Quantity assessment

2.3.1.

Data collection

After the rain events, the rainfall height and the amount of water collected were registered and the tank was emptied (rain events that did not generate runoff were not included). Several precipitation events were selected, the rainfall and ADWP range of which is shown in Table 3. The prevailing wind direction during the rain event was also recorded. Rainfall events that exceeded rainwater tank capacity were excluded.

2.3.2.

Data analysis. Determination of RC

In order to develop a model to estimate the runoff of each roof, a statistical analysis has been conducted with the aid of PASW Statistics 17, from the Statistical Package for the Social Sciences (SPSS) software. This analysis included a correlation analysis followed by lineal regression model, considering the amount of runoff depending on several independent variables (rainfall, ADWP and wind direction). In all cases the assumptions were verified. The technique of cross-validation (Snee, 1977) is used to assess how the results of the regression models will generalise to an independent dataset. Therefore, it is used to estimate how accurately the predictive model will perform in practice. The sample of data is divided into two complementary subsets (half and half at random division): the calibration and the validation sets. If the reduction in the cross-validation gives numbers smaller than 0.1, it is assumed that the model will accurately predict the runoff. Then, the regression model is used in order to estimate the initial abstraction, defined as the amount of rainfall that occurs prior to the start of direct runoff (McCuen, 2004).

Table 2 e Characteristics of the roof catchments. Roof

Roof type (slope)

Roughness Orientation

Rather smooth

None prevailing

Roof footprint (m2) 120.0

Clay tiles

Hip sloping roof (30 )

Metal sheet

Smooth Single pitch sloping roof (30 )

50 NE

40.6

Polycarbonate plastic

Smooth Single pitch sloping roof (30 )

230 SW

40.6

None prevailing

56.6

Flat gravel (particle diameter: w5 mm, Flat roof (1 ) gravel depth 15e20 mm)

Rough

Environment

UTM

Surrounded by forest. A few trees 424482.6 E overhanging the roof 4594896.6 N Urban environment (no trees 425349.6 E nearby) 4594851.4 N Urban environment (no trees 425347.1 E nearby) 4594848.6 N Urban environment (some trees 425476.6 E nearby) 4594733.4 N

3248

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

After obtaining the composite sample, the tank was emptied. Then, samples were immediately prepared for analyses and taken to the laboratory. Electrical conductivity (EC) and pH were measured using a Crison probe (mod. 401/L K1 and 5202, respectively). Phosphate (PO43), sulphate (SO42), chloride (Cl), nitrate (NO3) and nitrite (NO2) in filtered samples were measured with ionic chromatography (DIONEX ICS-2000 Integrated Reagent-Free IC System with an auto-sampler AS40). Mixed liquor total suspended solids (TSS) were analysed according to standard procedures (APHA, 1995). Total ammonium nitrogen (TAN) was analysed using Lange LCK302, LCK303 and LCK304 ammonium kits. Finally, Total inorganic carbon (TIC) and total organic carbon (TOC) were measured using a 1020A O-IAnalytical TOC analyser. Bicarbonates (HCO3), carbonates (CO32) and carbon acid (H2CO3) were calculated based on TIC and pH results. Heavy metals were not measured due to a lack of laboratory equipment and economic constraints. Base cations (Naþ, Kþ, Mg2þ, Ca2þ) were not measured since previous quality assessments indicated that their concentration was in the low range and no precipitates were observed in water samples.

Fig. 1 e Diagram of the experimental design.

The global RC for the different roofs is estimated by means of the obtained runoff model, taking into account the local rainfall profile. Data from the weather station of Cerdanyola del Valle`s (provided by the Meteorological Service of Catalonia, unpublished), where the roofs are located, has been used for the period 1999e2009. The calculation consists of, first, estimating the runoff of each rain event, and second, dividing the total runoff per year by the annual rainfall, according to Eq. (2): RC ¼ R=P

(2)

where R is the total height of runoff (L) and P is the total height of precipitations (L) in a yearly basis for each roof.

2.4.

Quality assessment

2.4.1.

Sample collection and physicalechemical analysis

A composite sample (V ¼ 0.6 L) of the content of the rainwater tank was taken after each monitored rain event (n ¼ 55). Several precipitation events were selected, with a rainfall which varied depending on the roof (Table 3) and an ADWP between 0.1 and 37.9 days for all roofs. The concentration of the composite sample represents the Event Mean Concentration (EMC) of that event. The EMC can be defined as the total mass load of a pollutant from a site during a storm divided by the total runoff water volume discharged during the storm (Bertrand-Krajewski et al., 1998).

2.4.2.

Data analysis

Descriptive statistics were obtained with the aid of PASW Statistics 17, from the SPSS software. Average values were expressed both in means (with standard error) and in medians, since some parameters do not have a lognormal distribution (Go¨bel et al., 2007). The data generated was subjected to appropriate statistical analyses including variance and correlation analyses. Variance analyses were used to determine if the differences in the mean/ median concentration of the roofs were statistically significant. Whenever possible, the one-way ANOVA test was the preferred option. For this reason, data about water quality were transformed with the aid of the power estimation procedure in order to meet the requirements of the ANOVA test (in particular, the normal distribution and the assumption of homogeneity of variance). When this was not possible, the KruskaleWallis test was used. Pairwise comparisons were carried out by means of either the Bonferroni method or the ManneWhitney U test using Bonferroni correction method, respectively. Correlation analyses were used to determine the degree of association between water quality parameters and storm characteristics (total rainfall height and ADWP) and between water quality parameters themselves, for the whole set of roofs and also for each particular roof (Spearman Rho correlation coefficient).

Table 3 e Rain events considered in the determination of the runoff coefficient (RC) and the samples for quality analysis. Abbreviations: ADWP: antecedent dry weather period. Roof

Clay tiles Metal Plastic Flat gravel

Quantity assessment

Quality assessment

# Monitored events

ADWP range (days)

Rainfall range (mm)

# Monitored events (samples)

Rainfall range (mm)

25 22 23 22

0.1e37.9 0.5e28 0.5e28 0.2e37.9

1e14 1e49 1e49 2e21

14 14 15 12

1.2e68 1e62 2.5e31.2 1e62

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

3.

Results and discussion

3.1.

Quantity assessment

The correlation between runoff and rainfall is high (Pearson coefficient > 0.95 and p < 0.05 for all roofs). The regression model (R ¼ mP þ n) between roof runoff (R) and precipitation height (P) for each roof is presented in Fig. 2. All the regression parameters are statistically significant ( p < 0.05), except for the y-intercept (n) in the equations for M and P roofs. The regression model has been successfully cross-validated, with reductions in the cross-validation in the range of 0.005 and 0.039. Therefore, it can be assumed that the inferences of these regression models to the whole population are valid.

3.1.1.

Initial abstraction

The texture of different roof materials causes different retention, different runoff behaviour and different weathering processes (Go¨bel et al., 2007). Therefore, each roof has its own characteristic initial abstraction volume, mostly explained by its slope and materials (roughness). The most important abstraction is interception, which can be defined as the rainfall that wets and sticks to aboveground objects until it is returned to the atmosphere through evaporation (Corbitt, 1998). The initial abstraction is estimated as the value of the xintercept (n/m) from the regression model (see Fig. 2). Thus, the estimates of the initial abstraction are 0.8 and 3.8 mm for CT and FG, respectively. Since the estimation of the y-intercept (n) is not significant ( p > 0.05) for M and P roofs, which means that it cannot be asserted that n s 0, it is assumed that their initial abstraction is also zero. Although it may be argued that there must be a certain amount of initial abstraction, it is so close to zero that it has not been possible to estimate it with the available data (despite us having included small rain events in the analysis). The highest initial abstraction of the FG roof is explained by the fact that the more porous or rough the roof surface, the more likely it will retain or absorb rainwater (Lancaster, 2006). Therefore, the gravel intercepts the first litres of water due to its high water retention capacity (greater interstitial pore space) together with the lack of slope in the roof. The results

3249

for the FG roof coincide with the predictions from WrighteMcLaughlin Engineers (Corbitt, 1998) who affirm that typical depression and detention values for flat roofs are between 2.5 and 7.5 mm. However, the results for the sloping roofs differ from their predictions since WrighteMcLaughlin Engineers assume that the initial abstraction for sloping roofs is between 1.3 and 2.5 mm.

3.1.2.

Global RC

Regression models are used to calculate the runoff for those rain events that exceed the initial abstraction. In the case of M and P roofs, it is considered that runoff never exceeds the event rainfall height (which would happen according to the runoff-rain function for the smaller rain events due to the positive y-intercept). Then, Eq. (2) is used to calculate the global RC. Depending on the local rainfall profile (either predominance of small or large rain events), the total annual losses related to initial abstraction will vary. The average global RC (period 1999e2009) for CT, M, P and FG roofs is 0.84  0.01, 0.92  0.00, 0.91  0.01 and 0.62  0.04, respectively. The lowest RC for FG roof is explained because of its high initial abstraction. This is relevant for stormwater management and, particularly, for flood prevention. Many cities are facing problems with the management of combined sewer overflows during and after rain events. As a solution to this problem, combined sewer overflow tanks are being constructed to reduce the runoff peak flows, which might be alternatively attained by having roofs with low RC instead. In contrast, higher RC values (corresponding to sloping smooth roofs) would be preferable in order to maximise the amount of rainwater harvested. In this context, it worth mentioning what RWH has a great potential in the case study area, despite its rather limited rainfall. Previous research (Fragkou, 2007) has shown that rainwater falling on the urbanised areas within the administrative region compiled of 27 coastal municipalities of the Barcelona metropolitan region (with a population density of more than 5000 inhabitants per km2) is about 1.4 times higher than its water consumption (including related losses due to leakage from piping). Therefore, rainwater management may have a substantial role in urban water supply schemes in Spain, for which the selection of the right roof is desirable. In addition, the selection of the most appropriate scale for its infrastructures (i.e. building vs. neighbourhood level) should be explored in order to implement the most cost-efficient strategy (Farreny et al., 2011). Since the rainfall profile affects the RC, the application of the model to other climatic characteristics would lead to different results. The greatest differences in roof RC between the set of roofs are found when the size of rain events is small (i.e. if all rain events were of 5 mm, the RC would be 0.79, 0.95, 0.96 and 0.23 for CT, M, P and FG roofs, respectively). However, if the size of rain events were greater, differences would be much smaller (i.e. if all rain events were of 15 mm, the RC would be of 0.89, 0.91, 0.90 and 0.71 for CT, M, P and FG roofs, respectively).

3.1.3. The effect of ADWP and wind direction in the regression model Fig. 2 e Regression model for roof runoff and rainfall height. Regression equations and corrected R2 are also shown. Each point represents a rain event.

Since the effective collection area of each roof depends on factors such as the direction of prevailing winds and orientation (Villarreal and Dixon, 2005), particularly for sloping roofs, it is

3250

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

suggested that wind direction could affect the runoff from M and P roofs (the only single pitch sloping roofs). However, statistical analyses show that there is no significant ( p > 0.05) relationship between runoff and wind direction for any of the roofs. However, we suggest that the effect of wind could partially explain the lack of significance of the y-intercept for M and P. Another variable that could affect runoff is the ADWP. Although there is no significant ( p > 0.05) correlation between runoff and ADWP for any roof, its incorporation in the regression model provides a more adjusted regression model in the case of FG roofs (corrected R2 ¼ 0.92):

Skaryska et al., 2007; Villarreal and Dixon, 2005), namely, characteristics of the surface, atmosphere conditions and properties of pollutants. In order to allow for comparisons, the water guidelines from the Drinking Water Directive 98/83/EC on the quality of water intended for human consumption (European Commission, 1998) are shown for the legislated parameters. Furthermore, the pollution levels of the currently used raw water sources for drinking purposes in the region e either surface or groundwater e previous to water purification are shown.

3.2.1. General assessment of the runoff quality for the whole set of roofs

R ¼ 0:937 P  0:083 ADWP  2:901

EC, which represents the samples’ total ion content, can be identified as a leading parameter (Go¨bel et al., 2007) and may be regarded, to a certain extent, as a measure of the concentration of dissolved matter (Deletic, 1998). All roofs are within a low range of EC (Table 4), particularly compared to current surface and groundwater sources. Generally, the pH of rainwater ranges from 4.5 to 6.5 but increases slightly after falling on the roof and during storage in tanks (Go¨bel et al., 2007; Meera and Ahammed, 2006). However, our results indicate higher pH values, in the range of 6.54 and 8.25. These results are consistent with those obtained by Melidis et al. (2007) in several roofs in Greece (among which there were CT and M roofs). This pH can be explained as a result of the neutralisation which takes place mainly because of high values of alkalinity and base cations in African rains, which are common in the region, compared to the rains of European origin (Avila and Alarco´n, 1999). The limited amounts of nitrates and sulphates also explain these high pH values. The measurement of the TSS contents in urban runoff is of major concern with respect to the transport of anthropogenic

The negative coefficient that accompanies the variable ADWP indicates that the longer the ADWP, the smaller the roof runoff. This is explained by the accumulation of water over longer periods in FG roofs, which is linked to their higher initial abstraction.

3.2.

Quality assessment

Descriptive statistics including minimum, maximum, mean and median concentration of the water quality variables for the whole set of roofs are presented in Table 4. Nitrites and phosphates were not detected at all in most samples (35 and 41 out of 55, respectively), for which a concentration of 0 mg/L was considered for further statistical analyses. Table 4 also compares the water quality results to the values from a review of rooftop runoff quality of data expressed in EMCs (Chang et al., 2004; Evans et al., 2006; Go¨bel et al., 2007; Melidis et al., 2007). The variety and quantity of individual pollutants present in roof runoff are affected by a number of factors (Chang et al., 2004; Sazakli et al., 2007; Simmons et al., 2001;

Table 4 e Rooftop runoff quality results and review of literature results. Parameter

Units

Case study roofs

DWGb

Roof review

a

Surface and groundwater sourcesc 1

2

3

Min

Max

Mean  S.E.

Med

Min

Max

Med

Limit

Mean

Mean

Mean

Physicalechemical parameters Conductivity mS/cm 15.4 pH upH 6.54

456 8.85

85.0  10.0 7.59  0.07

59.3 7.61

2.2 3.3

269 8.25

141 5.7

2500 6.5e9.5

1439 8.11

481 7.91

3169 7.52

Sum parameters TSS TOC TIC

mg/L mg/L mg/L

0 0.65 1.36

38.5 53.6 19.0

5.98  0.95 11.6  1.7 7.37  0.66

3.63 6.4 5.78

13 n.a. n.a.

120 n.a. n.a.

43 n.a. n.a.

e e e

n.a. 4.04 n.a.

n.a. 1.76 n.a.

n.a. 4.04 n.a.

Nutrients PO43 NH4þ NO3 NO2

mg/L mg N/L mg/L mg/L

0.00 0.04 0.01 0.00

6.60 2.42 9.34 3.45

0.32  0.14 0.50  0.07 1.75  0.26 0.13  0.05

0.00 0.42 1.16 0.00

n.a. 0.1 0.1 n.a.

n.a. 6.2 5.73 n.a.

n.a. 3.39 2.78 n.a.

e 0.5 50 0.1

0.51 0.91 9.23 0.55

0.11 0.15 7.84 0.13

0.18 4.89 0.05

Main ions SO42 Cl Total carbonates

mg/L mg/L mmol/L

0 0.15 0.12

11.5 119 1.62

3.54  0.39 8.86  2.38 0.63  0.06

2.59 3.38 0.49

0.01 5.73 n.a.

19.7 40.5 n.a.

46.7 7.74 n.a.

250 250 e

173.5 292.2 n.a.

58.8 47.1 n.a.

244.2 977.7 n.a.

a Medians are based on the review made by Go¨bel et al. (2007). b Drinking water guidelines (European Commission, 1998). c Sources 1, 2 and 3 correspond to quality data from Llobregat river, Ter river and Llobregat Delta Aquifer, respectively, (ACA, 2010).

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

pollutants, since pollutants are dominantly bound to particles (Moilleron et al., 2002). The amount of TSS is small in all roofs (median TSS <5 mg/L), particularly compared to the review. There is no guideline for TSS for drinking purposes. However, a content below 25 mg TSS/L is associated with excellent water quality (Davis and McCuen, 2005). The TOC content, which is not legislated for drinking purposes, is slightly higher than that in surface and groundwater (Table 4). However, it is considered that concentrations below 20 mg/L correspond to good water quality (Davis and McCuen, 2005). Inorganic nitrogen occurred mainly as NO3 and ammonium (NH4þ) while NO2 occurred in smaller proportions (Table 4). Because NO3 is a transformation product, it shows a reversed behaviour and increases in concentration as the NH4þ concentration decreases (Go¨bel et al., 2007). Sulphates and nitrates together represent the major ionic derivatives of industrial and traffic emissions (Evans et al., 2006), as a result of fossil fuels combustion (Mouli et al., 2005). Despite the proximity to a motorway, the concentrations of these pollutants are in the low range, compared to the review of roofs (Table 4). This can be partly explained by the relative position of the predominant direction of winds in the area and the motorway. Predominant winds come from westerly directions and the motorway is located to the east, which prevents its pollution from reaching the roofs. Ammonium concentrations are normally of natural origin e fermentation of nitrogenised products such as bird faeces e in areas with low industrial activity (Melidis et al., 2007). Therefore, bird excrements together with moss and lichens on the roofs can cause an increase in ammonium as well as phosphorus levels (Go¨bel et al., 2007). The runoff quality data is affected by the first flush phenomena (see Kus et al., 2010; Zhang et al., 2010), since the experimental design does not consider first flush diversion. However, a practical implementation of a RWH strategy would divert the very dirty runoff from the first few millimetres of

3251

rainfall away from the tanks to avoid contamination (Villarreal and Dixon, 2005), which is a practice followed globally (Sazakli et al., 2007). Thus, the water quality of the RW collected in the tank would be significantly improved (Kus et al., 2010) Comparing the quality (expressed in medians) of runoff from the case study roofs to the review made by Go¨bel et al. (2007) (Table 4), who reviews more than 300 references providing about 1300 pieces of data for different pollutants, the quality in the case study area is, in general, better for the parameters under study and with available data. The violations in water quality standards were most severe for NO2 and NH4þ (18% and 24% of samples exceeded the European drinking water standards (Table 4), respectively). However, the nitrite concentration should not be considered a concern since American and Australian legislations establish a limit of 1 and 3 mg/L, respectively (Australian Government, 2004; US EPA, 2009). On the other hand, median NH4þ concentration was within the guidelines. On the other hand, it can be stated that the physicochemical quality of the collected roof runoff is, in general, superior to the sources of surface and groundwater in the region. This can be partly explained by the degradation of current water sources in the region. This is consistent with van Roon (2007), who stated that the physical and chemical properties of rainwater are usually superior to sources of groundwater that may have been subjected to contamination.

3.2.2. The effect of rainfall height and ADWP on runoff quality The wide distribution of EMCs depends on total rainfall because of the dilution effect during a storm (Kim et al., 2007). Our results show that, in general (for the whole set of roofs), the higher the rainfall height, the smaller the pollution loads in the samples. The correlation is significant ( p < 0.05) and negative between rainfall height and the following parameters: EC, TIC, TOC, TC, total carbonates system, Cl, NO3 and SO42 (Spearman

Table 5 e Spearman Rho correlation coefficients between the water quality parameters for the whole set of roofs.

3252

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

Fig. 3 e Box plot diagram of rooftop runoff water quality for each roof. Abbreviations: CT [ clay tiles; M [ metal; P [ plastic; FG [ flat gravel.

correlation coefficient between 0.301 and 0.642). If the correlation analysis distinguishes between roofs, the highest significant correlation is found between rain and EC (coefficient between 0.843 and 0.940 among the several roofs), TC (between 0.564 and 0.814 among the several roofs), total carbonates system (0.723 for M roof), NO3 (0.591, 0.771 and 0.789 for CT, M and P roofs, respectively) and SO42 (0.618, 0.656 and 0.629 for CT, M and P roofs, respectively). On the other hand, it has been previously observed that ADWP can markedly affect the quality of runoff water (Fewtrell and Kay, 2007). However, no significant ( p > 0.05) correlations are found between ADWP and the quality parameters, except for NO2 (Spearman correlation coefficient ¼ 0.439). These results are consistent with those obtained by Kim et al. (2005c) who were disappointed because of a lack of correlation between water quality (from highways) and storm characteristics (such as ADWP). Correlation analyses between the water quality parameters (Table 5) show that the highest correlations are found between any combinations of the following parameters: TIC, TOC, TC and total carbonates system. The negative significant ( p < 0.01) correlation between NH4þ and some parameters (i.e. pH and TIC) can be explained by the fact that alkaline mediums foster ammonia (NH3) volatilisation and also because aerobic conditions encourage oxidation processes e which result in CO2, measured indirectly by means of TIC e and nitrification, both of which result in less amounts of NH4þ.

3.2.3.

Differences in water quality between roofs

The water quality results for the runoff collected during the study period in each roof are shown in Fig. 3 for the following parameters: pH, EC, TSS, TOC, SO42, NO3, NH4þ and total carbonates system (HCO3, CO32 and H2CO3). Statistical analyses of variances indicate that the differences in water quality within the whole roof set are not significant ( p > 0.05) for the following parameters: pH, Cl, NO3, SO42 and TSS. However, significant differences were found ( p < 0.05) between the quality in FG roof and the three pitched roofs for

EC, TIC, TOC, TC, total carbonates system and NH4þ, presenting higher pollution levels for all parameters except for NH4þ. The higher pollution load in FG roof can be explained by the weathering of the roof materials (gravels) and the accumulated deposits of particulates and associated flora on them. FG roofs are more predisposed to being colonised by a wide range of plants, notably mosses, algal crust and lichens. Besides this, the lack of slope aids the development of these processes. In contrast, lower levels of NH4þ in FG roof may be explained by the higher alkalinity and the greater oxidation processes (indirectly measured by means of TIC). Nevertheless, it is believed that other flat roofs with different materials other than gravel (such as concrete, asphalt, hot tar or tiles) would present different pollution results. On the other hand, M and P roofs do not show significant differences ( p > 0.05) for any of the parameters. This can be explained by the similarity of the hydraulic behaviour of both catchments (sloping smooth materials). The two studied roofs are actually the opposite sides of the same roof (although made of different material) but there are no differences in concentrations since neither side receives a distinctly higher net precipitation than the other. These results agree with those obtained by Chang et al. (2004) for similar roof materials (painted aluminium, galvanised iron and composite shingles), in which the orientation of the roof had no effect on the runoff water quality. In contrast, M and P roofs present differences with the CT roof for TOC, TC and NH4þ. The higher amount of TOC in CT roofs can be explained by the relatively high porosity of clay tiles, for which the material is exposed to greater physical, chemical and biological degradation. Besides this, the lichen’s presence and growth can cause several additional deterioration problems (Kiurski et al., 2005). From these results, one can affirm that runoff from M and P roofs is of the same or better quality than the other roofs for all parameters, except for NH4þ. On the other hand, runoff from FG roofs presents the best quality in terms of NH4þ but the worst in terms of EC, TOC and the system of carbonates. The runoff from CT roofs presents an in-between quality.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

4.

Conclusions

- The linear regression model developed for each roof (clay tiles e CT, metal sheet e M, plastic sheet e P and flat gravel e FG) shows that runoff depends greatly on the rainfall height. The initial abstraction, which depends on the roof’s slope and roughness, is highest in flat rough roofs (i.e. 3.8 mm for FG) while sloping roofs present much smaller abstractions (0.8 mm). - The selection of sloping smooth roofs (i.e. M and P roofs, with a RC > 0.9) implies a global RWH potential approximately 50% greater than flat rough roofs (i.e. FG roof, with RC ¼ 0.62). The promotion of roofs with low RC may be advisable in order to reduce the peak flow and minimise the problem of combined sewer overflows. - Quality analyses indicate that rainwater runoff samples are in the low range for EC (85.0  10.0 mS/cm), TSS (5.98  0.95 mg/L) and TOC (11.6  1.7 mg/L); and their pH is basic (7.59  0.07 upH). Thus, the quality of rainwater runoff in the case study area (north-eastern Spain) appears to be generally better than the average quality found for roof runoff in the literature review. - Differences in runoff water quality are relevant between sloping smooth and flat rough roofs. The FG roof presents higher levels of all pollutants (except for NH4þ) because of the processes of particle deposition, roof weathering and plant colonisation. In contrast, sloping roofs (such as CT, M and P roofs) present better quality. - These results have an important significance for local governments and urban planners in the design and planning of cities. With city planning policies that could establish guidelines regarding the slope and roughness of roofs (both for the existing city and for new developments), stormwater roof runoff could be promoted both in terms of resource availability and quality. Thus, sloping smooth roofs, which have proved to perform best, may be preferable in order to foster RWH.

Acknowledgments The authors wish to thank the Catalan Government and the European Social Fund for the FI scholarship received by Farreny R., and the grant from Colciencias (Administrative Department of Science, Technology and Innovation of Colombia) received by Morales-Pinzo´n T. With financial support from the Spanish Ministry for Science and Innovation through the project ‘Ana´lisis ambiental del aprovechamiento de las aguas pluviales’ (PLUVISOST CTM2010-17365). Contributors: Ramon Farreny is responsible for the collection of data and samples from the rainwater tanks. Besides, he has participated in the preparation of the article. Carlota Taya` has carried out the physicalechemical analyses at the laboratory. Tito Morales-Pinzo´n has carried out the statistical analyses. Albert Guisasola has participated in the experimental design, has supervised the laboratory tasks and has participated in the preparation of the article. Joan Rieradevall has participated in the analysis of the results and in the

3253

preparation of the article. Xavier Gabarrell have supervised the whole research and participated in the article preparation. All authors have approved the final article.

references

ACA, 2010. Xarxes de Control. Age`ncia Catalana de l’Aigua. Available from: http://aca-web.gencat.cat/aca (accessed November 2009). Adeniyi, I.F., Olabanji, I.O., 2005. The physico-chemical and bacteriological quality of rainwater collected over different roofing materials in Ile-Ife, southwestern Nigeria. Chemistry and Ecology 21, 149e166. Albrechtsen, H.J., 2002. Microbiological investigations of rainwater and graywater collected for toilet flushing. Water Science and Technology 46 (6e7), 311e316. APHA, 1995. Standard Methods for the Examination of Water and Wastewater, 19th ed. American Publishers Health Association, Washington DC, USA. Appan, A., 2000. A dual-mode system for harnessing roofwater for non-potable uses. Urban Water 1, 317e321. ASCE, 1969. Design and Construction of Sanitary Storm Sewers. In: ASCE Manuals and Reports on Engineering Practice, vol. 37. ASCE, New York. Australian Government, 2004. Australian Drinking Water Guidelines. Available from: http://www.nhmrc.gov.au/ publications/synopses/eh19syn.htm (accessed September 2010). Avila, A., Alarco´n, M., 1999. Relationship between precipitation chemistry and meteorological situations at a rural site in NE Spain. Atmospheric Environment 33, 1663e1677. Bertrand-Krajewski, J.-L., Chebbo, G., Saget, A., 1998. Distribution of pollutant mass vs volume in stormwater discharges and the first flush phenomenon. Water Research 32 (8), 2341e2356. Chang, M., McBroom, M.W., Scott Beasley, R., 2004. Roofing as a source of nonpoint water pollution. Journal of Environmental Management 73, 307e315. Corbitt, R.A., 1998. Standard Handbook of Environmental Engineering, second ed. McGraw-Hill, New York. Davis, A.P., McCuen, R.H., 2005. Stormwater Management for Smart Growth. Springer, United States. Deletic, A., 1998. The first flush load of urban surface runoff. Water Research 32 (8), 2462e2470. European Commission, 1998. Council Directive 98/83/EC of 3 November 1998 on the Quality of Water Intended for Human Consumption. Available from: http://eur-lex.europa.eu/ LexUriServ/LexUriServ.do?uri=CELEX:31998L0083:EN:NOT (accessed September 2010). Evans, C.A., Coombes, P.J., Dunstan, R.H., 2006. Wind, rain and bacteria: the effect of weather on the microbial composition of roof-harvested rainwater. Water Research 40 (1), 37e44. Farreny, R., Gabarrell, X., Rieradevall, J., 2011. Cost-efficiency of rainwater harvesting strategies in dense mediterranean neighbourhoods. Resources, Conservation and Recycling. doi: 10.1016/j.resconrec.2011.01.008. Fewkes, A., 2000. Modelling the performance of rainwater collection systems: towards a generalised approach. Urban Water 1, 323e333. Fewtrell, L., Kay, D., 2007. Microbial quality of rainwater supplies in developed countries: a review. Urban Water Journal 4, 253e260. Fletcher, T.D., Deletic, A., Mitchell, V.G., Hatt, B.E., 2008. Reuse of urban runoff in Australia: a review of recent advances and remaining challenges. Journal of Environmental Quality 37, S116eS127.

3254

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 4 5 e3 2 5 4

Fo¨rster, J., 1996. Patterns of roof runoff contamination and their potential implications on practice and regulation of treatment and local infiltration. Water Science and Technology 33 (6), 39e48. Fo¨rster, J., 1999. Variability of roof runoff quality. Water Science and Technology 39 (5), 137e144. Fragkou, M.C. 2007. Evaluation of Urban Sustainability Through a Metabolic Perspective. Application in a Mediterranean Coastal Region. PhD thesis, Department of Chemical Engineering, Universitat Auto`noma de Barceona. Ghisi, E., Tavares, D.d.F., Rocha, V.L., 2009. Rainwater harvesting in petrol stations in Brası´lia: potential for potable water savings and investment feasibility analysis. Resources, Conservation and Recycling 54, 79e85. Go¨bel, P., Dierkes, C., Coldewey, W.G., 2007. Storm water runoff concentration matrix for urban areas. Journal of Contaminant Hydrology 91, 26e42. Gromaire, M.C., Garnaud, S., Saad, M., Chebbo, G., 2001. Contribution of different sources to the pollution of wet weather flows in combined sewers. Water Research 35 (2), 521e533. Kim, R.-H., Lee, S., Kim, Y.M., Lee, J.H., Kim, S.K., Kim, J.G., 2005a. Pollutants in rainwater runoff in Korea: their impacts on rainwater utilization. Environmental Technology 26, 411e420. Kim, R.H., Lee, S., Lee, J.H., Kim, Y.M., Suh, J.Y., 2005b. Developing technologies for rainwater utilization in urbanized area. Environmental Technology 26, 401e410. Kim, L.-H., Kayhanian, M., Zoh, K.-D., Stenstrom, M.K., 2005c. Modeling of highway stormwater runoff. Science of The Total Environment 348, 1e18. Kim, L.-H., Ko, S.-O., Jeong, S., Yoon, J., 2007. Characteristics of washed-off pollutants and dynamic EMCs in parking lots and bridges during a storm. Science of The Total Environment 376, 178e184. Kiurski, J., Ranogajec, J.G., Ujhelji, A.L., Radeka, M.M., Bokorov, M. T., 2005. Evaluation of the effect of lichens on ceramic roofing tiles by scanning electron microscopy and energy-dispersive spectroscopy analyses. Scanning 27, 113e119. Kus, B., Kandasamy, J., Vigneswaran, S., Shon, H.K., 2010. Analysis of first flush to improve the water quality in rainwater tanks. Water Science and Technology 61 (2), 421e428. Lancaster, B., 2006. guiding principles to welcome rain into your life and landscape. In: Rainwater Harvesting for Drylands and Beyond, vol. 1. Rainsource Press, Tucson, Arizona. Liaw, C.H., Tsai, Y.L., 2004. Optimum storage volume of rooftop rain water harvesting systems for domestic use. Journal of the American Water Resources Association 40, 901e912. Magyar, M.I., Mitchell, V.G., Ladson, A.R., Diaper, C., 2007. An investigation of rainwater tanks quality and sediment dynamics. Water Science and Technology 56 (9), 21e28. McCuen, R., 2004. Hydrologic Analysis and Design, third ed. Pearson Education Inc, Upper Saddle River, NJ. Meera, V., Ahammed, M.M., 2006. Water quality of rooftop rainwater harvesting systems: a review. Journal of Water Supply Research and Technology-Aqua 55 (4), 257e268. Melidis, P., Akratos, C.S., Tsihrintzis, V.A., Trikilidou, E., 2007. Characterization of rain and roof drainage water quality in xanthi, greece. Environmental Monitoring and Assessment 127, 15e27. Moilleron, R., Gonzalez, A., Chebbo, G., The´venot, D.R., 2002. Determination of aliphatic hydrocarbons in urban runoff samples from the “Le Marais” experimental catchment in Paris centre. Water Research 36 (5), 1275e1285.

Mouli, P.C., Mohan, S.V., Reddy, S.J., 2005. Rainwater chemistry at a regional representative urban site: influence of terrestrial sources on ionic composition. Atmospheric Environment 39, 999e1008. Pacey, A., Cullis, A., 1989. Rainwater Harvesting: The Collection of Rainfall and Runoff in Rural Areas, London, UK. Polkowska, Z., Go´recki, T., Namiesnik, J., 2002. Quality of roof runoff waters from an urban region (Gdansk, Poland). Chemosphere 49, 1275e1283. Rahman, A., Dbais, J., Imteaz, M., 2010. Sustainability of rainwater harvesting systems in multistorey residential buildings. American Journal of Engineering and Applied Sciences 3, 889e898. Sazakli, E., Alexopoulos, A., Leotsinidis, M., 2007. Rainwater harvesting, quality assessment and utilization in Kefalonia Island, Greece. Water Research 41 (9), 2039e2047. Simmons, G., Hope, V., Lewis, G., Whitmore, J., Gao, W.Z., 2001. Contamination of potable roof-collected rainwater in Auckland, New Zealand. Water Research 35 (6), 1518e1524. Singh, V.P., 1992. Elementary Hydrology. Prentice Hall, Upper Saddle River, New Jersey. Skaryska, K., Polkowska, A., Namienik, J., Przyjazny, A., 2007. Application of different sampling procedures in studies of composition of various types of runoff waters e a review. Critical Reviews in Analytical Chemistry 37, 91e105. Snee, R.D., 1977. Validation of regression models: methods and examples. Technometrics 19, 415e428. TxDOT, 2009. Hydraulic Design Manual, Texas. Available from: http://onlinemanuals.txdot.gov/txdotmanuals/hyd/hyd.pdf (accessed July 2010). Uba, B.N., Aghogho, O., 2000. Rainwater quality from different roof catchments in the Port Harcourt district, Rivers State, Nigeria. Journal of Water Supply Research and TechnologyAqua 49, 281e288. US EPA, 2009. National Primary Drinking Water Regulations. Available from: http://www.epa.gov/safewater/contaminants/ index.html#4 (accessed September 2010). Van Metre, P.C., Mahler, B.J., 2003. The contribution of particles washed from rooftops to contaminant loading to urban streams. Chemosphere 52, 1727e1741. van Roon, M., 2007. Water localisation and reclamation: steps towards low impact urban design and development. Journal of Environmental Management 83, 437e447. Viessman, W., Lewis, G.L., 2003. Introduction to Hydrology, fifthth ed. Prentice Hall, Upper Saddle River, New Jersey. Villarreal, E.L., Dixon, A., 2005. Analysis of a rainwater collection system for domestic water supply in Ringdansen, Norrkoping, Sweden. Building and Environment 40, 1174e1184. Ward, S., Memon, F.A., Butler, D., 2010. Harvested rainwater quality: the importance of appropriate design. Water Science and Technology 61 (7), 1707e1714. Zhang, M.L., Chen, H., Wang, J.Z., Pan, G., 2010. Rainwater utilization and storm pollution control based on urban runoff characterization. Journal of Environmental Sciences e China 22, 40e46. Zhu, K., Zhang, L., Hart, W., Liu, M., Chen, H., 2004. Quality issues in harvested rainwater in arid and semi-arid Loess Plateau of northern China. Journal of Arid Environments 57, 487e505. Zobrist, J., Mu¨ller, S.R., Ammann, A., Bucheli, T.D., Mottier, V., Ochs, M., Schoenenberger, R., Eugster, J., Boller, M., 2000. Quality of roof runoff for groundwater infiltration. Water Research 34 (5), 1455e1462.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 5 5 e3 2 6 2

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Iron crystallization in a fluidized-bed Fenton process Nonglak Boonrattanakij a,b, Ming-Chun Lu c, Jin Anotai d,* a

International Postgraduate Programs in Environmental Management, Graduate School, Chulalongkorn University, Bangkok 10330, Thailand b National Center of Excellence for Environmental and Hazardous Waste Management (NCE-EHWM), Chulalongkorn University, Bangkok 10330, Thailand c Department of Environmental Resources Management, Chia-Nan University of Pharmacy and Science, Tainan 717, Taiwan d National Center of Excellence for Environmental and Hazardous Waste Management, Department of Environmental Engineering, Faculty of Engineering, King Mongkut’s University of Technology Thonburi, Pracha-u-tid Road, Thungkru, Bangkok 10140, Thailand

article info

abstract

Article history:

The mechanisms of iron precipitation and crystallization in a fluidized-bed reactor were

Received 15 November 2010

investigated. Within the typical Fenton’s reagent dosage and pH range, ferric ions as

Received in revised form

a product from ferrous ion oxidation would be supersaturated and would subsequently

22 March 2011

precipitate out in the form of ferric hydroxide after the initiation of the Fenton reaction.

Accepted 23 March 2011

These precipitates would simultaneously crystallize onto solid particles in a fluidized-bed

Available online 31 March 2011

Fenton reactor if the precipitation proceeded toward heterogeneous nucleation. The heterogeneous crystallization rate was controlled by the fluidized material type and the

Keywords:

aging/ripening period of the crystallites. Iron crystallization onto the construction sand

Advanced oxidation processes

was faster than onto SiO2, although the iron removal efficiencies at 180 min, which was

Crystallite

principally controlled by iron hydroxide solubility, were comparable. To achieve a high iron

Iron oxide

removal rate, fluidized materials have to be present at the beginning of the Fenton reac-

Iron removal

tion. Organic intermediates that can form ferro-complexes, particularly volatile fatty acids,

Nucleation

can significantly increase ferric ion solubility, hence reducing the crystallization perfor-

Precipitation

mance. Therefore, the fluidized-bed Fenton process will achieve exceptional performance with respect to both organic pollutant removal and iron removal if it is operated with the goal of complete mineralization. Crystallized iron on the fluidized media could slightly retard the successive crystallization rate; thus, it is necessary to continuously replace a portion of the iron-coated bed with fresh media to maintain iron removal performance. The iron-coated construction sand also had a catalytic property, though was less than those of commercial goethite. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

The conventional Fenton process, an advanced oxidation process (AOP), has been proven to be very effective in removing various organic contaminants with simple operation and low capital cost (Canizares et al., 2009). A major

disadvantage of most homogeneous Fenton processes is the production of large volumes of ferric hydroxide sludge upon pH neutralization after treatment. Even after dewatering, this bulky sludge still typically contains up to 80% water; hence, disposal expense is very costly. Non-electrochemical heterogeneous Fenton processes that use iron catalysts in a solid

* Corresponding author. Tel.: þ662 470 9166; fax: þ662 470 9165. E-mail address: [email protected] (J. Anotai). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.045

3256

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 5 5 e3 2 6 2

form are of interest because most of the iron remains in the solid phase in the form of iron oxides; however, these solid iron catalysts have lower oxidation performance compared to the catalysts used in homogeneous Fenton processes (Kwan and Voelker, 2002; Chou et al., 1999). The fluidized-bed Fenton process combines the advantages of the rapid oxidation rate of homogeneous Fenton reactions with the advantages of solid iron crystals of heterogeneous Fenton reactions by crystallizing the iron onto fluidized media surfaces. Nonetheless, it was found that the iron removal efficiency by the fluidized-bed Fenton process varied dramatically depending on the reaction conditions, although the performance with respect to pollutant oxidation was very impressive. Chou et al. (1999) obtained around 25e90% iron reduction in a fluidizedbed Fenton reactor treating 100 mg L1 benzoic acid using 200 mg L1 H2O2 in solutions with various pHs and containing various ferrous ion supplements. Anotai et al. (2009) could achieve only 50% removal of the total iron content from a fluidized-bed Fenton reactor treating nitrobenzene at pH 2.8. Therefore, a greater understanding of iron crystallization behavior and the factors affecting crystallization performance in the fluidized-bed Fenton process are needed. This study explored the basic characteristics of iron precipitation and crystallization in the fluidized-bed reactor and determined the factors controlling the crystallization of iron onto seed materials including construction sand (CS), silica dioxide (SiO2), and aluminium oxide (Al2O3). Iron crystallization behaviors in completely mixed, fluidized-bed, and fluidized-bed Fenton reactors in the presence and absence of organics were also investigated. The knowledge obtained from this study can be applied in the field to optimize the performance of both pollutant and iron removal of the fluidized-bed Fenton process.

2.

Materials and methods

2.1.

Reagents and fluidized materials

All chemicals including FeSO4$7H2O, H2O2, 2,6-dimethylaniline, aniline, oxalic, acetic, and formic acids supplied from Merck were reagent grade and were used without further purification. Catalyst-grade goethite (a-FeOOH) was purchased from SigmaeAldrich. All solutions were prepared using ultrapure water (18.2 MU cm; Barnstead). The CS, SiO2, and Al2O3 were used as fluidized materials. The diameters of CS and SiO2 were between 0.42 and 0.59 mm (passing through sieve #30 and retained on sieve #40), whereas that of Al2O3 was 2.5 mm (the only size that is available locally). All media were soaked in HCl solution at pH 1 for 24 h, washed with deionized water until the pH of the rinse water was 7, and then ovendried at 103  C.

2.2.

Experimental conditions

A 0.5 L beaker with a magnetic stirrer was used as a completely mixed reactor (CMR). For fluidized-bed experiments, the schematic apparatus is shown in Fig. 1. A 1.35 L bench-scale glass fluidized-bed reactor (FBR) of 5.2 cm 4  133 cm height was used in this study. Synthetic wastewater (1.3 L) and

Fig. 1 e Schematic diagram of the fluidized-bed reactor (not to scale).

fluidized media (300 g) were placed into the FBR. Bed expansion was kept constant at 50% by controlling the recirculation flow rate. The solubilities of Fe2þ and Fe3þ were studied as a function of pH between pH 2 and pH 12 in the CMR. The ionic strength of each solution was buffered with 0.1 M NaClO4. For iron crystallization study, the experiments were mainly conducted in the FBR. In the experiments with Fenton reaction, ferrous ion was added to the synthetic wastewater and the pH was adjusted to the desired value. Hydrogen peroxide was then added to the solution and the reaction was simultaneously started. At the selected time interval, a sample was taken and filtered immediately through a 0.2 mm filter paper (Advantec) before analysis. For residual organic measurement, filtered sample was added with an appropriate amount of 0.1 N NaOH to raise the solution pH to the alkaline range to stop any further Fenton reaction. The sample was filtered again through a 0.2 mm filter paper to remove ferric precipitates before further analysis. All experiments were carried out at 25  C and the solution pH was maintained at the desired value by adding either H2SO4 or NaOH.

2.3.

Analytical methods

The ferrous ion concentration was analyzed using light absorbance measurements taken with a UVeVIS spectrophotometer (Genesys 20, Thermo Scientific) at 510 nm after

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 5 5 e3 2 6 2

forming a complex with 1,10-phenanthroline following the Standard Methods (APHA et al., 1992). For soluble (filtered through 0.2 mm filter paper) and total iron analysis, samples were digested and reduced to the ferrous state using concentrated hydrochloric acid and hydroxylamine, respectively, prior to color formation with 1,10-phenanthroline. Residual H2O2 was analyzed by iodometric titration with a Na2S2O3 solution (Kolthof et al., 1969). Aniline and 2,6dimethylaniline were analyzed using a Shimadzu GC-17A gas chromatograph equipped with a flame ionization detector and a HewlettePackard HP-5 capillary column (0.53 mm 4  15 m length) following the procedures and operating conditions of Boonrattanakij et al. (2009). Crystal structures of CS and ironcoated CS were characterized by the X-ray powder diffraction spectroscopy (D8 Discover, Bruker-AXS) with Cu Ka radiation (l ¼ 0.15406 nm) in a 2q range of 20e80 and a scanning speed of 3 min1. Element composition was analyzed by the dispersive X-ray fluorescence spectrometer (WD-XRF S4 Pioneer, Bruker-AXS).

3.

Results and discussion

3.1.

Iron solubility

The aqueous solubilities of Fe2þ and Fe3þ at 25  C between pH 2 and pH 12 in 0.1 M NaClO4 buffer solutions are shown in Fig. 2. The solubility of Fe3þ, which is the product of Fe2þ oxidation by H2O2 in Fenton reaction, was much lower than that of Fe2þ. Within the typical pH range for the Fenton process of 2e4 (Pignatello, 1992; Sedlak and Andren, 1991), Fe2þ was highly soluble, but Fe3þ was not. The solubility of Fe3þ was only around 10 mg L1 at pH 3 and decreased tremendously as the pH was increased. This finding is in agreement with the theoretical calculation under ideal conditions, as shown in the log C-pH diagram (Fig. 3) for ferric hydroxide complexes in the presence of Fe(OH)3 precipitates prepared by using b1, b2, b3, b4, and Ksp values of 1011.81, 1022.33, 1028.40, 1034.40, and 1037.11, respectively, from Benjamin (2002). The result implies that ferric precipitates should be formed rapidly after the initiation of the Fenton reaction in the form of 2.5

Fe2+ Fe3+

Soluble iron (mM)

2.0

1.5

1.0

0.5

0.0 0

2

4

6

pH

8

10

12

Fig. 2 e pH dependence of Fe2D and Fe3D solubilities at 25  C.

14

3257

Fig. 3 e Log C-pH diagram for a system in equilibrium with Fe(OH)3(s) in an ideal solution at 25  C.

noncrystalline Fe(OH)3 or so called “ferrihydrite” (Nagano et al., 1992; Cudennec and Lecerf, 2006). The physical appearance of the ferric hydroxide precipitates obtained in this study at different pHs was significantly different, i.e., the precipitates were yellow and made up of discrete particles at acidic pHs and became orange and orange-red with a puffyfloc appearance at neutral and basic pHs, respectively. Amonette and Rai (1990) prepared ferric hydroxide by titrating FeCl3 solution with NaOH to pHs ranging from 5.5 to 13. They found noncrystalline Fe(OH)3 in the solutions with pHs between 5.5 and 8.5. The distinctive OH-bending doublet associated with a-FeOOH in the infrared absorbance spectrum could only be detected at pHs higher than 10 after 18 days of incubation. Differences in the crystalline structure/color might be due to the pH dependence of the crystallization rate, crystal form, and ripening period (Nagano et al., 1992; Amonette and Rai, 1990). According to Stumm and Morgan (1996) and Morel and Hering (1993), the solubility diagram of Fe(OH)3 (Fig. 3) can be divided into three areas, i.e., the undersaturated zone (the area below the Fe(III) solubility line), the metastable zone (the shaded area above the Fe(III) solubility line), and the supersaturated zone (the area above the metastable zone). Ferric ions will not precipitate out in undersaturated solution but will precipitate spontaneously in a strongly supersaturated solution via homogeneous nucleation, in which Fe(OH)3 nuclei will be formed rapidly enough to serve as the centers from which crystal growth can occur to form crystallites that simultaneously ripen to form stable crystals. In metastable oversaturated zone, however, the formation of Fe(OH)3 nuclei is limited due to the lower saturation ratio; hence, homogeneous nucleation proceeds very slowly. Nonetheless, if the metastable solution is seeded with foreign particles, Fe(OH)3 nuclei can be deposited on these foreign surfaces; this nucleation process is called “heterogeneous nucleation.” Because the initial ferric ion concentration in this solubility study was kept constant, increasing the solution pH would directly increase the OH concentration (10 fold per pH unit). As a result, the solution becomes more supersaturated as the pH increases toward the basic range, and the ferric hydroxide precipitation should move from

3258

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 5 5 e3 2 6 2

metastable zone to the supersaturated or labile zone. This would result in different nucleation, crystal growth, and ripening rates; hence, the Fe(OH)3 crystals prepared at different pHs should have different physical appearances. To verify this hypothesis, a supersaturated ferric ion solution at pH 3 was briefly purged with nitrogen gas to remove all dissolved oxygen and was immediately transferred to an airtight BOD bottle. The color of precipitates gradually changed from yellow to orange and finally to orange-red after 48 h, similar to those precipitated in the neutral and basic range, indicating that the crystal ripening time had an effect similar to those observed by others (Nagano et al., 1992; Amonette and Rai, 1990). Our finding is also supported by Cudennec and Lecerf (2006) and Loan et al. (2005); they found that Fe(OH)3 could be converted to either a-FeOOH or hematite (a-Fe2O3) depending on the pH, the ripening period, the temperature, and the presence of catalyst.

3.2. Ferric hydroxide crystallization in a fluidized-bed reactor The crystallization behavior of ferric hydroxide on a media surface in an FBR was investigated. Ferric ion solutions (1 mM) were prepared at pHs 3 and 7, the saturation ratios of which are different by a factor of approximately 1012. The ferric ion solution was immediately poured into the FBR, which was operated in a batch mode using either SiO2, Al2O3, or CS as the fluidized material. Iron crystallization onto the fluidized particles did not occur at pH 7 regardless of the identity of the fluidized material, as shown in Fig. 4. On the other hand, an appreciable amount of iron crystallized onto CS at pH 3 (80% reduction in 3 h). Under strongly supersaturated condition at pH 7, nucleation, crystal growth, and crystal ripening of Fe (OH)3 can occur very rapidly via homogeneous nucleation without external inducing requirement as mentioned previously. Therefore, the presence of fluidized materials that have

much less active surface area than Fe(OH)3 nuclei did not have any significant involvement in the overall precipitation process. Thus, iron crystallization onto fluidized media did not occur at pH 7, Fe(OH)3 precipitates being formed were suspended thoroughly in the aqueous phase; hence, total iron did not change. On the other hand, the ferric ion solution at pH 3 was in the metastable oversaturated zone, where homogeneous nucleation occurs slowly. Hence, the presence of foreign surfaces could notably induce iron crystallization via heterogeneous nucleation. As a result, most of the Fe(OH)3 precipitates at pH 3 tended to be deposited on the surface of the fluidized materials and the remaining iron was decreasing toward its observed solubility of 0.18 mM (from Fig. 2). Our result is supported by Diz and Novak (1998), they crystallized Fe(OH)3 in a fluidized-bed reactor using quartz sand as the media and found the optimum pH for iron crystallization to be between 3 and 4. These findings suggest that the success in removing iron or other metals from the aqueous phase by using a precipitation and crystallization mechanism in an FBR can be achieved if the system is operated under the metastable saturated conditions of the metal.

3.3. Ferric hydroxide crystallization in a fluidized-bed Fenton process The iron crystallization in the fluidized-bed Fenton process was investigated. Fenton’s reagent (Fe2þ and H2O2) was added simultaneously into the FBR, which was operated in a batch mode at pH 3 with either CS or SiO2 serving as the fluidized material. It is well documented that most of the Fe2þ will rapidly transform into Fe3þ after the initiation of the Fenton reaction. The results demonstrate that the generated Fe3þ significantly crystallized onto the fluidized media, as shown in Fig. 5. Triplicate runs with CS and duplicate runs with SiO2 provided very similar results, indicating the reliability and accuracy of the experimental setup and procedure. With

SiO SiO2 2#1

1

0.8 CS at pH 7 Al Al2O3 atpH pH77 2O3 at

0.6

SiO SiO22 at at pH pH 7 CS at pH 3

0.4

Total iron (C/C0)

Total iron (C/C0 )

1

SiO2#2 CS#1

0.8

CS#2 CS#3

0.6

0.4

0.2

0.2

0

0 0

30

60

90 120 Time (min)

150

180

Fig. 4 e Iron crystallization in the fluidized-bed reactor with various fluidized materials and at different pHs (initial conditions: 1 mM Fe3D, 230.77 g LL1 media, 25  C, and 50% bed expansion; CS at pH 7 was run for 1 h).

0

30

60

90 120 Time (min)

150

180

Fig. 5 e Iron crystallization in the fluidized-bed Fenton process with various fluidized materials (initial conditions: 1 mM Fe2D, 20 mM H2O2, 230.77 g LL1 media, pH 3, 25  C, and 50% bed expansion).

3259

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 5 5 e3 2 6 2

(OH)3 crystals were well formed (aging and ripening to some extent as those precipitating at pH 7), no deposition onto foreign particles would be observed because the crystal formation would proceed toward homogeneous nucleation. This experimental part aimed to explore crystallization behavior during the transient stage where crystallites are formed after nucleation as the result of nuclei growth but have not yet ripened to become stable crystals. In this part of the experiment, the Fenton reaction was allowed to proceed in a completely mixed reactor (CMR) in the absence of solid particles for a period of time before switching to an FBR using CS as the seed media. By using this procedure, a significant amount of Fe(OH)3 crystallites should develop in the CMR before contact with the seed surfaces in the FBR. It was found that the crystallization rate onto the CS surface was faster for the fully fluidized-bed Fenton process than for the 1-h preCMRþFBR process for all Fenton’s reagent dosages, as shown in Fig. 6. In addition, in the case of 3 mM Fe2þ and 60 mM H2O2 supplementation, 5-min pre-CMRþFBR provided faster surface crystallization than 1-h pre-CMRþFBR (Fig. 6(c)). Nonetheless, it was believed that the iron removal efficiencies of these two operating modes would have been similar if the surface crystallization had been allowed to reach equilibrium. This result supports the previous findings that to optimize the iron removal in the fluidized-bed Fenton process, the crystallization of Fe(OH)3 onto the surface of fluidized particle surface should occur instantly or as soon as possible to obtain

Fully FBR 1 hr pre-CMR+FBR

1 0.8 0.6 0.4

a 1 mM Fe

2+

+ 20 mM H2O2

b 2 mM Fe

2+

+ 40 mM H2O2

0.2 0 1 Total iron (C/C0 )

respect to the type of fluidized material, it was found that iron crystallization onto the CS was faster than crystallization onto the SiO2, although the iron removal efficiencies at 180 min were comparable. The difference in crystallization rates might be due to impurities on the CS surface and/or to differences in the surface properties between these two materials. According to the XRF result, major components of the fresh CS are SiO2, Al2O3, K2O, and Fe2O3 which are accounted for 82.8%, 8.8%, 5.6%, and 1.3% by weight, respectively. Comparing the results from this part with those of the Fe3þ solution in the FBR with CS, shown in Fig. 4 (the initial total iron concentration was 1 mM in both cases), it can be seen that the iron crystallization rate in the Fenton reaction was more rapid, although the final amounts of iron removed at 180 min were almost the same. This result is comprehensible because Fe3þ was generated in situ by the reaction of Fe2þ with H2O2 in the Fenton experiment, whereas all of the iron present in the FBR without the Fenton reaction was initially in the form of Fe3þ, which would lead to some differences in iron nucleation. In the ordinary FBR, although most of the Fe(OH)3 nuclei being formed would proceed toward heterogeneous nucleation, some were still able to agglomerate together via homogeneous nucleation because all iron nuclei were formed instantly. In contrast, in the fluidized-bed Fenton reactor, Fe3þ would be formed over a period of time (very short but not instantaneous) after the Fenton reaction has been initiated; hence, metastable oversaturated ferric ion conditions was initially more prevailed in the Fenton reaction, and Fe(OH)3 nuclei should be deposited on the CS surface via heterogeneous nucleation more rapidly than in the Fe3þ supplement case. Nonetheless, the final amounts of iron removed were the same because they were both controlled by similar ferric ion solubility chemistry. Chou et al. (2001) successfully coated iron oxide on a brick grain support using the Fenton reaction at pH 3.5 in a circulating fluidized-bed reactor under a similar configuration as used in this study. Chou et al. (1999) examined iron crystallization in a fluidized-bed Fenton reactor using 50 mg L1 Fe2þ and 200 mg L1 H2O2 and found that total iron removal in 52 min increased from about 25% at pH 2.8e90% at pH 4.5 and then dropped substantially to 40% at pH 5. This result is quite different from that of this study (using 1 mM or 55.85 mg L1 Fe2þ and 20 mM or 680 mg L1 H2O2), where 80% iron reduction could be obtained in 3 h at pH 3. The difference is believed to be due to the results of differences in the material type and surface characteristics as well as the crystallization time. The results from this part of the study indicate that the fluidized-bed Fenton process in which Fe3þ is generated in-situ and immediately exposed to solid particles is theoretically very promising for iron removal via heterogeneous crystallization if proper conditions such as pH, type and amount of fluidized media, and contact time are provided.

Fully FBR 1 hr pre-CMR+FBR

0.8 0.6 0.4 0.2 0 1

Fully FBR 1 hr pre-CMR+FBR 5 min pre-CMR+FBR

0.8 0.6

c 3 mM Fe

2+

0.4 0.2

+ 60 mM H2O2

CMR w/o CS FBR with CS

3.4.

Effect of Fe(OH)3 crystallites

0 -60 -30

Up to this point, the results suggested that iron crystallization onto fluidized materials should perform optimally if foreign particles were introduced into a metastable oversaturated Fe3þ solution. The stronger metastable condition, the better and faster the Fe(OH)3 deposition onto foreign surfaces via heterogeneous nucleation should be. On the other hand, if Fe

0

30

60 90 120 150 180 Time (min)

Fig. 6 e Effect of crystallite formation on iron crystallization in the fluidized-bed Fenton process (initial conditions: 230.77 g LL1 CS, pH 3, 25  C, and 50% bed expansion; CMR [ completely mixed reactor; FBR [ fluidized-bed reactor).

3260

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 5 5 e3 2 6 2

the highest surface crystallization rate. In addition, this result also suggested that, as long as the Fe(OH)3 precipitates did not age or ripen into stable crystals, surface crystallization onto foreign seeds was still possible, though at a decreasing rate as the precipitates transformed from nuclei into crystallites.

3.5.

Effect of organic compounds

The primary purpose of the Fenton reaction is to generate hydroxyl radicals to oxidize organic pollutants to form less toxic intermediates, some of which can be sequentially transformed into carbon dioxide. These organic intermediates might interfere with ferric ion solubility and surface crystallization. Fig. 7 shows the effect of organic compounds on iron crystallization in the fluidized-bed Fenton process. The presence of aniline and 2,6-dimethylaniline retarded/interfered with the crystallization of Fe(OH)3 on CS surfaces. In addition, the total iron in the system with these two aromatics increased after 60 min under full FBR operation. This increase in the iron concentration could occur because several intermediates, particularly volatile fatty acids (VFA) that were produced from oxidation of the organic compounds by hydroxyl radicals, could react with Fe3þ to form soluble ferroorganic complexes. After a certain reaction period, the first and second generations of aromatic intermediates, such as 2,6 dimethylnitrobenzene, 2,6-dimethylphenol, and 2,6-dimethylbenzoquinone from 2,6-dimethylaniline oxidation (Boonrattanakij et al., 2009) and nitrobenzene, phenol, and benzoquinone from aniline oxidation (Brillas et al., 1998; Sauleda and Brillas, 2001), could be further transformed via ring cleavage to form volatile fatty acids including formic, acetic, and oxalic acids, all of which can form complexes with Fe3þ (Safarzadehet-Amiri et al., 1996; Richter et al., 1982; Hislop and Bolton, 1999). The presence of these acids could have increased the Fe3þ solubility. As a result, Fe(OH)3

precipitates that had been deposited on CS surfaces had to dissolve into to the aqueous phase to establish a new equilibrium condition. To verify this hypothesis, another experiment was conducted by allowing the Fenton reaction to proceed for 1 h in the CMR before switching to the FBR mode. Using this procedure, most organics in the solution should already be transformed into volatile fatty acids at the time of the FBR initiation. The results revealed that iron crystallization onto CS surfaces was very poor since the beginning, as illustrated in Fig. 7, which confirmed the previous observation. In addition, the impact of ferro-organic complexes was directly evaluated by mixing 1 mM Fe3þ with formic, acetic, and oxalic acids at the concentrations determined by Boonrattanakij et al. (2009) during the oxidation of 1 mM 2,6dimethylaniline and 1 mM aniline by the Fenton reaction. No iron crystallization occurred throughout the experimental period of 180 min. As a result, it can be concluded that volatile fatty acids, which are the primary organic products prior to conversion to carbon dioxide, could significantly reduce iron crystallization in the fluidized-bed Fenton process because these compounds could form soluble complexes with Fe3þ and, hence, increase the iron solubility. Chou et al. (1999) also found a decrease in the iron removal of the fluidized-bed Fenton process in the presence of benzoic acid. Therefore, to remove both organic compounds and iron at appreciable levels, the fluidized-bed Fenton process should be operated with the goal of complete mineralization in order to eliminate the interference from volatile fatty acids.

3.6. Crystallization repeatability and catalytic activity of Fe(OH)3-coated sand In real practice, fluidized particles will be used repeatedly for a certain period of time in a fluidized-bed Fenton reactor. Therefore, it is important to characterize the repeatability of the surface crystallization of Fe(OH)3 in a long-term application. Iron crystallization was performed 101 times in a fluidized-bed Fenton reactor with a 1 h hydraulic retention time

1.0

0.6

0.8

0.4 Fully FBR w/o org Fully FBR + org 1 hr pre-CMR+FBR + org Fully FBR with 1 mM Fe3+ + VFA w/o H2O2

0.2

0 -60

-30

0

30

60 90 Time (min)

120

150

Total iron (C/C0 )

0.8

Total iron (C/C0)

Total iron (C/C0 )

1 1st-cycle 101st-cycle

1 0.8 0.6 0.4 0.2 0 0

0.6

30 60 90 120 150 180 Time (min)

0.4

0.2

Effluent

180

Fig. 7 e Effect of organic compounds on Fe(OH)3 crystallization in the fluidized-bed Fenton process (initial conditions: 1 mM Fe2D, 20 mM H2O2, 230.77 g LL1 CS, pH 3, 25  C, and 50% bed expansion; org [ 1 mM 2,6dimethylaniline D 1 mM aniline; VFA [ 2 mM formic acid D 0.5 mM acetic acid D 2 mM oxalic acid).

0.0 0

20

40

60

80

100

Cycle

Fig. 8 e Repeatability of iron crystallization on CS in the fluidized-bed Fenton process (initial conditions: 2 mM Fe2D, 20 mM H2O2, 230.77 g LL1 CS, pH 3, 25  C, 50% bed expansion, and 1 h crystallization period).

3261

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 5 5 e3 2 6 2

Goethite Goethite 0.075 g-1g/l L-1 Goethite g/lg L Goethite 11.0 Goethite Goethite 75 g/l g L-1

-1 Goethite 0.75 gg/lL Goethite 7.5 g/l g L-1 101st-cycle Fe(OH)3-coated CS 75 g L-1

[Aniline] (C/C0)

1 0.8 0.6 0.4 0.2

[2,6-dimethyl-aniline] (C/C0 )

(iron crystallization was found to occur most rapidly within the first hour); the results are shown in Fig. 8. The performance of iron crystallization slightly decreased as the cycle number increased. This decrease was due to a reduction in the crystallization rate, as shown by the time profiles of iron crystallization using fresh and 101st-cycle CS (Fig. 8). Hence, to maintain high iron removal performance, it may be necessary to continuously replace a portion of the Fe(OH)3-coated CS in the fluidized-bed Fenton reactor with fresh CS. From the mass balance of iron, the amount of iron coated on the 101st-cycle CS surface was approximately 3.5% by weight, and the color of the CS had changed tremendously, from the original light orange to dark-red at the end of the 101st cycle. The 101st-cycle ironcoated CS was removed from the FBR, air dried under room conditions in order to prevent any phase transformation by either heat or sunlight, and analyzed for the crystal structure. Fig. 9 shows XRD profiles of fresh CS and the 101st-cycle ironcoated CS. It reveals that fresh CS mainly consisted of SiO2 whereas the 101st-cycle iron-coated CS showed some peak intensities possibly corresponding to Fe(OH)3. No other iron oxide species was detected in the XRD profiles. In addition to the well-formed crystals, coated Fe(OH)3 could also be possibly present in noncrystalline structure as observed by others (Amonette and Rai, 1990; Nagano et al., 1992; Cudennec and Lecerf, 2006). Iron oxides in various forms are believed to have catalytic activity that can stimulate the decomposition of H2O2 to generate hydroxyl radicals. It was interesting to determine the catalytic activity of Fe(OH)3-coated CS and to compare it with that of commercial a-FeOOH. The results shown in Fig. 10 indicate that 75 g of the 101st-cycle Fe(OH)3coated CS had a catalytic activity approximately equivalent to that of only 2 g of a-FeOOH in terms of aniline and 2,6-dimethylaniline oxidation. Although the commercial a-FeOOH had a higher BET surface area than the 101st-cycle Fe(OH)3-coated CS (13.51 versus 8.63 m2/g, respectively), this enormous difference in the catalytic activity was mainly to the result of the effects of iron oxide species. It is well documented that Fe (OH)3 has a much lower catalytic activity than a-FeOOH (Valentine and Wang, 1998). Nonetheless, the main purpose of

0 1

0.8 0.6 0.4 0.2 0 0

30

60

90 120 Time (min)

150

180

Fig. 10 e Catalytic activity of 101st-cycle Fe(OH)3-coated CS and commercial a-FeOOH in the heterogeneous Fenton process (initial conditions: 1 mM aniline, 1 mM 2,6dimethylaniline, 20 mM H2O2, pH 3, and 25  C).

iron crystallization in the fluidized-bed Fenton process is to reduce the volume of puffy Fe(OH)3 sludge, which requires further costly treatment. Through crystallization onto solid surfaces, the amount of Fe(OH)3 can be reduced significantly because ordinary Fe(OH)3 cake after dewatering still contains up to 80% water, whereas Fe(OH)3 deposited on fluidized media contains almost no water after a short drying period. In addition, the activity of Fe(OH)3 is not of concern because most H2O2 will decompose to form hydroxyl radicals via the homogeneous Fenton reaction in the presence of Fe2þ from the external supplementation with inexpensive ferrous salts.

SiO2 Fe(OH)3

4.

Conclusions

Intensity

In this paper, the mechanisms of iron precipitation and crystallization in a fluidized-bed Fenton process were clearly revealed. The main conclusions of this work are as follows: 101st-cycle iron-coated CS

fresh CS

20

30

40

50 60 2-Theta (degree)

70

80

Fig. 9 e XRD profiles of fresh CS and 101st-cycle Fe(OH)3coated CS.

- Due to their very limited solubility, ferric ions as a product from ferrous ion oxidation will subsequently precipitate out in the form of ferric hydroxide after the initiation of the Fenton reaction and will simultaneously crystallize onto the fluidized media present in the reactor if the precipitation proceeds toward heterogeneous nucleation. - Prolong aging/ripening period of the crystallites will slow down the crystallization rate of ferric hydroxide onto solid surfaces; hence, the fluidized materials have to be present at the beginning of the Fenton reaction.

3262

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 5 5 e3 2 6 2

- To achieve superior performance on both organic and iron removals, the Fluidized-bed Fenton process should be operated with the goal of complete mineralization in order to eliminate the interference from organic intermediates that can significantly increase the ferric solubility by forming complexes with ferric ions. - It is necessary to continuously replace a portion of the ironcoated bed with fresh media to maintain the iron removal performance since the crystallized iron can retard the successive crystallization rate. - Coated iron also has a catalytic property, though was less than those of commercial goethite.

Acknowledgment This research was financially supported by the Thailand Research Fund through the Royal Golden Jubilee Ph.D. Program (Grant No. PHD/0056/2549), by the National Center of Excellence for Environmental and Hazardous Waste Management of Thailand, by the National Research University Project of Thailand’s Office of the Higher Education Commission, and by the National Science Council of Taiwan (Grant No. NSC96-2628-E-041-001-MY3).

references

Amonette, J.E., Rai, D., 1990. Identification of noncrystalline (Fe, Cr)(OH)3 by infrared spectroscopy. Clays Clay Miner. 38, 129e136. Anotai, J., Sakulkittimasak, P., Boonrattanakij, N., Lu, M.C., 2009. Kinetics of nitrobenzene oxidation and iron crystallization in fluidized-bed Fenton process. J. Hazard. Mater. 165, 874e880. APHA, AWWA, WEF, 1992. Standard Methods for the Examination of Water and Wastewater, eighteenth ed.. American Public Health Association, Washington D.C. Benjamin, M.M., 2002. Water Chemistry. McGraw-Hill Companies, Inc., New York. Boonrattanakij, N., Lu, M.C., Anotai, J., 2009. Kinetics and mechanism of 2,6-dimethyl-aniline degradation by hydroxyl radicals. J. Hazard. Mater. 172, 952e957. Brillas, E., Mur, E., Sauleda, R., Sanchez, L., Peral, J., Domenech, X., Casadi, J., 1998. Aniline mineralization by AOP’s: anodic oxidation, photocatalysis, electro-Fenton and photoelectroFenton processes. Appl. Catal. B. Environ. 16, 31e42. Canizares, P., Paz, R., Saez, C., Rodrigo, M.A., 2009. Costs of the electrochemical oxidation of wastewaters: a comparison with

ozonation and Fenton oxidation processes. J. Environ. Manage. 90, 410e420. Chou, S., Huang, C., Huang, Y.H., 1999. Effect of Fe2þ on catalytic oxidation in a fluidized-bed reactor. Chemosphere 39, 1997e2006. Chou, S., Huang, C., Huang, Y.H., 2001. Heterogeneous and homogeneous catalytic oxidation by supported g-FeOOH in a fluidized-bed reactor: kinetic approach. Environ. Sci. Technol. 35, 1247e1254. Cudennec, Y., Lecerf, A., 2006. The transformation of ferrihydrite into goethite or hematite, revisited. J. Solid State Chem. 179, 716e722. Diz, H.R., Novak, J.T., 1998. Fluidized bed for removing iron and acidity from acid mine drainage. J. Environ. Eng. 124, 701e708. Hislop, K.A., Bolton, J.B., 1999. The photochemical generation of hydroxyl radicals in the UVevis/ferrioxalate/H2O2 system. Environ. Sci. Technol. 33, 3119e3126. Kolthof, I.M., Sandell, E.B., Meehan, E.J., Buckstein, S., 1969. Quantitative Chemical Analysis, fourth ed. Macmillan, New York. Kwan, W.P., Voelker, B.M., 2002. Decomposition of hydrogen peroxide and organic compounds in the presence of dissolved iron and ferrihydrite. Environ. Sci. Technol. 36, 1467e1476. Loan, M., Parkinson, G.M., Richmond, W.R., 2005. The effect of zinc sulfide on phase transformations of ferrihydrite. Am. Mineral 90, 258e261. Morel, F.M.M., Hering, J.G., 1993. Principles and Applications of Aquatic Chemistry. John Wiley & Sons, Inc., New York. Nagano, T., Nakashima, S., Nakayama, K., Osada, K., Senoo, M., 1992. Color variations associtated with rapid formation of goethite from proto-ferrihydrite at pH 13 and 40 C. Clays Clay Miner 40, 600e607. Pignatello, J.J., 1992. Dark and photoassisted Fe3þ-catalyzed degradation of chlorophenoxy herbicides by hydrogen peroxide. Environ. Sci. Technol. 26, 944e951. Richter, H.W., Fetrow, M.A., Lewis, R.E., Waddell, W.H., 1982. Reactive species produced by the 5-methylphenazinium methyl sulfate/reduced b-nicotinamide adenine dinucleotide/ oxygen system in the hydroxylation of benzoic acid. J. Am. Chem. Soc. 104, 1666e1671. Safarzadehet-Amiri, A., Bolton, J.R., Cater, S.R., 1996. Ferrioxalatemediated solar degradation of organic contaminants in water. Solar Energy 56, 439e443. Sauleda, R., Brillas, E., 2001. Mineralization of aniline and 4chlorophenol in acidic solution by ozonation catalyzed with Fe2þ and UVA light. Appl. Catal. B. Environ. 29, 135e145. Sedlak, D.L., Andren, A.W., 1991. Oxidation of chlorobenzene with Fenton’s reagent. Environ. Sci. Technol. 25, 777e782. Stumm, W., Morgan, J.J., 1996. Aquatic Chemistry: Chemical Equilibria and Rates in Natural Waters, third ed. John Wiley & Sons, Inc., New York. Valentine, R.L., Wang, H.C.A., 1998. Iron oxide surface catalysed oxidaiton of quinoline by hydrogen peroxide. J. Environ. Eng. 124, 31e38.

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 6 3 e3 2 6 9

Available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/watres

Comparison of advanced oxidation processes for the removal of natural organic matter Rupa Lamsal, Margaret E. Walsh, Graham A. Gagnon* Department of Civil & Resource Engineering, Dalhousie University, Halifax, NS B3J 1Z1, Canada

article info

abstract

Article history:

This study examined the impact of UV, ozone (O3), advanced oxidation processes (AOPs)

Received 22 October 2010

including O3/UV, H2O2/UV H2O2/O3 in the change of molecular weight distribution (MWD)

Received in revised form

and disinfection by-product formation potential (DBPFP). Bench-scale experiments were

18 March 2011

conducted with surface river water and changes in the UV absorbance at 254 nm (UV254),

Accepted 21 March 2011

total organic carbon (TOC), trihalomethane and haloacetic acid formation potential

Available online 29 March 2011

(THMFP, HAAFP) and MWD of the raw and oxidized water were analyzed to evaluate

Keywords:

TOC and UV254 reduction than the individual processes. The O3/UV process was found to be

treatment performance. Combination of O3 and UV with H2O2 was found to result in more Advanced oxidation

the most effective AOP for NOM reduction, with TOC and UV254 reduced by 31 and 88%,

processes (AOPs)

respectively. Application of O3/UV and H2O2/UV treatments to the source waters organics

Ozone (O3)

with 190e1500 Da molecular weight resulted in the near complete alteration of the

Ultraviolet radiation (UV)

molecular weight of NOM from >900 Da to <300 Da H2O2/UV was found to be the most

Natural organic matter (NOM)

effective treatment for the reduction of THM and HAA formation under uniform formation

Disinfection by-products (DBPs)

conditions. These results could hold particular significance for drinking water utilities with

High performance size exclusion

low alkalinity source waters that are investigating AOPs, as there are limited published

chromatography (HPSEC)

studies that have evaluated the treatment efficacy of five different oxidation processes in parallel. ª 2011 Elsevier Ltd. All rights reserved.

1.

Introduction

Natural organic matter (NOM) is a complex heterogeneous mixture of different organic compounds with varying molecular size and properties. A common drinking water treatment goal is to remove NOM as it is a precursor for unwanted disinfection by-products (DBPs) during chemical disinfection processes, such as chlorine (Edzwald et al., 1985; Mosteo et al., 2009) and ozone (Gagnon et al., 1997; Schechter and Singer, 1995). NOM has also been shown to contribute to fouling on membrane surfaces (e.g., Hong and Elimelech, 1997; Her et al., 2008), the production of biologically unstable water (Rittmann and Snoeylink, 1984) and other unwanted water quality issues

such as metal complexes (Ravichandran et al., 1998; Schmitt et al., 2002). The application of advanced oxidation processes (AOPs) has gained significant interest in the drinking water industry as an additional tool for removing NOM and minimizing the formation of DBPs in drinking water (Zhou and Smith, 2001; Chin and Be´rube´, 2005). Previous studies have focused on ozone (O3) (e.g. Gagnon et al., 1997), ultraviolet radiation (UV) (e.g. Chin and Be´rube´, 2005; Thomson et al., 2002) and AOPs including hydrogen peroxide (H2O2) in combination with UV (H2O2/UV) (e.g. Toor and Mohseni, 2007; Wang et al., 2006), O3 in combination with UV (O3/UV) (Amirsardari et al., 2001; Chin and Be´rube´, 2005) and H2O2 in combination with O3 (H2O2/O3)

* Corresponding author. Tel.: þ1 902 494 3268; fax: þ1 902 494 3108. E-mail address: [email protected] (G.A. Gagnon). 0043-1354/$ e see front matter ª 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.watres.2011.03.038

3264

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 6 3 e3 2 6 9

(e.g. Kleiser and Frimmel, 2000) to evaluate the potential for NOM reduction and the mitigation of DBP formation in finished water. Matilainenm and Sillanpa¨a¨ (2010) have provided a thorough review of published oxidation and AOPs studies that have been conducted on both natural and synthetic test waters. However, these studies have primarily focused on evaluating one or two oxidation or AOPs for NOM reduction. This study goes beyond the previously published studies by directly comparing the treatment efficacy of five different oxidation processes in parallel, in terms of changes to molecular weight distribution (MWD) of the source water, NOM reduction and subsequent minimization of disinfection by-product formation potential (DBPFP) from a low turbidity, highly colored surface water. During advanced oxidation treatment, hydroxyl radicals (HO) are formed which act as a strong oxidant and transform NOM. Westerhoff et al. (2007) directly measured the rate constants for reactions between HO radicals and seven dissolved organic matter (DOM) isolates from different sources and observed rate constants in range from 1e5  108 M1S1, which is three to four orders of magnitude higher than for chlorine and ozone (Crittenden et al., 1999). Hydroxyl radicals produced during AOPs are capable of reducing total organic carbon (TOC) concentrations and DBPFP of raw water (Amirsardari et al., 2001; Chin and Be´rube´, 2005; Kusakabe et al., 1990; Sierka and Amy, 1985; Glaze et al., 1982). Under strong advanced oxidation conditions (i.e. long irradiation time and/or higher H2O2 concentrations) NOM is mineralized, indicated by a decrease in TOC and DBPFP (Kleiser and Frimmel, 2000; Wang et al., 2006; Toor and Mohseni, 2007). However, such strong treatment conditions may not be economically feasible, and in commercial applications, low or moderate advanced oxidation conditions are applied. Under these conditions, NOM is partially oxidized and higher molecular weight compounds are transformed into smaller and more biodegradable compounds such as aldehydes and carboxylic acids (Backlund, 1992; Edwards and Benjamin, 1992; Gagnon et al., 1997; Sarathy and Mohseni, 2007). Such changes in the chemical characteristic of NOM also result in reducing TOC concentrations and/or alter the characteristics of the DBP precursor material potentially reducing its reactivity with chlorine. The objective of this study was to compare O3, UV and three AOPs including H2O2/O3, H2O2/UV and O3/UV for NOM

removal and assess the impact on modifying the MWD of NOM following treatment. This study was conducted using laboratory-controlled conditions with a natural surface water source that has a low alkalinity (<5 mg/L as CaCO3) and moderate level of total organic carbon (TOC of 3e4 mg/L). The effectiveness of each treatment process was evaluated by traditional metrics for NOM; namely, UV absorbance at 254 nm (UV254), TOC concentration, specific UV absorbance (SUVA), trihalomethane formation potential (THMFP), and haloacetic acid formation potential (HAAFP). In addition, the MWD following each treatment was assessed using high performance size exclusion chromatography (HPSEC) analysis. HPSEC has been demonstrated to be an effective technique for determining the MWD of NOM (Pelakani et al., 1999). Determination of the MWD of NOM provides information on the specific fraction of NOM that plays important role in DBP formation (Amy et al., 1987; Chang and Young, 2000) and membrane fouling potential during water treatment (e.g. Her et al., 2008).

2.

Materials and methods

2.1.

Source water characterization

Surface water collected from the French River, which provides the drinking water in a northern shore community in Nova Scotia, Canada, was used for the bench-scale study. The French River water is characterized by its low alkalinity (<5 mg CaCO3/L), low turbidity (<1.5 NTU), and high color level (>35 PteCo). The French River has general characteristics that are similar to other surface water sources in Nova Scotia and Atlantic Canada (Waller et al., 1996).

2.2.

Experimental set-up

The laboratory scale batch set-up for the ozone experiments used in this study is shown in Fig. 1. It consists of a compressed air system, ozone generator, a contactor (reactor) and off-gas collection system. The reactor was a glass tank with a working volume of 10 L (0.305 cm diameter  0.41 cm height). The inflow and outflow of the ozone gas line in the reactor was fitted with a laboratory stopper (Fisher scientific # 14141R) at the top of the reactor and sample was

Fig. 1 e Schematic of laboratory set-up for ozone experiment.

3265

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 6 3 e3 2 6 9

taken from the bottom of the reactor. Compressed air with a flow rate of 2 L/min was passed into the ozone generator (VMUS-4), where high voltage corona discharge causes break down of oxygen molecules into radicals that combine with oxygen molecules to form ozone. Ozone was bubbled into the base of the reactor using a fine bubble diffuser at a flow rate of 2 L/min and pressure of 15 psi. A potassium iodide solution (20 g KI in 1 L water) was used to collect the residual ozone in the off-gas from the reactor. The ozone experiments were conducted in a semi-batch mode by continuously passing O3 gas in a 10 L reactor directly with 3 L sample volume at room temperature (23  C) for 30 min. Ozonation for 30 min showed better performance than for lower treatment times (e.g., 5 and 15 min) and similar performance to higher treatment times (45 min and 60 min). The concentration of ozone consumed during the 30 min reaction time was 4.04  0.11 mg/L. The detailed procedure for ozone dose calculation is provided in the supplementary material. For the H2O2/O3 experiments, 23 mg/L of hydrogen peroxide solution (50% Fisher Scientific) was mixed with 3 L of the raw water for approximately 5 min and the mixture was ozonated for 30 min, similar to the treatment times used in the ozone experiments. The concentration of H2O2 was chosen following previous studies using surface water (Sarathy and Mohseni, 2007; Goslan et al., 2006; Toor and Mohseni, 2007). Toor and Mohseni (2007) have reported that at lower concentrations of H2O2, the H2O2/UV AOP was not effective for reducing DBP precursors. A low pressure ultraviolet lamp (Trojan UV Max.) with 43 W power was used during the UV experiment. The dimension of the chamber assembly was 0.495 m  0.09 m, and the length of the lamp (i.e., sleeve length) was 0.405 m. The UV reactor is a glass tube with a working volume of approximately 2 L. Raw water was pumped into the reactor at a flow rate of 167 mL/ min using a masterflex pump to achieve the maximum UV dose delivered by the lamp (e.g.,>1000 mJ/cm2). The delivered UV dose in the UV and UV based AOP experiments was 1140 mJ/cm2, which was determined by using potassium ferrioxalate actinometer. Additional information on the UV dose calculation is provided in the supplementary material. For the H2O2/UV AOP experiments, 23 mg/L of H2O2 was first mixed with 3 L of the raw water for 5 min. The mixture was then pumped through the UV reactor at the same flow rate as that of UV process experiments. For the O3/UV combined AO process, the raw water sample was ozonated for 30 min and then pumped through the UV reactor to achieve the 1140 mJ/ cm2 UV dosage. The oxidation processes evaluated in this study utilized higher dosages of oxidants than typically applied in drinking water treatment for optimum removal of

NOM and DBPFP, consistent with earlier studies (Chin and Be´rube´, 2005; Toor and Mohseni, 2007).

2.3.

Analytical methods

In this study, NOM was quantified by measuring UV254, TOC concentrations, and dissolved organic carbon (DOC) concentrations. Specific UV absorbance (SUVA) which can be used as a surrogate parameter to monitor the changes in aromatic nature of NOM in water was calculated from UV254 and DOC as outlined by Edzwald et al. (1985). In addition, DBPFP was determined for THMs and HAAs using the uniform formation conditions (UFC) methodology (Summers et al., 1996). Finally, NOM characterization included analysis using high performance size exclusion chromatography (HPSEC) (PerkinElmer, Series 200) with a UV/VIS detector to determine the molecular weight distribution of NOM.A detailed description of the analytical test procedures used for this research is provided in the supplementary material. Also, included in the supplementary material is a methodology for the general water quality parameters (e.g., pH, turbidity) that were measured as part of this study.

3.

Results and discussion

3.1.

Characterization of the source water

The majority of organic carbon of the French River was in the dissolved fraction, as demonstrated by the TOC (3.10  0.3 mg/ L) and DOC (2.85  0.13 mg/L) measurements. The UV254 value was 0.090  0.003 cm1 (Table 1). The SUVA for the French River water was 3.2 m1/(mg/L), which indicates that the source water contained a mixture of hydrophobic and hydrophilic NOM fractions (Owen et al., 1995). The chromatogram used to determine the MWD of the source water is presented in Fig. 2. The total area of the sample was integrated using Totalchrom software (PerkinElmer, Ontario, Canada) to obtain the entire MWD of NOM in the sample (Fig. 3). The relationship between the molecular weight of organic compounds and their retention time was determined by log-linear regression between log molecular weight and retention time. HPSEC analysis showed that the French River raw water consists of four different MW fractions: 1246, 690, 478 and 292 Da. The highest percentage area of chromatogram (i.e., 66%) was observed with the 1246 Da MW fraction. Earlier studies have proposed that compounds having 1000e1500 Da MW range likely represent

Table 1 e Mean and standard deviations of water quality parameters and dosages used in each treatment processes. Raw UV254 (cm1) TOC (mg/L) DOC (mg/L) SUVA (m1/(mg/L)) Dosages

0.092  3.10  2.85  3.16 

0.003 0.325 0.131 0.170

UV 0.080  0.003 2.99  0.049 2.73  0.072 2.93  0.131 1140 mJ/cm2

O3 0.039  2.92  2.79  1.43  4.04 

0.007 0.053 0.175 0.023 0.110 mg/L

O3/UV 0.011  0.004 2.12  0.116 2.0  0.723 0.55  0.006 4.04  0.110 mg/L 1140 mJ/cm2

3266

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 6 3 e3 2 6 9

Fig. 2 e High performance size exclusion chromatography (HPSEC) chromatograms of raw and oxidized waters.

humic and fulvic acids (e.g. Huber and Frimmel, 1996), which is supported by the measured SUVA value (3.19 m1/mg/L) in the French River. The lower proportion of intermediate and low MW fractions present in the French River water may represent simple aromatic compounds or fulvic acids as described by Her et al. (2002, 2008).

3.2.

Impact of advanced oxidation on TOC and SUVA

In the AOP experiments, the reduction of NOM was attributed to chemical oxidation of NOM present in the raw water by hydroxyl radicals (HO). However, these HO can also react with carbonate and bicarbonate ions which are typically present in raw water. This reaction significantly reduces the amount of HO radicals available for oxidation of NOM (Gottschalk et al., 2000). Since the source water used in this study has low alkalinity, the impact of carbonate and bicarbonate ions on the resulting concentration of HO was expected to be negligible. A reduction in UV254 absorbance was observed after each treatment process is presented in Table 1. UV radiation at 254 nm is mainly absorbed by aromatic compounds and conjugated double bonds (Singer, 1999). Therefore, reduction in UV254 indicates a loss of aromatic and conjugated double bond structures of NOM (Owen et al., 1995). However, the observed impact of the treatment processes on TOC concentration was less because of the partial oxidation of NOM to other intermediate by-products (Table 1). UV treatment on its own had minor impact upon the UV254 absorbance and almost no impact on TOC concentration. The impact of UV or H2O2

alone has been found to be negligible for NOM reduction in other studies (Chin and Be´rube´, 2005). The rate of UV254 and TOC reduction increased significantly when H2O2 was combined with UV. UV in combination with H2O2 promotes the formation of HO, as reported by other researchers (Wang et al., 2006; Toor and Mohseni, 2007). The UV254 absorbance and TOC concentrations decreased from 11 to 60% and 3e23%, respectively, with UV treatment in the presence of H2O2 compared to UV treatment on its own. Sarathy and Mohseni (2007) also observed significant reduction of UV254 without TOC reduction in experiments with H2O2/UV AOP with a UV dose of 1500 mJ/cm2 and H2O2 concentrations up to 20 mg/L. The ozone and H2O2/O3 treatment processes reduced the raw water TOC concentration by 6 and 10%, respectively. However, as presented in Table 1, UV254 absorbance was observed to be reduced by 57% with O3 treatment and 59% with H2O2/O3 treatment. The higher reduction of UV254 absorbance with less reduction of TOC demonstrates the removal of conjugated double bonds with minimal mineralization. Increased NOM oxidation in H2O2/O3 process, as compared to the O3 process alone, was a result of more HO formation. Combination of ozone with UV (O3/UV) reduced TOC and UV254 by 31 and 88%, respectively. The higher reduction of UV254 achieved can be explained by the NOM reaction with O3. In all O3 experiments, the samples were treated with high doses of O3, and as a result, double bonds were oxidized and UV254 was reduced. Moreover, since the ozone concentration evaluated in this study is much higher than hydroxyl radical levels, ozone is expected to be the main degradation pathway for NOM. In O3/UV AOP systems evaluated in other studies, mineralization of organic carbon was also observed (Amirsardari et al., 2001; Kusakabe et al., 1990; Glaze et al., 1982; Sierka and Amy, 1985). Chin and Be´rube´ (2005) evaluated the O3/UV AOP with an O3 dose of 4 mg/L and a UV dose of 0.13 W/cm2 on raw water characterized with 1.3e3.2 mg/L TOC concentrations. That study found approximately 15% mineralization of the TOC in the raw water after O3/UV treatment, although the UV dose evaluated (approx. 130 mJ/ cm2) was much lower than that used in this study (i.e., 1140 mJ/cm2). The increased mineralization observed in O3/ UV AOP compared to H2O2/O3 and H2O2/UV AOPs may be due to a larger yield of hydroxyl radical per oxidant compared to other advanced oxidation processes (Gottschalk et al., 2000; Oh et al., 2003). Since production yields of HO in each oxidation process were not measured during this study, further work would be required to verify this theory.

3.3.

Fig. 3 e Chromatogram area counts for raw and oxidized waters for different molecular weight compounds.

Impact of oxidation on NOM molecular weight

HPSEC chromatograms for the AOP test waters are presented in Fig. 2. Higher MW organics are eluted from the column first and lower MW organics are eluted later. The peak area of chromatogram represents the intensity of UV absorbance of the sample detected by the UV detector at 254 nm. Therefore, these peaks are indication of the presence of aromatic or double bond organic compounds. Prior to application of the treatments evaluated in this study, the HPSEC chromatogram of raw water featured a large peak, and the total area under the HPSEC chromatogram decreased with the application of the different

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 6 3 e3 2 6 9

treatment processes. These results demonstrate the oxidation of aromatic or double bond organic matter into lower molecular weight compounds after treatment. The MWD of each process observed in Fig. 2 was translated into quantitative terms using Totalchrom software available with the HPSEC instrument and is presented in Fig. 3. The UV treatment process showed minor impact on the MWD of the source water NOM. Ozone was found to reduce the >900 Da and 600e900 Da MW fractions of organics by 25 and 68%, respectively. As described earlier, UV254 and TOC were found to be reduced by 57 and 6%, respectively, after O3 treatment, demonstrating that NOM oxidation occurred with removal of conjugated double bonds with minimal mineralization. Frimmel et al. (2000) also observed that ozone treatment decreased the absorbance of Ruhr River water with minimal mineralization and found decreases in the higher MW fractions with concomitant increase in the lower MW fractions. However, minimal increase in the lower MW (i.e. 13% of 300e600 Da) fraction of NOM was observed during this study. Fig. 3 shows that the reduction of larger MW NOM was higher than that of lower MW NOM. The preferential reduction of larger MW organic matter, in comparison to lower MW organic matter, could be a result of the higher reaction rate constant between HO and the larger MW compounds (Thomson et al., 2004). Higher MW compounds tend to be more aromatic in nature, so they may have a larger number of reaction sites than smaller MW compounds. Thomson et al. (2004) also explained that higher MW compounds react fastest as they have higher molar absorptivities than lower MW compounds. Westerhoff et al. (1999) observed the positive correlation between molecular weights and aromaticity and the reaction rate constant between HO and NOM. Ozone in combination with UV showed complete removal of MW NOM >900 Da. However, there was no observed increase in the formation of lower MW NOM. Similarly, when H2O2 was combined with the UV process, the H2O2/UV AOP reduced the >900 Da MW fractions by 85% and the 600e900 Da MW fractions by 100% without any observed increase in lower MW fractions of the NOM. This is in contrast to previous studies that have shown significant reduction of larger MW NOM in combination with an increase in lower MW NOM (Sarathy and Mohseni, 2007). However, that study performed HPSEC analysis at 260 nm to detect the chromophoric NOM only. Observation of 23% TOC reduction versus 60% UV254 reduction with the H2O2/UV AOP implies increase in lower MW NOM. However, the HPSEC analysis used in this study did not provide further information for single bond organic carbon since the UV detector of the HSPEC instrument only measures the aromatic or double bond organics, making the direct relationship and quantification between HPSEC and TOC results difficult.

3.4.

3267

Fig. 4 e THMs for raw and oxidized waters (Vertical bars represent 2s levels).

are aliphatic as discussed in detail in Bond et al. (2009) and Hong et al. (2009). The treatment by AOPs tends to decrease the aromaticity of NOM, therefore, the decrease in THMFP is greater than that of HAAFP caused by the larger removal of THMs precursor materials. In both the raw and oxidized waters, chloroform formed the majority of trihalomethane species, followed by dichlorobromomethane and dibromochloromethane. The concentrations of bromoform were below the detection limit (zero) in all samples. The majority of HAA species measured in the raw water were dichloroacetic acid, bromochloroacetic acid, chloroacetic acid. Dibromoacetic acid and bromodiacetic acid concentrations were found to be below detection limit. The results of the UV treatment process showed 15% reduction of THMFP and no reduction of HAAFP. The little to no reduction of THMFP and HAAFP may be due to the minor impact of UV radiation on UV254 reductions and changes in MWD of organics. These observations are consistent with those found in other studies (e.g., Chin and Be´rube´, 2005), where it has also been demonstrated that UV treatment on its own is ineffective at reducing THMFP and HAAFP. When H2O2 was combined with UV, THMFP and HAAFP were reduced by 77 and 62%, respectively. The increased reduction of THMFP and HAAFP observed with the H2O2/UV AOP agrees with the increased reduction of UV254 achieved with H2O2/UV treatment (e.g., 60%) compared to the moderate 11% reduction in

Impact of advanced oxidation on DBP formation

The total trihalomethane formation potential (THMFP) and haloacetic acid formation potential (HAAFP) of the raw and oxidized waters are shown in Figs. 4 and 5, respectively. In general, THMFP removal was greater than HAAFP in the oxidation processes studied. The precursor materials for THMs tend to be aromatic whereas HAAs precursor materials

Fig. 5 e HAAs of raw and oxidized waters (Vertical bars represent 2s levels).

3268

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 6 3 e3 2 6 9

UV254 achieved with UV treatment alone. MWD results also showed that H2O2/UV AOP resulted in increased reduction of >900 Da MW fractions by 85% and the complete reduction of 600e900 Da MW fractions of NOM. Such a decrease in THMFP and HAAFP was also observed with 23 mg/L initial H2O2 concentration and UV dose higher than 1500 mJ/cm2 in a study conducted by Toor and Mohseni (2007) with similar source water qualities. Liu et al. (2002) reported reduction of both THMFP and HAAFP with UV dose of 1000 mJ/cm2 or higher and initial H2O2 concentration of 100 mg/L. These studies have suggested that a combination of high UV dose and H2O2 concentration is required for the potential generation of higher levels of HO radicals and hence the reduction of THMFP and HAAFP. The study conducted by Toor and Mohseni (2007) also demonstrated a significant reduction of H2O2 concentration in the solution, indicating the generation of HO radicals that consequently oxidized DBP precursors and reduced the THMFP and HAAFP of the source water. Significant reduction of THMFP and HAAFP with H2O2/UV treatment was also observed in this study, which indicates a decrease in H2O2 concentration in the solution. In contrast to UV treatment alone, ozone treatment showed a higher reduction of THMFP and HAAFP (i.e., 69 and 8%, respectively). The increased percent reduction of THMFP with the O3 process is supported by increased reduction of UV254 (57%), indicating strong correlation between UV254 and THMFP reductions (Edzwald et al., 1985). The MWD results also showed the increased reduction of UV absorbing organics. These results are consistent with previous studies (e.g., Hu et al., 1999; Westerhoff et al., 1999; Galapate et al., 2001). The reduction of THMFP and HAAFP of the ozonated samples can be explained by the reaction pathway for DBPs. Similar to chlorine, ozone reacts by addition to the aromatic system and once the aromatic double bonds are consumed by ozone, fewer sites are available for chlorine addition. Chlorine addition to the double bond is a main pathway for DBP production. In the H2O2/O3 AOP experiments, the THMFP and HAAFP were reduced by 70 and 31%, respectively. For treatments involving O3 in combination with UV, THMFP was reduced by 75% and HAAFP was reduced by 52%. Glaze et al. (1982) observed that the combined application of O3 and UV was more effective than ozone alone for the destruction of THM precursors in two southern U.S. surface water sources. Other researchers have also reported significant reduction of THMFP and HAAFP during treatment with the O3/UV process (Sierka and Amy, 1985; Chin and Be´rube´, 2005). Overall, the results of this study found that H2O2/UV showed improved precursor reduction of 77% for THMFP and 62% for HAAFP, compared to the reduction of 75% for THMFP and 52% for HAAFP in O3/UV, and the reduction of 70% for THMFP and 31% for HAAFP in H2O2/O3.

4.

Conclusions

This study evaluated O3, UV, and three advanced oxidation processes including H2O2/O3, H2O2/UV and O3/UV for the removal of natural organic matter and reduction in DBP formation potential of the treated source water. Bench-scale experiments demonstrated that the ozone and UV treatment

processes alone showed less impact on TOC reduction compared to the combined AOPs of H2O2/O3 and H2O2/UV. However, O3 showed significant reduction of UV254. The O3/UV AOP showed increased performance reducing UV254 by 88% and TOC by 31% compared to the other oxidation processes evaluated. The H2O2/UV process reduced UV254 by 60% and TOC by 23%, achieving somewhat lower reductions than the O3/UV process. Further study with measurement of product yield in each oxidation process would help for better explanation of the results. The HPSEC analysis showed that the molecular weight (MW) of the organic compounds that are able to absorb UV light at 254 nm in the source water ranged from 190 to 1500 Da. Overall, the application of the oxidation processes evaluated in this study resulted in the reduction of higher MW NOM, with the O3/UV and H2O2/UV AOPs having the largest impact on MW transformation of the source water. Treatment with the H2O2/UV AOP resulted in the largest reduction of THMFP (77%) and HAAFP (62%) compared to the other treatment processes evaluated. Similarly, treatment with the O3/UV AOP showed comparable reduction of THMFP (75%) and HAAFP (52%). Results from this study suggest that O3/UV and H2O2/UV are viable options for maximum reduction of NOM from low alkalinity drinking water sources characterized with low turbidity and medium SUVA, and could hold particular significance for plants that are investigating alternative AOPs currently available in the drinking water marketplace. However, further studies that focus on measurement of product yield and include cost analysis for each oxidation process would be necessary for appropriate selection of AOPs.

Acknowledgments The authors acknowledge the technical support provided to this project by allowing access to the water treatment plant in Tatamagouche by the Municipality of the County of Colchester, Nova Scotia. In addition, the authors acknowledge the funding support provided by Natural Sciences and Engineering Research Council of Canada (NSERC).

references

Amirsardari, Y., Yu, Q., Williams, P., 2001. Effect of ozonation and UV irradiation with direct filtration and disinfection byproduct precursors in drinking water treatment. Environ.Technol. 22, 1015e1023. Amy, G.L., Collins, M.R., Kuo, C.J., King, P.H., 1987. Comparing gel permeation chromatography and ultrafiltration for the molecular weight characterization of aquatic organic matter. J. Am. Water Works Assn. 79 (1), 43e49. Backlund, O., 1992. Degradation of aquatic humic materials by ultraviolet light. Chemosphere 25 (12), 1869e1878. Bond, T., Henriet, O., Goslan, E., Parsons, S.A., Jefferson, B., 2009. Disinfection by-product formation and fractionation behaviour of natural organic matter surrogates. Environ. Sci. Technol. 43 (15), 5982e5989. Chang, P.B.L., Young, T.M., 2000. Kinetics of methyl tert-butyl ether degradation and by-product formation during UV/

w a t e r r e s e a r c h 4 5 ( 2 0 1 1 ) 3 2 6 3 e3 2 6 9

hydrogen peroxide water treatment. Water Res. 34, 2233e2240. Chin, A., Be´rube´, P.R., 2005. Removal of disinfection by-product precursors with ozone-UV advanced oxidation process. Water Res. 39 (10), 2136e2144. Crittenden, J.C., Hu, S., Hand, D.W., Green, S.A., 1999. A kinetic model for H2O2/UV process in a completely mixed batch reactor. Water Res. 33, 2315e2328. Edwards, M., Benjamin, M.M., 1992. Transformation of NOM by ozone and its effect on iron and aluminum solubility. J. Am. Water Works Assn. 84 (6), 56e66. Edzwald, J.K., Becker, W.C., Wattier, K.L., 1985. Surrogate parameters for monitoring organic matter and trihalomethane precursors. J. AWWA. 4, 122e128. Frimmel, F.H., Hesse, S., Kleiser, G., 2000. In natural organic matter and disinfection byproducts - Characterisation and control in drinking aater. ACS Symposium Series. 761. Gagnon, G.A., Booth, S.D., Peldszus, S., Mutti, D., Smith, F., Huck, P.M., 1997. Carboxylic acids: formation and removal in full-scale plants. J. Am. Water Works Assn. 89, 88e97. Galapate, R., Baes, A., Okada, M., 2001. Transformation of dissolved organic matter during ozonation: effects on trihalomethane formation potential. Water Res. 35, 2201e2206. Glaze, W., Peyton, G., Lin, S., Huang, R., Burieson, J., 1982. Destruction of pollutants in water with ozone in combination with ultraviolet radiation. 2. Natural trihalomethane precursors. Environ. Sci. Technol. 16, 454e458. Goslan, E.H., Gurses, F., Banks, J., Parsons, S.A., 2006. An investigation into reservoir NOM reduction by UV photolysis and advanced oxidation processes. Chemosphere 65 (7), 1113e1119. Gottschalk, C., Libra, J., Saupe, A., 2000. Ozonation of Water and Wastewater: A Practical Guide to Understanding Ozone and Its Application. WileyeVCH, New York. Her, N., Amy, G., Foss, D., Cho, J., Yoon, Y., Kosenka, P., 2002. Optimization of method for detecting and characterizing NOM by HPLC-size exclusion chromatography with UV and on-line DOC detection. Environ. Sci. Technol. 36 (5), 1069e1076. Her, N., Amy, G., Chung, J., Yoon, J., Yoon, Y., 2008. Characterizing dissolved organic matter and evaluating associated nanofiltration membrane fouling. Chemosphere 70 (3), 495e502. Hong, S., Elimelech, M., 1997. Chemical and physical aspects of natural organic matter (NOM) fouling of nanofiltration membranes. J. Membr. Sci. 132, 159e181. Hong, H.C., Wong, M.H., Liang, Y., 2009. Amino acids as precursors of trihalomethane and haloacetic acid formation during chlorination. Arch. Environ. Contam. Toxicol. 56 (4), 638e645. Hu, J., Wang, Z., Ng, W., Ong, S., 1999. Disinfection byproducts in water produced by ozonation and chlorination. Environ. Monit. Assess. 59, 81e93. Huber, S.A., Frimmel, F.H., 1996. Size-Exclusion Chromatography with Organic Carbon Detection (LC-OCD): A Fast and Reliable Method for the Characterization of Hydrophilic Organic Matter in Natural Waters. Vom Wasser 86, 277e290. Kleiser, G., Frimmel, F.H., 2000. Removal of precursors for disinfection by-products (DBPs)-differences between ozoneand OH-radical-induced oxidation. Sci. Total Environ. 256 (1), 1e9. Kusakabe, K., Aso, S., Hayashi, J., Isomura, K., Morooka, S., 1990. Decomposition of humic acid and reduction of trihalomethane formation potential in water by ozone with UV irradiation. Water Res. 24, 781e785. Liu, W., Andrews, S.A., Sharpless, C., Stefan, M., Linden, K.G., Bolton, J.R., 2002. Bench-scale investigations into comparative evaluation of DBP formation from different UV/H2O2 technologies. Proceedings, AWWA Water Quality Technology Conference, Seattle, WA.

3269

Matilainenm, A., Sillanpa¨a¨, M., 2010. Removal of natural organic matter from drinking water by advanced oxidation processes. Chemosphere 80 (4), 351e365. Mosteo, R., Miguel, N., Martin-Muniesa, S., Ormad, M.P., Ovelleiro, J.L., 2009. Evaluation of trihalomethane formation potential in function of oxidation processes used during the drinking water production process. J. Hazard. Mater. 172 (2e3), 661e666. Oh, H.J., Kim, W.J., Choi, J.S., Gee, C.S., Hwang, T.M., Kang, J.G., Kang, J.W., 2003. Optimization and control of ozonation plant using raw water characterization method. Ozone Sci. Eng. 25, 383e392. Owen, D.M., Amy, G.L., Chowdhury, Z.K., Paode, R., McCoy, G., Viscosil, K., 1995. NOM characterization and treatability. J. Am. Water Works Assn. 87, 46e63. Pelakani, C., Newcombe, G., Snoeyink, V.L., Hepplewhite, C., Assemi, S., Beckett, R., 1999. Characterization of natural organic matter using high performance size exclusion chromatography. Environ. Sci. Technol. 33, 2807e2813. Ravichandran, M., Aiken, G.R., Reddy, M.M., Ryan, J.N., 1998. Enhanced dissolution of cinnabar (mercury sulfide) by dissolved organic matter isolated from the Florida Everglades. Environ. Sci. Technol. 32, 3305e3311. Rittmann, B.E., Snoeylink, V.L., 1984. Achieving biologically stable drinking water. J. Am. Water Works Assn. 76 (10), 106e114. Sarathy, S.R., Mohseni, M., 2007. The impact of UV/H2O2 advanced oxidation on molecular size distribution of chromophoric natural organic matter. Environ. Sci. Technol. 41, 8315e8320. Schechter, D.S., Singer, P.C., 1995. Formation of aldehydes during ozonation. Ozone Sci. Eng. 17 (1), 53e59. Schmitt, D., Taylor, H.E., Aiken, G.R., Roth, D.A., Frimmel, F.H., 2002. Influence of natural organic matter on the adsorption of metal ions onto clay minerals. Environ. Sci. Technol. 36, 2932e2938. Singer, P.C., 1999. Humic substances as precursors for potentially harmful disinfection byproducts. Water Sci. Technol. 40 (9), 25e30. Sierka, R., Amy, G., 1985. Catalytic effects of UV light and/or ultrasound on the ozone oxidation of humic acid and THM precursors. Ozone Sci. Eng. 7, 47e62. Summers, R.S., Hooper, S.M., Solarik, G., Owen, D., 1996. Assessing DBP yield: uniform formation conditions. J. Am. Water Works Assn. 81 (7), 80e93. Thomson, J., Roddick, F., Dirkas, M., 2002. UV photooxidation facilitating biological treatment for the removal of NOM from drinking water. J. Water Supply Res. Technol. AQUA 51, 297e306. Thomson, J., Parkinson, A., Roddick, F.A., 2004. Depolymerization of chromophoric natural organic matter. Environ. Sci. Technol. 38, 3360e3369. Toor, R., Mohseni, M., 2007. UV/H2O2 based AOP and its integration with biological activated carbon treatment for DBP reduction in drinking water. Chemosphere 66, 2087e2095. Waller, D.H., Macphee, M.J., Prendiville, P.W., McCurdy, R.F., Gates, A.W., D’Eon, W.J., 1996. Characteristics of Nova Scotia surface waters and treatment options for removal of color and trihalomethane precursors. Can. J. Civ. Eng. 23, 1316e1325. Wang, G.S., Liao, C.H., Chen, H.W., Yang, H.C., 2006. Characteristics of natural organic matter degradation in water by UV/H2O2 treatment. Environ. Technol. 27 (3), 277e287. Westerhoff, P., Debroux, J., Aiken, G., Amy, G., 1999. Ozone induced changes in natural organic matter (NOM) structure. Ozone Sci. Eng. 21, 551e570. Westerhoff, P., Mezyk, S.P., Cooper, W.J., Minakata, D., 2007. Electron pulse radiolysis determination of hydroxyl radical rate constants with Suwannee River fulvic acid and other dissolved organic matter isolates. Environ. Sci. Technol. 41 (13), 4640e4646. Zhou, H., Smith, D., 2001. Advanced technologies in water and wastewater treatment. Can. J. Civ. Eng. 28, 49.