Computers in Fisheries Research

Computers in Fisheries Research

Bernard A. Megrey

l

Erlend Moksness

Editors

Computers in Fisheries Research Second Edition

13

Editors Bernard A. Megrey U.S. Department of Commerce National Oceanic and Atmospheric Administration National Marine Fisheries Service Alaska Fisheries Science Center 7600 Sand Point Way N.E. Seattle, WA 98115 USA [email protected]

ISBN: 978-1-4020-8635-9

Erlend Moksness Institute of Marine Research Flødevigen Marine Research Station 4817 His Norway [email protected]

e-ISBN: 978-1-4020-8636-6

DOI: 10.1007/978-1-4020-8636-6 Library of Congress Control Number: 2008935557 # Springer ScienceþBusiness Media B.V. 2009 No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work. Printed on acid-free paper 9 8 7 6 5 4 3 2 1 springer.com

Preface

The first edition of this book was published by Chapman and Hall Ltd. in 1996. The first edition contained nine chapters and, for all except one chapter, the original chapter authors agreed to update their chapter. Comparing these chapters gives the reader an idea of the development over a time span of more than 10 years between the two editions. In the preparation of the second edition we decided to add more chapters reflecting some important fields with significant contributions to present day fishery research. These are the use of internet for searching of information (Chapter 2), and the present state and use of remote sensing (Chapter 5), ecosystem modeling (Chapter 8) and visualization of data (Chapter 10). This second edition provides a valuable sampling of contemporary applications. Scientists have an opportunity to evaluate the suitability of different computer technology applications to their particular research situation thereby taking advantage of the experience of others. The chapters that follow are the fruition of this idea. The history behind this book started in 1989 when we were asked by Dr. Vidar Wespestad (previously: Alaska Fisheries Science Center, Seattle, USA) to prepare and convene a session at the 1992 World Fishery Congress in Athens, Greece on computer applications in fisheries. We agreed that the idea was a good one and the computer session in 1992 turned out to be very successful. The computer session was organised in three parts: training classes, informal demonstrations accompanied by posters, and oral presentations of scientific papers. We were both amazed by the high level of interest and the high quality of contributions presented at the paper session. The following year we organised together with Dr. John Ramster (Previously: Fishery Laboratory, Lowestoft, England) a theme session on the topic ‘‘Computers in Fisheries Research’’ at the ICES (International Council for the Exploration of the Sea) statutory meeting in Dublin, Ireland. The response we received from the call for papers exceeded our most optimistic expectations. A total of 62 abstracts were submitted. Nigel J. Balmforth (then at Chapman and Hall Ltd.) who attended the World Fisheries Congress, asked us to consider preparing an outline for a book on the topic. Based on our two recent experiences, we knew that the interest level in the international fisheries community was high and we were convinced that there was a need for such a book and that the idea was timely v

vi

Preface

since we determined that a book that reviews current and future computer trends in fisheries science applications did not exist. Individuals are quick to realize the potential of computers in fisheries and more scientists are taking advantage of these new tools. We believe this book will be of interest to any quantitative resource management course, fisheries course, or application of computers to fisheries course. It should also be useful as a background reading requirement for graduate and undergraduate students taking the above mentioned courses. Academic institutions with agriculture, fisheries or resource management programs, national fisheries laboratories, and library systems should find the book useful. The book will also prove useful to administrators, managers, research scientists, field biologists, university researchers, university teachers, graduate and undergraduate students, consultants, government researchers, and laypersons involved in the fisheries or natural resource disciplines. Finally we would like to say that we are very grateful for the positive response we received from all the chapter authors during the preparation of this book. USA, Norway October 2008

Bernard A. Megrey Erlend Moksness

Contents

1

2

3

4

Past, Present and Future Trends in the Use of Computers in Fisheries Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bernard A. Megrey and Erlend Moksness

1

The Consumption and Production of Fisheries Information in the Digital Age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Janet Webster and Eleanor Uhlinger

31

Extended Guide to Some Computerized Artificial Intelligence Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Saul B. Saila

69

Geographical Information Systems (GIS) in Fisheries Management and Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Geoff Meaden

93

5

Remote Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Olav Rune Godø and Eirik Tenningen

121

6

Quantitative Research Surveys of Fish Stocks . . . . . . . . . . . . . . . . . . . Kenneth G. Foote

145

7

Geostatistics and Their Applications to Fisheries Survey Data: A History of Ideas, 1990–2007. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pierre Petitgas

191

Ecosystem Modelling Using the Ecopath with Ecosim Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Marta Coll, Alida Bundy and Lynne J. Shannon

225

8

9

Image Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thomas T. Noji and Ferren MacIntyre

293

vii

viii

10

Contents

Visualization in Fisheries Oceanography: New Approaches for the Rapid Exploration of Coastal Ecosystems . . . . . . . . . . . . . . . Albert J. Hermann and Christopher W. Moore

317

11

Computers in Fisheries Population Dynamics. . . . . . . . . . . . . . . . . . . Mark N. Maunder, Jon T. Schnute and James N. Ianelli

337

12

Multispecies Modeling of Fish Populations . . . . . . . . . . . . . . . . . . . . Kenneth A. Rose and Shaye E. Sable

373

13

Computers and the Future of Fisheries . . . . . . . . . . . . . . . . . . . . . . . . Carl J. Walters

399

Species Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

413

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

415

Contributors

Alida Bundy Fisheries and Oceans, Canada Bedford Institute of Oceanography, Dartmouth, PO BOX 1006, N.S., B2Y 4A2, Canada Marta Coll Institute of Marine Science (ICM-CSIC), Passeig Marı´ tim de la Barceloneta, 37-49, 08003 Barcelona, Spain Kenneth G. Foote Woods Hole Oceanographic Institution, Woods Hole, MA 02543, USA Olav Rune Godø Institute of Marine Research, Nordnes, 5817 Bergen, Norway Albert J. Hermann Joint Institute for the Study of the Atmosphere and Ocean, University of Washington, Seattle, WA 98115, USA James N. Ianelli U.S. Department of Commerce, National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Alaska Fisheries Science Center, REFM Div., 7600 Sand Point Way NE, Seattle, WA 98115-6349, USA Ferren MacIntyre Expert-center for Taxonomic Identification, U. Amsterdam, NL-1090 GT Amsterdam, The Netherlands; National University of Ireland, University Road, Galway, Ireland Mark N. Maunder Inter-American Tropical Tuna Commission, 8604 La Jolla Shores Drive, La Jolla, CA 92037-1508, USA Geoff Meaden Department of Geographical and Life Sciences, Canterbury Christ Church University, North Holmes Road, Canterbury, Kent, CT1 1QU, UK Bernard A. Megrey U.S. Department of Commerce, National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Alaska Fisheries Science Center, 7600 Sand Point Way NE, Seattle, WA 98115, USA Erlend Moksness Institute of Marine Research, Flødevigen Marine Research Station, 4817 His, Norway ix

x

Contributors

Christopher W. Moore Joint Institute for the Study of the Atmosphere and Ocean, University of Washington, Seattle, WA 98115, USA Thomas T. Noji U.S. Department of Commerce, National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Northeast Fisheries Science Center, Sandy Hook, NJ 07732, USA Pierre Petitgas IFREMER, Department Ecology and Models for Fisheries, BP. 21105, 44311 cdx 9, Nantes, France Kenneth A. Rose Department of Oceanography and Coastal Sciences, Louisiana State University, Baton Rouge, LA 70803, USA Shaye E. Sable Department of Oceanography and Coastal Sciences, Louisiana State University, Baton Rouge, LA 70803, USA Saul B. Saila 317 Switch Road, Hope Valley, RI 02832, USA Jon T. Schnute Fisheries and Oceans Canada, Pacific Biological Station, 3190 Hammond Bay Road, Nanaimo, B.C., V9T 6N7, Canada Lynne J. Shannon Marine and Coastal Management, Department of Environmental Affairs and Tourism, Private Bag X2, Rogge Bay 8012, South Africa; Marine Biology Research Centre, Department of Zoology, University of Cape Town, Private Bag, Rondebosch, 7701, South Africa Eirik Tenningen Institute of Marine Research, Nordnes, 5817 Bergen, Norway Eleanor Uhlinger Dudley Knox Library, Naval Postgraduate School, Monterey, CA 93943, USA Carl J. Walters Fisheries Centre, University of British Columbia, Vancouver, B.C., V6T1Z4, Canada Janet Webster Oregon State University Libraries, Hatfield Marine Science Center, 2030 Marine Science Drive, Newport, OR 97365, USA

Chapter 1

Past, Present and Future Trends in the Use of Computers in Fisheries Research Bernard A. Megrey and Erlend Moksness

I think it’s fair to say that personal computers have become the most empowering tool we’ve ever created. They’re tools of communication, they’re tools of creativity, and they can be shaped by their user. Bill Gates, Co-founder, Microsoft Corporation Long before Apple, one of our engineers came to me with the suggestion that Intel ought to build a computer for the home. And I asked him, ‘What the heck would anyone want a computer for in his home?’ It seemed ridiculous! Gordon Moore, Past President and CEO, Intel Corporation

1.1 Introduction Twelve years ago in 1996, when we prepared the first edition of Computers in Fisheries Research, we began with the claim ‘‘The nature of scientific computing has changed dramatically over the past couple of decades’’. We believe this statement remains valid even since 1996. As Heraclitus said in the 4th century B.C., ‘‘Nothing is permanent, but change!’’ The appearance of the personal computer in the early 1980s changed forever the landscape of computing. Today’s scientific computing environment is still changing, often at breathtaking speed. In our earlier edition, we stated that fisheries science as a discipline was slow to adopt personal computers on a wide-scale with use being well behind that in the business world. Pre-1996, computers were scarce and it was common for more than one user to share a machine, which was usually placed in a public area. Today, in many modern fisheries laboratories, it is common for scientists to use multiple computers in their personal offices, a desktop

B.A. Megrey (*) U.S. Department of Commerce, National Oceanic and Atmospheric Administration, National Marine Fisheries Service; Alaska Fisheries Science Center, 7600 Sand Point Way NE, BIN C15700, Seattle, WA 98115, USA

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_1, Ó Springer ScienceþBusiness Media B.V. 2009

1

2

B.A. Megrey and E. Moksness

personal computer and a portable laptop is often the minimum configuration. Similarly, in many lab offices, there are several computers, each dedicated to a specific computational task such as large scale simulations. We feel that because of improvements in computational performance and advances in portability and miniaturization, the use of computers and computer applications to support fisheries and resource management activities is still rapidly expanding as well as the diversity of research areas in which they are applied. The important role computers play in contemporary fisheries research is unequivocal. The trends we describe, which continue to take place throughout the world-wide fisheries research community, produce significant gains in work productivity, increase our basic understanding of natural systems, help fisheries professionals detect patterns and develop working hypotheses, provide critical tools to rationally manage scarce natural resources, increase our ability to organize, retrieve, and document data and data sources, and in general encourage clearer thinking and more thoughtful analysis of fisheries problems. One can only wonder what advances and discoveries well known theorists and fisheries luminaries such as Ludwig von Bertalanffy, and William Ricker, or Ray Beverton and Sidney Holt would have made if they had had access to a laptop computer. The objective of this book is to provide a vehicle for fisheries professionals to keep abreast of recent and potential future developments in the application of computers in their specific area of research and to familiarize them with advances in new technology and new application areas. We hope to accomplish this by comparing where we find ourselves today compared to when the first edition was published in 1996. Hopefully, this comparison will help explain why computational tools and hardware are so important for managing our natural resources. As in the previous edition, we hope to achieve the objective by having experts from around the world present overview papers on topic areas that represent current and future trends in the application of computer technology to fisheries research. Our aim is to provide critical reviews on the latest, most significant developments in selected topic areas that are at the cutting edge of the application of computers in fisheries and their application to the conservation and management of aquatic resources. In many cases, these are the same authors who contributed to the first edition, so the decade of perspective they provide is unique and insightful. Many of the topics in this book cover areas that were predicted in 1989 to be important in the future (Walters 1989) and continue to be at the forefront of applications that drive our science forward: image processing, stock assessment, simulation and games, and networking. The chapters that follow update these areas as well as introduce several new chapter topic areas. While we recognize the challenge of attempting to present up to date information given the rapid pace of change in computers and the long time lines for publishing books, we hope that the chapters in this book taken together, can be valuable where they suggest emerging trends and future directions that impact the role computers are likely to serve in fisheries research.

1

Past, Present and Future Trends in the Use of Computers

3

1.2 Hardware Advances It is difficult not to marvel at how quickly computer technology advances. The current typical desktop or laptop computer, compared to the original monochrome 8 KB random access memory (RAM), 4 MHz 8088 microcomputer or the original Apple II, has improved several orders of magnitude in many areas. The most notable of these hardware advances are processing capability, color graphics resolution and display technology, hard disk storage, and the amount of RAM. The most remarkable thing is that since 1982, the cost of a high-end microcomputer system has remained in the neighborhood of $US 3,000. This statement was true in 1982, at the printing of the last edition of this book in 1996, and it holds true today.

1.2.1 CPUs and RAM While we can recognize that computer technology changes quickly, this statement does not seem to adequately describe what sometimes seems to be the breakneck pace of improvements in the heart of any electronic computing engine, the central processing unit (CPU). The transistor, invented at Bell Labs in 1947, is the fundamental electronic component of the CPU chip. Higher performance CPUs require more logic circuitry, and this is reflected in steadily rising transistor densities. Simply put, the number of transistors in a CPU is a rough measure of its computational power which is usually measured in floating point mathematical operations per second (FLOPS). The more transistors there are in the CPU, or silicon engine, the more work it can do. Trends in transistor density over time, reveal that density typically doubles approximately every year and a half according to a well know axiom known as Moore’s Law. This proposition, suggested by Intel co-founder Gordon Moore (Moore 1965), was part observation and part marketing prophesy. In 1965 Moore, then director of R&D at Fairchild Semiconductor, the first large-scale producer of commercial integrated circuits, wrote an internal paper in which he drew a line though five points representing the number of components per integrated circuit for minimum cost for the components developed between 1959 and 1964 (Source: http://www.computerhistory.org/semiconductor/ timeline/1965-Moore.html, accessed 12 January 2008). The prediction arising from this observation became a self-fulfilling prophecy that emerged as one of the driving principals of the semiconductor industry. As it related to computer CPUs (one type of integrated circuit), Moore’s Law states that the number of transistors packed into a CPU doubles every 18–24 months. Figure 1.1 supports this claim. In 1979, the 8088 CPU had 29,000 transistors. In 1997, the Pentium II had 7.5 million transistors, in 2000 the Pentium 4 had 420 million, and the trend continues so that in 2007, the Dual-Core Itanium 2 processor has 1.7 billion transistors. In addition to transistor density, data

4

B.A. Megrey and E. Moksness

Intel 4004

10000000000

Intel 8008 Intel 8080

1000000000

Intel 8088

log(Number of Transistors)

100000000

Intel 80286 Intel 80386

10000000

Intel 80486 Pentium

1000000

AMD K5 Pentium II

100000

AMD K6 Pentium III

10000

AMD K6-III AMD K7

1000

Pentium 4 Itanium

100

AMD K8 Itanium 2

10

Core 2 Duo

1 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006 2010 Year

Core 2 Quad G80 POWER6 Dual-Core Itanium 2

Fig. 1.1 Trends in the number of transistors placed on various CPU chips. Note the y-axis is on the log scale (Source: http://download.intel.com/pressroom/kits/IntelProcessorHistory.pdf, accessed 12 January 2008)

handling capabilities (i.e. progressing from manipulating 8, to 16, to 32, to 64 bits of information per instruction), ever increasing clock speeds (Fig. 1.2), and the number of instructions executed per second, continue to improve. The remarkable thing is that while the number of transistors per CPU has increased more than 1,000 times over the past 26 years, and another 1,000 times since 1996, performance (measured with millions of instructions per second, MIPS) has increased more than 10,000 times since the introduction of the 8088 (Source: http://www.jcmit.com/cpu-performance.htm, accessed 12 January 2008). Scientific analysts, who use large databases, scientific visualization applications, statistics, and simulation modeling need as many MIPS as they can get. The more powerful computing platforms described above will enable us to perform analyses that we could not perform earlier (see Chapters 8, 11 and 12). In the original edition we predicted that ‘‘Three years from now CPU’s will be four times faster than they are today and multi-processor designs should be commonplace.’’ This prediction has generally proven to be true. CPU performance has continued to increase according to Moore’s Law for the last 40 years, but this trend may not hold up in the near future. To achieve higher transistor densities requires the manufacturing technology (photolithography) to build the transistor in smaller and smaller physical spaces. The process architecture of

1

Past, Present and Future Trends in the Use of Computers

5

Maximum Intel CPU Clock Speed (GHz)

4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Year

Fig. 1.2 Trends in CPU clock speed (Source: http://wi-fizzle.com/compsci/cpu_speed_Page_3.png, accessed 12 January 2008)

CPUs in the early 1970s used a 10 micrometer (mm, 10 6m) photolithography mask. The newest chips use a 45 nanometer (nm, 109m) mask. As a consequence of these advances, the cost per unit of performance as measured in gigaflops has dramatically declined (Fig. 1.3).

log(Cost per GFLOP $USD)

$100,000.00 $10,000.00 $1,000.00

$100.00 $10.00 $1.00 $0.10 1996

1998

2000

2002 Year

2004

2006

2008

Fig. 1.3 Trends in the cost ($USD) per gigaflop (109 floating point instructions s–1) of CPU performance. Note y-axis is on the log scale (Source: http://en.wikipedia.org/wiki/Teraflop, accessed 12 January 2008)

6

B.A. Megrey and E. Moksness

Manufacturing technology appears to be reaching its limits in terms of how dense silicon chips can be manufactured – in other words, how many transistors can fit onto CPU chips and how fast their internal clocks can be run. As stated recently in the BBC News, ‘‘The industry now believes that we are approaching the limits of what classical technology – classical being as refined over the last 40 years – can do.’’ (Source: http://news.bbc.co.uk/2/hi/science/nature/4449711.stm, accessed 12 January 2008). There is a problem with making microprocessor circuitry smaller. Power leaks, the unwanted leakage of electricity or electrons between circuits packed ever closer together, take place. Overheating becomes a problem as processor architecture gets ever smaller and clock speeds increase. Traditional processors have one processing engine on a chip. One method used to increase performance through higher transistor densities, without increasing clock speed, is to put more than one CPU on a chip and to allow them to independently operate on different tasks (called threads). These advanced chips are called multiple-core processors. A dual-core processor squeezes two CPU engines onto a single chip. Quad-core processors have four engines. Multiple-core chips are all 64-bit meaning that they can work through 64 bits of data per instruction. That is twice rate of the current standard 32-bit processor. A dual-core processor theoretically doubles your computing power since a dual-core processor can handle two threads of data simultaneously. The result is there is less waiting for tasks to complete. A quad-core chip can handle four threads of data. Progress marches on. Intel announced in February 2007 that it had a prototype CPU that contains 80 processor cores and is capable of 1 teraflop (1012 floating point operations per second) of processing capacity. The potential uses of a desktop fingernail-sized 80-core chip with supercomputer-like performance will open unimaginable opportunities (Source: http://www.intel.com/ pressroom/archive/releases/20070204comp.htm, accessed 12 January 2008). As if multiple core CPUs were not powerful enough, new products being developed will feature ‘‘dynamically scalable’’ architecture, meaning that virtually every part of the processor – including cores, cache, threads, interfaces, and power – can be dynamically allocated based on performance, power and thermal requirements (Source: http://www.hardwarecentral.com/hardwarecentral/ reports/article.php/3668756, accessed 12 January 2008). Supercomputers may soon be the same size as a laptop if IBM brings to the market silicon nanophotonics. In this new technology, wires on a chip are replaced with pulses of light on tiny optical fibers for quicker and more power-efficient data transfers between processor cores on a chip. This new technology is about 100 times faster, consumes one-tenth as much power, and generates less heat (Source: http://www.infoworld.com/article/07/12/06/IBM-researchers-build-supercomputeron-a-chip_1.html, accessed 12 January 2008). Multi-core processors pack a lot of power. There is just one problem: most software programs are lagging behind hardware improvements. To get the most out of a 64-bit processor, you need an operating system and application programs that support it. Unfortunately, as of the time of this writing, most

1

Past, Present and Future Trends in the Use of Computers

7

software applications and operating systems are not written to take advantage of the power made available with multiple cores. Slowly this will change. Currently there are 64-bit versions of Linux, Solaris, and Windows XP, and Vista. However, 64-bit versions of most device drivers are not available, so for today’s uses, a 64-bit operating system can become frustrating due to a lack of available drivers. Another current developing trend is building high performance computing environments using computer clusters, which are groups of loosely coupled computers, typically connected together through fast local area networks. A cluster works together so that multiple processors can be used as though they are a single computer. Clusters are usually deployed to improve performance over that provided by a single computer, while typically being much less expensive than single computers of comparable speed or availability. Beowulf is a design for high-performance parallel computing clusters using inexpensive personal computer hardware. It was originally developed by NASA’s Thomas Sterling and Donald Becker. The name comes from the main character in the Old English epic poem Beowulf. A Beowulf cluster of workstations is a group of usually identical PC computers, configured into a multi-computer architecture, running a Open Source Unix-like operating system, such as BSD (http://www.freebsd.org/, accessed 12 January 2008), Linux (http://www.linux.org/, accessed 12 January 2008) or Solaris (http://www.sun.com/software/solaris/index.jsp?cid=921933, accessed 12 January 2008). They are joined into a small network and have libraries and programs installed that allow processing to be shared among them. The server node controls the whole cluster and serves files to the client nodes. It is also the cluster’s console and gateway to the outside world. Large Beowulf machines might have more than one server node, and possibly other nodes dedicated to particular tasks, for example consoles or monitoring stations. Nodes are configured and controlled by the server node, and do only what they are told to do in a disk-less client configuration. There is no particular piece of software that defines a cluster as a Beowulf. Commonly used parallel processing libraries include Message Passing Interface; (MPI, http://www-unix.mcs.anl.gov/mpi/, accessed 12 January 2008) and Parallel Virtual Machine, (PVM, http://www.csm.ornl.gov/pvm/, accessed 12 January 2008). Both of these permit the programmer to divide a task among a group of networked computers, and recollect the results of processing. Software must be revised to take advantage of the cluster. Specifically, it must be capable of performing multiple independent parallel operations that can be distributed among the available processors. Microsoft also distributes a Windows Compute Cluster Server 2003 (Source: http://www.microsoft.com/windowsserver2003/ccs/ default.aspx, accessed 12 January 2008) to facilitate building a high-performance computing resource based on Microsoft’s Windows platforms. One of the main differences between Beowulf and a cluster of workstations is that Beowulf behaves more like a single machine rather than many workstations. In most cases client nodes do not have keyboards or monitors, and are

8

B.A. Megrey and E. Moksness

accessed only via remote login or through remote terminals. Beowulf nodes can be thought of as a CPU + memory package which can be plugged into the cluster, just like a CPU or memory module can be plugged into a motherboard. (Source: http://en.wikipedia.org/wiki/Beowulf_(computing), accessed 12 January 2008). Beowulf systems are now deployed worldwide, chiefly in support of scientific computing and their use in fisheries applications is increasing. Typical configurations consist of multiple machines built on AMD’s Opteron 64-bit and/or Athlon X2 64-bit processors. Memory is the most readily accessible large-volume storage available to the CPU. We expect that standard RAM configurations will continue to increase as operating systems and application software become more full-featured and demanding of RAM. For example, the ‘‘recommended’’ configuration for Windows Vista Home Premium Edition and Apple’s new Leopard operating systems is 2 GB of RAM, 1 GB to hold the operating system leaving 1 GB for data and application code. In the previous edition, we predicted that in 3–5 years (1999–2001) 64–256 megabytes (MB) of Dynamic RAM will be available and machines with 64 MB of RAM will be typical. This prediction was incredibly inaccurate. Over the years, advances in semiconductor fabrication technology have made gigabyte memory configurations not only a reality, but commonplace. Not all RAM performs equally. Newer types, called double data rate RAM (DDR) decrease the time in takes for the CPU to communicate with memory, thus speeding up computer execution. DDR comes in several flavors. DDR has been around since 2000 and is sometimes called DDR1. DDR2 was introduced in 2003. It took a while for DDR2 to reach widespread use, but you can find it in most new computers today. DDR3 began appearing in mid-2007. RAM simply holds data for the processor. However, there is a cache between the processor and the RAM: the L2 cache. The processor sends data to this cache. When the cache overflows, data are sent to the RAM. The RAM sends data back to the L2 cache when the processor needs it. DDR RAM transfers data twice per clock cycle. The clock rate, measured in cycles per second, or hertz, is the rate at which operations are performed. DDR clock speeds range between 200 MHz (DDR200) and 400 MHz (DDR-400). DDR-200 transfers 1,600 megabits per second (Mb s1:106 bits s1), while DDR-400 transfers 3,200 MB s1. DDR2 RAM is twice as fast as DDR RAM. The bus carrying data to DDR2 memory is twice as fast. That means twice as much data are carried to the module for each clock cycle. DDR2 RAM also consumes less power than DDR RAM. DDR2 speeds range between 400 MHz (DDR2-400) and 800 MHz (DDR2-800). DDR2-400 transfers 3,200 MB s1. DDR2-800 transfers 6,400 MB s1. DDR3 RAM is twice as fast as DDR2 RAM, at least in theory. DDR3 RAM is more powerefficient than DDR2 RAM. DDR3 speeds range between 800 MHz (DDR3-800) and 1,600 MHz (DDR3-1600). DDR3-800 transfers 6,400 MB s1; DDR3-1600 transfers 12,800 MB s1. As processors increased in performance, the addressable memory space also increased as the chips evolved from 8-bit to 64-bit. Bytes of data readily

1

Past, Present and Future Trends in the Use of Computers

9

accessible to the processor are identified by a memory address, which by convention starts at zero and ranges to the upper limit addressable by the processor. A 32-bit processor typically uses memory addresses that are 32 bits wide. The 32-bit wide address allows the processor to address 232 bytes (B) of memory, which is exactly 4,294,967,296 B, or 4 GB. Desktop machines with a gigabyte of memory are common, and boxes configured with 4 GB of physical memory are easily available. While 4 GB may seem like a lot of memory, many scientific databases have indices that are larger. A 64-bit wide address theoretically allows 18 million terabytes of addressable memory (1.8 1019 B). Realistically 64-bit systems will typically access approximately 64 GB of memory in the next 5 years.

1.2.2 Hard Disks and Other Storage Media Improvements in hard disk storage, since our last edition, have advanced as well. One of the most amazing things about hard disks is that they both change and don’t change more than most other components. The basic design of today’s hard disks is not very different from the original 5¼’’ 10 MB hard disk that was installed in the first IBM PC/XTs in the early 1980s. However, in terms of capacity, storage, reliability and other characteristics, hard drives have substantially improved, perhaps more than any other PC component behind the CPU. Seagate, a major hard drive manufacturer, estimates that drive capacity increases by roughly 60% per year (Source: http://news.zdnet.co.uk/communications/ 0,100,0000085,2067661,00.htm, accessed 12 January 2008). Some of the trends in various important hard disk characteristics (Source: http://www.PCGuide.com, accessed 12 January 2008) are described below. The areal density of data on hard disk platters continues to increase at an amazing rate even exceeding some of the optimistic predictions of a few years ago. Densities are now approaching 100 Gbits in2, and modern disks are now packing as much as 75 GB of data onto a single 3.5 in platter (Source: http://www. fujitsu.com/downloads/MAG/vol42-1/paper08.pdf, accessed 12 January 2008). Hard disk capacity continues to not only increase, but increase at an accelerating rate. The rate of technology development, measured in data areal density growth is about twice that of Moore’s law for semiconductor transistor density (Source: http://www.tomcoughlin.com/Techpapers/head&medium.pdf, accessed 12 January 2008). The trend towards larger and larger capacity drives will continue for both desktops and laptops. We have progressed from 10 MB in 1981 to well over 10 GB in 2000. Multiple terabyte (1,000 GB) drives are already available. Today the standard for most off the shelf laptops is around 120–160 GB. There is also a move to faster and faster spindle speeds. Since increasing the spindle speed improves both random-access and sequential performance, this is likely to continue. Once the domain of high-end SCSI drives (Small Computer System Interface), 7,200 RPM spindles are now standard on mainstream desktop and

10

B.A. Megrey and E. Moksness

notebook hard drives, and a 10,000 and 15,000 RPM models are beginning to appear. The trend in size or form factor is downward: to smaller and smaller drives. 5.25 in drives have now all but disappeared from the mainstream PC market, with 3.5 in drives dominating the desktop and server segment. In the mobile world, 2.5 in drives are the standard with smaller sizes becoming more prevalent. IBM in 1999 announced its Microdrive which is a tiny 1 GB or device only an inch in diameter and less than 0.25 in thick. It can hold the equivalent of 700 floppy disks in a package as small as 24.2 mm in diameter. Desktop and server drives have transitioned to the 2.5 in form factor as well, where they are used widely in network devices such as storage hubs and routers, blade servers, small form factor network servers and RAID (Redundant Arrays of Inexpensive Disks) subsystems. Small 2.5 in form factor (i.e. ‘‘portable’’) high performance hard disks, with capacities around 250 GB, and using the USB 2.0 interface are becoming common and easily affordable. The primary reasons for this ‘‘shrinking trend’’ include the enhanced rigidity of smaller platters. Reduction in platter mass enables faster spin speeds and improved reliability due to enhanced ease of manufacturing. Both positioning and transfer performance factors are improving. The speed with which data can be pulled from the disk is increasing more rapidly than positioning performance is improving, suggesting that over the next few years addressing seek time and latency will be the areas of greatest attention to hard disk engineers. The reliability of hard disks is improving slowly as manufacturers refine their processes and add new reliability-enhancing features, but this characteristic is not changing nearly as rapidly as the others above. One reason is that the technology is constantly changing, and the performance envelope is constantly being pushed; it’s much harder to improve the reliability of a product when it is changing rapidly. Once the province of high-end servers, the use of multiple disk arrays (RAIDs) to improve performance and reliability is becoming increasingly common, and multiple hard disks configured as an array are now frequently seen in consumer desktop machines. Finally, the interface used to deliver data from a hard disk has improved as well. Despite the introduction to the PC world of new interfaces such as IEEE-1394 (FireWire) and USB (universal serial bus) the mainstream interfaces in the PC world are the same as they were through the 1990s: IDE/ATA/SATA and SCSI. These interfaces are all going through improvements. A new external SATA interface (eSATA) is capable of transfer rates of 1.5–3.0 Gbits s1. USB transfers data at 480 Mbits s1 and Firewire is available in 400 and 800 Mbits s1. USB 3.0 has been announced and it will offer speeds up to 4.8 Gbits s1. Firewire will also improve to increases in the range of 3.2 Gbits s1. The interfaces will continue to create new and improved standards with higher data transfer rates to match the increase in performance of the hard disks themselves. In summary, since 1996, faster spindle speeds, smaller form factors, multiple double-sided platters coated with higher density magnetic coatings, and improved recording and data interface technologies, have substantially increased hard disk storage and performance. At the same time, the price per

1

Past, Present and Future Trends in the Use of Computers

11

$1,000.00

Cost per GB ($USD)

$100.00

$10.00

$1.00

$0.10 1994

1996

1998

2000

2002

2004

2006

2008

Year

Fig. 1.4 Trends in cost per GB of hard disk storage ($USD) (Source: http://www. mattscomputertrends.com/harddiskdata.html, accessed 12 January 2008)

unit of storage has decreased (Fig. 1.4). In 1990, a typical gigabyte of storage cost about $US 20,000 (Kessler 2007). Today it is less than $US 1. The total hard disk capacity shipped as of 2003 (Fig. 1.5) indicates exponentially

16000 14000

14000

Petabytes

12000 10000

8500

8000 6000

4900

4000

2700 1500

2000 0

100

200

400

1000

1995

1996

1997

1998

1999 Year

2000

2001

2002

2003

Fig.1.5 PC hard disk capacity (in petabytes-1015 B or 1,000 TB) shipped as of 2003 (Data from 1999 Winchester Disk Drive Market Forecast and Review Table C5, International Data Corporation) (Source: http://www2.sims.berkeley.edu/research/projects/how-mucþh-info/ charts/charts.html, accessed 12 January 2008)

12

B.A. Megrey and E. Moksness

increasing capacity through time. Today 2.5’’ 250 GB hard disks are common and multiple terabyte hard disks collected together in RAID configurations provide unprecedented storage capacity. The trends continue as recently Seagate announced research into nanotube-lubricated hard disks with capacities of several terabits per square inch, making possible a 7.5 TB 3.5 in hard disk (Source: http://www.dailytech.com/article.aspx?newsid¼3122&ref¼y, accessed 12 January 2008). Hard disks are not the only available storage media. Floppy disks, formerly a mainstay of portable storage, have become a thing of the past. Today computers are rarely shipped with floppy disk drives. At one time, Iomega’s portable ZIP drives looked promising as a portable device to store about 200 MB of data. In 1996, we predicted that ‘‘Newer storage media such as read-write capable CD-ROM’s and WORM’s (write once read many times) will eventually displace floppy disks as the storage medium of choice’’. This has taken place and today even the CD-ROM, which in the past held promise for large capacity storage (700 MB) has been replaced with the ubiquitous ‘‘thumb drive’’ memory sticks. These marvels of miniaturization can accommodate 8–16 GB of data, use very fast USB 2.0 transfer interfaces, easily connect to any computer with a USB port, and are unusually inexpensive. As of the time of this writing a 4 GB USB 2.0 memory stick costs around $US 40. Double-sided rewritable DVD media are increasingly being used to easily store data in the 4–6 GB range.

1.2.3 Graphics and Display Technology In 1996, we predicted that in 3–5 years (1999–2001), support for 24-b color, full 3-D acceleration, broadcast quality video, and full-motion near-lifelike virtualreality capabilities would be commonplace. This forecast has proven to be true. The very first video card, released with the first IBM PC, was developed by IBM in 1981. The MDA (monochrome display adapter) only worked in text mode representing 2580 lines in the screen. It had a 4 KB video memory and just one color. Today’s graphic cards offer radically improved capabilities. Modern video cards have two important components. The first is the GPU (graphics processing unit). This dedicated microprocessor, separate from the main CPU, is responsible for resolution and image quality. It is optimized for floating point calculations, which are fundamental to 3D graphics rendering. The GPU also controls many graphic primitive functions such as drawing lines, rectangles, filled rectangles, polygons and the rendering of the graphic images. Ultimately, the GPU determines how well the video card performs. The second important component is the video RAM (or vRAM). In older graphics cards, system RAM was used to store images and textures. But with a dedicated video card, built-in vRAM takes over this role, freeing up system RAM and the main CPU for other tasks. When it comes to vRAM, there are a variety of options. If

1

Past, Present and Future Trends in the Use of Computers

13

you’re just doing simple tasks, 64 MB is adequate. If you’re editing video, 128 MB should be the minimum, with larger amounts up to 512 MB – 1 GB available for more demanding tasks. Also as a general rule, the more powerful the GPU, the more vRAM it will require. Modern video cards also incorporate high speed communication channels that allow large amounts of graphic data to pass quickly through the system bus. Today’s video cards also contain multiple output options including S-video, super VGA (SVGA), Digital Video Interface (DVI), and High Definition Multimedia Interface (HDMI) connections are common as well as options for up to 32-b and 64-b colors, and resolutions approaching 2,5601,600 at very fast refresh rates in the range of 85 Hz. The very newest cards include television tuners, some even offering the newly emerging Highdefinition standard. This feature is mainly relevant to home computer systems for those who want to turn their computer into a personal video recorder. We are convinced a scientific application for this feature will become useful in the years to come. The ability to produce graphics is just one piece of the graphics system, with the other being the display device. Old large and power hungry analog monitors are slowly being replaced by digital Liquid Crystal Display (LCD) panels, the latter appearing sometimes in large (19–22 in) formats. LCD monitors are sleeker than bulky cathode-ray tube models and they are more energy efficient. Some LCD monitors consume 1/2 to 2/3 the energy of traditional monitors. Since Windows XP was released, with its expanded desktop feature, dual LCD monitor desktop computers have become more common. The increased popularity of multi-display systems has to do with advances in technology as well as economics. Though Windows 98 first allowed for dual display configurations, the bulky analog CRTs that sat on most desks and workspaces simply could not accommodate more than one monitor. Flat-panel displays solved the space problem. Originally expensive they were considered a luxury, with prices often exceeding $US 1000. Resolution increased along with the ability to pack more and more transistors into the LCD panel, and today’s monitors, by contrast, are just a fraction of original cost. Today a good quality 22 in LCD monitor costs around $US 300. That means adding a second or third monitor is comparable to the cost of some of the original models. Research shows that there is a productivity benefit that is almost immediate. Numerous studies estimate productivity increases of anywhere from 10 to 45% (Russel and Wong 2005; Source: http://www.hp.com/sbso/solutions/ finance/expert-insights/dual-monitor.html, accessed 12 January 2008). Efficiency experts suggest that using two LCD monitors improves efficiency by up to 35% and researchers at Microsoft also found similar results, reporting that workers increased their productivity 9–50% by adding a second or third monitor. (Source: http://www.komando.com/columns/index.aspx?id¼1488 accessed, 12 January 2008).

14

B.A. Megrey and E. Moksness

1.2.4 Portable Computing Another recent trend is the appearance of powerful portable computer systems. The first portable computer systems (i.e. ‘‘luggables’’) were large, heavy, and often portability came at a cost of reduced performance. Current laptop, notebook, and subnotebook designs are often comparable to desktop systems in terms of their processing power, hard disk, and RAM storage and graphic display capabilities. In 1996, we observed that ‘‘It is not unusual, when attending a scientific or working group meeting, to see most participants arrive with their own portable computers loaded with data and scientific software applications.’’ Today, it is unusual to see scientists attending technical meetings arrive without a portable computer. Since 1996, the performance and cost gap between notebooks and desktops capable of performing scientific calculations has continued to narrow, so much so, that the unit growth rate of notebook computers is now faster than for desktops. With the performance gap between notebooks and desktop systems narrowing, commercial users and consumers alike are beginning to use the notebooks more and more as a desktop replacement since the distinction between the two as far as what work can be accomplished is becoming more and more blurred. Moreover, the emergence of notebook ‘‘docking stations’’ allows the opportunity to plug notebooks into laboratory network resources when scientists are in their office and then unplug the notebook at the end of the day to take it home or on the road, all the while maintaining one primary location for important data, software, working documents, literature references, email archives, and internet bookmarks. We have seen that miniaturization of large capacity hard disk storage, memory sticks, printers, and universal access to email made available via ubiquitous Internet connectivity (see below) all contribute to a portable computing environment, making the virtual office a reality.

1.3 Coping with Oceans of Data The information explosion is well documented. Information stored on hard disks, paper, film, magnetic, and optical media doubled from 2000 to 2003, expanding by roughly 5 EB (exabytes: over 5 billion gigabytes) each year or about 800 MB per person per year (Lyman and Varian 2003). These authors present, as of 2003, an intriguing look into the volume of digital information produced worldwide, where it originates and interesting trends through time. For example, in the United States we send, on average, 5 billion instant messages and 31 billion emails each day (Nielsen 2006). The trend is clear for scientific pursuits; the growth of data is one of the biggest challenges facing scientists today. As computer software and hardware improve, the more sensors we place into the biosphere, the more satellites we

1

Past, Present and Future Trends in the Use of Computers

15

put into orbit, the more model runs we perform, the more data that can be – and is being – captured. In fact, Carlson (2006) tells us, ‘‘Dealing with the ‘data deluge,’ as some researchers have called it, along with applying tested methods for controlling, organizing and documenting data will be among the great challenges for science in the 21st century.’’ Unfortunately, having more data does not mean we are able to conduct better science. In fact, massive volumes of data can often become detrimental to scientific pursuits. Data originating from different sources can sometimes be conflicting and certainly require ever increasing resources, hardware and maintenance. Someone once said: ‘‘We are drowning in data, but starving of information’’. We feel this is particularly true for fisheries data. In addition to the ever increasing quantity of data we add the vexing problems of harmonizing heterogeneous data collected on different spatial and temporal scales and the ever present problem of inappropriate use of online data because the metadata are missing (Chapter 5). Documenting data by writing metadata is a task scientists are reluctant to undertake, but a necessary step that will allow efficient data discovery as volumes of data continue to grow. Scientists have been struggling with this issue for years and metadata software solutions are scarce and often inadequate. Metadata will be a major issue in the coming decade. Below we present some current examples of the increasing amounts of data we are required to accommodate. Current generation general ocean circulation models using the Regional Ocean Modeling System (ROMS; Source: https://www.myroms.org/, accessed 24 December 2007) linked to lower trophic level (NPZ) ecosystem models (see Chapter 10) using typical grid spacing (3 km horizontal and 30 vertical levels, giving 462  462 horizontal gridpoints) over a typical ocean domain such as the central Gulf of Alaska (Hermann, in press, Chapter 10) generates 484 MB of output a day (where all the physical and biological variables are saved at every horizontal/vertical gridpoint). Hence a full model year of daily output from this model generates up 484 MB  365 = 176 GB (Albert Hermann, pers. comm., NOAA, Pacific Marine Environmental Laboratory). If a relatively short time series of model simulations (say 10 years) were permanently archived, it would require almost 2 TB (TB, 1,000 GB) of storage. Data collection rates for a typical shipboard acoustic echosounder system (see Chapter 5), such as the Simrad EK60 using 3 frequencies (3 frequencies at 1 ms pulse to 250 m contains 3,131 pings; 1 frequency ¼ 16.7 MB) generates about 50 MB of data. A hypothetical acoustic mooring designed to measure down to 250 m will generate about 4 MB h1 or about 50 MB day1. In the case of a typical groundfish survey, the echosounder will generate about 95 MB1.2 GB h1, depending on the ping rate (Alex deRoberts, pers. comm. NOAA, Alaska Fisheries Science Center). Finally, newer multibeam systems, such as the Simrad ME70 will collect 10–15 GB h1 for typical applications (see e.g. Ona et al. 2006). Many of our ‘‘standard’’ field data collection devices (i.e. measuring boards, scales, and net sampling equipment) are now digital and, interact with other

16

B.A. Megrey and E. Moksness

on-board ship sensors (i.e. GPS) providing large amounts of additional quality controlled information. For example, our old paradigm of net measurement (i.e. spread and height) has improved over the last decade with the use of depth sensors and bottom contact sensors. The amount of potential additional net information is about to explode with the use of automated net mensuration systems capable of providing height, spread, bottom contact, temperature, depth, net symmetry, speed, geometry of the codend, fish density, distance and angle of the net relative to the boat and even net damage reporting. In addition, there is a rapidly expanding flow of information from sounders and enhanced sonar devices capable of providing many data streams regarding the sea bottom condition and hardness, currents and other sea states. This means that even traditional data sources have the potential to rapidly expand in quantity and metadata requirements. Cabled ocean observing systems, such as VENUS (Victoria Experimental Network Under the Sea) (Source: http://www.martlet.ca/view.php?aid¼38715, accessed 12 January 2008) and NEPTUNE (North-East Pacific Time-Series Undersea Networked Experiments) (Source: http://www.neptunecanada.ca/ documents/NC_Newsletter_2007Aug31F.pdf, accessed 12 January 2008), off the North American west coast, are some of the world’s first regional permanent ocean observatories. The observing system consists of undersea nodes to regulate and distribute power and provide high bandwidth communications (4 GBs1) through fiber-optic cable, connecting more than 200 instruments and sensors, such as video cameras, a 400 m vertical profiler (to gather data at various ocean depths) and a remotely operated vehicle, as they collect data and imagery from the ocean surface to beneath the seafloor. The existing VENUS node is similarly configured and collecting data at a rate of 4 GB per day. John Dower (pers. comm.), affiliated with the NEPTUNE and VENUS cabled observing system, characterized the problems associated with coping with the vast amounts of data being delivered from ‘‘always on’’ data streams such as these new cabled system as trying to take a ‘‘Drink from a fire hose’’. The Intergovernmental Panel on Climate Change (IPCC) coordinated scientists at 17 major climate modeling centers throughout the world to run a series of climate models under various standard prescribed climate scenarios to examine the anticipated affect of factors contributing to climate change. They then prepared climate assessment reports, the most recent being the Fourth Assessment Report or AR4 (IPCC 2007). The massive output files are archived at the Lawrence Livermore National Laboratory (Source: http://www-pcmdi.llnl.gov/, accessed 12 January 2008) and are made available to the scientific community for analysis. These data consist of 221 output files from different ‘‘experiment scenario/model’’ combinations and the data volume totals approximately 3 TB. Remote sensing equipment such as the ARGO system is a global array of about 3,000 free-drifting profiling ARGO floats (Fig. 1.6) that measures the temperature and salinity of the upper 2,000 m of the ocean. The floats send their data in real-time via satellites to ARGO Global Data Acquisition Centers (GADC). Data from 380,472 individual profiles are instantly available at the

1

Past, Present and Future Trends in the Use of Computers

17

Fig. 1.6 Location of 3071 active ARGO floats which have delivered data within the past 30 days, as of 25 December 2007) (Source: http://www.argo.ucsd.edu/Acindex.html, accessed 25 December 2007)

GDACs including 168,589 high quality profiles provided by the delayed mode quality control process. Google Earth can be used to track individual floats in real-time (Source: http://w3.jcommops.org/FTPRoot/Argo/Status/, accessed 12 January 2008). This amazing resource allows, for the first time, continuous monitoring of the temperature, salinity, and velocity of the upper ocean, with all data being relayed and made publicly available within hours after collection (Source: http://www.argo.ucsd.edu/Acindex.html, accessed 12 January 2008). Satellites offer another example of broadband high capacity data delivery systems. The Advanced Very High Resolution Radiometer (AVHRR) data set is comprised of data collected by the AVHRR sensor and held in the archives of the U.S. Geological Survey’s EROS Data Center. AVHRR sensors, carried aboard the Polar Orbiting Environmental Satellite series, consist of a 4- or 5-channel broad-band scanning radiometer, sensing in the visible, nearinfrared, and thermal infrared portions of the electromagnetic spectrum (Source: http://edc.usgs.gov/guides/avhrr.html, accessed 12 January 2008). The AVHRR sensor provides for global (pole to pole) on board collection of data from all spectral channels. Each pass of the satellite provides a 2,399 km (1,491 mi) wide swath. The satellite orbits the Earth 14 times each day from 833 km (517 mi) above its surface. The objective of the AVHRR instrument is to provide radiance data for investigation of clouds, land-water boundaries, snow and ice extent, ice or snow melt inception, day and night cloud distribution, temperatures of radiating surfaces, and sea surface temperature. Typical data file sizes are approximately 64 MB per a 12 min (in latitude-longitude coordinates) sampling swath per orbit. The Sea-viewing Wide Field-of-view Sensor (SeaWiFS) is another example of a satellite system designed to provide quantitative data on global ocean bio-optical properties to the Earth science community (Source: http://oceancolor.gsfc.nasa. gov/SeaWiFS/, accessed 12 January 2008). Subtle changes in ocean color, and

18

B.A. Megrey and E. Moksness

in particular surface irradiance on every band, signify various types and quantities of marine phytoplankton (microscopic marine plants), the knowledge of which has both scientific and practical applications. Since an orbiting sensor can view every square kilometer of cloud-free ocean every 48 h, satelliteacquired ocean color data provide a valuable tool for determining the abundance of ocean biota on a global scale and can be used to assess the ocean’s role in the global carbon cycle and the exchange of other critical elements and gases between the atmosphere and the ocean. The concentration of phytoplankton can be derived from satellite observations of surface irradiance and quantification of ocean color. This is because the color in most of the world’s oceans in the visible light region, (wavelengths of 400–700 nm) varies with the concentration of chlorophyll and other plant pigments present in the water (i.e. the more phytoplankton present, the greater the concentration of plant pigments and the greener the water). A typical SeaWiFS SST file of sea surface temperature for 1 day from the MODUS sensor can be as large as 290 MB. On-line databases of compiled and quality controlled data are another source of large quantities of information. Examples include biological databases such as a comprehensive database of information about fish (FishBase) that includes information on 29,400 species (Source: http://www.fishbase.org/, accessed 12 January 2008), a database on all living cephalopods (octopus, squid, cuttlefish and nautilus) Cephbase (Source: http://www.cephbase.utmb.edu/, accessed 12 January 2008), Dr. Ransom Myer’s Stock Recruitment Database consists of maps, plots, and numerical data from over 600 fish populations (over 100 species) from all over the world (Source: http://www.mscs.dal.ca/myers/welcome.html, accessed 12 January 2008), Global Information System about fish larvae (LarvalBase) (Source: http://www.larvalbase.org/, accessed 28 December 2007), the FAO Statistical Database consists of a multilingual database currently containing over 1 million time-series records from over 210 countries (Source: http://www.fao.org/waicent/ portal/statistics_en.asp, accessed 12 January 2008), not to mention the numerous catch and food habits databases, often consisting of tens of millions of records. Even given the sometimes overwhelming quantity of data, one trend that has definitely happened in the last decade is the movement of data from flat ASCII files and small ad-hoc databases (i.e. EXCEL spreadsheets) into relational databases with designs based on actual data relationships and collection methodology. This has been a very important and powerful step towards control of data quality. Hopefully, the problems mentioned at the beginning of this section can be addressed with the tremendous advancements in hardware mentioned above as well as software advances covered in the next section.

1.4 Powerful Software At the time of the last writing of this book, application software was only available from commercial sources. Since 1996, a remarkable development has taken place – Open Source software (free source code) is widely available

1

Past, Present and Future Trends in the Use of Computers

19

for almost any purpose and for almost any CPU platform. Open Source software is developed by an interested community of developers and users. As Schnute et al. (2007) eloquently put it, ‘‘Open source software may or may not be free of charge, but it is not produced without cost. Free software is a matter of liberty, not price. To understand the concept, one should think of free as in free speech, not as in free beer’’. To our knowledge, no one has attempted to estimate the true cost of Open Source software. Some notable examples of Open Source or no cost software includes operating systems such as Fedora Linux (Source: http://fedoraproject.org/, accessed 12 January 2008); web sever software by Apache (Source: http://www.apache.org/, accessed 12 January 2008); high level numerical computing software such as Octave (Source: http://www.gnu.org/software/octave/, accessed 12 January 2008) and SciLab (Source: http://www.scilab.org/, accessed 12 January 2008) which is similar to MATLAB; statistical software such as R (source: http:// www.r-project.org/, accessed 12 January 2008) and WinBUGS; (Lunn et al. 2000; Source: http://www.mrc-bsu.cam.ac.uk/bugs/winbugs/contents.shtml, accessed 12 January 2008) for implementing Bayesian statistics; compilers such as the GNU Family of Compilers for C (Source: http://gcc.gnu.org/, accessed 12 January 2008) and FORTRAN (Source: http://www.gnu.org/software/ fortran/fortran.html, accessed 12 January 2008); plotting software, also from the GNU development team (Source: http://www.gnu.org/software/ fortran/fortran.html, accessed 12 January 2008 and www.gnuplot.info/ accessed 12 January 2008); database software such as MySQL (Source: http://www.mysql.com/, accessed 12 January 2008); software business productivity programs such as OpenOffice (Source: http://www.openoffice.org/, accessed 12 January 2008); ecosystem modeling software such as Ecopath with Ecosim (Source: http://www.ecopath.org/, accessed 12 January 2008) and the newly released fisheries library in R (FLR, Kell et al. 2007) (Source: http:// www.flr-project.org/, accessed 12 January 2008). Many other offerings can be located at the Free Software Foundation (Source: http://www.fsf.org/, accessed 12 January 2008). Similar to our previous observation, we still see software functionality and growing feature sets advancing in lockstep with improvements in computer hardware performance and expanded hardware capability. Today’s application software packages are extremely powerful. Scientific data visualization tools and sophisticated multidimensional graphing applications facilitate exploratory analysis of large complex multidimensional data sets and allow scientists to investigate and undercover systematic patterns and associations in their data that were difficult to examine several years ago. This trend enables users to focus their attention on interpretation and hypothesis testing rather on the mechanics of the analysis. Software that permits the analysis of the spatial characteristics of fisheries data are becoming more common. Programs to implement geostatistical algorithms (see Chapter 7) and Geographic Information System (GIS) software (see Chapter 4) have made significant advances that offer the fisheries biologist the ability to consider this most important aspect of natural

20

B.A. Megrey and E. Moksness

populations in both marine and freshwater ecosystems. Image analysis software (see Chapter 9) also offers promise in the areas of pattern recognition related to fisheries science such as identification of taxa, fish age determination and growth rate estimation, as well as identifying species from echo sounder sonar records. Highly specialized software such as neural networks and expert systems (see Chapter 3), which in the past have received limited application to fisheries problems, are now becoming commonplace. Very advanced data visualization tools (see Chapter 10) offer exciting new research opportunities heretofore unavailable to fisheries scientists. Whole ecosystem analysis tools (see Chapter 8) allow the simultaneous consideration of the entirety of the biological components that make up these dynamic systems. The marriage of powerful computer systems to remote sensing apparatus and other electronic instrumentation continues to be an area of active research and development (see Chapter 5). The area of population dynamics, fisheries management, stock assessment and statistical methodology software (see Chapters 11 and 12), long a mainstay of computer use in fisheries, continues to receive much attention.

1.5 Better Connectivity No aspect of our scientific lives remains untouched by the World Wide Web and Internet connectivity. The explosive growth of the Internet over the last decade has led to an ever increasing demand for high-speed, ubiquitous Internet access. The Internet is the fastest growing communication conduit and has risen in importance as the information medium of first resort for scientific users, basically achieving the prominence of a unique, irreplaceable and essential utility. How did we do our jobs without it? In 1996, we predicted that ‘‘the Internet, other network, and Wide-AreaNetwork connectivity resources held great promise to deliver global access to a vast and interactive knowledge base. In addition, the Internet would provide to the user a transparent connection to networks of information and more importantly people.’’ This has largely proven to be true and compared to today’s Internet resources, it may seem as a bit of an understatement. Compared to 12 years ago, access has improved, speed has increased, content has exploded providing significantly more resources available over the web. We feel it is true to say that the Internet is considered the method of choice for communication in scientific circles. O’Neill et al. (2003) present a nice summary of trends in the growth of the web, current as of 2003. Figure 1.7 depicts the steadily increasing trend in the number of Internet host servers on line (Source: http://www.isc.org/index.pl?/ops/ds/host-count-history.php, accessed 12 January 2008) and the large and growing community of users (Fig. 1.8) (Source: http://www.esnips.com/doc/f3f45dae-33fa-4f1f-a780-6cfbce8be558/Internet-Users, accessed 12 January 2008; 2007 statistics from: http://www.internetworldstats.com/ stats.htm, accessed 12 January 2008). These data show that the number of hosts

1

Past, Present and Future Trends in the Use of Computers

21

Number of Servers (millions)

600 500 400 300 200 100 0 1996

1998

2000

2002 Year

2004

2006

2008

Fig. 1.7 Trends in the number of servers that make up the World Wide Web (Source: http://www.isc.org/index.pl?/ops/ds/host-count-history.php, accessed 12 January 2008)

and users have increased 5.071% and 3.356%, respectively since 1996. Lyman and Varian (2003) estimate that the web accounted for 25–50 TB of information. Most electronic communication flows through four main channels: radio and television broadcasting, telephone calls and the Internet. The Internet, the

Number of Internet Users (millions)

1400 1244

1200 1018

1043

1000 817

800

719 553

600

605

451

400 248

200

150 16

36

70

0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Year

Fig. 1.8 Trends in the number of internet users (millions) (Source: http://www.esnips.com/ doc/f3f45dae-33fa-4f1f-a780-6cfbce8be558/Internet-Users, accessed 26 November 2007; 2007 statistics from: http://www.internetworldstats.com/stats.htm, accessed 12 January 2008)

22

B.A. Megrey and E. Moksness

newest electronic information medium, has proven to be capable of subsuming all three other communication channels. Digital TV stations are broadcast over IPTV (Internet Protocol Television is a system where digital television service is delivered by using Internet Protocol over a network infrastructure; See Section 1.6). We believe this relatively quick transition is a prelude of what opportunities we can expect in the near future. Today, there are three basic options available to access the Internet: The oldest method is dial-up access. Back in 1996, dial-up was one of the few options for many scientists to connect to the Internet, mainly through commercial Internet providers such as America On Line. During this era, typical fisheries laboratories did not have broadband connections or significant within-lab Internet resources (i.e. application servers, distributed electronic databases, web pages, etc.). Even email connectivity was slow by today’s standards. As the Internet grew, with its emphasis on visual and multimedia delivery, it presented a problem for the dial-up community. So much so that web pages offered ‘‘image free’’ versions of their content with the intention to speed up access to the core of their content. Ironically, in today’s telecommunication environment, this is the exact situation for the fledgling web-enabled cell phone and Personal Digital Assistant (PDA) customers. Eventually, the speed of dial-up modems simply could not accommodate the abundant digital content desired by web users. The progression of modem speed was impressive at first: 1,200 baud, 3,600, 9,600 baud. No end was in sight. But the expanding and band-width consuming Internet content continued to push the envelope. Eventually, even the speedy 56,000 baud modem was too slow. It was simply not fast enough to carry multimedia, such as sound and video except in low quality. In modern digital societies, dial-up is the method of last resort. If you are using dial-up, it is clear that either broadband access is not available, or that broadband access is too expensive. We suggest, that today, the main communication link to the web and the Internet for fisheries and resource scientists within their offices is via high-speed connections such as T1 or T3 lines (often connected via fiber optic cable) or access to proven high-speed technologies such as cable modems and digital subscriber lines (DSL). While it is true that individual situations vary, we feel confident saying that within-laboratory internet connectivity has come a long way compared to 12 years ago and we expect it will continue to improve with alarming speed. The most recent major change in Internet interconnectivity developments since 1996 involves the almost ever-present wireless Internet access. Trends in wireless communications today are vast and exciting and accelerating at the high speeds they employ. Twelve years ago, we felt privileged as scientists, if we had the opportunity to attend a meeting or working group venue that had wireless connectivity to the Internet. At the time, these services were supplied by visionary venue hosts and often our laptops were not even capable of accessing wireless signals without external adapters connected to USB or PCMCIA ports.

1

Past, Present and Future Trends in the Use of Computers

23

Most laptops manufactured today come with Wi-Fi cards already built-in. Once the laptop’s Wi-Fi capability is turned on, software usually can detect an access point’s SSID – or ‘‘service set identifier’’ – automatically, allowing the laptop to connect to the signal without the user having to intervene. Today’s portable laptops, almost without exception, utilize some variation of the Intel Centrino CPU, which has a wireless mobile chipset, wireless interface, and WiFi adapter embedded directly onto the computational CPU. Thus, Wi-Fi capability is already built in. Wireless extends beyond Internet connectivity. We have wireless computers; wireless Internet, WANs, and LANs; wireless keyboards and mice; pagers and PDAs; and wireless printers, scanners, cameras, and hubs. The potential is very real for our children to say, ‘‘What is a cable?’’ Today, we don’t feel it is an over exaggeration to say that scientists expect SOME level of Internet connectivity when they are outside of their laboratory environment. This includes meeting locations, workshop venues, alternate work locations, not including public access points located at places such as airports, local hot spots, hotels. A minimum is access to email – and the expectation is that both hardware and software tools will be available to accomplish communication with the parent laboratories or distant colleagues. Even better would be to have access, via wired or wireless connections to files at the home lab for interactions or demonstrations for such things as examination of virtual databases, access to simulation model animations, or the ability to instantaneously access working documents, PDF publications, or library resources. Since the last edition of this book, not only has wireless connectivity become commonplace, it has gone through several iterations of improvements. The Institute of Electrical and Electronic Engineers (IEEE) standard or protocol, known as 802.11, began with the 802.11B version (a data transfer rate of around 11 Mbits s1 (Mbps) using the 2.4 G HZ band with a range of 38 m), then progressed to 802.11 G (a data transfer rate of 54 Mbps using the same 2.4 GHz band as 802.11 B with similar range) and now the emerging standard is 802.11 N (over 248 Mbps using the 5 GHz and 2.4 GHz spectrum bands and a wider range of 70 m). This is just about as fast as can be experienced over a hard-wired network. With each new iteration of the Wi-Fi standard, transmission speed and range are generally improved. Lyman and Varian (2003) report the number of users who connect wirelessly has doubled from 2002. They estimate that 4% or roughly 1.4 million users now access the Internet without wires. Just 1 year later, the estimate was updated to be over 40,000 hot spots catering to over 20 million users. Hot spots are Wi-Fi locations setup to provide Internet access through a wireless network to nearby computers. This extraordinary explosion of access points and users is a testimony to the utility and demand for Wi-Fi Internet access. Current estimates suggest that there are 100,000 Wi-Fi hot spots worldwide (Source: http://www.jiwire.com/about/ announcements/press-100k-hotspots.htm, accessed 12 January 2008). Data indicate that Europe has the fastest annual growth (239%: Source: http://ipass.com/

24

B.A. Megrey and E. Moksness

pressroom/pressroom_wifi.html, accessed 12 January 2008), with most of that annual growth (255%) taking place in hotel venues (Source: http://ipass.com/ pressroom/pressroom_wifi.html, accessed 12 January 2008). The wide adoption of Wi-Fi and its rapid growth help scientists gradually become familiar with this new means of Internet access. We typically search for Wi-Fi hot spots while we are away from the home or office. Many airports, coffee bars, hotels and motels now routinely provide these services, some for a fee and some for free. According to projections, over 90 million laptops and personal digital assistants (PDAs) will have ready to access Wi-Fi LANs and local hot spots. This technology is not without problems. The main problem with wired broadband access is that it requires a cost-based subscription and it does not reach all areas. The main problem with Wi-Fi access is that hot spots are physically small, so coverage is sparse and spatially very localized. WiMAX (Worldwide Interoperability for Microwave Access) was designed to solve these problems. WiMAX, which is based on the IEEE 802.16 standard, is a new telecommunications technology designed to provide wireless data over long distances. The expectation is that WiMAX will be considered an alternative to wired broadband like cable and DSL since it will provide universal wireless access, where almost everywhere is a ‘‘hot spot’’. In practical terms, WiMAX would operate similar to Wi-Fi but at higher speeds, over greater distances and for a greater number of users. WiMAX can provide broadband wireless access up to 30 mi (50 km) for fixed stations, and 3–10 mi (5–15 km) for mobile stations. In contrast, the Wi-Fi/802.11 wireless local area network standard is limited in most cases to only 100–300 feet (30–100 m). WiMAX could potentially erase the suburban and rural blackout areas that currently have no broadband Internet access because phone and cable companies have not yet run the necessary wires to those remote locations. Another trend that is beginning to impact scientific activity is the growth of smaller devices to connect to the Internet and the convergence of cellular phones, other portable devices, and computers. Small tablets, pocket PCs, smart phones, and even GPS devices are now capable of tapping into the web, further advancing the realization of true mobile computing. Output from real time data streams from Ocean Observing Systems such as the Alaska Ocean Observing System (Source: http://ak.aoos.org/op/data.php?region¼AK, accessed 12 January 2008) can now be viewed on web-enabled hand-held mobile devices.

1.5.1 Security Increased levels of unwanted and malevolent computer intrusions are a regrettable outcome of better connectivity. They have grown so rapidly in recent years that they are no longer just an aggravation. A recent report estimates that in 2000, hacking attacks and computer viruses cost global businesses around 40,000 human years in lost productivity equaling about $US1.6 trillion dollars

1

Past, Present and Future Trends in the Use of Computers

25

(Source: http://www.vnunet.com/vnunet/news/2113080/hackers-viruses-costbusiness-6tn, accessed 12 January 2008). The Internet was not designed with security in mind. It follows the democratic philosophy of open sharing. Because everything is interconnected, everything is vulnerable. For example, SPAM (unsolicited bulk email) is considered a threat because SPAM often contains malicious attachments. If the attachment is opened, it may unleash a virus. Because of overly aggressive SPAM filters placed on our email systems to protect us from malicious SPAM, email has become an unreliable form of communication. All too many times legitimate emails go undelivered or unread because they are incorrectly identified as SPAM. According to figures reported in Lyman and Varian (2003), unsolicited bulk email make up 40% of all email traveling over the Internet. In addition there are other serious attacks and threats delivered via email such as viruses, worms (software that attach to and take control of a normal program then spread to other computers). Very aggressive viruses can spread quickly. For example, the Sapphire/Slammer worm, released in February 2003, required roughly 10 min to spread worldwide. In the early stages of infection, the number of compromised hosts doubled every 8.5 s. At its peak infection rate, achieved approximately 3 min after it was released, the virus scanned over 55 million IP address per second and infected 1 in 12 emails on the Internet (Source: http://www.caida.org/research/security/sapphire/, accessed 12 January 2008). Other threats include, phishing (using email to search for valuable information such as a credit card number), trojan horses (software that appearing to be useful but is in reality malicious), and system monitors (software that tracks everything the user does, then emails it back to the author of the malicious software). Adware (software to display unwanted adds) and Spyware (software to secretly monitor and record what a user types) can also be serious as they tend to degrade system performance. Many of the hacker tools that required in-depth knowledge a few years ago to implement these threats have been automated and are much easier to use. A consequence to fisheries computer users is the unanticipated constraints or restrictions information technology administrators place on our use of computers because of real or perceived threats to system security. We need increased security to protect us, our computers and the computer network resources we rely on. Typically, computer security efforts focus on keeping ‘‘outsiders’’ out, through physical and technical measures such as gates, guards, passwords, locks and firewalls. In today’s computing environment, it is absolutely necessity to use virus protection software with current virus definition files. We are now often required to use ‘‘strong’’ passwords that have to be changed on a frequent schedule, use Virtual Private Network (VPN) software to remotely gain secure access to computer network resources contained behind firewalls, or encrypt entire laptop computer hard disks. As unfortunate as this situation appears, the reality is that serious computer hackers are in a cat-and-mouse dance with security professionals. We expect security issues to remain a part of the computing landscape for many years to come.

26

B.A. Megrey and E. Moksness

1.6 Collaboration and Communication The old paradigm of scientific communication consisted of publishing research results in peer-reviewed printed publication outlets such as journals and books. Collaborations took place mainly through face-to-face meetings and attending scientific workshops and conferences. Today, these reliable and still pervasive outlets remain important, but we have many other alternatives for scientific communication and collaboration. It is not uncommon today, some would say required, for a scientist to prepare their research results in several different mediums and formats such a written paper for publication in a published journal or one of the newer open access digital journals, a PowerPoint presentation published on the web, and a personal or laboratory web page. Email is currently one of the most widespread and pervasive methods of communication. Extrapolated from data supplied by the International Data Corporation, and reported by Lyman and Varian (2003), it is estimated that the current worldwide email volume amounts to about 60 billion emails a day or the equivalent of 1,829 TB daily or 3.35 PB (PB: 1015 B) annually. Lyman and Varian (2003) offer more intriguing statistics that were valid 4 years ago: email ranks second behind the telephone for the largest information flow; 60% of workers with email access receive 10 or fewer messages on an average day, 23% receive more than 20, and 6% more than 50. They estimate that workers spend an hour or less per day on email. They also report that 78% of workers send 10 or fewer messages on an average day and that 11% send more than 20. In our 1996 edition, we mentioned the benefits of using electronic communication and the benefits of the Internet to help us organize the book, especially email. By our current thinking and expectations, it would be unthinkable to contemplate organizing and planning an endeavor such as the second edition of this book without the benefits of email and the Internet. The decreasing cost of video capture and display technology along with widely available high speed Internet connectivity has fostered the increased use of personal video teleconference systems based on webcams, personal computer systems, and software compression. Use of video conferencing saves valuable time and reduces the cost of collaborations since often it removes or reduces the need to travel. The hardware used for this technology continues to improve in quality, prices have dropped dramatically, and the availability of freeware (often as part of chat programs) has made software based videoconferencing accessible to many. Voice over Internet Protocol (VOIP) is a common and widely used protocol developed for the transmission of voice though the Internet. Software products have been built around this new technology allowing instantaneous voice communications between two or more people using computers connected to the Internet. Typically these computers are portable laptops with a built in or externally attached digital video camera. There are many capable software programs to facilitation communication and collaboration via VOIP. Video

1

Past, Present and Future Trends in the Use of Computers

27

conferencing services that use VOIP are common today on most free instant message clients, such as Yahoo! Messenger (Source: http://messenger.yahoo.com/, accessed 12 January 2008) and MSN Messenger (Source: http://join.msn.com/messenger/ overview2000, accessed 12 January 2008) or the newer Windows Live Messenger (Source: http://get.live.com/messenger/features, accessed 12 January 2008). Skype is another widely used free software program using VOIP that allows users to make telephone calls from their computer to other Skype users free of charge, or to landlines and cell phones for a fee (Source: http://www.skype.com/, accessed 12 January 2008). All tools provide very similar features sets including video conferencing, instant messaging, text messaging, PC-mobile messaging, file transfer, and an ability to circumvent firewalls. One new developing medium to facilitate communication are blogs or on-line diaries where commentaries or news on a particular subject are placed for users to read. Communication is the key focus and readers can leave comments in an interactive format. A typical blog combines text, images, and links to other blogs, web pages, and other media related to its topic. As of December 2007, blog search engine Technorati was tracking more than 112 million blogs (Source: http:// en.wikipedia.org/wiki/Technorati, accessed 12 January 2008). Blogs related to the aquatic sciences exist as well. Some examples of marine and freshwater blogs are: ‘‘Thoughts from Kansas’’ (Source: http://scienceblogs.com/tfk, accessed 12 January 2008), MarineBio Blog (Source: http://marinebio.org/blog/?cat¼2, accessed 12 January 2008), blogs about fisheries (Source: http://www.blogtoplist. com/rss/fisheries.html, accessed 12 January 2008), What is Your Ecotype (Source: http://whatsyourecotype.blogspot.com/, accessed 12 January 2008), and John’s Marine and Environmental Blog (Source: http://jmcarroll-marinebio. blogspot.com/2007/10/letter-to-national-marine-fisheries.html, accessed 12 January 2008). LISTSERV, a pervasive mailing list manager, is also a newer medium for communication. Recent estimates suggest that LISTSERV’s sends approximately 30 million messages a day using approximately 300,000 mailing lists (Lyman and Varian 2003). Finally, we remind readers of the pervasiveness of PowerPoint or similar presentation software as a medium of communication. In the past, presentations at scientific meetings were made predominately using 35 mm slides and overhead transparencies. Now, it is rare to see these approaches used. Almost without exception, scientific results are communicated using presentation software such as PowerPoint, running on a computer connected to a digital computer projector. On many university campuses, open lectures are delivered using PowerPoint presentations. Often PowerPoint presentations given at scientific meetings are published to the web as a record of the meeting and a resource for those unable to attend the meeting. For example, the North Pacific Marine Science Organization (PICES) routinely performs this service (Source: http:// www.pices.int/publications/presentations/default.aspx, accessed 12 January 2008).

28

B.A. Megrey and E. Moksness

1.7 Summary We will see further miniaturization of the microcomputer, increased capacity both in terms of processing speed, memory and storage capacity, and relatively stable prices. Although we eagerly anticipate these changes, we probably will never be satisfied. As performance levels markedly improve so to will the complexity of our work, the size of the data sets we are expected to handle, and our expectations of ‘‘minimum system configuration’’ and ‘‘adequate performance’’. The desire for faster machines, better graphics, and larger hard disks will not go away in the near future. Computer software will continue to provide expanded analysis capabilities to fisheries professionals. We now have microcomputer operating systems that surpass what were commonly available on large mainframe computers many years ago, communication tools that make world-wide communication not only possible but fun, great suites of consistent, powerful productivity applications, personal database applications with sophisticated development and data management tools, and wider availability of highly specialized fisheries analysis programs. Better connectivity allows fisheries scientists to more easily communicate with colleagues within the scientific community who they might not otherwise have the opportunity to meet as well as external clients. Increased microcomputer connectivity holds great promise for research collaboration and synthesis activities. Opportunities abound for scientists to effectively communicate the value and relevance of their research findings to all audiences. The Internet is a proven technology, already traveled by billions of users around the world, and it will continue to open new frontiers that are difficult to imagine today. Moreover, it will provide expanded capabilities to share databases, computer programs, and experience. Carl Walters realized the benefits of an expansion of networked microcomputers and predicted this outcome back in 1989 (Walters 1989). Storage and sharing of fisheries database and historic information products is encouraging the systematic analysis of past fisheries information and facilitating inter-regional comparison of fisheries experience. Very often data sets from a scientific perspective become valuable when they become longer in length and one is able to put the data into perspective. Environmental data sets are most valuable after they are long enough to sample the target phenomenon multiple times. It is only when we start collecting data more regularly and routinely that we gain immense understanding into various processes. We see opportunities for scientists to ask increasingly broader questions as they focus their attention on processing and analyzing large volumes of data. Our ability to understand the fast paced and considerable changes we observe gives us a distinct advantage when dealing with complex multidisciplinary issues. Improvements in hardware and software performance will help investigators sift through increasing amounts of data in a much more powerful way or

1

Past, Present and Future Trends in the Use of Computers

29

execute computationally intensive statistical models such as the 550 parameter stock assessment model that includes a Bayesian Markov Chain Monte Carlo (MCMC) analysis (Iannelli et al. 2007). Through these activities, a deeper understanding of the data will improve the design of studies that look for associations between components of the biosphere on spatial and temporal scales relevant to biological and ecological systems. New and upcoming technologies will soon permit the ability to built truly integrated fisheries ecosystem models. For example, fine-scale biogeochemical-physical models dynamically coupled with spatially-explicit multi-trophic level ecosystem models will soon be commonplace, thus allowing holistic examination of the ecosystem and the response of its species-to-species interrelationships to management or climate change scenarios. Common patterns and trends may emerge as data sets are shared, made available for quick retrieval and analysis, and accompanied by companion data sets such as vast oceanographic and meteorological data libraries. One of the ongoing challenges will be the development of tools to provide this capacity to process large volumes of raw data and to make it available in real time over the Internet. Computers, fisheries research, and the resulting synergism are indeed exciting. We think that it is true that in any branch of science computers are essential. You can often set yourself apart as a valuable scientist and researcher by learning and using advanced computational tools (i.e. going beyond using basic computer applications). A necessary and valuable skill of future scientists will be to have some knowledge of how to solve problems and being able to look at a problem and come up with a computer solution. We hope the chapters in this book contribute to the development of this skill set. The following collection of papers, while not intended to be comprehensive, do characterize the breadth and sophistication of the application of computers to modern fisheries analysis, and documents the progression in advances in technology and their application to fisheries and resource management problems. We believe that the topics covered here are a prelude to new future opportunities and we anticipate with enthusiasm the challenges ahead.

References Carlson S (2006) Lost in a sea of science data. The Chronicle of Higher Education, Information Technology Section 52(42):A35 (Source: http://chronicle.com/free/v52/i42/42a03501.htm, accessed 12 January 2008). Hermann AJ, Hinckley S, Dobbins EL, Haidvogel DB, Mordy C (in press) Quantifying crossshelf and vertical nutrient flux in the Gulf of Alaska with a spatially nested, coupled biophysical model. Progress in Oceanography. Iannelli JN, Barbeaux S, Honkalehto T, Kotwicki S, Aydin K, Williamson N (2007) Eastern bering sea pollock. In: National Marine Fisheries Service Stock Assessment and Fishery Evaluation Report for the Groundfish Resources of the Bering Sea/Aleutian Islands Region in 2007. North Pacific Fishery Management Council, Anchorage, AK.

30

B.A. Megrey and E. Moksness

IPCC (2007) Climate Change (2007) Synthesis Report. Intergovernmental Panel on Climate Change. Kell LT, Mosqueira I, Grosjean P, Fromentin J-M, Garcia D, Hillary R, Jardim E, Mardle S, Pastoors MA, Poos JJ, Scott F, Scott RD (2007) FLR: an open-source framework for the evaluation and development of management strategies. ICES Journal of Marine Science 64:640–646. Kessler M (2007) Days of officially drowning in data almost upon us. USA Today, Technology News, March, 05, 2007 (Source: http://www.usatoday.com/tech/news/2007-03-05-data_ N.htm accessed 12 January 2008). Lunn DJ, Thomas A, Best N, Spiegelhalter D (2000) WinBUGS – a Bayesian modelling framework: concepts, structure, and extensibility. Statistics and Computing 10:325–337. Lyman P, Varian HR (2003) How much information? School of Information Management and Systems, University of California at Berkeley (Source: http://www2.sims.berkeley. edu/research/projects/how-much-info-2003/, accessed 12 January 2008). Moore GE (1965) Cramming more components onto integrated circuits. Electronics Magazine 38(8) (April 19, 1965). Nielsen JL (2006) Thoughts from Kansas. President’s hook. Fisheries 31(10):480, 514–515. Ona E, Dalen J, Knudsen HP, Patel R, Andersen LN, Berg S (2006) First data from sea trials with the new MS70 multibeam sonar. Journal Acoustic Society of America 120:3017–3018. O’Neill ET, Lavoie BF, Bennett R (2003) Trends in the evolution of the public web. D-Lib Magazine, 9(4), April 2003 (Source: http://www.dlib.org/dlib/april03/lavoie/04lavoie.html, accessed 12 January 2008). DOI: 10.1045/april2003-lavoie. Russell SE, Wong K (2005) Dual-Screen monitors: a qualitative analysis of their use in an academic library. The Journal of Academic Librarianship 31(6):574–577. Schnute JT, Maunder MN, Ianelli JN (2007) Designing tools to evaluate fishery management strategies: can the scientific community deliver? ICES Journal of Marine Science 64:1077–1084. Walters CJ (1989) Development of microcomputer use in fisheries research and management. In: Edwards EF, Megrey BA (eds.) Mathematical Analysis of Fish Stock Dynamics. American Fisheries Society Symposium 6:3–7.

Chapter 2

The Consumption and Production of Fisheries Information in the Digital Age Janet Webster and Eleanor Uhlinger

2.1 The Fisheries Information Life Cycle Fisheries scientists persistently create, communicate, and use information. In fact, if they did not, there would be no fisheries science. To exist, science must be part of a continuum where shared information, from casual hallway communications to rigorously reviewed articles, documents the questions asked and the solutions suggested. Relevant information is critical to the success of basic and applied fisheries research projects. Identifying the relevant at the beginning of a project and then communicating what is important out of the project are elements of the life cycle of fisheries information. Both have become simultaneously easier and more difficult as the amount of information increases within the digital environment. The access to information is simpler and yet more nuanced. As producers and consumers, we sustain the life cycle of fisheries information. We learn to consume information as students, often modeling our behavior from our professors. They give us a stack of reprints to read, and those articles become the foundation for our exploration into fisheries sciences. Or, we start with a pivotal article and work back through its references and forward through its sphere of influence defined by citations. Now, new alerting tools and search engines broaden our information horizons, enriching our perspectives while obscuring the relevant through the deluge. Consumption can be a feast of delectable facts, theories, datasets and findings or an orgy of the same leaving indigestion rather then satisfaction. This changing information environment also affects scientists as producers of information. We are faced with a plethora of publishing options where once there were only a few selective journals. We can publish in highly specialized titles with limited audiences, target the mainstream with high impact journals, issue findings electronically through blogs or web sites, or present at conferences where all becomes part of a streaming video record. The decisions we make when J. Webster (*) Oregon State University Libraries, Hatfield Marine Science Center, 2030 Marine Science Drive, Newport, OR 97365, USA

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_2, Ó Springer ScienceþBusiness Media B.V. 2009

31

32

J. Webster and E. Uhlinger

producing information are no longer straightforward, but require thought and preparation so the information produced is consumable and not half-baked or forgotten on a back shelf. The information life cycle has not changed fundamentally with the advent of computers, the Internet and digital resources. However, the environmental factors affecting how we produce and consume information have changed. A major factor is the growth of the digital network and how that shapes the ways information is published, disseminated and accessed. We should consider other factors as well when thinking about how to effectively consume and produce information. Fisheries science is no longer just about natural science; we need to increase our awareness of the information from the social sciences as the problems we address often have significant human components to them. The scale we work within has expanded as long term datasets become available, as genetic work reveals finer granularity, and as geographic limits diminish with extended global networks. The breadth of sources widens and we look to non-scientists for assistance with local information and insight. All these factors shape how we use information in our work. All enrich, yet none make it easier as they demand more decisions throughout the scientific process. The following attempts to assist with that decision making by describing approaches, options and challenges to consuming relevant information and then producing, or communicating, the same. On the consumption side, we will discuss how to identify, obtain and manage fisheries information. As tools change, the focus will be on strategies with specific examples of current tools. On the production end, we will explain the decisions to be made regarding intended audiences and possible outlets, publishing options, copyright considerations, access points, and archiving responsibilities. Finally, we will return to the electronic information environment to put the consumption strategies and publishing decisions into a larger context. Here we will touch on the economics of publishing and access, possible legal issues, the concept of the digital library, and information integrity and preservation.

2.2 Consuming Information 2.2.1 Identifying Fisheries Information There is a Chinese proverb that states: ‘‘Void of a long-term plan will bring you trouble soon.’’ This proves applicable to that point when you are starting a project. You need to consider your question and then your strategy for finding the answer. Uncovering the pertinent literature is a critical strategic step. Starting by typing keywords into Googleä returns reams of information, but often with a degree of randomness that may leave a queasy feeling of missing the right pieces.

2

The Consumption and Production of Fisheries Information

33

Before starting to look for information, it is useful to carefully think about what types of information you are looking for, who may produce it, and where it might appear. Then, you can tackle how to find it. Here are examples of questions to consider at this step in your research strategy.

 Broad or narrow topic?











The level of specificity may indicate where to start and where to look for information. The bigger or less focused the question, typically the more broadly you will need to look to identify as much relevant information as possible. It is difficult to answer a broad fisheries question such as the effect of global warming on salmon populations, by only referring to the work of population dynamics researchers. Limited geographic scope or global? If the problem is highly localized, you will want to concentrate on local information, yet with an eye on how others may have addressed the same problem. If global, the sources will be multinational and perhaps multi-lingual. Applied or basic research? The research continuum from basic to applied is paralleled by an information continuum. As research moves towards the more applied, different sources of information become more useful such as trade publications, patents, and government documents. Science or policy? Many fisheries questions have policy implications. So, it is smart to be aware of information that may be outside the normal scientific communication channels. Who? Understanding who has worked on the question provides a starting point as does considering who may have funded research or be interested in the outcome. Organizations as well as individuals may have a vested interest in the issue. Where? Related to the Who question is where the topic may be discussed. This suggests not only which journals may contain articles, but also which conferences or electronic discussion forums may address the topic. Considering where the conversation is generated may provide insight into where to look for current information as well as possible audiences for future communication.

2.2.2 The Tools Another proverb suggests that ‘‘A long march starts from the very first step.’’ That step after considering the types of information is identifying what tools may be helpful. These range from the general to the very specialized, from classic to contemporary, and from free to very expensive. All have a place in the information gathering process, but some will prove easier to use, more relevant,

34

J. Webster and E. Uhlinger

or more accessible depending on your circumstances and need. Tools change over time; some may become obsolete while new ones are developed. In the following, some specific tools are described with a discussion of their strengths and weaknesses in terms of content and access. They are grouped to help you identify the types of tools and then which ones you may be able to access given individual circumstances. 2.2.2.1 General Science Indexes These broad, science indexes generally cover the core fisheries literature adequately. They are solid starting points as you will find the major fisheries journals as well those in related fields such as ecology, biology and zoology. They are not the complete universe of fisheries literature, though. Additionally, almost all of those described are accessible through paid subscriptions only. Pricing usually depends on the size of the institution (e.g. number of FTEs) and size of the database (e.g. number of years covered). Web of Scienceâ Formerly known as the ISI Science Citation Index, the current electronic iteration continues to provide access to a broad suite of science journals in multiple disciplines. (A master journal list is available from the Thomson Scientific web site – www.thomsonscientific.com.) First published in the early 1960s, its continuing strength is its capacity to relate articles through citations allowing a user to investigate who is citing whom, who is working on related topics, and what are a topic’s core papers. Other resources such as CiteSeer, Google Scholarä and Scopusä are beginning to track citation patterns but currently not with the same accuracy (Roth 2005; Jacso´ 2006a). Its greatest weakness is the lack of coverage of monographs, conference proceedings, and report literature. It is also one of the most expensive general science databases so access may be very limited unless your institution subscribes. Subscriptions to Web of Scienceâ are priced in part by number of 5 year blocks of records; access to a complete range of years covered by the index increases the cost. While powerful, the search interface is not clean using some jargon that for occasional users makes searching challenging. The display of results can be cryptic until familiarity is gained with use. Web of Scienceâ remains the deepest general science index in chronological coverage and consistency of sources indexed. Its sister index, Web of Social Scienceâ, shares the same interface and is similar in construction and purpose. It is useful for delving in to the social and economics sides of fisheries. BIOSIS The tomes of Biological Abstracts are now electronically accessible as Biosis. This classic index for biological information covers over 6,500 journals

2

The Consumption and Production of Fisheries Information

35

including the core fisheries titles. Coverage includes some conference proceedings and reports. Its strength is its longevity (in print since 1927 with electronic access from the 1970s) and the depth of the indexing making it very searchable by subject and keyword for the power user. Its weakness is the lack of consistent coverage of non-mainstream publications including foreign language material and trade titles. Access is through subscription and is expensive. It can be purchased through a variety of vendors who then offer access to it through their search interface. Scopusä Scopusä is Elsevier Publishing’s foray into the general scientific index arena complete with citation tracking. It is strong competition to Thomson’s Web of Scienceâ, but may do it less consistently with noticeable gaps in coverage (Jacso´ 2007). The types of publications covered are broad including: journal articles, conference proceedings, patents, books and trade journals. The journal literature makes up its core. The depth of coverage in temporal terms varies depending on the subject area; life and health sciences coverage extends back to 1966 while the social sciences are covered from 1996 forward. For fisheries, the major journals are covered but not all the book series or potentially useful trade publications. The search interface is straight forward and the results display versatile and readable. Scopusä is competition to Web of Scienceâ, yet remains an expensive alternative resource. 2.2.2.2 Specialized Indexes Fisheries scientists are fortunate to have subject-specific indexes providing deeper access to the published literature than the more general ones. Often starting broadly and then working to the specific is recommended as you may find material that is tangentially related in the broad searching and then can hone in on the very specific. The down side of this approach is the duplication you will encounter. The following two examples are primarily accessible through paid subscriptions. While not as expensive as the general science indexes, these still represent a sizable investment for an organization. Aquatic Sciences and Fisheries Abstracts (ASFA) In the late 1950s, fisheries scientists at the Food and Agriculture Organization of the United Nations (FAO) began compiling a bibliography of documents ‘‘which contribute to knowledge of living resources of the seas and inland waters’’ (Food and Agriculture Organization of the U.N. 1958). The goal was, and remains, to provide coverage of the world literature through an international cooperative effort of monitoring and entering relevant documents. This effort is administered by the ASFA Secretariat located within the FAO Fisheries Department who then partners with Cambridge Scientific Abstracts

36

J. Webster and E. Uhlinger

(CSA), a commercial publisher, to enhance and produce the database. The current database contains over one million citations from the early 1970s to the present; older ones are added selectively. Coverage ranges from the mainstream science journals to conference proceedings to national documents. Over 50 partners including international organizations (e.g. International Council for the Exploration of the Seas and Network of Aquaculture Centres of Asia-Pacific) and national institutions (e.g. CSIRO Marine Research and IFREMER) contribute to the database making it rich in content. The official list of partners is maintained on the ASFA Secretariat web site (ASFA Secretariat 2006). The geographic diversity and variety of research foci of the contributing partners are strengths of ASFA. For some, this diversity is distracting as there is considerable non-English material as well as citations to documents difficult to access (e.g. limited distribution). The traditional subject scope was on living resources and a more applied perspective. That has broadened as more ecological journals are now monitored. ASFA is inconsistent in its coverage of the social science side of fisheries and living resources; management documents are not always included due to the reliance on local partners to contribute what they deem important. CSA does not regularly include material from social science and development journals, instead adding more science citations. ASFA consists of five subsets:

    

Biological Sciences and Living Resources; Ocean Technology, Policy and Non-Living Resources; Aquatic Pollution and Environmental Quality; Aquaculture Abstracts; Marine Biotechnology Abstracts.

To many users, these subsets are transparent. To database vendors, the subsets are useful as they can be packaged separately or in multiple configurations depending on the audience. CSA packages the complete ASFA and allows users to select subsets to search. National Information Services Corporation (NISC), another database publisher, packages the Biological Sciences and Living Resources subset with other databases to create its popular product, Aquatic Biology, Aquaculture & Fisheries Resources. Most institutions subscribe to the online version of the database through CSA or NISC for a significant annual fee. Those who contribute to the database as a partner receive free access through the internet or by CD available from the ASFA Secretariat. Institutions in low income food deficit countries are also eligible for free access. ASFA remains an excellent specialized index for fisheries scientists.

Fish and Fisheries Worldwide National Information Services Corporation (NISC) created this citation database by combining various existing databases, some ongoing and some ceased. These include:

2

The Consumption and Production of Fisheries Information

     

FISHLIT (from the J.L.B. Smith Institute of Ichthyology) U.S. Fish and Wildlife Reference Service database A fish subset of MedLine South Africa’s Fishing Industry Research Institute Database Castell’s Nutrition References NOAA’s Aquaculture database

37

This approach retains the value of older databases that are no longer maintained and enhances them with the addition of new material from other sources. Too often, older indexes become inaccessible as nobody sees the value of transforming them from a stand-alone database or a print bibliography. NISC attempts to capture such historic citation caches and build with them. Fish and Fisheries Worldwide is smaller than ASFA (less than 600,000), but very useful for its coverage of taxonomic records, sub-tropical freshwater fish, and U.S. local and federal government material. It also tends to cover some geographic areas more thoroughly than ASFA, Africa in particular. It is focused on fish and fisheries rather than the aquatic environment making it a useful tool for fisheries scientists. It is not as expensive as CSA’s ASFA making it attractive to institutions not needing the breadth of the full ASFA and looking for more specificity in some areas. Its interface is simple and quite intuitive for all levels of users.

2.2.2.3 The Worldwide Web as an Index The rapid growth of digital information builds the wealth of information available through web search engines. The Web still is a morass of information, good, bad and ugly. The search engines such as Googleä, Yahooâ and Askä are useful tools for sorting through the vast amount of digital information. As these engines evolve, their differences become more apparent and users should expect to see more differentiation in how they search and display results. Scientists need to know what sources they are searching. The established indexes such as BIOSIS and ASFA clearly explain what journals and sources they draw from; the web search engines are rarely as clear, and never as focused. However, they tend to cast a broad net, useful for establishing the scope of a project or trying to find something specific fast or with little effort. Whatever the reason for using a web search engine, it is how many start and end the quest for information. It has obvious and not so obvious problems, yet can yield satisfactory results. Fisheries scientists should recognize the limitations of web searches and know when to use indexes that will go deeper into the literature. This entails checking the ‘‘about’’ on each search engines home page. Rarely does a company specify exactly how they are searching and ranking the results of the search. However, a user can get an idea and recognize why different engines come up with different results. Googleä was the first, and holds the patent, on the search and ranking system referred to as page-ranking (Page et al. 1998).

38

J. Webster and E. Uhlinger

The algorithm considers how many pages are linking to the specific page as well as relative importance of the referring page. Askä tweaks the page-ranking by attempting to cluster like pages and analyses the relationship among those pages, thus returning pages that link within a topic area and not those random linkages. Some, such as Yahooâ, integrate paid or sponsored sites into the rankings; while this practice probably does not affect search results for fisheries science information, it could for fisheries trade information. In contrast, searches within the indexes described earlier are worked through a closed set of citations with field tags (e.g. author, keyword, title) so results are ranked by matches to the contents of the fields searched and not by the complexities of relative importance among the citations. It is a controlled information environment as compared to the wide open Web. Yet, the convenience of a simple interface and direct links to the full text of articles make web search engines attractive. The rest of the information world – database vendors and libraries included – scrambles to package their resources with as simple an interface. They are also tailoring what is searched and how to provide the scholarly audience search tools that integrate with existing work patterns and computer desktops. Scirus from Elsevier and Google Scholarä are examples of free multidisciplinary indexing and abstracting databases.

Google Scholarä (http://scholar.google.com/) Googleä launched this service in 2004 with much fanfare. In essence, it is a subset of the Web providing access to ‘‘peer-reviewed papers, theses, books, abstracts and articles, from academic publishers, professional societies, preprint repositories, universities and other scholarly organizations’’ (Google 2005). Yet, it does not specify which publishers and institutions participate leaving the user to guess or take it on faith that the coverage is broad and wide (Jacso´ 2005a). Additionally, it is unclear how often and how deep various sites are mined for results, leaving gaps in coverage revealed if the publisher’s site is searched directly (Jacso´ 2005a). Research conducted on its coverage and utility suggest that it is stronger in the sciences than social sciences and has a definite English language bias (Neuhaus et al. 2006). The search interface is familiar and simple with an advanced option that increases its utility. The links to full text articles (if the user’s institution has implemented the service) make searching and getting items more efficient. With the addition of citations to the search results, some suggest that Google Scholarä can replace Web of Scienceâ or the newer, Scopus (Pauly and Stergiou 2005) while others urge scholars to use it in addition to the more structured databases (Bauer and Bakkalbasi 2005). The fisheries scientist will find it an easy place to start, but should continue exploring the literature in one of the specialized indexes for more thorough coverage of the field’s varied information.

2

The Consumption and Production of Fisheries Information

39

Scirus (http://www.scirus.com/) Elsevier Publishing started this free search service focused on its deep database of articles and over time has added other sources such as patent data and electronic theses and dissertations (Pruvost et al. 2003). Unlike Google Scholarä, Scirus is open about what is covered within its scope providing direct links to those partners. The search interface includes the familiar simple box as well as an advanced option that helps the user narrow results by terms, years, format and source. The strength for fisheries people is the coverage of Elsevier’s journals, some of the most widely cited in the field. Its weakness is the hype as Elsevier claims that it is ‘‘the world’s most comprehensive science-specific index’’ (Elsevier Ltd. 2004). Again, Scirus is more structured than Google Scholarä and more transparent giving it greater credibility. It is a decent resource as long as it is used in conjunction with others.

2.2.3 Searching Effectively The myriad of tools available to the fisheries scientists adds confusion to identifying information. The tools described above represent some of the most accessible or most useful. In deciding which to use, what you are investigating can suggest were to look. Broad, inter-disciplinary questions need to be investigated using indexes that are temporally and topically deep while geographically inclusive and covering multiple disciplines. One tool is not adequate for a thorough search for information. Each has its particular strengths in terms of coverage and search sophistication. Any sophisticated searcher should be aware of the scope of content of the database or span of coverage of a web search service. Also, the user will eventually know when to go deeper for information and when the obvious is good enough. Another consideration in choosing an index or a search engine is the search interface and the results display. Features are constantly being refined by all; however, there are basic ones that make a tool usable (e.g. searching within a field such as title) and those that increase its value (e.g. linking to full text). Some users will always execute simple searches and not experience some of the satisfaction that results from refining a search or ferreting out a resource not readily searchable by keyword. The following discusses three functions inherent in search interfaces that can reveal differences which may influence use. 2.2.3.1 Searching Options The ubiquitous search box presents the simple option of entering in a single keyword and getting results. To some, a complex search is adding more keywords. Any database or search engine should have this basic search option as there are times that a single term or a simple phrase is adequate,

40

J. Webster and E. Uhlinger

and more choices confusing or extraneous. However, there are times that a simple keyword search does not produce any results or does not reveal the relevant. One obvious possibility is misspelling; not all databases have a spell checking facility. Other possibilities to consider are the structure and the scope of the resource being searched, and the structure of the search query. Scope has been discussed earlier; however, it is useful to briefly discuss it again along with structure. The various indexes will return different results from the same search strategy. The differences reflect their scope and content. Different web search engines return varying results as they use slightly different searching algorithms and relevancy factoring (Spink and Cole 2006). Tools exist to visualize the overlap (and lack of it) between various search engines (Jacso´ 2005b). An efficient approach to the overlap issue is the ability to search across resources. Some web search engines use this approach (e.g. Dogpile). Within the citation databases, some vendors allow you to search multiple databases simultaneously, so you expand what you are searching and usually increase your results (although you also increase the duplicated citations.) Librarians are developing federated search tools so the user can generate a simple query that is executed across a wide suite of information resources (Avrahami et al. 2006). This concept is quite powerful as web search engines do not penetrate the ‘‘Deep Web’’, material protected by passwords, licenses or structure. An example of the later are library catalogues that while openly searchable are not mined by the typical web search engine as their records are within a database that is not probed by the web crawlers. The same premise holds true for a structured database such as BIOSIS or the Web of Scienceâ. Basic searches using Googleä will return many results, but will not necessarily search deeply into specialized indexes or resources. The structure of the search query is another consideration for effective searching. A simple keyword search can build into a query with multiple field-specific terms. Adding synonyms or related terms can increase search results as can searching across all fields in the resources. For example, if the basic search in a given system is limited to selected fields such as title and author, it will not return citations where the keyword is embedded in the abstract. Building effective search queries involves the above as well as informed use of phrasing and the Boolean terms (‘‘and’’, ‘‘or’’ and ‘‘not’’). Some web search engines assume multiple keywords have ‘‘and’’ between each rather than ‘‘or’’; this approach tends to restrict results. If adjacency of keywords is important, such as ‘‘population dynamics’’ or ‘‘freshwater aquaculture’’, using quotes is usually a trigger for a search engine to search for the phrase rather than the individual words. Boolean terms allow users to build sets, narrowing or expanding results, and helping them find the most relevant information. An obvious time that a fisheries scientist would use Boolean terms is searching a particular species where it is important to use the scientific and common names to retrieve all pertinent references (Fig. 2.1).

2

The Consumption and Production of Fisheries Information

Oncorhynchus tshawytscha

41

Chinook

Fig. 2.1 Simple Boolean search indicating the possibilities of expansion (considering both sets so references with either term), and narrowing (considering references containing both terms)

Some search interfaces build Boolean searching into their advanced options using multiple boxes with connecting terms (Fig. 2.2). The option is almost always available even if query and connecting terms must be manually entered. Figure 2.3 illustrates how Boolean terms work conceptually when combining search terms. Using a basic search, each concept is placed in quotes or parentheses and searched to form a set of results. These sets are then combined with ‘‘and’’ to narrow the search to a subset. Using ‘‘or’’ as the combing terms would expand the results to include all sets. The term, ‘‘not’’, is used to exclude a concept that interferes with the results; for example, to find information on the marine phase of salmonids, the sets of keywords or phrases are searched and then the freshwater set excluded.

Fig. 2.2 Cambridge Scientific Abstracts’ Illumina search interface integrates Boolean search terms

42

J. Webster and E. Uhlinger

Fig. 2.3 Boolean search illustrating the variety of combinations possible with thoughtful searching of terms (* is a common sign for truncation and in this example will retrieve results containing the root of salmon such as salmonids)

freshwater

marine or ocean

salmon*

A basic search option is very powerful if used thoughtfully. However, a well designed advanced search option is critical as it allows for more specific and often more efficient searching. Few besides those who search daily and librarians regularly use advanced features; in fact most who use web search engines rarely exploit the great potential of more advanced searches (Jansen et al. 2000). Databases and web search engines of value offer advanced search features. Some of the features are described below:

 Field searching

 Example: if you only want to retrieve documents written by a particular author and not those containing citations to that author, you would limit your search to the author field.  Example: if you are looking for articles that have a primary focus on a topic, you may limit a keyword search to the title rather than the entire record including the abstract.

 Limiting

 Example: if you want only the most recent references, you limit your search to the current year within the publication date field.

 Example: if you only want articles from a certain journal, you specify that journal in the source field hence limiting the range of publications searched.  Example: if you want to find all articles published by authors in three countries, you add those countries within the author affiliation field to your search.

 Format

 Example: you only want those references that are readily available as full text, so you limit your search to full text.

 Example: you want a review article, so use the publication type field to refine your search.

 Example: you may want to find images so will want to limit your search by file extension such as jpeg or gif.

2

The Consumption and Production of Fisheries Information

43

 Thesaurus or keyword list

 Example: you are looking for a scientific name but cannot remember exactly how to spell it.

 Example: you are not finding anything using a particular keyword, so want to find other synonyms.

 Search history

 Example: you executed a complex search for a species that you want to combine with an earlier search on habitat and life history.

 Example: after a long search session, you want to retrieve an earlier search that had some references you forgot to note. Search interfaces constantly evolve as their creators integrate user feedback into making a better mousetrap. While laudable, it is also disconcerting as you get used to working in certain ways. Too many bells and whistles become distracting without adding much utility. So, when deciding on tools to use, it is perfectly acceptable to use those that present the most understandable and easy for you to use interface. Mastering the basic search using Boolean logic will greatly improve search results. Adding an understanding of field limiting and using controlled vocabulary will enhance efficiency and efficacy.

2.2.3.2 Displaying Results The display of results can affect their utility to the searcher. Too much information slows down the ability to scan for relevancy, yet too little leads to guessing and perhaps missing important documents. A well designed interface allows the user to tailor, to some degree, the results display showing more or less detail as desired. For instance, a simple list of titles can be easily scanned for interesting citations; yet, a more complete record with the abstract is valuable if looking for something particular. Web search engines do not currently have the same capacity for manipulating the display of results. A decent interface will also allow the user to sort the results by date or relevancy if not other factors. Again, web search engines do not currently allow this as they are not working with a controlled and limited database of citations. There are certain obvious elements of any display of results. These include the following:

 Title of the resource  Author(s) including first initials if not complete name  Basic citation information such as the journal title, volume, date and pages or conference name

 Abstract or simple description

44

J. Webster and E. Uhlinger

The last element in the list, the abstract, is often critical in deciding whether something is useful. Many citation databases have complete abstracts as written by the authors or database editors while web search engines automatically create a summary using various strategies. The content of the summary should help the user decide if the resource will be useful or relevant to the current search. Fisheries scientists are accustomed to the classic abstract so can read through a well-written one and grasp the research question, the methodology and the results. Web summaries can be problematic as they do not have a consistent structure and being short, do not always provide enough context or information (White et al. 2003). On the positive side, it is often simple to click through to the document itself or a more complete description of the item. Additional display features, while not critical, can be useful. These are usually of two types: the first group being elements that provide more information about the item and the other type being connections to additional information or the item itself. The former are most visible in citation databases with structured records. The value of the records is increased with the addition of more complete publication information including publisher information and a complete citation as well as more information on the author such as affiliation and contact information. Often subject headings or descriptors have been assigned; these allow you to search for other records with the same descriptors, a useful tactic when exploring a topic. The latter type of elements, external linkages, is a newer development as linkages to full text of articles and other resources have evolved. With web search engine results, the greatest feature is the link to the full text of an item, although too often that link is to an incomplete citation or reference buried within another document. Linking to full text is not assured as the full text of an article may be restricted to those with licenses or authority. The citation databases can be integrated with an institution’s journal databases so linkages are automatic if the institution has a subscription to that journal. This is done through implementation of an OpenURL resolver, a software that gathers the information about a user, the institution’s licenses and the information resources, and then matches the access rights (McDonald and Van de Velde 2004). Even with the limitations to access, linking out to full text resources is a boon to the fisheries scientist providing faster access to information. Another form of link is to related records or similar pages which can lead to resources of interest. Sometimes these linked resources are related only through payment to the search engines, and sometimes they are related through shared keywords or source. Within a scientific information database, the relatedness may be through shared references or shared subject descriptors. The results display in many citation databases give the user more complete information about a resource and allow some manipulation of the results set. The results display of web search engines can reveal a wealth of information not covered by the citation databases and usually provide some kind of direct

2

The Consumption and Production of Fisheries Information

45

access to the full text if available. So, the differences in display once again reinforce that one tool does not satisfy every information need or every user’s expectations.

2.2.3.3 Using Results Finally, there are differences in how to use the results. Linkages to more information including the full text exemplify one use. Others involve manipulating the results for further use. Effective use of results can ease the research process. Tracking what has been searched and found relevant allows compilation of sources in a logical manner. The web search engines are not as conducive to this more structured information search; rather than marking a list of references and then checking as a batch, you must click back and forth between the results page and possible documents of interest. When using a web search engine, one strategy is to maintain a research log and cut and paste relevant or interesting web page addresses along with the date accessed so you can return to the site. The citation databases allow the user to mark references of interest as they are perused, compiling them into a subset. Then the user can print, download, email or simply review the subset.

2.2.4 Managing Information Another Chinese proverb states ‘‘Once on a tiger’s back, it is hard to alight.’’ Ferreting out the information can become addictive and the consumer of information becomes consumed with the task. Knowing when to stop searching and start reading and synthesizing is as critical as knowing how to start searching. It is almost impossible in this age of rapid information transfer and burgeoning information resources to feel that you have found everything on a topic. However, you can be confident if you have worked through your information searching logically and systematically. The logic can be temporal – starting with the historic pieces and working forward or vice versa. Or, it can be database-centric – executing similar searches across multiple databases. Over time, you will devise your own methods and process. Maintaining a research log can be useful for managing the process. This entails simply noting what databases you have searched when and what search strategies you used. You can then re-execute those strategies at a later date if working on a long term project. You will also remember what you have already done if you get interrupted or return to a project. Another important component of managing the process and the information gathered involves organizing what you find. Random citations jotted down on slips of paper or emailed to your mailbox are easily lost and have little context. It is not enough to copy or print off various articles; you need to keep them organized so you can use them. One method is the old-fashioned list compiled as

46

J. Webster and E. Uhlinger

information is gathered with the corresponding reprint file. This method has been updated with the advent of easy-to-use bibliographic software such as EndNoteä or Zotero. One way of looking at this type of software is that it replaces the old card files; however, it has much more potential as a highly useful research tool (Mattison 2005; Webster 2003). Most bibliographic software allows the user to enter records with the typical fields of author, title, source, add personal annotations through keywords, notes and abstracts, and even link to digital versions of the item. The resulting personal citation database is searchable and serves as a tool to manage your research. Beyond that, the most valuable aspects of bibliographic software are its ability to import records you have identified while searching the citation databases and its capacity to format those citations in a variety of styles as you use them in your writing. Some see this as just another software package to learn, so procrastinate. Those that do make the effort to use one of the many available bibliographic software packages available find it a valuable tool for managing information from consumption through production.

2.2.5 Obtaining Information It is one thing to identify information resources and yet another to actually get them to read and review. This step is made easier with the increase in digital information and the integration of links to articles from the citation databases and within the Web. Those of us working within research, governmental and educational institutions often enjoy broad access to digital information and well-stocked libraries of print material. Obtaining material is not always perceived as an issue. However, we enjoy that access because the digital material is either freely available through open access repositories or web sites, or purchased by the institution. The institutional entity usually responsible for maintaining adequate access to information is the library. Remove the licenses the library has negotiated, purchased and maintained, and a fisheries scientist would be frustrated by the lack of seamless access to electronic journals in particular. So, the library should be a researcher’s first means of obtaining information whether virtually or physically. A core principle of librarianship is to connect the user with the information needed (Ranganathan 1963). The format, topic or source does not matter, but access does. If stymied in obtaining information, work with your librarian to secure electronic access or to facilitate a loan or purchase. Not all fisheries scientists have a librarian or a library. Exploring if the material is freely available in electronic format is currently the favored approach. This entails looking beyond an initial search of the Web to investigating the digital holdings of relevant organizations. For example, the Food and Agriculture Organization of the U.N. has a large digital document

2

The Consumption and Production of Fisheries Information

47

repository that is available to all; however, most web searches will not penetrate this rich source of full text documents as it is not structured to be readily mined by the search engines (Food and Agriculture Organization of the U.N. 2006). One strategy for finding electronic documents is to look to the organization responsible for publishing the document in question or funding the research. The growing trend towards institutional repositories that capture the digital output of an organization increases access. However, often these repositories must be searched individually by going to the institution’s web site. If a freely available digital version of a piece of information is not readily available, the next step is to request it from the author or producing organization. This used to be a common practice and why authors continue to get a stack of reprints from their publishers (although often at a cost.) More authors are advocating for electronic reprints usually as a pdf that can be posted on an institutional web site for sharing with colleagues. Before posting to a site that is publicly available, the authors should verify that they secured that right as part of the copyright agreement with the publisher. If this is not the case, another way of sharing electronic reprints with requestors is to put it on an FTP site that is either password protected or time limited so access is restricted to those authorized. The final option is paying for the information, something libraries do daily, but the individual researcher does rarely. Many publishers of scientific articles and reports have simplified paying for individual articles. There will be times when reviewing citations that you will link to an article that your institution does not subscribe to, or you have linked to that article in a way that the publisher does not recognize that you are affiliated with a subscribing institution. At that point, most systems will request a login and password, or your credit card number. Before despairing, check with your librarian to see if you should have access. If not, then you will have to decide if the article is worth the cost.

2.2.6 Staying Current with Information Given the perceived deluge of information, it can be daunting to stay current with research, policy changes and management decisions. Various tools are available to help address the challenge. These include electronic tables of contents, personalized alerts, discussion lists and RSS feeds. Each has its strengths and weaknesses, but all provide ways to stay informed. Browsing the tables of contents of relevant journals is a tried and true method. It is an easy way to see what is being published as well as a means of discovering information that you may overlook in a search. Most publishers maintain journal web pages containing the tables of contents by issue. These are easily browsed when accessed. A more effective method is to subscribe to email

48

J. Webster and E. Uhlinger

alerts either through the publisher or through a compiler. Setting up alerts involves going to the publisher’s web site, registering and selecting those journals which interest you. Then, you will receive an email whenever a new issue is published. You will need to go to a variety of publisher sites to cover all of the publications you may want. An alternative if available to you is to use a service such as Current Contentsä or Ingentaä that allow you to set up an account and select journals by a variety of publishers. The strength of such services is the ability to manage one account for access to multiple publishers and their journals. The drawback is the cost; both the examples above involve a substantial annual subscription fee that your institution may or may not choose to pay. In addition to table of contents alerts, many publishers and citation databases include a feature for search alerts. The concept is that you may have a search that you want to conduct regularly such as a species, an author, or citations to your own publications; a search alert provides a mechanism for running these search strategies on a regular basis and having any results emailed to you. Even if a favorite citation database or publisher does not have the alert capability, it may have a way to store your search strategies so you can easily retrieve them and run at a later date. This alleviates reconstructing a search that was productive. Some alerts are automatically run and sent weekly even if there are no new items; others only generate an alert when there is something to send. Either way, it is a simple way to keep informed on new publications by certain authors or on a particular topic. Another way to stay informed is to subscribe to relevant electronic discussion lists. Some generate too much traffic in your email box, but others may be a valuable resource for learning about new developments in your field. Lists seem to be a particularly useful for announcements of new books and reports as publishers or authors find them a useful way to generate interest in a publication. LISTSERVâ, one of the major software tools used for creating discussion lists, maintains a searchable list of those lists thus providing one tool for identifying appropriate discussion lists (http://www.lsoft.com/lists/listref.html). Asking colleagues which lists they subscribe to is often the most effective way of finding relevant lists. Most professional organizations also maintain email lists that can be useful ways to stay informed. RSS (real simple syndication or rich site summary) feeds are one more tool to mention in this day and age. Many web sites incorporate this tool as a means to ‘‘push’’ new information to those interested. A typical way that such feeds are encountered is at the bottom of many web sites where a stream of news is constantly changing; this is an RSS feed. Subscribing to RSS feeds allows you to monitor changes in a web site of interest such as a blog on marine fisheries management or a particular site that lists fisheries jobs. A simple way to do so is by using an aggregator such as Bloglines or NetVibes. A web service that allows subscribers to set up a personalized web site that monitors selected web sites and blogs.

2

The Consumption and Production of Fisheries Information

49

2.2.7 Information Consumed Searching effectively entails all of these steps.

 Learning how to structure searches.  Investigating options for displaying and using results.  Selecting the most appropriate resource to search. The last may be the most important. The best search interface is meaningless if the suite of information being searched is irrelevant to the searcher. The broadest citation database is worthless if it does not cover the discipline being investigated. The World Wide Web is multi-dimensional and searching its most accessible dimension is not adequate for scientific research. So, select your information tool carefully and search intelligently. As the Web, information resources and computing evolve, more tools will become available for consuming information.

2.3 Producing Information After completing the analysis of research findings, the penultimate step of the scientific process is communicating the results. Scientists present their findings to others for ratification, verification, and discussion, consequently contributing to a discipline’s body of literature. After building upon the work of others by ‘‘consuming’’ information, we scientists produce information. At this step, you make decisions that can help shape the body of scientific literature through effective scholarly communication.

2.3.1 Audience Various modes of communication are available to scientists and choosing the proper one begins with determining the intended audience for the work. The intended audience often shapes the focus of the content, the style and the venue. For example, the elements necessary to explain a scientific finding are different than those for recommending changes to fisheries policy; one may require more text while another may depend heavily on data presented in graphical format. Traditionally, fisheries scientists wrote for other scientists. In contemporary society, they also may need to communicate to the lay person, policy makers or students. Each audience responds best to communication directed to their information needs and use patterns (e.g. regular reading of scientific journals versus browsing of fisheries web pages). With the advent of electronic delivery, it is easy to lose sight of intention. Scientists may read research summaries on public websites rather than seek out the peer-reviewed paper. Or, students may stumble on the paper when a summary or simpler explanation may better fit their needs. This blurring is driven by practical considerations of time and effort

50

J. Webster and E. Uhlinger

(it is fast and relatively simple to find information on the Web versus sorting through the peer-reviewed journals even when available electronically). Yet, it does not negate considering audience when producing information. The blurring suggests that information once produced for a single audience of scientists can now be used by more than one audience; so, thoughtful production is necessary. Scientists write technically for other scientists, and tend to follow a prescribed structure that reflects the scientific method. The outlets are scientific journals and conference proceedings with the peer-reviewed journal article being the most credited communication piece. While individual journals have different styles and requirements for authors, all require common elements such as an introduction, an explanation of materials and methods, and a discussion and analysis of data and results. Additionally, fisheries scientists use common terminology such as internationally accepted scientific binomial names for the organisms described, international units of measurement, and technical abbreviations and acronyms that are often used without explanation. These standards facilitate the communication among scientists as readers can maneuver through the common structure. By contrast, communicating fisheries science to general audiences requires less technical language that describes the subject matter in an understandable manner as these readers do not share the common language of peer-reviewed science. Illustrations become an essential means of explaining the issues and the process for addressing them. Organisms may be referred to by their local vernacular or common names rather than their binomial scientific names. This common practice makes literature more accessible to local readers and those unfamiliar with scientific names. The methodology may be the focus of the writing rather than the findings (e.g. explaining how a pit tag works) if that addresses the interest of the audience. The purpose is usually more education and information rather than the drive to document and validate found in peer-reviewed communication. Policy communications blend the popular and the scientific. When fisheries scientists work with policy makers, they are usually providing an expert opinion or scientific findings. Policy makers are not scientists although many may have extensive scientific experience and credentials. Consequently, scientific language is adapted so concepts and findings are well articulated and understandable to the lay person. Fisheries scientists when working with policy makers decide what role they are playing – scientist or advocate – and shape their writing to reflect the decision (Lackey 2006). Some would say that this decision is arbitrary, yet the communication will be shaped by the nature of the language, the tone and the viewpoint. This makes policy communication challenging.

2.3.2 Publishing Venues Once the audience is recognized, you select a publishing venue that addresses your audience, its needs and its information seeking behavior. The growth of

2

The Consumption and Production of Fisheries Information

51

the Web adds new venues as well as expands the reach of existing ones. The borders between venues blur. Peer-reviewed articles are available electronically so become elements of websites rather than limited to a bound journal. Policy statements are posted to web sites in a timely manner so edits and revised drafts are immediately open to scrutiny by the interested public and those affected by the decision. The electronic environment opens communications in terms of access and timeliness. While venues blur, fisheries scientists still need to focus on a primary one when crafting their communication. The publishing venue is shaped by those who contribute to it, those who read what is published and then by the venue itself. 2.3.2.1 Peer-Reviewed Journals Scientists prefer peer-reviewed journals for most scholarly scientific publications as their primary audience reads them and the authors usually get greater credit for their career. Peer-review is a collaborative process whereby papers are submitted to an Editor who in turn solicits anonymous review of the work by other scientists working in the field. These anonymous reviewers assure quality control by evaluating the materials and methods; the veracity and repeatability of the findings; and the contribution, if any, that the new work makes to the field. Peer-reviewed journals can be broad in scope (Science and Nature) or more specialized (Fisheries Oceanography). They may be produced by commercial publishers (Elsevier, Wiley), societies (American Fisheries Society, World Aquaculture Society), institutions (Journal of the Marine Biological Association of the UK ), or governments/non-government organizations (National Research Council of Canada, the International Whaling Commission). The cost of peer-reviewed journals is highly variable, with commercial publications often being extremely expensive and government publications being less expensive or even free. This is an important point to consider when selecting a publishing outlet; a high cost journal may have a more limited readership than a freely available publication. Again, the intended audience is one consideration in selecting the appropriate publishing venue. Scientists debate the quality and ‘‘impact’’ of peer reviewed scientific journals. ‘‘Impact factors’’ are one method for determining the ‘‘value’’ of a journal and such factors are considered by many institutions for purposes of conferring promotion, tenure, and grant monies to those who publish in ‘‘high impact’’ journals. The term ‘‘impact factor’’ was coined by Dr. Eugene Garfield and applies only to those journals indexed by Thomson Reuters Scientific in the Web of Scienceâ (described in Section 2.2.2.1). The impact factor is determined by a simple mathematical formula that divides the number of published articles in a 2-year period of a journal title, by the number of citations to those same articles in a different 2-year period (Garfield 1994). The impact factor is highly controversial, widely misunderstood, and frequently irrelevant in fisheries science for two reasons. First, the Web of Scienceâ does not index many relevant fisheries publications that fall outside of the mainstream. Second, many

52

J. Webster and E. Uhlinger

fisheries scientists publish outside of fisheries journals, so the relative impact of the 40 titles in the fisheries cluster does not reflect the importance of an article in Conservation Biology, for example. The Web of Scienceâ does not quickly add titles due to changing research interests; for example, Fisheries Oceanography, first published in 1992, was not indexed in the Web of Scienceâ until the last issue of the 4th volume published in 1995, and Journal of Cetacean Research and Management first published in 1999 is still not covered in 2008. All the above suggests that the peer-reviewed journals have a definite place in documenting and communication fisheries science, but often too much emphasis is put on this sector of the information web (Lange 2002; Webster and Collins 2005). 2.3.2.2 Professional and Trade Journals and Newsletters Essential to fisheries science are the publications produced by scientific societies and industry organizations. These often address applied research issues and results such as stock assessment, policy discussion, and trade information and trends. Such publications also take many forms including professional journals that may be peer-reviewed (Fisheries), trade journals (National Fisherman, World Fishing) or even popular magazines (Blue Planet, Oceanus). These publications are focused on specific aspects of a discipline, or may seek to bring varying viewpoints together around a particular technology or policy issue. Because much fisheries literature is ‘‘applied’’ rather than ‘‘experimental’’ in nature, these publications provide an important outlet for best practices, describing new gear or technologies, and stimulating debates by creating a forum for policy discussions. 2.3.2.3 Grey Literature Another form of publication common to fisheries literature is the so called ‘‘grey literature.’’ As the term suggests, this venue is not obvious and often not accessible to all, yet critical as it encompasses much that is not commercially published. Finding or consuming grey literature can be problematic because not enough attention is paid by authors producing it. A prime example is a technical report containing datasets and observations that are not distilled into a format suitable for publication in a several page article. Master’s theses or doctoral dissertations may be considered grey literature, as are data sets or time series. Grey literature may be individual reports or comprise parts of long standing series (such as the many series published by FAO). It is often published by government entities, non-governmental organizations and international organizations. Distribution may be extremely limited, yet critical to those seeking to understand a particular issue or search for a specialized dataset. This limited distribution, coupled with lack of peer-review, means that grey literature is under-represented or excluded from many general abstracting and indexing services. Specialized databases (such as ASFA and Fish and Fisheries Worldwide), however, specialize in identifying such literature, which further extends

2

The Consumption and Production of Fisheries Information

53

the value and importance of the data to scholars and policy makers. Grey literature may also be assembled into aggregated databases such as the National Sea Grant Library (National Sea Grant Library 2006) or in collections at government agencies (Office of Scientific and Technical Information (U.S.) 2006) or organizational repositories (Food and Agriculture Organization of the U.N. 2006). Even though distribution is uneven and publications may be difficult to locate, the grey literature in fisheries science provides a large reservoir of important information. Some classic studies were originally published in government series (Beverton and Holt 1957). Practical management guidelines often appear in this venue. Even the fodder for ongoing debates can first appear in the grey literature (Food and Agriculture Organization of the U.N. 1995; Pew Oceans Commission and Panetta 2003). For many fisheries scientists, this is their venue as their agencies and organizations expect reports and not finely tuned journal articles; or, their annual reports or technical handbooks are more appropriate means of communicating to their audience. Grey literature takes many forms, with varying styles and differing purposes. As a venue, though, it is important to recognize and use it.

2.3.3 Copyright Copyright is an important but subtle and often confusing aspect of scientific publishing. It used to be a peripheral concern when making decisions about publishing venue. Now, copyright can be a deciding factor in whether an article is accessible to all readers and usable by the authors. It is worthwhile to have a working knowledge of copyright so authors can make thoughtful decisions. Copyright laws differ from country to country, but all seek to protect the intellectual property of an author. The World Intellectual Property Organization (WIPO) and treaties such as the Berne Convention (signed by 162 countries since its inception in 1886) work towards collaborative and shared recognition and enforcement of member nations’ copyright laws. At the most basic level, copyright confers to the copyright owner specific privileges:

 The right to reproduce the work;  The right to prepare or authorize derivative works based upon the copyrighted work;

 The right to distribute copies and collect royalties;  The right to display or perform copyrighted works. Copyright typically resides with the creator of a work. One significant exception is that the work of US Federal Government employees is not copyrighted and is in the ‘‘public domain’’ where it is freely usable by anyone anywhere in the world. This is also true for many state employees, although the law varies from state to state and institution to institution.

54

J. Webster and E. Uhlinger

A perplexing trend has taken place in scholarly scientific publishing over the years. Publishers usually require authors to sign over their copyrights to the publisher in order to have the work published in a scientific journal. Publishers claim that this right is necessary for them to protect and responsibly manage that piece of intellectual property for the legal term of the copyright. In the US, copyright currently lasts for the life of the author plus 70 years, which seems an inordinately long term for a scientific work to need protection or to be managed. Further, while copyright transfer is common practice in scientific publishing, it is not common with many other types of publishing (such as fiction and legal publishing). Because copyright assignment gives exclusive rights to the copyright holder, authors (aka creators of the work) may be prohibited from using their own work in other format or forum – such as classroom teaching, distributed learning, for inclusion in other works, or posting on a personal web site. Alternative copyright models are emerging and understanding of constraints of current practice is growing. For example, in the United Kingdom authors can assign their copyright to a publisher, while at the same time retaining the moral rights (as opposed to economic rights) to their intellectual property. Moral rights involve ‘‘the right to claim authorship of a work, and the right to oppose changes to it that could harm the creator’s reputation’’ (World Intellectual Property Organization 2006). More authors are refusing to sign away their copyright, and instead choosing to give non-exclusive rights to publishers for first publication of their work, while retaining for themselves other rights (such as classroom and instructional uses). There are also new copyright models such as the Science Commons and Creative Commons models that enable authors to retain their copyright while assigning various levels of uses of their work (Creative Commons 1999; Creative Commons 2005) (Fig. 2.4). SHERPA/RoMEO Service provides information on the copyright policies of various publishers (SHERPA and University of Nottingham 2006). It grew out of the 2002/03 RoMEO Project (Rights MEtadata for Open Archiving) of the Joint Information System Committee of the UK at the University of Loughborough (Joint Information Systems Committee 2006). The project correctly identified a need to document publisher policies as interest in self-archiving grows. Most publishers allow authors to post their work online; however many have restrictions to how this is done and what can be posted (e.g. pre-print, post-print, article pdf). This service assists authors who want to better understand their rights before or after publishing as well as others who may want to use a copyrighted article (Fig. 2.5). A wealth of copyright resources exists for authors. At times, there is too much information so we tend to ignore it and hence are faced with consequences that can be problematic. As an author, you should check your organization’s guidelines (if they exist) so you know what your rights may be. If none exist, use other available resources such as a university’s or a government’s copyright site (Table 2.1). Also, read the publisher’s copyright agreement and amend it to address your need to archive and access your work. Ignorance is not bliss when it comes to copyright in the digital age.

2

The Consumption and Production of Fisheries Information

55

ADDENDUM TO PUBLICATION AGREEMENT 1. THIS ADDENDUM hereby modifies and supplements the attached Publication Agreement concerning the following Article: _________________________________________________________________________ (manuscript title) _________________________________________________________________________ (journal name) 2.

The parties to the Publication Agreement as modified and supplemented by this Addendum are: ____________________________________(corresponding author) ____________________________________________________ ____________________________________________________ ____________________________________________________ (Individually or, if one than more author, collectively, Author)

_______________________________________ (Publisher)

3. This Addendum and the Publication Agreement, taken together, allocate all rights under copyright with respect to all versions of the Article. The parties agree that wherever there is any conflict between this Addendum and the Publication Agreement, the provisions of this Addendum are paramount and the Publication Agreement shall be construed accordingly. 4. Author’s Retention of Rights. Notwithstanding any terms in the Publication Agreement to the contrary, AUTHOR and PUBLISHER agree that in addition to any rights under copyright retained by Author in the Publication Agreement, Author retains: (i) the rights to reproduce, to distribute, to publicly perform, and to publicly display the Article in any medium for noncommercial purposes; (ii) the right to prepare derivative works from the Article; and (iii) the right to authorize others to make any non-commercial use of the Article so long as Author receives credit as author and the journal in which the Article has been published is cited as the source of first publication of the Article. For example, Author may make and distribute copies in the course of teaching and research and may post the Article on personal or institutional Web sites and in other open-access digital repositories. 5. Publisher's Additional Commitments. Publisher agrees to provide to Author within 14 days of first publication and at no charge an electronic copy of the published Article in a format, such as the Portable Document Format (.pdf), that preserves final page layout, formatting, and content. No technical restriction, such as security settings, will be imposed to prevent copying or printing of the document. 6. Acknowledgment of Prior License Grants. In addition, where applicable and without limiting the retention of rights above, Publisher acknowledges that Author’s assignment of copyright or Author’s grant of exclusive rights in the Publication Agreement is subject to Author’s prior grant of a non-exclusive copyright license to Author’s employing institution and/or to a funding entity that financially supported the research reflected in the Article as part of an agreement between Author or Author’s employing institution and such funding entity, such as an agency of the United States government. 7. For record keeping purposes, Author requests that Publisher sign a copy of this Addendum and return it to Author. However, if Publisher publishes the Article in the journal or in any other form without signing a copy of this Addendum, such publication manifests Publisher’s assent to the terms of this Addendum. AUTHOR ___________________________________________ (corresponding author on behalf of all authors)

PUBLISHER ____________________________________

_______________________________________(Date)

___________________________________(Date)

Neither Creative Commons nor Science Commons are parties to this agreement or provide legal advice. Please visit www.sciencecommons.org for more information and specific disclaimers.

SPARC (the Scholarly Publishing and Academic Resources Coalition) and the Association of Research Libraries (ARL) are not parties to this Addendum or to the Publication Agreement. SPARC and ARL make no warranty whatsoever in connection with the Article. SPARC and ARL will not be liable to Author or Publisher on any legal theory for any damages whatsoever, including without limitation any general, special, incidental or consequential damages arising in connection with this Addendum or the Publication Agreement. SPARC and ARL make no warranties regarding the information provided in this Addendum and disclaims liability for damages resulting from the use of this Addendum. This Addendum is provided on an “as-is” basis. No legal services are provided or intended to be provided in connection with this Addendum.

Access-Reuse 1.0 www.sciencecommons.org

Fig. 2.4 An example of a copyright addendum from Science Commons

SPARC Author Addendum 3.0 www.arl.org/sparc/

56

J. Webster and E. Uhlinger

Fig. 2.5 Examples of a publisher profile from SHERPA/RoMEO

Title Author’s rights (SPARC)

Table 2.1 Copyright resources Web address

Description

http://www.arl.org/sparc/author/ addendum.html

Explains your rights as an authors and includes copyright addendum Scholar’s Copyright Project http://sciencecommons.org/ Another example of a (Science Commons) projects/publishing copyright addendum Copyright Management http://www.copyright.iupui.edu/ Explains U.S. copyright, Center (Indiana University) the concept of Fair Use, and other concepts Copyright (World Intellectual http://www.wipo.int/copyright/en/ A discussion of Property Organization) copyright from an international perspective

2.3.4 Access Producing scientific information translates into providing access to it. This used to be straight forward; yet now, in the electronic environment, access issues present another set of decisions for authors. Scholarly communication as we now know it ‘‘began’’ in the 17th century when reports of scientific discovery or

2

The Consumption and Production of Fisheries Information

57

observation were presented to scholars by reading them before the assembled members of scientific societies. The resulting papers were printed in compilations, the first being the Philosophical Transactions of the Royal Society (London) in 1665, and thus ‘‘the scientific journal’’ was born (Vickery 2000). For centuries printed journals were the norm, with the number of journal titles growing as new disciplines and sub-specialties of science developed. Scholarly societies and organizations along with commercial ventures were all publishers serving different audiences and roles. The societies tended towards printing papers that had been presented and vetted within their meetings while others sought publishing as a means to broaden communication among scientists rather than within organizations (Henderson 2002; Vickery 2000). As scholarly communication grew, its audience expanded and diversified, and the cost of producing, delivering and archiving scientific information increased as well (Prosser 2005). Today, we expect producers of scientific journals to offer multiple modes of access (print and electronic), more content (e.g. pages) and archives of all. Electronic full text access to current and old items should be within one or two clicks of a search. Our broader audiences also want ready access to the information fisheries scientists produce. They want it in a digestible format that is easily accessed. They do not subscribe to the scientific journals, so electronic delivery makes sense. Thanks to the widespread adoption of personal computers, standardized software and stable file formats, nearly all scientific publications are ‘‘born digital.’’ Digital content can easily be converted into appropriate styles or formats, and delivered on the Web through the sophisticated web sites of major scientific publishers to workable government and academic sites as well as a myriad of other web presences. Digital files, distributed across the Internet, have drastically altered the reach and potential markets for scientific literature. In fact, the print journal we have known since the 1600s is considered by many to be nearing extinction or at best to being an archival format. Online journals developed from the expansion of ‘‘born digital’’ information and the global spread of Internet technologies. These are available in multiple forms and collections with associated differences in how potential readers must access the content. The two primary types are the traditional journals that are now available in electronic form (e.g. Reviews in Fisheries Science) and the open access journals (e.g. Scientia Marina). The electronic equivalent of print journals must still be purchased for the most part. They may be distributed on an individual basis, gathered into collections by the publisher, or aggregated into bundles of journals from multiple publishers. A particular journal title might even be available via any or all of these mechanisms, and such compilations may be ‘‘full text’’ cover-to-cover or just select portions of journals (for example research articles but not news or letters to the editor). Multiple mechanisms allow users to pick the one that fits their budget; however, it also means that libraries may have to duplicate purchases to capture all the content. As an author, you want to explore how your article is marketed as that affects how people access it. If too expensive or if included in an aggregated package that

58

J. Webster and E. Uhlinger

many find too confusing to purchase, your work may be essentially ‘‘lost’’ to parts of the audience. New initiatives promote ‘‘open access’’ to scientific literature, whereby articles are either published in an ‘‘open access’’ journal or posted to an electronic repository. One of the clearest definitions of ‘‘open access’’ is found in the Budapest Open Access Initiative: By ‘open access’ to this literature, we mean its free availability on the public internet, permitting any users to read, download, copy, distribute, print, search, or link to the full texts of these articles, crawl them for indexing, pass them as data to software, or use them for any other lawful purpose, without financial, legal, or technical barriers other than those inseparable from gaining access to the internet itself. The only constraint on reproduction and distribution, and the only role for copyright in this domain, should be to give authors control over the integrity of their work and the right to be properly acknowledged and cited Chan et al. (2002)

This concept and the initiatives it has spawned change the landscape of scholarly communication and access. Authors can retain control over their copyright and hence exercise more control of the access to their work. Such radical change does not come easily, quickly or smoothly. Yet, change does seem to be happening. Driven by initiatives such as the Open Archives Initiative and the Berlin Declaration, efforts vary from country-to-country (Van de Sompel and Lagoze 2000; Gruss and German Research Organizations 2003). The Open Access movement suggests new responsibilities for authors including making sure that they retain their rights to give their work to the public and that it is stored in a stable manner. New requirements by funders and institutions are one impetus for the growth of open access. For example, in 2008 the US National Institute of Health started requiring all grantees to deposit their findings in PubMedCentral. However, compliance is spotty and slow. Others have also found authors slow to put their material in publicly accessible sites; but persistence and mandates work over time (Sale 2006). It is important to note that while open or public access is a fairly recent phenomenon, studies show that the ‘‘impact’’ and citedness of such articles is as high as or better than articles published in traditional journals (Kousha and Thelwall 2006; Antelman 2004; Harnad and Brody 2004). Considering the alternatives to commercial journals and traditional publishing outlets is valid when producing information; ‘‘open access’’ journals and e-repositories may provide better access to your work for your intended audience. Beyond the traditional scientific journal and the growing open access movement, we are learning how to share our research in other ways. No longer dependent on the mail and correspondence, we share research findings, collaborate in real time across tremendous distances, and participate in ‘‘live’’ debates with just a few keystrokes on the Internet. Technological advances and relatively inexpensive gadgets make it possible to talk (using Voice Over Internet Protocol or VOIP), participate in video conferences, and hold interactive distributed online seminars called ‘‘webinars’’. Blogs (web logs or online diaries), wikis, and other

2

The Consumption and Production of Fisheries Information

59

collaborative authoring tools are also drastically changing how scientists do business. Such low-cost tools make it possible for scientists, students, and policy-makers anywhere on the globe (or even in space) to quickly communicate with others and quickly disseminate information. Despite reports by Christopher Columbus in 1492, it appears that the world really is flat.

2.3.5 Archiving Access and archiving are linked in the electronic environment; you cannot access an electronic document unless it has been stored in some logical, stable manner. File drawers and library shelves historically held the archives of the scientific debate but the digital millennium changes this norm. Now, you are more likely to post a pdf of your article to your web site than to order a box of reprints from the publisher. Or, you rely on the publisher to maintain an electronic copy on its server. Both options are tenable, but may have legal implications or monetary limitations. You can usually satisfy the former if you reserved the right to post a copy of the pdf to a publicly accessible site when signing your copyright statement. The latter can be more complex. With access licenses comes the vulnerability that access to information ends when one stops paying ‘‘the rent.’’ Usually, the library pays ‘‘the rent’’ through contractual agreements with the publishers for access. Publishers make their digital archives available, some at a high cost to libraries and institutions, and others more altruistically, making them publicly available at no or low cost. An example of a commercial model is ScienceDirect, Elsevier’s online journal collection; an institution can subscribe to some or all of the content of Elsevier’s vast suite of scientific journals depending on the strength of its budget and the need of its researchers. PubMed Central sponsored by the US National Institute of Health is a different archiving model; here all articles are freely accessible to all. As publishing mergers continue, archives change hands and access costs and rights can change with new owners. Unstable budgets can interrupt service and libraries lose access to previously licensed archives. And, authors lose access to their work. Identifying how your work will be archived is yet another step in the production cycle. Again, there are choices and consequences of those choices. In an attempt to archive and secure ongoing access to their contribution to the scholarly process, many entities are creating institutional repositories (IR) as a digital preservation space (University of Houston Libraries, Institutional Repository Task Force and Bailey 2006). Institutional repositories provide a service to collect, archive and provide access to the information produced by members of a defined community such as a university or a discipline (Lynch 2003). They create a virtual and intellectual environment for the community’s digital output. They are an attempt to address the challenges of digital archiving, the expectations of the campus and research community for better access to

60

J. Webster and E. Uhlinger

information, and the inadequacies of the current cumbersome model for scholarly communication. Various organizational models, software and hardware are emerging as more universities and agencies implement IRs (Crow 2004).

2.3.6 Information Produced As fisheries scientists, we want to share our work with colleagues, policy makers and the public. Computers make it easier to produce work by streamlining our writing and editing. With the Web, we can now also easily publish our work making it accessible to all. However, producing quality information still involves multiple steps that affect its credibility and use. Scholarly communication is changing, and you need to recognize where and how you can change your actions to improve the information landscape. Consider your audience and its information consumption behavior. Also, consider your future audience. Such consideration will help you decide where you publish as it will suggest how your work will be identified and accessed by readers today and tomorrow.

2.4 The Future of Fisheries Information While the life cycle of fisheries information remains constant through consumption and production, its environment is changing. Much of the change is driven by the integration of technology into how we ‘‘do’’ science. The change is inevitable; however, as fisheries scientists, we can shape the environment by making the communication of science better – more timely and accessible – while maintaining our credibility and honesty. This takes effort and a willingness to modify some of our ways of consuming and producing information. Engagement in the discussion about scholarly communication is imperative, followed by action. Open and efficient access to fisheries information requires shifts in how we finance production of information. This encompasses the debate over journal pricing, the open access principles and the future of the scientific journal. Ease and stability of access require us to work with those who design and maintain search systems, databases, and archives so the systems respond to our needs.

2.4.1 Changing Economics of Fisheries Information The old system of scholarly publication cannot be sustained given changing user expectations and economics. Pricing continues to escalate with great variability among publishers. For example, in a 2004 study, median overall journal prices vary from £124 (Cambridge University Press) to £781 (Elsevier) (White and Creaser 2004). Price increases from 2000 to 2004 ranged from 27% (Cambridge University Press) to 94% (Sage), well above any inflation factor

2

The Consumption and Production of Fisheries Information

61

(White and Creaser 2004). Yet we continue to struggle to implement a new publishing model that improves access and archiving for all. Commercial publishers expect profit margins and must often pay dividends to their stockholders. Professional societies generate income from subscriptions, and use the income to provide benefits to members. All publishers need to cover costs. At issue is how to do that in an equitable manner as well as one that promotes open and efficient scholarly communication (Edwards and Shulenburger 2003). Authors provide and consume the product. Yet, the costs are usually born by their institutions. Print subscriptions are sold to individuals (or perhaps are included as a benefit of membership in a particular society) for their personal use. These same journals are sold to libraries at a higher cost because they are accessible to many potential readers. Publishers have discovered that scientific articles are themselves discrete information commodities that can be sold in a collection, bundled into packages of often unrelated journals, or one-by-one. Unlike print journals, publishers have many different market models for pricing online subscriptions, for example, charging based on the number of ‘‘FTE’’ (full time equivalents) of faculty, staff, and students; or by the total amount of grant dollars received; or the number of advanced degrees conferred in a particular subject by an institution. Publishers may offer a subscription at one price to a small marine laboratory and the same publication at a completely different price to a neighboring university. Standard pricing appears to have disappeared as ‘‘deals’’ and ‘‘negotiations’’ have become the norm (Frazier 2001). Access to the article-level is also possible via alternative means and costs, including by subscription to an entire journal or on a pay-per-view basis. In addition to the highly variable subscription prices of scientific journals, authors may face additional costs. Author fees (typically called ‘‘page charges’’ or ‘‘color charges’’) are commonly found in society journals. The charges offset the expense of printing and allow societies to sell subscriptions at a ‘‘subsidized’’ or lower cost. Author charges may or may not be payable with grant funds, or an institution may pay on behalf of its authors. Emerging models that allow ‘‘open access’’ may also come at a cost borne by the author or her/his institution. Simultaneously, open access publications such as the Public Library of Science, have been subsidized by grants and are provided to readers free. Some open access publications offer institutional subscriptions that afford authors at the institution with reduced page charge fees. Even so, the market continues to evolve and access may be ‘‘embargoed’’ whereby current articles are closed, but older articles are ‘‘open access’’ or there may be a mix of access types within current issues (for example Limnology and Oceanography where an article can be ‘‘unlocked’’ or made open access by payment of an additional ‘‘page charge’’). Theoretically, online publication should reduce costs because there are fewer steps and ‘‘consumables’’ (paper and ink) used in the production process as well as reduced costs formerly associated with postage, shipping and handling. However, in many cases the move to electronic delivery and access has

62

J. Webster and E. Uhlinger

significantly increased the cost to acquire scientific literature. Publishers insist that online publishing has raised their costs due to the need to upgrade and maintain servers and authentication mechanisms for online subscribers. So, while digital publishing increases the timeliness of access, it also compounds the ways users can access the material and the ways publishers can sell the product (Quandt 2003). It complicates things for all, just when we believe that scholarly communication should be easier, faster and cheaper. Scientific publishing is rapidly evolving and unsettled, driven by technology and the growth of Internet-based services. For centuries, libraries preserved the scientific record by purchasing journal subscriptions, binding loose issues into complete volumes, cataloguing and preserving them and making collections available to current and future generations of students and scholars. But by the late 20th century, the information moved off the printed page, and access and archiving are no longer assumed with the purchase of a subscription to a journal or electronic book. One approach to the problem is more funding for institutional purchases of electronic information; that is not going to happen at most institutions and still leaves those fisheries scientists unaffiliated with a strong library or research institution unable to get full access to the information needed. A more realistic approach requires government funded research to be published in a publicly accessible venue (Edwards and Shulenburger 2003). An immediate step authors can take is to deposit their publications in a stable electronic repository that is openly accessible and searchable (OhioLINK Governing Board 2006). Change in the publishing landscape is happening rapidly; changing our behavior as consumers and producers is slower, and we need to remedy that to maintain quality fisheries science.

2.4.2 Ensuring Access to and Preservation of Fisheries Information Ease and stability of access to information relate to changes in the publishing landscape, yet have unique issues as well. Ease of access implies improved search interfaces and algorithms as well as more connectivity among sources of information. This challenge seems overwhelming, but realistically can be addressed at various scales and by a range of users. Locally, scientists can work with their librarians and computer scientists to make sure information created and stored locally is easy to search, find and use. A concrete example is to examine how you store your article reprints; are they in a secure and searchable place, or merely tucked on your own computer? Another example is considering how you construct and host a web site for your research project; is the metadata up to standards so it is indexed by web search engines or is the coding something you had not considered? Within professional societies, you can advocate for simple and intuitive interfaces to your organization’s information and publications. Scientists should be willing to participate in studies on the

2

The Consumption and Production of Fisheries Information

63

usability of search systems, patterns of searching behavior and use of information. The more input on how search systems are used by those who really use them, the better the systems will eventually become. Stability of access dictates if future fisheries scientists will be able to find and use the information created today. With the evolution of publishing from print-only to print plus online models, there has been a cultural shift from ‘‘ownership’’ (whereby each library or individual purchases a subscription to the journal) toward an ‘‘access model’’ whereby libraries and publishers enter into contractual and license agreements that define the terms of access to and use of online content (Quandt 2003). Under this new model, instead of owning journals, content is ‘‘leased’’ and made accessible under specific terms for a specific period of time. Under the terms of contracts and licenses, when libraries cancel a subscription, they may lose access to all content they have leased in the past, thus ending up with nothing to show for their investment over time. The stable print archive the library used to represent has disappeared. So, now we debate how to preserve scholarly information that we may or may not own, and do not really understand its technical life expectancy. CDRoms, once thought to be a good preservation medium, have been shown to fail much earlier than anticipated. Publishers have rushed to digitize past volumes of scientific journals, converting millions of print pages into bits and bytes stored on computers. In fact, that first scientific journal has been digitized as part of the JSTOR initiative so that all articles from 1665 are searchable, retrievable, and printable via any Internet connected computer if the searcher is accessing the resource through an institution with a subscription to this archive (JSTOR 2000–2006). Even when digitized, where is that article stored, in what format, and will we will be able to refresh it as software and hardware changes? Fisheries scientists are not going to solve the digital preservation quandary. However, awareness of the fragility of digital information may make all of us more diligent with our decisions about storing our publications and data. Simple steps are critical, such as using standard formats for digital documents and adding basic metadata to datasets. More complex ones take greater effort and often specific expertise. These include building robust data repositories and experimenting with new ways of storing and accessing files. The keys to change here are involvement and collaboration. Waiting for the publishers to improve search interfaces and provide permanent archives may be waiting for an outcome that is untenable.

2.4.3 Checklist for Consumers and Producers In the end, the cycle of science continues. The information that feeds new ideas and questions continues to be produced and consumed. Maintaining the vigor of fisheries science in the changing environment requires attention by all who are part of the information cycle.

64

J. Webster and E. Uhlinger

When consuming information:

   

Consider your question before feasting on the information. Select the right tool and search strategy for your need. Try multiple tools and strategies. Do not assume that nothing exists on the topic. Remember that not everything is found by Googleä. Science happened before computers were invented.  Evaluate your sources. Everything you find is not true, accurate or timely.  Keep track of the sources you find so you can use them accurately and ethically.  Ask for help from an expert – a librarian or a colleague. When producing information:

 Think about your audience when writing.  Consider the practices of publishers when selecting one.

 How do they establish prices?  What is their policy for posting publicly accessible sites?  Do they allow users in developing countries free access to their publications?

 Do they charge you or your institution?  How will they store your work?  Modify your copyright agreement to retain the rights you want.  Deposit your publications in an open access repository.  As a reviewer, consider the practices of the journal that asks for your time and expertise.

 As a member of a professional society, know your organization’s policies and change the ones that inhibit the free flow of information.

 As a colleague and mentor, encourage others to join the discussion and change how we communicate.

 Check the SPARC site for current information on scholarly communication trends (Association of Research Libraries and Scholarly Publishing and Academic Resources Coalition 2006).

References Antelman K (2004) Do open access articles have a greater research impact? College & Research Libraries 65(5):372–82 ASFA Secretariat (2006) List of ASFA partners [Web Page]. Located at: ftp://ftp.fao.org/FI/ asfa/asfa_partner_list.pdf. Accessed 2006 Aug. Association of Research Libraries, Scholarly Publishing and Academic Resources Coalition (2006) CreateChange: Change & you [Web Page]. Located at: http://www.createchange. org/changeandyou.html. Accessed 2006 Sep 7 Avrahami TT, Yau L, Si L, Callan J (2006) The FedLemur project: federated search in the real world. Journal of the American Society for Information Science and Technology 57(3):347–58

2

The Consumption and Production of Fisheries Information

65

Bauer K, Bakkalbasi N (2005). An examination of citation counts in a new scholarly communication environment. D-Lib Magazine 11(9):1–7 Beverton RJH, Holt SJ (1957). On the dynamics of exploited fish populations. London, UK: Her Majesty’s Stationery Office; (Great Britain. Ministry of Agriculture, Fisheries and Food. Fishery Investigations: ser. 2, v. 19) Chan L, Cuplinskas D, Eisen M, Friend F, Genova Y, Gue´don J-C, Hagemann M, Harnad S, Johnson R, Kupryte R, La Manna M, Re´v I, Segbert M, Souza S, Suber P, Velterop J (2002) Budapest Open Access Initiative [Web Page]. Located at: http://www.soros.org/openaccess/read.shtml. Accessed 2006 Sep 7 Creative Commons (1999) About Creative Commons [Web Page]. Located at: http:// creativecommons.org/. Accessed 2006 Sep 1 Creative Commons (2005) Scholar’s copyright project [Web Page]. Located at: http:// sciencecommons.org/literature/scholars_copyright. Accessed 2006 Sep 1 Crow, R (2004) A guide to institutional repository software. Second Edition. Open Society Institute: New York Edwards R, Shulenburger D (2003) The high cost of scholarly journals (and what to do about it). Change 35(6):10–9 Elsevier Ltd (2004) Scirus White Paper: how Scirus works. Amsterdam, Netherlands: Elsevier Ltd. Food and Agriculture Organization of the U.N. (1958) Current Bibliography for Fisheries Science. Rome, Italy Vol. 1 Food and Agriculture Organization of the U.N. (1995) Code of conduct for responsible fisheries. Rome, Italy: FAO Food and Agriculture Organization of the U.N. (2006) FAO Corporate Document Repository [Web Page]. Located at: http://www.fao.org/documents/. Accessed 2006 Sep 1 Frazier K (2001) The librarian’s dilemma: contemplating the costs of the ‘‘Big Deal’’. D-Lib Magazine 7(3):10.1045/march2001-frazier Garfield E (1994) The ISI impact factor. Current Contents: Agriculture, Biology, & Environmental Sciences 25(25):3–7 Google (2005) About Google Scholarä [Web Page]. Located at: http://scholar.google.com/ intl/en/scholar/about.html. Accessed 2006 Aug. Gruss P, German Research Organizations (2003) Berlin Declaration on open access to knowledge in the sciences and the humanities [Web Page]. Located at: http://www.zim. mpg.de/openaccess-berlin/berlindeclaration.html. Accessed 2006 Sep 7. Harnad S, Brody T (2004) Comparing the Impact of Open Access (OA) vs. Non-OA Articles in the same journals. D-Lib Magazine 10(6):doi:10.1045/june2004-harnad Henderson A (2002) Diversity and the growth of serious/scholarly/scientific journals. [in] Abel RE, Newlin LW, ed. Scholarly publishing: Books, journal, publishers, and libraries in the Twentieth Century. US: John Wiley & Sons, Inc. pp 133–62 Jacso´ P (2005a) Google scholar: the pros and cons. Online Information Review 29(2):208–14 Jacso´ P (2005b) Visualizing overlap and rank differences among web-wide search engines: some free tools and services. Online Information Review 29(5):554–60 Jacso´ P (2006a) Savvy searching: deflated, inflated and phantom citation counts. Online Information Review 30(3):297–309 Jacso´ P (2007) Scopus. Pe´ter’s Digital Reference Shelf [Web Page]. Located at: http://www. gale.cengage.com/reference/peter/200711/scopus.htm. Accessed 2008 Oct 10 Jansen BJ, Spink A, Saracevic, T (2000) Real life, real users, and real needs: a study and analysis of user queries on the web. Information Processing and Management 36(2000):207–27 Joint Information Systems Committee (2006). About JISC – Joint Information Systems Committee [Web Page]. Located at: http://www.jisc.ac.uk/. Accessed 2006 Sep 1 JSTOR (2000) About JSTOR [Web Page]. Located at: http://www.jstor.org/about/. Accessed 2006 Jan Kousha K, Thelwall M (2006) Google Scholar citations and Google Web/URL citations: A multidiscipline exploratory analysis. [in] Proceedings International Workshop on Webometrics,

66

J. Webster and E. Uhlinger

Informetrics and Scientometrics & Seventh COLLNET Meeting Nancy, France. Located at: http://eprints.rclis.org/archive/00006416/01/google.pdf Accessed 2006 Sep 1 Lackey RT (2006) Axioms of ecological policy. Fisheries 31(6):286–90 Lange LL (2002) The impact factor as a phantom: is there a self-fulfilling prophecy effect of impact? The Journal of Documentation 58(2):175–84 Lynch CA (2003) Institutional repositories: essential infrastructure for scholarship in the digital age. ARL Bimonthly Report 226 Mattison D (2005) Bibliographic research tools round-Up. Searcher 13(9):10704795 McDonald J, Van de Velde EF (2004) The lure of linking. Library Journal 129(6):32–4 National Sea Grant Library (2006) National Sea Grant Library [Web Page]. Located at: http://nsgd.gso.uri.edu/. Accessed 2006 Sep 1 Neuhaus C, Neuhaus E, Asher A, Wrede C (2006) The depth and breadth of Google Scholar: an empirical study. Portal: Libraries and the Academy 6(2):127–41 Office of Scientific and Technical Information (U.S.) (2006) GrayLIT Network: A science portal to technical papers [Web Page]. Located at: http://www.osti.gov/graylit/. Accessed 2006 Sep 1 OhioLINK Governing Board (2006) OhioLINK Library Community recommendations on retention of intellectual property rights for works produced by Ohio faculty and students [Web Page]. Located at: http://www.ohiolink.edu/journalcrisis/intellproprecsaug06.pdf. Accessed 2006 Sep 7 Page L, Brin S, Montwani R, Winograd T (1998) The PageRank citation ranking: bringing order to the Web. Technical Report, Stanford University Database Group Pauly D, Stergiou KI (2005) Equivalence of results from two citation analyses: Thomson ISI’s Citation Index and Google’s Scholar service. Ethics in Science and Environmental Politics December 2005:33–5 Pew Oceans Commission, Panetta LE (2003) America’s living oceans: charting a course for sea change: a report to the nation: recommendations for a new ocean policy. Arlington, VA: Pew Oceans Commission Prosser DC (2005) Fulfilling the promise of scholarly communication – a comparison between old and new access models. [in]: Nielsen EK, Saur KG, Ceynowa K, eds. Die innovative Bibliothek: Elmar Mittler zum 65. Geburtstag. K G Saur. pp 95–106 Pruvost C, Knibbs C, Hawkes R (2003) About Scirus [Web Page]. Located at: http://www. scirus.com/srsapp/aboutus. Accessed 2006 Aug Quandt RE (2003) Scholarly materials: Paper or digital? Library Trends 51(3):349–75 Ranganathan SR (1963) The five laws of library science. [Ed. 2, reprinted with minor amendments] Bombay, New York: Asia Publishing House Roth DL (2005) The emergence of competitors to the Science Citation Index and the Web of Science. Current Science 89(9):1531–6 Sale A (2006) The acquisition of open access research articles. First Monday 11(10) [Web Page]. Located at: http://eprints.utas.edu.au/388/ SHERPA, University of Nottingham. (2006) SHERPA/RoMEO Publisher copyright policies & self-archiving [Web Page]. Located at: http://www.sherpa.ac.uk/romeo.php. Accessed 2006 Sep 1 Spink A, Cole C (2006) Human information behavior integrating diverse approaches and information use. Journal of the American Society for Information Science and Technology 57(1):25–35 University of Houston Libraries, Institutional Repository Task Force, Bailey CW (2006) Institutional repositories. Washington, DC: Association of Research Libraries, Office of Management Services Van de Sompel H, Lagoze C (2000) The Santa Fe Convention of the Open Archives Initiative. D-Lib Magazine 6(2):DOI: 10.1045/february2000-vandesompel-oai Vickery BC (2000) Scientific communication in history. Lanham, MD: Scarecrow Press Webster JG (2003) How to create a bibliography. Journal of Extension 41(3)

2

The Consumption and Production of Fisheries Information

67

Webster JG, Collins J (2005) Fisheries information in developing countries: support to the implementation of the 1995 FAO Code of Conduct for Responsible Fisheries. Rome, Italy: Food and Agriculture Organization of the U.N.; (FAO Fisheries Circular No. 1006) White RW, Jose JM, Ruthven I (2003) A task-oriented study on the influencing effects of query-biased summarisation in web searching. Information Processing and Management 39(2003):707–33 White S, Creaser C (2004) Scholarly journal prices: Selected trends and comparisons. Leicestershire, UK: Library and Information Statistics Unit, Loughborough University; (LISU Occasional Paper: 34) World Intellectual Property Organization [2006]. Copyright FAQs: What rights does copyright provide? [Web Page]. Located at: http://www.wipo.int/copyright/en/faq/faqs.htm#rights. Accessed 2006 Sep 1

Chapter 3

Extended Guide to Some Computerized Artificial Intelligence Methods Saul B. Saila

3.1 Introduction The purpose of this chapter is to reassess and extend some of the earlier concepts and developments related to computerized artificial intelligence methods described initially (Saila, 1996). These new concepts and developments were derived from about a decade of increasing activity in this subject area. Due to this increased activity a certain amount of subjectivity has been required in order to keep this review from becoming too detailed and lengthy. Therefore, most examples of various newer developments are restricted to those believed to have fishery science applications; new methodologies deemed especially useful in fishery science applications are described in more detail. During this past decade of rapid developments in artificial intelligence there has also been considerable effort to combine some computing paradigms, such as fuzzy set theory, neural networks, genetic algorithms, rough sets, and casebased reasoning as well as other methodologies for the purpose of generating more effective hybrid systems. These are now termed soft computing. In soft computing the individual tasks act synergistically, not competitively, enhancing the application domain of the other. The purpose of these combined soft computing methods is to develop flexible information processing systems that can exploit tolerance for imprecision, uncertainty, approximate reasoning, and partial truth in order to achieve tractability and close resemblance to human decision making. These methods may also provide a reduction in solution costs. The term soft computing, in contrast to conventional (hard) computing could be characterized as automated intelligent estimation. It is intended to provide an alternative to conventional computing that allows for the formal and systematic treatment of problems which, due to their complexity, size and/or uncertainty, are not practical to solve in a conventional manner. Soft computing is thought to arise from a recognition that some complex problems do not lend themselves to solution by any conventional (hard) computing methods. Soft computing attempts to emulate and articulate the techniques used by intelligent humans to S.B. Saila (*) 317 Switch Road, Hope Valley, RI 02832, USA

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_3, Ó Springer ScienceþBusiness Media B.V. 2009

69

70

S.B. Saila

deal adequately and quickly with complex problems and causes; in effect, to mimic what is often called ‘‘human intuition’’. Unfortunately, there appears to be confusion regarding some of the terminology applied to soft computing. This is believed to be due to the relatively rapid developments of this concept (Corchado and Lees, 2001). For example, although the terms case-based reasoning and rough sets apply soft computing generally, they may also be included as methodologies applied to data mining. Herein, data mining is defined as the process of extracting information and patterns, often previously unknown, from large quantities of data using techniques such as machine learning and various statistical procedures. For the purposes at hand, case-based reasoning and rough sets are considered to be soft computing techniques which may be effectively utilized in data analysis from multiple sources or with variable hypotheses. I believe that case-based reasoning and rough sets are also tools useful to and included in the more general term data mining. ‘‘Meta-analysis’’ is another term, similar to or synonymous with data mining, frequently used by fisheries scientists and ecologists to describe what I suggest is some form of data mining. The free encyclopedia wikipedia (http://en.wikipedia.org/wiki/ metaanalysis) states that ‘‘In statistics, a meta-analysis combines the results of several studies that address a set of related research hypotheses.’’ Data mining has been defined as the nontrivial extension of implicit, previously unknown, and potentially useful information from data by the above-mentioned source. On the other hand, in case-based reasoning systems expertise is embodied in a library of past cases, rather than being included in classical rules. Each case typically contains a problem description as well as a solution or outcome. These are utilized to form a solution to new data. Some examples of published work in fisheries and ecology utilizing the term meta-analysis include Myers and Mertz (1998) who utilize this approach in an effort to reduce uncertainty in the biological basis of fisheries management. Taylor and White (1992) studied the hooking mortality of nonanadromous trout from 18 studies using meta-analysis methodology. Englund et al. (1999) offer some recommendations regarding the conclusions from meta-analysis which, he asserts, could be colored by the data selection process. Myers et al. (2001) describe statistical methods to allow many data sets to be analyzed simultaneously. These methods provide some further strength to the analyzed data. A short summary of some meta-analysis applications suggests to me that the perhaps better and more generic term (data mining) might be used in the future in order to embrace all the methodologies related to the more effective utilization of complex data that address some related research hypotheses with statistical and analytical tools. This definition seems to coincide with the definition of Hastie et al. (2001). In addition, data mining is a widely used term in business applications. The recent book by Walters and Martell (2004) clearly and concisely describes the derivation, use, and abuse of various mathematical models which have been utilized for decisions regarding the management of harvested aquatic ecosystems. It is evident from this book that quantitative modeling

3

Guide to Some Computerized Artificial Intelligence Methods

71

methods have become a central tool in the management of harvested fish and invertebrate populations. These models are complex and they include structural diversity as well as dynamic complexity with feedbacks. The conclusion by the authors is that it is impossible to fully capture the rich behavior of ecosystems in mathematical models. However, they also offer cogent arguments for continuing efforts to build useful mathematical models in the future. Although I agree with the conclusions of the above authors regarding the need for further developments in this kind of modeling (hard computing), it also seems to me that the versatile tools of computerized artificial intelligence, now known as soft computing, should not be ignored. Indeed, they may offer some useful alternatives to conventional models in the search for effective and adaptive fishery management at both the species and system levels.

3.2 Working with Some Forms of Fisheries Data Some fisheries data can be disorganized, ambiguous, often incomplete, and yet quite detailed. These data can contain an abundance of information, but data can also be imprecise. Generally data has been examined using equations, algorithms, decision tables, and decision trees. Although many deterministic and stochastic models of fisheries systems exist, the overall quality of the input data is often inadequate; and thus, the input to the models may not adequately meet the assumptions and requirements of the models. It seems clear that these models deserve careful and thoughtful validation as well as further study, with more active consideration of other approaches to fishery management. In fishery science, as in other areas of science, there is a tendency for one to commit oneself to a particular methodology in the belief that it is that methodology alone which matters. Professor Lotfi Zadeh, an outstanding scientist, has enunciated two principles related to methodological tunnel vision. The first is the hammer principle, which states that when the only tool one has is a hammer, then everything looks like a nail. The second is the vodka principle, which states that no matter what the problem is, vodka will solve it! It seems apparent that there is a substantial amount of the ‘‘one size fits all’’ mentality in fishery science, and it therefore seems important to suggest means for countering this situation. Professor Zadeh has also made a profound statement concerning the mathematical complexity of systems. It is quoted as follows: ‘‘As the complexity of a system increases, our ability to make precise and significant statements about its behavior diminishes until we reach a threshold beyond which precision and significance (or relevance) become almost exclusive characteristics.’’ More recently, Ulanowicz (2005) has reinforced the above quotation of Zadeh by stating that emergent attributes of complex living systems render the conventional Newtonian postulates as inappropriate for ecosystem

72

S.B. Saila

dynamics. He suggested the use of network analysis to describe complex systems about two decades ago (Ulanowicz, 1986), but this has not received much application to date. Frequently, knowledge can be expressed as logical rules, in spite of the fact that using rules makes representation of this knowledge less precise. An example is provided by the so-called ‘‘rule of thumb.’’ However, in spite of the loss of precision, using rules has an advantage in making knowledge easier to explain to others. We do not always need detailed data and/or mathematical models to reason about facts, events, and problems. An excellent example is provided by physicians who diagnose a patient’s condition on the basis of general data, such as whether or not the patient’s temperature is elevated. This is derived from a simple thermometer reading. Indeed, the utility of coarse data is supported by our everyday experiences. Too much data may be confusing or disturbing and may prevent us from organizing the data into useful patterns. One approach to dealing with organizing data into useful patterns will be briefly described. It is suggested that it is sometimes reasonable to analyze an empirical data set to derive the maximum amount of information from the data itself instead of developing a functional model which may often be based on unrealistic assumptions.

3.3 Rough Set Concepts, Introduction, and Example A relatively new formal framework (rough sets) for discovering facts from imperfect data has been developed by Pawlak (1982). Rough sets provide tools for data analysis and rule discovery from imprecise and ambiguous data. Some advantages of rough set theory described in Pawlak (1999) are as follows: (a) It provides for both qualitative and quantitative descriptions of data; (b) It accommodates data inconsistencies as well as different distributions; (c) It allows for selection of the most important attributes, and the generation of ‘‘if. . .then. . .’’ rules as well as classification of future events. Rough sets may sometimes be confused with fuzzy sets. Both rough sets and fuzzy sets address the need to effectively use imperfect knowledge. However, fuzzy sets address the vagueness of information, whereas rough sets address indiscernability, or imprecision and ambiguity of data. The material which follows includes a brief review and some background including an example of a fishery related utilization of rough sets. This is only a crude example of the basic concepts and the reader interested in further details is referred to Pawlak (1991) and Slowinski (1992) for a much better understanding of rough set concepts and applications. Other useful reference books include Polkowski (2002), which deals with the mathematical foundations of rough sets, and Orlowska (1998), which deals with both theory and applications.

3

Guide to Some Computerized Artificial Intelligence Methods

73

Rough set theory has attracted considerable attention within the scientific community for more than a decade. Although rough set theory provides a mathematical approach to deal with problems which involve uncertainty and vagueness, I am not aware of any direct applications to date in fishery science. Motivation for the use and development of this approach seems justified for fishery scientists because rough sets can be developed with both qualitative and quantitative descriptions of data, rough sets do not require specific data distributions, and the method can accommodate data inconsistencies. Other advantages of rough set theory involve a method for the selection of the most important attributes of a data set, easily interpreted ‘‘if. . .then’’ rules and classification of future events. The advantage of rough set theory according to Grzymala-Busse (1988) is that it does not need any preliminary or additional information about data, such as probability distributions as in statistics, basic probability assignment as in Dempster-Shafer theory, or grade of membership or value of possibility as in fuzzy set theory. Several contributions have already been made to the development and application of rough sets in diverse areas of science. Some specific examples include Slowinski (1995), Lin and Wildberger (1995), Ziarko (1995), Rossi et al. (1999), and Che`vre et al. (2003). These examples of rough set applications range in diversity from decision analysis to industrial and engineering applications. Of particular interest and relevance to fishery science are studies related to pollution control in aquatic ecosystems. The material which follows, largely derived from the work of Rossi et al. (1999), and Pawlak et al. (1995), is included because I believe this application is relevant to some problems in fishery science. Briefly, rough set philosophy is founded on the assumption that every object of a universe of discourse is associated with some information. Objects characterized by the same information are indiscernible (that is, they can’t be recognized as different from each other) in light of the available information about them. The indiscernability which is generated in this manner seems to provide the mathematical basis of rough set theory. Objects which are indiscernible from one another with respect to some attributes are termed an elementary set, and it forms a basic ‘‘granule’’ of knowledge about the universe being considered. Any set of objects which may be represented as a union of some elementary sets is referred to as crisp (precise) in the context of the given attributes, or otherwise the set is rough (imprecise). As a consequence, each rough set has some borderline cases. These are objects that cannot be classified with certainty as members of the set or its complement. Rough set theory handles rough sets by replacing each object with a pair of crisp sets. These are called the lower and upper approximations. The lower approximation consists of all objects which belong to the set with certainty, and the upper approximation contains objects which possibly belong to the set. An elementary example follows. Information about the real world is provided in the form of an information or decision table. Rows of the table correspond to the objects (examples, sites, entities) and the columns of the table describe

74

S.B. Saila Table 3.1 Example information table for concept development using juvenile brook trout habitat variables and estimated abundance Site FN VD GS JBT s1 Y Y Low Low s2 Y Y Med High s3 Y Y High High s4 N Y Low Low s5 N N Med Low s6 N Y High High Note: FN refers to a Froude number <0.20, VD refers to a Velocity/Depth ratio <1.25, GS refers to a Gravel/Sand ratio, and JBT refers to the relative numerical abundance of juvenile brook per unit area.

properties of these objects. These are termed attributes or features. Two kinds of attribute are distinguished, and they are condition attributes and decision attributes. Characteristically, the information table contains a single decision attribute and several condition attributes. The simple example provided herein is contained in an information table (Table 3.1) which describes a hypothetical assessment of juvenile stream habitat for the brook trout, (Salvelinus fontinalis). In this table the examples are distinct sites sampled (6 in this case). The condition attributes are properties of the sites (3 properties), and the decision attribute consists of the estimated levels of juvenile brook trout found at each site of similar dimension. A major concept of rough set theory is an indiscernability relation defined for a given subset of attributes. Consider a set consisting of two condition attributes of Table 3.1 (FN – the Froude number and VD – the velocity/depth ratio). The attribute value is Y (yes) for these attributes. Sites s1, s2 and s3 are characterized by the same values of both attributes. As a result, sites s1, s2, and s3 will be referred to as indiscernible according to attribute set {FN, VD}. In addition, sites s4 and s6 are also indiscernible according to attribute set FN and VD. This indiscernability relation is clearly an equivalence relation. Thus the set of attributes FN and VD define the following elementary sets: ES1 = {s1, s2, s3}, ES2 = {s4, s6}, and ES3 = {s5}. The latter is a single element elementary set. Any finite union of elementary sets is called a definable set. An example set of objects {s1, s2, s3, s5}, is defined for attribute set {FN,VD} to represent it as a union of ES1 and ES3, since it is possible to represent it as a union of elementary sets. In this case it is a union of ES1 and ES3. The other set {s4, s6} is defined by the attribute FN equal to no and the attribute VD equal to yes. The concept of the indiscernability relation permits one to simply define redundant (dispensable) attributes and to introduce the notion of attribute redundancy. If any set of attributes and it’s superset define the same indiscernibility relation (that is, if elementary sets of both relations are identical) then any attribute that belongs to the superset and not to the set is redundant. In the example from Table 3.1, let the set of attributes be the set Froude number (FN)

3

Guide to Some Computerized Artificial Intelligence Methods

75

and the gravel-sand ratio (GS), and let the superset be the set of all three attributes; that is, the set (FN), (VD), and (GS). Elementary sets of the indiscernibility relation defined by the set Froude number (FN) and the gravel-sand ratio (GS) are singletons, that is, sets {s1}, {s2}, {s3}, {s4}, {s5}, and {s6}, and so are elementary sets of the indiscernibility relation defined by the set of all three attributes. Therefore, the attribute velocity-depth ratio (VD) is redundant. On the other hand, the set Froude number (FN) and the gravel-sand ratio (GS) does not contain any redundant attribute since elementary sets for attribute sets Froude number (FN) and the gravel-sand ratio (GS) are not singletons. Such a set of attributes with no redundant attribute is called minimal or independent. The set P of attributes is the reduct of another set Q of attributes if P is minimal and the indiscernibility relations defined by P and Q are the same. This last condition indicates that elementary sets, determined by indiscernibility relations defined by P and Q are identical. In our example, the set (FN, GS) is a reduct of the original set of attributes (FN, VD, and GS). Table 3.2 illustrates a new information table based on this reduct. So far in this discussion we have not addressed decision attributes. By analogy with the condition attributes, it is possible to define elementary sets associated with the decision attribute as subsets of the set of all examples with the same value of the decision. Such subsets are called concepts. From Tables 3.1 and 3.2 the concepts are {s1, s4, s5} and {s2, s3, s6}. The first concept corresponds to all samples with higher than average juvenile trout abundance and the second to all samples with less than average juvenile abundance. The important question is whether it is possible to determine greater or less than average trout abundance based on attribute values in Table 3.2. To answer this question recall that in terms of rough set theory a decision on abundance greater than average depends on attributes FN and GS since all elementary sets of indiscernibility relations associated with FN and GS are subsets of some concepts. It is possible to induce the following rules from Table 3.2. [gravel-sand ratio (GS) medium] ! JBT low [Froude number (FN) no and gravel-sand ratio (GS) high] ! JBT low [Froude number (FN) yes and gravel-sand ratio (GS) high] ! JBT high [gravel-sand ratio (GS) very high] ! JBT high

Table 3.2 Revised information table derived from Table 3.1 Site FN GS JBT S1 S2 S3 S4 S5 S6

Y Y Y N N N

Low Med High Low Med High

Low High High Low Low High

76

S.B. Saila Table 3.3 Inconsistent information table derived from Table 3.1 Site FN GS JBT s1 s2 s3 s4 s5 s6 s7 s8

Y Y Y N N N N N

Low Med High Low Med High Med High

Low High High Low Low High High Low

Now, enhance data from Table 3.2 by two additional samples s7 and s8 as shown in Table 3.3. Elementary sets of the indiscernibility relation defined by attributes FN and GS are {s1}, {s2}, {s3}, {s4}, {s5, s7}, and {s6, s8} while concepts defined by decision JBT are: {s1, s4, s5, s8} and {s2, s3, s6, s7} In Table 3.3 it is evident that the decision JBT does not depend on attributes FN and GS since neither {s5, s7}} nor {s6, s8} are subsets of any concept. Table 3.3 is described as inconsistent because samples s5 and s7 are conflicting. They are inconsistent since for both samples the value of any attribute is the same, and yet the decision value is different. Examples s6 and s8 are also conflicting. Rough set theory offers a tool to deal with inconsistencies. For each concept of X the greatest definable set contained in X and the least definable set containing X are computed. The former set is called the lower approximation of X and the latter is called an upper approximation of X. In the case of Table 3.3, for the concept {s2, s3, s6, s7} describing juvenile trout relative abundance the lower approximation is equal to the set {s2, s3} and the upper approximation is equal to the set {s2, s3, s5, s6, s7, s8}. For any concept, rules induced from its lower approximation are certainly valid (hence such rules are certain). Rules induced from the upper approximation of a concept are possibly valid. They are called possible. For Table 3.3, certain rules are: [GS medium] ! JBT low [FN yes] and [GS medium] ! JBT high [FN yes] and [GS high] ! JBT high Possible rules are: [FN no] ! JBT low [GS low] ! JBT low [GS med] ! JBT high [GS high] ! JBT high Measures of uncertainty have been developed within rough set theory. A frequently used one is the quality of lower approximation and the quality

3

Guide to Some Computerized Artificial Intelligence Methods

77

of upper approximation. For a given set X of examples not necessarily definable by a set P of attributes, the quality of lower approximation is a ratio of the number of all elements in the lower approximation of X to the total number of examples. Similarly, the quality of upper approximation is the ratio of the number of all elements in the upper approximation of X to the total number of examples. In the example from Table 3.3 for the concept X= {s1, s4, s5, s8} the quality of lower approximation is 0.25 and the quality of upper approximation is 0.75. Procedures for the induction of decision values from decision tables have been presented by Grzymala-Busse (1992, 1997) and others. Induction algorithms utilize one of the following strategies; (a) generation of a minimal set of values covering all objects from a decision table, (b) generation of an exhaustive set of rules and consisting of all possible rules for a decision table, and (c) generation of a set of ‘‘strong’’ decision rules, covering many but not necessarily all objects from the decision table. The application of the rough sets approach involves the following sequence of operations: (1) (2) (3) (4) (5) (6)

Data acquisition, selection of attributes Creation of a decision table Discretation of continuous data Reduction of decision table Induction of decision rules Validation.

3.4 Case-Based Reasoning In the earlier version of this chapter (Saila, 1996) no mention was made of casebased reasoning (CBR) because this soft computing tool had not been utilized to any substantial extent by fishery scientists or oceanographers. However, several papers as mentioned in the introduction, have already been published by fishery scientists which utilize data mining in order to obtain more information from a data base. It appears that these reports did not fully utilize certain formal data mining strategies, such as rough sets or case-based reasoning. Pal and Shiu (2004) clearly demonstrated the utility of CBR and it is therefore introduced here as a potentially valuable contribution to fishery science. A brief definition of CBR is that it is a methodology for solving problems by utilizing previous experiences. It can be more precisely defined as a model of reasoning that incorporates problem solving, understanding, and learning; and integrates them all with reasoning processes. This suggests that the abovementioned concepts touch upon some of the basic issues relative to knowledge representation, reasoning, and learning from experience. A major contribution of Pal and Shiu (2004) involved an extension of CBR, termed soft case-based reasoning (SCBR). SCBR is a combination of methodologies which collectively

78

S.B. Saila

provide a foundation for the conception, design, and integration of intelligent systems. In general, a case-based reasoner will be presented with a problem, either by a user or a program or system. The case-based reasoner then searches its memory for past cases looking for one that has the same problem specification as the case under analysis. The case-based reasoner solves new problems by adopting solutions to other problems. In many CBR systems the case-based reasoning mechanism has an internal structure divided into two parts: the case retriever and the case reasoner. The case retriever’s task is to find the appropriate case(s) in the case data base. The case reasoner then uses the retrieved cases to find a solution to the problem description provided. This generally involves determining the difference between the retrieved case and the concept case, and modifying the solution to reflect the difference appropriately. The essential parts of the CBR system involve the following: (a) (b) (c) (d)

Retrieving previously examined cases, Reusing the case by copying or integrating solutions from retrieved cases, Revising or adapting retrieved solutions in an effort to solve a new problem, Returning the new solution to the case database once it has been validated.

In addition to the book by Pal and Shiu (2004), a very readable tutorial on CBR has been provided by Main et al. (2001). Most recently Chao (2006) has published a data base development and management book which demonstrates the use of many software tools suitable for PC implementation of data base processing systems. Another very readable book on CBR is provided by Kolodner (1993). This book contains an extensive bibliography as well as an appendix which contains a 47-page library of case-based reasoning systems with a brief description of each one. It should be recognized that CBR is not suitable for certain types of problems. It is suggested that CBR may be applied to situations where: (a) There is little information available in the underlying model, (b) There are abundant prior cases available, (c) There is substantial benefit in adapting past solutions. In some instances CBR systems are a suitable alternative to rule-based (expert) systems. One of the more difficult aspects of rule-based systems is the knowledge acquisition task. Case-based reasoning usually requires less knowledge acquisition since it involves combining a set of past experiences without the necessity of developing a formal domain model.

3.4.1 Selected Applications Several applications of CBR to fishery and oceanographic situations are available in the scientific literature. Cui et al. (2003) have developed a fishing

3

Guide to Some Computerized Artificial Intelligence Methods

79

condition and forecasting analysis for the neon flying squid (Ommastrepes bartrami) using CBR. Yunyan et al. (2004) applied CBR in order to analyze and extract ocean mesoscale spatial similarity information. A final example by Corchado et al. (2001) is found in a book describing hybrid models which integrate in a single problem-solving mechanism, cycles of neural network and case-based reasoning processes.

3.5 Expert System Developments 3.5.1 Introduction A significant increase in the number of references relating to expert systems was found in this literature search in comparison to the initial search. A total of 347 references (1995–2005) were located by a search procedure identical to the initial one. Also it became evident that many of these references involved socalled hybrid systems, which are a combination of expert systems with other soft computing methodologies such as neural networks and genetic algorithms. Unfortunately most of these hybrid systems related primarily to the physical sciences such as climatology, meteorology, and physical oceanography. These hybrid systems consisted of expert systems combined with neural networks, genetic algorithms, and/or simulated annealing. It was suggested in the prior review of expert systems (Saila, 1996) that an integrated expert system (more recently termed hybrid system) would become an increasingly powerful and versatile tool for a number of application areas. This speculation was clearly justified on the basis of current increases in the use of expert systems combined with other artificial intelligence tools.

3.5.2 Expert System Application Extensions Table 3.4 lists and briefly describes examples of some more recent applications of expert systems, alone and in combination with other soft computing methodologies. The contents of Table 3.4 have been chosen to illustrate fishery and oceanography related problems for the most part. This results in an abbreviated list. The number of hybrid systems illustrated in this table is small due to the selection process which was employed, namely, emphasis on fishery related examples. However, it seems clear that there is an increasing tendency to combine expert systems with other soft computing methodologies because this seems to improve the performance of the entire system. Table 3.4 illustrates only a few of the novel hybrid systems which involve other methodologies combined with expert systems. I believe that there will be a continuing trend in the increase of hybrid systems, and the combination of

80

S.B. Saila

Table 3.4 Expert system extensions (1995–2005) related to fisheries and oceanography Reference Hybrid Summary Maloney et al. (2004)

no

Knud-Hansen et al. (2003)

no

Fang (2003)

no

Chen and Mynett (2003)

yes

Newell and Richardson (2003)

no

Li et al. (2002)

no

Handee and Berkelmans (2002) Miller and Field (2002) Mackinson (2000) Hu and Chen (2001) Zeldis and Prescott (2000) Lee et al. (2000) Hernandes-Llamas and Villereal-Colomares (1999) Chen et al. (1999)

no no yes no yes yes no

Painting and Korrubel (1998) Korrubel et al. (1998)

no yes

Handee (1998) Mackinson (2001) Saila (1997)

no yes yes

Pawlak et al. (1993)

no

no

anchovy recruitment strength prediction based on hydroacoustic surveys expert system to determine fish pond fertilization rates ocean community analysis related to Taiwan Strait tunnel combined data mining with heuristic knowledge for algal bloom prediction expert system for combining physical factors with soft shellfish production web-based intelligent diagnosis system for fish diseases expert coral reef early warning system to detect coral bleaching rule-based model for predicting recruitment fuzzy logic expert system for herring prediction expert system for chub mackerel prediction fish disease diagnosis system control system for denitrification analysis shrimp farming expert system stock assessment for hairtail, chub mackerel and filefish anchovy recruitment forecast recruitment success estimates for South African anchovy environmental monitoring expert system fuzzy logic expert system combined neural net and fuzzy logic system for quota based lobster management uncertainty analysis in expert systems

fuzzy logic with expert systems will be found to be particularly useful in fishery related problem areas.

3.5.3 Expert System Summary In general, it seems clear that expert systems are increasingly becoming a part of a hybrid system involving other soft computing methodologies. This trend is expected to continue and only a limited use of expert systems alone is predicted for the long term. The importance of combining various soft computing methodologies cannot be overemphasized. In combination these supply scientists with a more effective tool box to deal with complex, dynamic, and uncertain real world problems such as those encountered in fisheries and ecosystem management.

3

Guide to Some Computerized Artificial Intelligence Methods

81

3.6 Neural Networks 3.6.1 Background A brief introductory background to artificial neural networks was provided in the earlier version of this section (Saila, 1996). Therefore, emphasis will be placed herein almost exclusively on new applications during the past decade. The earlier review cited above suggested that the potential application of neural networks in fisheries and oceanography would be great. This prediction has apparently been realized. A search of the Aquatic Sciences and Fisheries Abstracts database revealed more than 1,000 citations during the past decade involving neural networks. These included use of neural networks alone as well as in conjunction with other soft computing methods. Saila (2005) has recently described neural networks used for classification purposes in fisheries and oceanography. A brief description of neural network tasks which include classification was provided in this reference. In addition, a listing of 32 references using neural network classification was provided. Therefore, only a few new applications of classification by neural networks will be provided herein. The large number of references related to neural network applications has been provided by authors from countries mainly outside the North American continent. It appears to me that scientists in Asia and Europe have been more active in applying neural nets to solving complex problems.

3.6.2 Extended Examples A very small sample of references was chosen from the large reference list to indicate the spectrum of applications of artificial neural networks in fisheries and oceanographic studies during the past decade. These are hopefully indications of the progress which has been made during the past 10 years. It is apparent that not only are artificial neural networks the most popular soft computing methodology at present, but also that there have been rapidly increasing uses of neural networks with other soft computing methods. At this time the majority of the applications of neural networks are in the physical sciences with a relatively small contribution from the life sciences. This is thought to be an anomalous situation because biological data are consistently more imprecise and vague than physical data, which should suggest a greater use of soft computing methods. The following material briefly describes some applications with an emphasis on the biological sciences. Mwale et al. (2005) utilized a genetic algorithm neural network disaggregation model to link scales of seasonal climate and hydrology for the prediction of weekly annual stream flow without modeling the rainfall-runoff process. Chua and Holz (2005) describe a neural network finite element river flow model which solves the two-dimensional shallow water

82

S.B. Saila

equations. A comparison was made by Salas et al. (2004) which showed that a recurrent neural network had better performance than a feed-forward neural network in the forecasting of multiple wave parameters. On the biological side Leffaile et al. (2004) used a neural network to classify ecological profiles, showing that eel densities were significantly related to three major influencing variables. Hsieh (2004) has demonstrated an optimal bioremediation design by means of a hybrid genetic-simulated annealing algorithm. His analysis of ocean and atmospheric data showed that linear methods were too simplistic to effectively describe certain real world systems. Classification of fish to nursery areas was accomplished by Hansen et al. (2004) using a parametric discriminant function analysis and with a neural networks simulation. Otolith composition analysis was shown to be a valid technique for classifying juvenile gag (Mycteroperca microlepis) to estuarine habitats. Discriminant functions and neural network analysis were also applied to herring scale data to predict age at maturity for several year classes by Engelhard and Heino (2004). It was demonstrated by Hernandez-Borges et al. (2004) that species like limpets can be classified according to their level of n-alkanes when an artificial neural network was utilized. Pei et al. (2004) demonstrated that the artificial neural network is an effective procedure for forecasting the concentration of chlorophyll-a and can provide a basis for the control of eutrophication in a lake. In order to predict the occurrence of each species in a site from a common set of predictor variables, Joy and Death (2004) utilized a multi response neural network to produce a single model to predict the entire fish and decapod assemblage in one procedure.

3.6.3 Summary The above-mentioned citations represent a small but useful sample to indicate the value and diversity of neural network applications undertaking during the past decade. This review clearly indicates substantial increases in the use of neural nets in scientific research activities. It is expected that this trend will continue in the future especially in the case of enhanced (hybrid) systems with neural networks as one important component of the system.

3.7 Genetic Algorithms 3.7.1 Background In contrast to earlier developments regarding a very limited use of genetic algorithms (Saila, 1996), the current situation indicates a substantial increase in utilization during the past decade. In the earlier version of this section it was suggested that genetic algorithms would be increasingly used, especially in the development of hybrid systems which combine this tool with neural networks,

3

Guide to Some Computerized Artificial Intelligence Methods

83

expert systems, and other soft computing tools. This prediction has been fulfilled during the past decade as will be evident in the cited references. No further attempt will be made herein to repeat introductory material presented in the earlier version of this section. However, a few additional statements will be made regarding the utility of genetic algorithms in general as well as in a fishery related context. Genetic algorithms are a very important problem-solving methodology. They are especially useful for solving complicated problems with several independent variables and a large number of possible outcomes. In general, I believe that the ideal method for solving a complex problem is to find a closed form solution. A simple example involves linear programming which can exactly solve certain specific types of linear problems. However, it is not difficult to demonstrate that the exact solution is sometimes elusive. For example, finding the minimum distance for a so-called traveling salesman problem may be a formidable undertaking for a closed form solution. Assume that there are 16 separate locations for lobster pot buoys in a lobster fishing area. The problem is to visit and tend traps at each buoy only once and return to port in a manner which minimizes total distance traveled, a savings of time as well as fuel cost. This problem seems like a relatively simple permutation of the number of vessel tracks until it is realized that there is an extremely large number of possible tracks which the vessel could take. This number is exactly 1,307,174,368,000. Clearly, the time it would take to solve this problem exactly would be prohibitive. However, this type of problem can be quickly solved utilizing a genetic algorithm. Another important difficulty to finding closed form solutions is that special purpose algorithms require certain assumptions regarding the nature of the problem to be met. Real world ecosystem level problems are often simplified to the point where a computer model no longer reflects a realistic situation. It is recognized that simulation is often utilized when it is not possible to obtain a closed form solution. It should be kept in mind that simulation is not an optimization technique. However, intuition and background knowledge are often used to try different solutions in the hope of finding an optimum. The situation mentioned above changes for the better when a search algorithm is used in conjunction with simulation. This allows the computer to search through a range of solutions. Genetic algorithms are thought to be the most powerful and flexible for this purpose. The real power of genetic algorithms is their ability to simultaneously search through numerous solutions and to combine the best ones in order to progressively achieve a better solution. This permits a convergence to an optimal or near optimal solution quickly. It seems evident to me that genetic algorithms should be applied more frequently in fishery related problem solving.

3.7.2 Application Examples Some references which seem to be closely related to fishery and oceanographic studies are briefly described. These references have been selected from an extensive list of genetic algorithm applications derived from the Aquatic

84

S.B. Saila

Sciences and Fisheries Abstract data base which was also searched in a similar manner in the earlier version of this section. Some fishery related citations will be considered first and it should be made clear that these are not a comprehensive set of references. The most recent article I have encountered which deals with a fishery application of a genetic algorithm is one by Drake and Lodge (2006) found in Fisheries. This report deals with forecasting potential distributions of nonindigenous fish species. Iguchi et al. (2004) applied an ecological niche model incorporating a genetic algorithm to predict potential distributions of two species of North American basses in Japan. Perez-Lasada et al. (2004) used a genetic algorithm in a heuristic search technique among several others for the analysis of molecular and morphologic data on Thoracican barnacles. The effects of density dependence on the diel vertical migration of krill was modeled utilizing a genetic algorithm (Burrows and Tarling, 2004). Guinand et al. (2004) demonstrated the effectiveness of a genetic algorithm for discovering multi-locus combinations that provide accurate individual decisions and estimates of mixture composition based on likelihood classification of lake trout (Salvelinus namaycush). A phytoplankton-zooplankton model and a genetic algorithm were utilized to determine parameter values that would maximize the value of certain goal functions in a report described by Kristensen et al. (2003). Dagorn (1994) utilizes a genetic algorithm to find optimal movement and schooling behavior for synthetic tuna populations in heterogenous and dynamic environments. These data were then compared with real schooling behavior to provide models for comparisons with real world experiments involving tropical tunas. In studies of the vertical migration of fish, Strand (2003) presents several models of vertical migration which are based on genetic algorithms. The remainder of these references are more general but I believe that they may be of interest to oceanographers and environmentalists. Jesus and Caiti (1996) described the estimation of geo-acoustic bottom properties from towed array data. This work involved an adaptive genetic algorithm. Rogers et al. (1995) have presented an approach to nonlinear optimization that combines an artificial neural network and a genetic algorithm to effectively search for costeffective groundwater remediation strategies. The versatility and effectiveness of genetic algorithms can be seen from this small sample, whether in combination with other soft computing methods or in isolation.

3.8 Simulated Annealing (SA) 3.8.1 Background The earlier brief description of simulated annealing did not list fishery or oceanographic applications of this methodology (Saila, 1996) because none seemed to be unavailable at that time. However, a current search of the literature using the Aquatic Sciences and Fisheries Abstracts search engine listed 122 records of

3

Guide to Some Computerized Artificial Intelligence Methods

85

publications which utilized simulated annealing. The references found from the search were primarily applications in physical oceanography and hydrology with a few interesting applications in fishery science.

3.8.2 SA Applications Material which follows is but a small sample of available published information in this subject area in which there are only a few abstracts related to fisheries and oceanography. Watters and Deriso (2000) have utilized a simulated annealing algorithm to summarize regression tree results to partition fishing grounds for near-term trends by catch per unit effort studies. They also indicated that simulated annealing can be useful for designing spatial strata in future sampling programs. A simulated annealing algorithm coupled with a recruitment model was utilized by Jager and Rose (2003) to discover flow regimes that minimize either the number of out-migrant smolt recruits or the variation in spawning time among recruits. They conclude that there is a potential for managing species with a special conservation status by combining state of the art optimization methods with nonparametric ecological models. Some oceanographic and environmental applications follow. Shieh and Peralta (2005) combined a genetic algorithm with simulated annealing to provide a cost-benefit analysis to determine the optimum number of stations to include in a new design. This optimization was solved by simulated annealing. Failkowski et al. (2003) utilized a simulated annealing optimization method and a coordinate rotation technique to characterize sediment properties derived from geoacoustic data. Teegavarapu and Simonovic (2002) utilized simulated annealing to optimize the operation of multiple reservoirs. Results obtained by the authors suggest that simulated annealing can be used to obtain near optimal solutions for multi-period reservoir operations that are computationally intractable. Simulated annealing was used by Brus et al. (2002) to find sampling designs with a minimum sampling variance for a fixed budget. Lawson and Godday (2001) developed a computer based design tool. The optimization of the objectives was achieved using a simulated annealing algorithm. It seems clear that the last decade has justified the previously made statement (Saila, 1996) that the role of simulated annealing in the solution of tough optimization problems should not be underestimated by fishery scientists. It is believed that the discipline will benefit measurably from applications by using this versatile soft computing tool in future studies.

3.9 Summary Statements and Some Further Speculation 3.9.1 Summary I believe that effective fishery management involves an understanding of a very complex system composed of several strongly interacting components and

86

S.B. Saila

subsystems. Fishery scientists seem to share a belief about complexity which is not necessarily documented by the nature of fishery science. They recognize somewhat that these systems are nonlinear, meaning that an effect is not proportional to a cause. Indeed, because of nonlinearity (due largely to the fact that fishery data contain lots of noise) the same cause (say fishing effort) can have very different effects on yield depending on circumstances. Major goals of studying nonlinearity are the measurement of the factors affecting a fishery system and, therefore, the prediction of future yields. Most of the present activity related to the construction and application of simulation models in fishery management address only a portion of the overall problem. It is my opinion that the greater challenge involves the data and model parameter estimates. Fishery data and model parameters can generally be characterized as noisy. In any case they are ‘‘fuzzy’’ from causes such as uncertainty. Fuzzy logic is a problem-solving technology which produces a useful way to draw definite conclusions from vague, un certain, and/or imprecise information. Imprecision refers to the lack of specificity of the contents of an information element. For example, the annual adult natural mortality rate of species X is believed to be between 10 and 25 percent. It should be recognized more clearly that in the fishery management system there are variable recruitment and growth rates, which lead to a nonlinear model of growth, exploitation, and decay. The resultant system may have considerably more than one possible steady state. This is not an easy problem to solve. Walters and Martell (2004) clearly and concisely define models for single species, multi-species, as well as models at an ecosystem level. The limitations of current conventional models and their parameters are also described. Much of what they discuss is summarized on page 286 of their book. They state that the real issue is not the quality of assumptions but the quality of approximations. I agree fully with them and believe that there is much room for improvement.

3.9.2 Some Speculation The phrase ‘‘artificial life’’ has not been previously used in this report nor in the earlier version of it. However, this phrase has been in common use by the scientific community for several decades. Langton (1989) defines artificial life as ‘‘The study of man-made systems that exhibit behavior characteristic of natural living systems. It complements the traditional biological sciences concerned with the analysis of living organisms by attempting to synthesize life-like behavior within computers and other artificial media.’’ More than a decade ago French fishery scientists utilized artificial life concepts in fishery-related scientific studies. These include Dagorn (1992, 1994) and Le Page (1996). Le Page demonstrated that models of artificial life represent an effective means for assessing characteristics of population dynamics models. Dagorn developed an artificial life model to study tuna migrations and behavior, and compared them to real world observations of tuna migration and movements.

3

Guide to Some Computerized Artificial Intelligence Methods

87

The study of fishery management methods has traditionally been expressed formally as systems of algebraic or differential equations. It should be recognized that equational models are subject to several limitations. For example, in many models it is common to refer to the derivative of the variable with respect to population size N. This implies the assumption of a very large population in order for such a derivative to make sense. Such an assumption has the effect of limiting small population effects, such as extinction. In my opinion, another serious difficulty lies in the fact that even a simple model of behavior as a function of genetic and environmental variables would require a very large number of equations, and there appear to be no appropriate tools for dealing with equations of this complexity. Additionally, equations are not good at dealing with highly nonlinear effects such as if-then-else conditionals. Artificial life provides insight into the development of an alternative population modeling paradigm that dispenses with equations entirely. Instead, the population or system is identified procedurally as a set of programs, one for each organism or some part of it. Generally speaking this representation of organisms by programs is the defining feature of artificial life models of behavior. I suggest artificial life will become quite useful in addressing issues fundamental to fisheries science and oceanography. These issues include evolution and stability of ecosystems. In such a scenario it is suggested that in an artificial life system parameters can be manipulated and compared with real world observations. In such a system measurements can be made that may be difficult or even impossible to obtain in natural systems. Taylor and Jefferson (1994) as well as Adami (2002) clearly identify the advantages and limitations of artificial life programs for studying changes in real time. The possibilities of soft computing in artificial life applications remain to be further tested by fishery scientists and oceanographers. Terzopoules et al. (1995) have already developed artificial life algorithms which emulate both individual as well as group behavior of fishes in a simulated physical world. Results from future studies of this nature will probably greatly exceed our current expectations. To repeat, I believe that fishery scientists have not been sufficiently concerned with the fact that conventional models may have serious limitations and may even be misleading when applied to complex systems, such as those encountered in ecosystem level management. It seems to me that future mathematical developments should deal more explicitly with the organization and evolution of system behavior and not merely with descriptions of static systems. Although the calculus of derivatives and integrals has been useful in describing relations between continuous quantities and their rates of change, its value is limited by the assumptions involved in its use. This caution might also apply to statistics, which has helped us in providing insights into the behavior of large populations. I conclude that the commonly used mathematical and statistical tools may be of limited value when it comes to the further understanding of complex living systems. I believe that soft computing may provide a toolbox of approaches to a better understanding of such systems. It is my profound hope that these will be given further attention.

88

S.B. Saila

References Adami L (2002) Ab initio modeling of ecosystem with artificial life. Natural Resource Modeling 15(1):135–145 Brus DJ, Jansen MJW, de Gruijter JJ (2002) Optimizing two- and three-stage designs for spatial inventories of natural resources by simulated annealing. Environmental and Ecological Statistics 9(1):71–88 Burrows MJ, Tarling G (2004) Effects of density dependence on diel vertical migration of populations of northern krill: a genetic algorithm model. Marine Ecology Progress Series 277:209–220 Chao L (2006) Database development and management. Auerbach Publications, Boca Raton, Florida Chen Q, Mynett RE (2003) Integration of data mining techniques and heuristic knowledge in fuzzy logic modeling of eurtrophication in Taibu Lake. Ecological Modeling 162 (1–2):55–67 Chen W, Li C, Hu F, Cui X (1999) The design and development of the expert system for fish stock assessment. Journal of Fisheries in China 23(1):343–349 Che`vre N, Gagni F, Gagnan P, Blaise C (2003) Application of rough sets to identify polluted sites based on a battery of biomakers: a comparison with classical methods. Chemosphere 51:13–23 Chua LHC, Holz KP (2005) Hybrid neural network – finite element river flow model. Journal of Hydraulic Engineering 131(1):52–59 Corchado J, Diken J, Rees N (2001) Artificial intelligence models for oceanographic forecasting. Plymouth Marine Laboratory, Plymouth, United Kingdom, 211 pp Corchado JM, Lees B (2001) Adaptation of cases for case-based forecasting with neural network support. In: Pal SK, Dillon TS, Yenng DS (eds) Soft computing in case-based reasoning. Springer-Verlag, London, pp. 253–319 Cui X, Fan W, Shen X (2003) Development of the fishing condition analysis and forecasting system of Ommastrephes bartrami in the Northeast Pacific Ocean. Journal of Fisheries of China 27(6):600–605 Dagorn L (1992) The emergence of artificial intelligence: application to tuna populations. Collective volume of scientific papers. International Commission for the Conservation of Atlantic Tunas 39(1):385–389 Dagorn L (1994) The behavior of tropical tuna. Modeling using the artificial-life concept. Ecole Nationale Superior d’Agrunomie, Rennes, France. Thesis 250 pp Drake JM, Lodge DM (2006) Forecasting potential distributions of nonindigenous species with a genetic algorithm. Fisheries 31(1):9–16 Engelhard GH, Heino M (2004) Maturity changes in Norwegian spring-spawning herring, before, during and after a major population collapse. Fish and Research 66(2–3):299–310 Englund G, Sernalle O, Cooper SD (1999) The importance of data selection criteria: meta-analysis of stream production experiments. Ecology 80(4):1132–1141 Failkowski LT, Dacol DK, Lingevitch JF, Kim E (2003) Rapid geoacustic inversion with a curved horizontal array. Journal of the Acoustical Society of America 113(4):2216 Fang H-Y (2003) The ocean community. Marine Georesources and Geotechnology 21(3–4):135–166 Grzymala-Busse JW (1988) Knowledge acquisition under uncertainty – a rough set approach. Journal of Intelligent Robotic Systems 1:3–16 Grzymala-Busse JW (1992) LERS: a system for learning from examples based on rough sets In: Slowinski R (ed) Intelligent decision support, handbook of applications and advances of rough sets theory. Kluwer, Dordrecht, The Netherlands, pp. 3–18 Grzymala-Busse JW (1997) A new version of the rule induction system LERS. Fundamata Informatica 31:27–39 Guinand B, Scribner KT, Topchy A, Page KS, Punch W, Barnhem-Curtis MK (2004) Sampling issues affecting accuracy of likelihood-based classification using genetical data. Environmental Biology of Fishes 69(1–4):245–259

3

Guide to Some Computerized Artificial Intelligence Methods

89

Handee JC (1998) An expert system for marine environmental monitoring in The Florida Keys National Marine Sanctuary and Florida Bay. Environmental Coastal Regions pp. 56–66 Handee JC, Berkelmans R (2002) Expert system generated coral bleaching alerts for Myrmidon and Agincont reefs, Great Barrier Reef, Australia. Proceedings of the Ninth International and Reef Symposium, Bali 23–27 October 2:1089–1104 Hansen PJ, Koenig CC, Zdanewicz US (2004) Elemental composition of otoliths used to trace estuarine habitats of juvenile gag Mycteroperca microlepis along the west coast of Florida. Marine Ecology Progress Series 267:253–265 Hastie T, Tibshirani R, Friedman J (2001) The elements of statistical learning – data mining Inference and prediction. Springer-Verlag, New York Hernandes-Llamas A, Villereal-Colomares H (1999) TEMA: a software reference to shrimp Litopenaeas vannami farming practice. Aquaculture Economics and Management 3(3):267–280 Hernandez-Borges J, Corbella-Tena R, Rodrigues-Delgardo MA, Garcia-Montelongo FJ, Havel J (2004) Content of alephatic hydrocarbons in limpets as a new way for classification of species using artificial neural networks. Chemosphere 54(8) 1059–1069 Hsieh WW (2004) Nonlinear multivariate and time series analysis by neural network methods. Reviews of Geophysics 42(1):875–1209 Hu F, Chen W (2001) Catch prediction of chub mackerel in the East China Sea by using fish stock assessment expert system. Journal of Fisheries of China 25(5):465–473 Iguchi K, Matsurra K, McNyset K, Kristina M, Peterson A, Scachetti-Pereira R, Powers, K, Vieglais D, Wiley E, Yudo T (2004) Predicting invasions of North America basses in Japan using native range data and a genetic algorithm. Transactions of the American Fisheries Society 133(4):245–254 Jager HI, Rose KA (2003) Designing optimal flow patterns for fall Chinook salmon in a Central Valley, California, River. North American Journal of Fisheries Management 25(1):1–21 Jesus SM, Caiti A (1996) Estimating geoacoustic bottom properties from towed array data. Journal of Computational Acoustics 4(3):273–290 Joy MK, Death RG (2004) Predictive modeling and spatial mapping in freshwater fish and decapod assemblages using GIS and neural networks. Freshwater Biology 49(1):1036–1052 Knud-Hansen CF, Hopkins KD, Guttman H (2003) A comparative analysis of the fixed input, computer modeling, and algal bioassay approaches for identifying pond fertilization requirements for semi-intensive aquaculture. Aquaculture 228(1–4):189–214 Kolodner J (1993) Case-based reasoning. M. Kauffmann, San Mateo, California Korrubel JL, Bloomer SF, Cochrane KL, Hutchings L, Field JG (1998) Forecasting in South African pelagic fisheries management: the use of expert and decision support systems. South African Journal of Marine Science 19:415–423 Kristensen NP, Gabric A, Braddock R, Cropp R (2003) Is maximizing resilience compatible with established goal functions? Ecological Modeling 169(1):61–71 Langton CG (1989) Artificial life. Proceedings of an interdisciplinary workshop on the synthesis and simulation of living systems. Addison-Wesley Publishing, Redwood City, New York Lawson K, Godday P (2001) Marine reserves: designing cost effective options. Economics of marine protected areas: a conference held at the UBS Fisheries Centre, July 2000. Fisheries Research Report 9(8):114–120 Le Page C (1996) Population dynamics and artificial life. Methodes d’etude des systemes halieutiques et aquacoles. Orstom, Paris (France) Colloques et seminares. Instat Francais de Researche Scientifigue pour de Developpement en Cooperation/Orstom, Paris, pp. 205–209 Lee PG, Lee RN, Prebilsky W, Turk DE, Ying H, Whitson JL (2000) Denitrification in aquaculture systems: an example of a fuzzy logic control problem. Aquacultural Engineering 23(1–3):37–59

90

S.B. Saila

Leffaile P, Baisez A, Rigend E, Feunteun E (2004) Habitat preferences of different European eel size classes in a reclaimed marsh: a combination to species and ecosystem conservation. Wetlands 24(3):642–651 Li D, Fu Z, Duan Y (2002) Fish-Expert: a web-based expert system for fish disease diagnosis. Expert Systems with Applications 23(3):311–320 Lin TY, Wildberger A (eds) (1995) Soft-computing: rough sets, fuzzy logic. neural networks, uncertainty management. Uncertainty Management Knowledge Discovery Simulation Councils, Inc., San Diego, California Mackinson S (2000) An adaptive fuzzy expert system for predicting structure, dynamics, and distribution of herring shoals. Ecological Modelling 126(2–3):155–178 Mackinson S (2001) Integrating local and scientific knowledge: an example in fisheries science. Environmental Management 27(4):533–545 Main J, Dillon TS, Shiu SCR (2001) A tutorial on case-based reasoning. In: Pal SK, Dillon TS, Young DS (eds) Soft computing in case-based reasoning. Springer-Verlag, London, pp. 2–27 Maloney CL, Vanderingen CD, Hutchings C, Field GJ (2004) Contributions of the Benguela ecology programme to pelagic fisheries management in South Africa (2004) South African Journal of Marine Science 26:37–51 Miller DCM, Field JS (2002) Predicting recruitment in South African anchovy-an expert system approach. Southern African Marine Science Ssposium (SAMSS 2002) Mwale D, Shen SSP, Gan JY (2005) Hilbert transforms, neural network genetic algorithms and disaggregation for the prediction of weekly annual streamflow from seasonal oceanic variability. American Meteorological Society Conference on Hydrology 19 Myers RA, MacKenzie BR, Bowen KC, Barrowman MJ (2001) What is the carrying capacity for fish in the ocean? A meta-analysis of population dynamics of North Atlantic cod (2001). Canadian Journal of Fisheries and Aquatic Sciences 58(7):1464–1476 Myers RA, Mertz G (1998) Reducing uncertainty in the biological basis of fisheries management by meta-analysis of data from many populations: a synthesis. Fisheries Research 37(1–3):51–60 Newell C, Richardson J (2003) An expert system for the optimization of shellfish raft culture. Journal of Shellfish Research 22(1):347 Orlowska E (ed) (1998) Incomplete information: rough set analysis. Physica-Verlag, Heidelberg Painting SJ, Korrubel JL (1998) Forecasts of recruitment in South African anchovy from SARP field data using a deterministic expert system. South African Journal of Marine Science 14:245–261 Pal SK and Shiu SCK (2004) Foundations of soft case-based reasoning. Wiley, Hoboken, New Jersey Pawlak Z (1982) Rough sets. International Journal of Informatics and Computer Science 11:341–356 Pawlak Z (1991) Rough sets, theoretical aspects of reasoning about data. Kluwer, Dordrecht, The Netherlands Pawlak Z (1999) Rough set theory for intelligent industrial applications. Proceedings of the second international conference on intelligent processing and manufacturing materials (IPM-95). IECF Press, Piscateway, New Jersey Vol. 1, pp. 37–44 Pawlak Z, Grzymala-Busse JM, Slowinski RM, Ziarko W (1993) Managing uncertainty in expert systems. Kluwer, Dordrecht, The Netherlands Pawlak Z, Grzymala-Busse JM, Slowinski RM, Ziarkow W (1995) Rough Sets. Communications of the ACM 38(11):89–95 Pei H, Lao N, Jiang Y (2004) Applications of back propagation neural network for predicting the concentrations of chlorophyll-a in West Lake. Acta Ecologica Sinica 24(2):246–251 Perez-Lasada M, Hoeeg JT, Crandall KA (2004) Unraveling the evolutionary radiation of the Thoracican barnacles using molecular and morphological evidence: a comparison of several divergent time estimation approaches. Systematic Biology 33(2):244–264 Polkowski L (2002) Rough sets. Physica-Verlag, Heidelberg

3

Guide to Some Computerized Artificial Intelligence Methods

91

Rogers LL, Dowle FS, Johnson VM (1995) Optimal field-scale groundwater remediation using neural networks and the genetic algorithm. Environmental Science and Technology 29(5):1145–1155 Rossi L, Slowinski R, Susmaga R (1999) Rough set approach to the evaluation of stormwater pollution. International Journal of the Environment and Pollution 12(2/3):232–250 Saila, SB (1996) Guide to some computerized artificial intelligence methods. In: Megrey B, Moksness E (eds) Computers in fisheries research. Chapman and Hall, New York, pp. 8–40 Saila SB (1997) Fuzzy control theory applied to American lobster management. Developing and sustaining world fisheries resources, the state of science and management, 3rd World Fisheries Congress Proceedings, pp 204–208 Saila SB (2005) Neural networks used in classification with emphasis on biological populations. In: Cadrin S, Friedland KD, Waldman J (eds) Stock identification methods applications in fishery science. Elsevier, Amsterdam, pp. 553–569 Salas CF, Koc L, Bales L (2004) Predictions of missing wave data by recurrent neural nets. Journal of Waterway, Port, Coastal and Ocean Engineering 130(5):256–265 Shieh HI, Peralta RC (2005) Optimal insitu bioremediation design by hybrid genetic algorithmsimulated annealing. Journal of Water Resources Planning and Management 131(1):61–78 Slowinski R (ed) (1992) Intelligent decision support handbook of applications and advances of the rough set theory. Kluwer, Derdrecht, The Netherlands Slowinski R (1995) Rough set approach to decision analysis. AI Expert 10(3):19–25 Strand E (2003) Adaptive models of vertical migration in fish. Dissertation, University of Bergen, Department of Fisheries and Marine Biology. Bergen, Norway, 213pp Taylor C, Jefferson D (1994) Artificial life as a tool for biological inquiry. Artificial Life 1:1–13 Taylor MJ, White KR (1992) A meta-analysis of hooking mortality of nonanadramous trout. North American Journal of Fisheries Management 12(4):760–767 Teegavarapu RV, Simonovic SP (2002) Optimal operation of reservoir systems using simulated annealing. Water Resources Management 11(5):401–428 Terzopoules DK, Tu X, Grzeszczuk R (1995) Artificial fishes: autonomous locomotion, perception, behavior and learning in a simulated physical world. Artificial Life 1:327–351 Ulanowicz RE (1986) Growth and development: ecosystems phenomenology. Springer-Verlag, New York Ulanowicz RE (2005) Ecological network analysis: an escape from the machine. In: Belgravo A, Schafer WM, Dunne J, Ulanowicz R (eds) Aquatic food webs, an ecosystem approach. Oxford University Press, New York Walters CJ, Martell SJD (2004) Fisheries ecology and management. Princeton University Press, Princeton, New Jersey Watters G, Deriso R (2000) Catches per unit of effort of bigeye tuna: a new analysis with regression trees and simulated annealing. Bulletin Inter-American Tropical Tuna Commission 21(8):531–552 Yunyan D, Le L, Su F, Tianyu Z, Xiaomei Y (2004) CBK spatial similarity analysis on mesoscale ocean eddies with remote sensing data. Indian Journal of Marine Sciences 33(4):319–338 Zeldis D, Prescott S (2000) Fish disease diagnostic program – problems and some solutions. Aquacultural Engineering 23(1–3):3–71 Ziarko W (ed) (1995) Rough sets, fuzzy sets, and knowledge discovery (RSKD ’93). Workshop in Computing Series, Springer-Verlag, London

Chapter 4

Geographical Information Systems (GIS) in Fisheries Management and Research Geoff Meaden

4.1 Introduction and Early History of GIS Geographical Information Systems (GIS) may be defined as a collection of computer hardware, software, data and personnel designed to collect, store, update, manipulate, analyse and display geographically referenced information (Rahel 2004). As a tool for terrestrial spatial analysis, GIS functionality first developed in the 1960s. This development occurred in Canada, emerging here as a result of the convergence of various needs and capabilities, e.g. a large spatial area having a rich resource base; insufficient capacity to map and record resource distribution by old cartographic means; developments in computing and computer graphics and computer assisted design; plus a number of individuals with a vision of how cartography might be automated and applied. During the 1970s considerable resources were put into further research into computer mapping by institutes such as Harvard University with their SYMAP initiative, plus GRID and GEOMAP (Tomlinson 1989). The 1980s were chiefly characterised by the move into GIS of the first private commercial software developers who could appreciate the great potential of this technology. Since the early 1990s GIS has continued to grow at an average compound rate of about 14% p.a., and sales world-wide today of GIS related hardware and software are measured in billions of dollars, with unquantified data sales on top of this. Most of this GIS success has been manifest in terrestrial applications. With the planet becoming an ever more crowded arena in terms of sourcing human needs, then ‘‘conflicts’’ over access to, and use of, space has become ever more prevalent. Conflict requires management and management requires decision making, and this is where GIS comes into its own. In almost all spheres of human activity GIS is now playing a useful role. Early GIS applications were mostly in the public domain in areas such as forestry, the emergency services, in public health and utilities, and in the defence arena, but now applications and use by private companies has probably equalled or overtaken public domain G. Meaden (*) Department of Geographical and Life Sciences, Canterbury Christ Church University, North Holmes Road, Canterbury, Kent, CT1 1QU, UK

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_4, Ó Springer ScienceþBusiness Media B.V. 2009

93

94

G. Meaden

applications. This is largely a result of cost reductions and positive perceptions of the utility of GIS, allied to the fact that GIS is now becoming an inbuilt component of some emerging technologies, e.g. vehicle navigation systems. So, terrestrial GIS is playing an important role in planning, managing and decision making, forecasting and modelling, reporting, education, and research, in a host of institutions, organisations and corporations world-wide. The emergence of GIS has not occurred in a vacuum. In fact, as shown in Box 4.1, it has only been possible through a series of parallel developments. Although many of these developments are in the computing or IT and digital areas, there is an interesting range in that some of the parallel developments are in fields that might not be obvious such as visualisation and geostatistics. Undoubtedly a fundamental promoter of all the parallel developments has been the huge cost reductions associated with computing, allied to the vast range of application developments that are proliferating from the likes of Silicon Valley. Box 4.1 GIS – Links to Parallel Technologies, Developments or Associated Disciplines For optimal operation it is essential that GIS does not function in a vacuum. Its development to date, plus the milieu in which it functions, is heavily dependent upon an array of associated disciplines, developments or capabilities. These include:

 The internet. This is required for information and for data downloads, and increasingly for interactive GIS functionality.

 Remote sensing. Satellite and aerial remotely sensed data provide possibly the largest source of data for GIS.

 Environmental modelling. Output from models provide a source of data and modelling itself is performed using GIS as a platform.

 Software developments. Not only are there many varied GIS packages  





but linkages between GIS and other software functionality is essential for many system’s tasks. Hardware. As well as forming the computer-based platform for GIS operations, numerous other pieces of hardware may form part of a complete GIS, e.g. scanners, plotters, digitisers, data loggers, GPS, sonar, etc. Visualisation. Mapping output has to be optimised in perceptual terms if this output is to be easily understood. There is an increasing range of visualisation considerations such as animations, time series data, 3D, graphs, multi-media, etc. Geostatistics. Much of the output from fisheries GIS depends upon the application of geostatistics to model various projections or distributions, e.g. estimating fish stocks or the relationship between fish species and environmental variables. Computer-Aided Design (CAD) and graphics. This represents an area having similar input/output requisites to GIS, and it has thus contributed significantly to its methodology.

4

GIS in Fisheries Management and Research

95

 Digital cartography. Whilst most cartography is not concerned with analysis per se, the output from digital cartography will share exact requirements to those from GIS.  Photogrammetry. This is the technique of measuring objects from photographs, electronic imagery, videos, satellite images, etc. These provide important sources of spatial information. If GIS can be successfully applied to the resolution of terrestrial problems, then why not applications in the marine domain? That many of the world’s commercial fisheries are in trouble and are facing resource depletion and economic decline needs little attention to readers of this book (Caddy et al. 1998; Christensen et al. 2003; Garcia and de Leiva Moreno 2001; Mullon et al. 2005; Pauly et al. 2002). These same readers will recognise that the problem has a geographic dimension in that the demise of the industry is widespread, and stock depletions have been recorded for seas surrounding all major continental areas (FAO 2004). They might also perceive of a temporal dimension to the problem in that, for some sought-after species, stocks have been threatened over a number of decades, or there have been cycles of high and low fisheries productivity causing alarms regarding supply continuity. There is now an awareness that so many stocks are presently threatened that a prevailing discussion in the fisheries literature concerns the closing of substantial marine areas to fishing by means of marine protected areas (MPAs) or similar No-Take Zones (NTZs) or marine reserves (Agardy 1994; Allison et al. 1998; Dugan and Davis 1993; Lubchenco et al. 2003) – a spatial solution to spatial problems! What is less well recognised is that the causes for the demise of fisheries are multi-faceted and complex. This is not the place to discuss these causes but frequent attention has been paid, often by scientists or conservationists, to harvesting technologies, over-fishing or poor management, and by fishers to factors such as climate change, abnormal conditions, pollution, poor scientific advice and predation by a range of top carnivores. What is important here is that all of these causes also have both geographic and temporal dimensions, all of which can be mapped in relevant space and time. Until comparatively recently it has not been a simple task to accomplish even the most basic of mapping of these ‘‘stock depletion causes’’, and there has certainly been no convenient method by which causative analyses could be successfully pursued. The advent of GIS has changed all of this.

4.2 GIS Software and Data for Fisheries Management and Research After specialist GIS software was first developed in the 1960s, there was an exponential growth in such system’s, until by the early 1990s when there were over 100 different GIS’s. Clearly the functionality of these was hugely

96

G. Meaden

varied, with many having been developed for specific operational areas or themes. The vast majority of GIS software was developed in North America, though there were a number of European initiatives. After peaking in the early 1990s the numbers of different GIS software packages began to rapidly decline. This occurred as certain companies were beginning to dominate the market through having produced GISs that could be adapted for many purposes. Clearly there were also mergers and many systems could not compete so they ceased production. Generally, there is now a situation where a few very dominant ‘‘players’’ are marketing worldwide multi-capable GIS, plus a large number of more specialist systems that are occupying niche market positions, plus a significant number of GISs that have been developed for particular institutions or market sectors such as defence, the utilities or navigation systems. This is not the place to go into detail on the actual functions or analyses that GIS’s might perform, but Box 4.2 provides a brief synopsis.

Box 4.2 Typical Range of GIS-Based Functions and Analyses Each GIS software product will support a range of GIS functions and analyses. Functions are data manipulations that ensure that the data can be modified to suit various purposes. Functions include such things as: aggregation, classification, editing, merging or integration, projection change, clipping, dissolving, structure conversion, data validation. At a more sophisticated level GIS software performs a range of analyses. Examples of these include:

 Buffering – defining zones of given dimensions around or along objects.  Overlaying and data integration – combining or joining varied thematic map layers for a given area or purpose.

 Network analyses – calculating connectivity or optimum routes along any network.

 Interpolation – locating the position of missing data points or lines on a surface or in a volume.

 Proximity analyses – establishing distances of objects relative to a theme or to other objects.

 Optimum location analyses – calculating the best location for a given activity or function.

 Digital elevation modelling – construction of 2.5D surfaces usually via the use of Triangulated Irregular Networks (TINs).

 Geostatistical analyses – the application of spatial statistics to create new or modelled data surfaces.

 Measurement – this includes simple length measurement as well as more complex areal or volumetric measurement.

 Contiguity analyses – determines the degree of relationship among neighbouring features across a surface, e.g. spatial autocorrelation.

4

GIS in Fisheries Management and Research

97

4.2.1 Marine or Fisheries GIS Software When compared to developments in the terrestrial GIS area, the situation is very different in the area of marine applications. Here, until the late 1980s there were no applications. Those that gradually then followed used proprietary GIS developed for terrestrial use, and this was the situation throughout the 1990s (Meaden 2000; 2001; Wright 2000; Valavanis 2002). At the same time a few workers were applying self developed systems that typically allowed for a limited range of marine/fishery GIS functionality (e.g. see Gill et al. 2001, or ‘‘Mike Marine GIS’’ developed by DHI Water & Environment in Denmark), or they were adapting niche software to marine problems, e.g. to show 2.5D representations of bathymetry, to perform interpolations of marine data, to handle satellite information, or to match marine and fisheries databases to mapping functionality. Recently Wood (2004) and Wood and Dunn (2004) have reported some interesting work in which an assortment of freely available tools such as operating systems, databases, analysis tools, mapping and traditional GIS tools visualisation software and office software have been ‘‘integrated’’ to fisheries data as a means of analysing and displaying this data for research and management purposes. The full sourcing of the various software is given in their references. A number of systems have also been developed that integrate fisheries electronic logbooks with GIS as a means of linking catches to spatial coordinates (Kemp and Meaden 2002; Barkai 2004; Mikol 2004). Clearly these systems allow the means of providing additional information through their ability to aggregate catches by species, time periods, or for vessels or groups of vessels in specific selected areas, and then matching these aggregations to other data as required. However, specific marine or fisheries related GIS has been slow to materialise, and it is not difficult to see why. Firstly, for any marine GIS to operate adequately it should have 3D functionality. Though 3D GIS exist they have largely been utilised in the sphere of geology where mapped structures are static, so these systems may not have the functionality to handle ‘‘movement’’. Secondly, why go to the trouble of developing a marine GIS when terrestrial systems are capable of doing the majority of the requisite functions? Thirdly, there is not presently the market for a commercial GIS product that can perform the desired range of functions. In other words the range of requirements for fisheries/marine research or management GIS is so large that any package developed may be presently unaffordable, certainly by most of the small, resource scarce research establishments that dominate the fisheries area. Fourthly, it is extremely difficult to create adequate output from a GIS when the system has to deal almost exclusively with data having high space/time variability. And finally, who would a marine or fisheries GIS be aimed at? The range of objectives of anyone researching or managing in this area is likely to be vast, so a multi-functional marine GIS might be of restricted utility.

98

G. Meaden

Notwithstanding these barriers to fisheries GIS software development, three advances are worthy of mention. Firstly, a major, multi-functional, marine fisheries software (Marine Explorer) has been produced by the Environmental Simulation Laboratory (ESL) in Saitama, Japan (see Itoh and Nishida 2001; Environmental Simulation Laboratory Inc. 2004, and www.esl.co.jp). This software has been developed over the past decade as a joint public/private sector project. The aims for this GIS are to integrate the most frequently needed functionality for fisheries and/or oceanographic analyses into one system, to produce a system that greatly reduces the time spent on spatio-temporal analyses, and to promote ‘‘the ecosystem safe sustainable fisheries management approach’’ (Itoh and Nishida 2001, p. 429). Marine Explorer can be used to input, store, manipulate and display a full range of fisheries or oceanographic data in both vertical and horizontal dimensions, with output being achieved in mapping, chart, or other graphical forms. The system has an integrated and simple spreadsheet format for data storage though it can be linked to external databases and to remotely sensed data such as satellite or acoustic data. Figure 4.1 illustrates the main menu items plus a typical dialogue box. Marine Explorer is likely to be the most advanced fisheries dedicated GIS presently available.

Fig. 4.1 Illustrative output from Marine Explorer

4

GIS in Fisheries Management and Research

99

A second recent advance is the work carried out on developing Arc Marine (or the ArcGIS Marine Data Model). To quote from the developer’s website (http://dusk.geo.orst.edu/djl/arcgis/ArcMarine_Tutorial/) ‘‘This is a geodatabase model tailored specifically for the marine GIS community. Created by researchers from Oregon State University, Duke University, NOAA, the Danish Hydrologic Institute and Environmental Systems Research Institute (ESRI), work on the data model began in 2001 in response to three major needs by the marine GIS community: (1) to provide an application-specific geodatabase structure for assembling, managing, storing and querying marine data in ArcGIS 8/9, (2) to provide a standardized geodatabase template upon which develop and maintain marine applications, and (3) to provide a better understanding of ESRI’s new geodatabase data structure.’’ The full functionality of the Marine Data Model is too detailed to be examined here, but for fisheries scientists wishing to use their data in a robust way, it is worth visiting and following the tutorial provided on the website. The third major advance is the recent release of Fishery Analyst. To quote the distributor (www.mappamondogis.it/products.htm) ‘‘This is an ArcGIS 9.1 extension developed to effectively analyze and visualize temporal and spatial patterns of fishery dynamics. The main functions are quantitative estimation and visualization of catch and effort and their variation in space and time, analysis of fishing vessel utilization, data quality control, and deriving information on the location of important economic and threatened species. The application provides a user friendly analysis interface allowing for easy and diverse output production. The interface allows the user to choose the analysis to perform (effort, catch density, catch per unit of effort, etc.) and to select data on criteria such as year, vessel name or size class, and fish species caught. Output can be generated as yearly, monthly, quarterly or user-defined date interval plots, and results can be plotted in pre-defined map layouts and saved in quantitative GIS data file formats (raster and vector) or as static maps.’’

4.2.2 Marine or Fisheries Data for GIS Although considerations of software are vital to successful GIS functionality and implementation, it could be argued that data considerations are equally if not more important. There are a number of reasons for this. For most users, considerations regarding software may be limited to a relatively small choice of packages, and software selection is typically made based upon system’s acquaintance and familiarity. Software purchase may be a one-off incident occurring at infrequent and well spaced intervals. However, for data the situation is very different. Costs may be extremely high with one data gathering marine survey perhaps costing an order of magnitude more than the GIS software cost. Then data needs to be constantly updated and invariably many data sets will be required even for a single project. Requisite data may already be in

100

G. Meaden

existence but it can be time consuming to establish this. And there are numerous other data considerations such as metadata existence, data standards, data formatting and structure, data verification and editing, copyright issues, etc. In fact, in this author’s experience, such are the considerations regarding data that it is often the case that the GIS project is designed around data availability – a poor reason indeed for most research projects! A further factor accentuating the importance of data is that, as has been indicated, GIS fisheries research is usually carried out by small scale institutions or organisations, and they typically have relative little funding to purchase what is relatively expensive data. It is likely that most fisheries research will utilize at least some primary data. For fisheries work this will typically, though not always, be gathered at sea. Useful terrestrial derived, primary fisheries data may be obtained from sources such as fish landings or sales data, data concerned with vessel registrations and environmental, social or economic data available from the fishers themselves (Neis et al. 1999). Although there are still data that may be gathered using preprinted (hard-copy) survey sheets, almost all primary data would now be gathered via the use of a wide range of digital equipment. Space prohibits a detailed listing of these, but it is easy to envisage that this equipment encompasses a hierarchy varying from simple to complex, or inexpensive to costly. At the lower end there is a range of perhaps measuring meters, e.g. CTD loggers, that for individual points will record temperature, salinity, depth, or other water quality parameters. These may be located aboard fixed or floating buoys, or they may be deployed on survey vessels to take readings along some pre-arranged strategic sampling route. There is then a range of middle cost equipment such as electronic log-books, positioning systems (including VMS and electronic charting systems), acoustic sonar, etc., and at the sophisticated end of the data gathering hierarchy there are systems such as airborne or satellite-based remote sensing systems, or underwater autonomous vehicles that will travel large distances underwater often following pre-programme routes to collect pre-programmed data. In fisheries research it is difficult to estimate the balance between primary data inputs and data being obtained from secondary sources, and in any case this would vary greatly from project to project. Secondary data will nearly always supply the basic mapping outlines, be they hydrographic, thematic, topographic, etc., and most of this mapping is now obtainable in digital format – even if it is sometimes expensive to purchase! For specific tasks it might still be necessary for the researcher to digitise his/her own map outlines, but for smaller areas this can usually be quickly accomplished using scanning and on-screen digitising facilities. Other secondary data inputs will consist largely of tabular data, collected and then stored on spreadsheets or on databases. Clearly researchers are likely to have access to substantial quantities of their own data covering any of a range of thematic areas. But, increasingly marine and fisheries data is obtainable via the Internet from a seemingly infinite variety of sources. Given the efficiency of search engines we need not exemplify these here, though

4

GIS in Fisheries Management and Research

101

additional information can be found from Meaden (2004) and especially from Valavanis (2002). Managers or researchers seeking secondary digital data should be wary of what they may be using because this author has to admit that he has yet to find a single piece of adequate secondary tabular data. The reasons for this are concerned with factors such as levels of aggregation, sampling techniques used, the lack of adequate metadata, doubtful georeferencing, copyright problems, costs problems, and simply locating the requisite data. It is of interest to note that attempts are being made to improve this situation through a combination of more data sources, through the growth of specialist data sources, e.g. sources that cover particular themes or geographic areas, through price reductions and through data sharing.

4.3 Main Centres of Fisheries GIS Development Activity and Support 4.3.1 Centres of Activity for Fisheries Related GIS Development It would be true to say that the use of GIS in fisheries research or management is a relatively low key activity in the sense that there are few large institutions where Centres of Excellence in this subject have been fostered and developed. But having said this it is also true that this work is very widespread and the volume of work being carried on in terms of the numbers of institutions and individuals involved must be quite considerable. Thus, because the subject area of ‘‘fisheries management or research’’ is an activity that is carried out in a large number of often isolated and small scale institutions, then it follows that GIS use exhibits this trend. This is especially the case since GIS nowadays lends itself quite easily to individual applications. In a sense this is both desirable and undesirable. The former because it means that a large human intellectual resource is being directed here, an input that should accrue much valuable output, but the latter because undoubtedly there will be much wasteful repetition. In Box 4.3 we provide some illustrations of institutions where the use of GIS for fisheries management or research is being pursued. Box 4.3 Examples of Institutions where Fisheries related GIS-Based Activities are Being Pursued Clearly this box can only be illustrative and it is still impossible to give any indication of the exact range or number of institutions carrying out fisheriesbased GIS work:

 Fisheries GIS Unit, Canterbury Christ Church University, Canterbury, UK. This is a small-scale specialist academic unit undertaking applied research and consultancy work.  IFREMER (France). This French government research agency has a large number of GIS initiatives, many of them being highly innovative.

102

G. Meaden

 University of British Columbia (Fisheries Centre). Arguably the premier    

   

academic institution for fisheries research, GIS is integrated into much of their output. Many projects integrate Ecopath and Ecosims modelling. Environmental Simulation Laboratory (ESL). This Japanese private company has worked with government fisheries research institutes to produce an advanced marine fisheries GIS (see Section 4.2.1). University of Aberdeen. In the university’s Zoology Department there is a small but thriving fisheries research team who have made innovative use of GIS on a range of projects. National Marine Fisheries Service of NOAA (USA). The US government sponsors a large number of research initiatives that utilise GIS, e.g. the mandate to develop Essential Fish Habitats. The large scale VIBES (Viability of exploited pelagics in the Benguela Ecosystem) uses GIS to research and manage this extremely productive marine area off the SW African coast. It is largely a combined French government and South African universities integrated project. Hellenic Centre for Marine Research, Greece. Amongst many fishery related activities, they have made significant advances in GIS, mainly from their Marine GIS laboratory in Heraklion, Crete. University of Miami. Here the Fisheries Ecosystems Modelling and Assessment Research Group are doing very advanced GIS-based fishery and ecosystems related research work. CEFAS, UK. This is the government fisheries research agency in the UK having a specialist GIS office at their Lowestoft headquarters. Food & Agricultural Organisation of the UN (FAO). Supportive work being undertaken at their Rome headquarters, and a number of international projects incorporate GIS as a means of better developing fisheries in some less developed areas. GISFish and COPEMED are examples of their supportive work.

As Box 4.3 illustrates, GIS-based fisheries applications take place in a wide range of institutions. In universities most of the applications are not ‘‘blue-sky’’ research in the sense of developing GIS software for the use in the spatiotemporal management or study of fisheries per se. Instead it is nearly all concerned with applying proprietary GIS in contracted projects usually associated with studying a particular fishery or aspects of a fisheries area. The research institutions would usually be carrying out similar types of projects. The national or international agencies are usually concerned with specific problems or problem areas, or they may wish to initiate particular projects associated with perhaps new laws or with helping particular sectors of society. The institutions listed in Box 4.3 may all employ a number of GIS-related workers, but this may only be in the range of perhaps 3–10 (varying with project demands). But there would be perhaps several hundred institutions world-wide

4

GIS in Fisheries Management and Research

103

where some fisheries GIS-based work was in progress, and the vast majority of these would be employing from 1 to 3 people having GIS capabilities. In many instances, the stage has now been reached in GIS use such that GIS is just another software tool and, unlike say five years ago, the term ‘‘GIS’’ does not appear in the list of ‘‘Keywords’’ near the start of publications, including academic papers.

4.3.2 Guidance and Support for Fisheries Related GIS Activity For those working with GIS for fisheries management or research the need for guidance and support will be ever present. This need largely results from the fact that changes in the ITC arena occur at a rapid rate, but it must also be remembered that anyone working in this field is, in many senses, also working in conjunction with some or all of those areas listed in Box 4.1. It takes a particularly skilled operative to pursue work in fisheries GIS! To make the situation easier there are many means of acquiring guidance and support. Valavanis (2002) lists some of the consortia and organisations that promote and sustain GIS work on an inter-disciplinary basis. Table 4.1 lists a small selection of such organisations. As well as the more general support listed in Table 4.1, there are various sources of guidance and information that are more specifically related to fisheries GIS. These take the form of conferences, books, and academic papers, plus a number of dedicated centres, and delivery might be both hard copy and digital. Room prohibits more than a brief mention of a few important sources. Conferences in the field of fisheries GIS did not commence until 1999. That year, in Seattle, the First International Symposium on GIS in Fishery Science was held, and this was the first in what is now a triennial event (Nishida et al. 2001). The second symposium took place in Brighton, Sussex, UK in 2002, the third in Shanghai, China in 2005 and the most recent occurred in Rio de Janeiro, Brazil in 2008. Further details of these are available from www.esl.co.jp/Sympo/outline.htm. Also in 1999 a similar event was held in Anchorage, i.e. the 17th Lowell Wakefield Fisheries Symposium on Spatial Processes and Management of Fish Populations. Since then there have been a number of Symposia especially devoted to Fisheries GIS, but most of these have been ‘‘low-key’’ events. It is important to mention this because the fact that GIS is now more simply a tool for fisheries science, management and research means that dedicated GIS-based events are seldom organised. The leading books on fisheries related GIS are those written or edited by Meaden and Kapetsky (1991); Meaden and Do Chi (1996); Valavanis (2002) and Fisher and Rahel (2004). There are also a number of books that have individual chapters devoted to fisheries GIS, and a few books that are concerned with similar themes (e.g. Kruse et al. 2001). It would be pointless to list

104

G. Meaden

Table 4.1 Some Major sources of GIS/fisheries advice and guidance (adapted and updated from Valavanis 2002) Name Information Website Association for Geographic Information (AGI) Association of Geographic Information Laboratories in Europe (AGILE) Centre for International Earth Science Information Network (CIESIN) Consortium for Ocean Leadership

California Department of Fish & Game

ESRI Conservation Programme – Marine and Coastal

Network of Marine Research Institutions and Documents (MareNet) European Platform for Coastal Research Coordination Action (ENCORA) Davey Jones Locker

FAO – Fisheries Information Centre FAO – Fisheries Global Information System

Supplies a wide range of support in the private and public sectors Promotes teaching and research in GeoSciences within Europe Promotes information to better understand the changing world An association of 66 US institutions representing the core of marine research and education Assists in the collection, documentation, and analysis of spatial data for effective conservation decision making A site of interest for mapping/ GIS, scholarly papers and ESRI Conference Proceedings relating to marine spheres An Internet network enabling marine scientists to communicate effectively worldwide A European platform for sharing knowledge and experience in coastal science policy and practice A portal for advice on seafloor mapping, marine and coastal GIS Provides direct access to a range of fisheries related data centres A network of integrated fisheries information

www.agi.org.uk

http://www.agile-online.org/

www.ciesin.org

http://oceanleadership.org/

www.dfg.ca.gov/biogeodata/ gis/imaps_about.asp

www.conservationgis.org/ links/marine.html

http://marenet.unioldenburg.de/MareNet

www.encora.org

http://seafloormapping.net/

www.fao.org/fishery/topic/ 2017/en www.fao.org/fishery/topic/ 3456/en

even a sample of the academic papers that have mentioned or discussed the use of GIS in fisheries research, and there are a variety of retrieval systems in place for locating most of these. The principal institution that is taking a leading role in disseminating information on the use of GIS for fisheries related activities (management, research, modelling, etc.) is the Food & Agriculture Organization of the UN (FAO). Since first becoming involved in fisheries GIS work

4

GIS in Fisheries Management and Research

105

nearly two decades ago, the FAO has made huge efforts to raise the profile of this tool. Much of their consultancy work has been to strive to put in better food production management systems and they have seen GIS as a vital aid here. The FAO have sponsored numerous GIS related workshops and publications, and have recently produced a specific fisheries GIS training and technical manual (de Graff et al. 2003) that successfully integrates fisheries GIS applications with GIS instruction based on the ubiquitous ArcView (by ESRI) software product. They are now in the process of setting up GISFish, which will be a dedicated ‘‘one-shop’’ Internet portal to fisheries GIS in all its aspects (Kapetsky and Aguilar-Manjarrez 2005). Other national and international institutions e.g. government fisheries institutions, are using GIS in fisheries research though they may not be actively promoting its use and worth.

4.4 Exemplars of GIS Applications for Fisheries Research Because there have now been so many examples of the use of GIS for fisheries research (and management), it is difficult to exemplify even a small range of the possibilities. However, this chapter would not be complete without an attempt to illustrate some typical uses of GIS in the fisheries research domain and to show a small selection of the visual output that has been recently obtained. The selection of case studies has been made on the basis of variety in terms of types of fishery, quality of visualisation and specific GIS functionality. Box 4.4 illustrates a number of main thematic areas in which GIS has been adopted and utilised. Box 4.4. Typical Uses of GIS in the Fisheries Domain There are a large number of ways in which GIS might be utilised to assist in fisheries research, and some of these include:

 Distribution displays – this is simply cartographic visualisation to show    

the distribution of any feature or combination of marine or fisheries features. Marine Habitat mapping and analysis, e.g. the work of Rubec et al. (1998) in establishing ‘‘essential fish habitats’’ in Florida, plus other work on fisheries oceanography. Resource analyses – to quantify and display the disposition of any marine resource or combination of resources. Modelling, i.e. these functions include work on illustrating themes, usually in a simplistic or general way, or there may be predictive modelling to show the outcome of potential decisions or actions. Monitoring management policies, e.g. optimising the disposition of fishing effort, perhaps via the help of electronic log-book or VMS tracking data.

106

G. Meaden

 Ecosystems relationships, e.g. predator/prey relationships, or relationships between fish distributions and any environmental parameter.

 Stock enhancement, e.g. the timing and selection of sites for artificial stocking, or the optimal siting for mariculture activities.

 Marine Reserve allocation, i.e. both identifying suitable areas for species protection and analysing the results achieved by these areas. This work is complex given the wide number of interested parties (often with conflicting ideals) plus the variety of spatial considerations involved.  The creation of economic surfaces, i.e. allowing researchers to model the likely income derived from fishery products based on alternative management and resource extraction scenarios.  Fishing fleet disposition and behaviour, i.e. to best sustain fish yields, vessels need to be optimally deployed throughout a management or ecosystem’s area.

More details on GIS usage in fisheries research, plus many specific case studies, can be obtained from the reading sources noted in Section 4.3.2 above plus the various Proceedings of the International Symposia on GIS in Fishery Science. More specific details for five case studies are provided below.

4.4.1 Developing Fish Habitat Models for the Eastern English Channel Eastwood and Meaden (2004) describe how GIS is used in their efforts to identify suitable habitats for sole (Solea solea) in the eastern English Channel. Figure 4.2 illustrates the stages involved in this modelling exercise. Fisheries survey data is used to provide quantitative numerical information within the survey area, and clearly this information could relate to different life stages for any species. Environmental data is also obtained for all major variables considered to influence sole habitat preferences. Development of the model is based on quantile regression estimates of the upper bounds of sole catch densities, this yielding the most comprehensive information on the likely habitat carrying capacity. The regression quantile parameter estimates are used to recode the environmental data (in raster format) to produce maps for each environmental parameter. The final habitat suitability map was calculated as the geometric mean of spatially coincident raster cells for the various environmental input layers. Note that the survey date frequently only covers one temporal period (e.g. month or season), and thus to produce a temporally holistic habitat suitability map year round survey data would be required. This might be important in the context of

4

GIS in Fisheries Management and Research

107

Fig. 4.2 Modelling procedure adopted to identify spawning habitat for Solea solea in the eastern English Channel

designating conservation areas. A more complete explanation of this work is given in Eastwood et al. (2001).

4.4.2 Mapping Fishing Effort in the Portuguese Crustacean Trawl Fishery Here the use is described (Afonso-Diaz et al. 2006) of GeoCrust 2.0, a dedicated GIS developed by the University of Algarve specifically for use in the Portuguese fisheries, and which allows for the storage, analysis and display of data obtained through Vessel Monitoring Systems (VMS). Figure 4.3 depicts fishing effort for the south-western Portuguese crustacean trawl fishery. Fishing effort data is obtained from GPS data provided by the Portuguese national VMS. Thus for each vessel, a geo-referenced point location reading is taken every 10 min. These point locations for the whole fleet for 1 year (2003) have been aggregated and mapped into 0.2  0.2 nautical mile cells. It can be seen that the number of times individual cells have been trawled varies from zero to >115. In some cases individual trawling paths are discernable, and the relationship between trawl depth and trawl frequency is easily established. In order to visualise the distribution of catches, the trawled areas have been divided into 7 specific zones for which landing data is recorded. If required the proportion of different catch species for each zone, and the catch rate per

108

G. Meaden

Fig. 4.3 Fishing effort in the SW Portuguese crustacean trawl fishery during 2003 as registered by VMS location data

hour fished, can be shown as proportional circles superimposed on a base map of the seven main fishing zones. Further details are provided in Afonso-Diaz et al. (2004).

4.4.3 Estimates of Catch Abundance for Emperor (Lethrinus) in Omani Waters Lacking a wide range of resources, including suitable soil and climatic conditions for agriculture, the Gulf state of Oman is heavily reliant on fisheries as a source of employment, local incomes and food. Figure 4.4 shows the total catch (traditional plus commercial) and distribution for emperor (Lethrinus) for 1996–2004 (Al-Kharusi 2006). When catch is related to bathymetry it is clear that the majority of the catch is made in coastal shelf waters (>200 m) in the south east. This is mainly a consequence of the Somalia upwelling that brings nutrients to these waters during most of the year but particularly during the summer southwest monsoon. The author has used GIS to determine and plot maps showing temporal sequences for various species. These maps relate either catch or CPUE to variables such as sea surface temperature or bathymetry, and thus show the seasonal spatial ‘‘evolution’’ of the catches.

4

GIS in Fisheries Management and Research

109

Fig. 4.4 Spatial distribution of total emperor catch (kg) in Omani waters – 1996–2004

4.4.4 Preliminary Estimates of the Sea Cucumber Abundance Around Ilha Grande, Rio de Janeiro, Brazil This case study concentrates on the very local scale, one that may typify a subsistence fishery activity. The authors (Miceli and Scott 2005) have sought to establish a preliminary estimate of the abundance of Isostichopus badionotus (sea cucumber) around the Ilha Grande near Rio de Janeiro, Brazil. It is important to do this because the animal is being illegally exploited to service a high demand from Far East Asian markets. I. badionotus is now on a Brazilian endangered species list.

110

G. Meaden

If its exploitation could be regularised and stocks could be properly managed then there is the potential to supply a good living for large numbers of local subsistence fishers. In their study the authors used a low cost, raster-based GIS, ‘‘IDRISI Kilimanjaro’’ (supplied by Clark University in Massachusetts, USA). They identified the controlling parameters to local sea cucumber production as:

    

Living in coastal waters of <20 m depth, Being within 300 m of the shoreline, Living in dominantly rocky coastal areas, Preferring areas protected from strong SW winds, Being homogenously distributed in suitable areas.

Figure 4.5a shows the bathymetry for the area around Ilha Grande, and Fig. 4.5b shows the areas protected from SW winds (in green) and those areas within 300 m of the shore being less than 20 m in depth (in light blue). This

Fig. 4.5 (a) Bathymetry around Ilha Grande, Rio de Janeiro, Brazil; and (b) area of Ilha Grande protected from the SW wind and optimum habitats for sea cucumber exploitation

4

GIS in Fisheries Management and Research

111

exploitation area is about 1630 ha, and with an assumed typical sea cucumber of one animal per 10 m2 this area could yield 1,630,000 individuals.

4.4.5 Quantitative Interactions Between Sardine and Anchovy in South African Waters In this case study Drapeau et al. (2004) ‘‘describe, represent and quantify some spatio-temporal interactions between several pairs of species using basic indices derived from the distribution maps of 13 key species, drawn from various and heterogeneous sources of information (commercial and scientific data)’’ (p. 142). Figure 4.6 shows the interaction between sardine and anchovy. To acquire the information for this figure six different sources of data were combined on a 10’10’ cell grid in a GIS (ArcView 3.2), and the resultant relative biomass of each species was then mapped using five abundance classifications. In this case the species interaction exemplified is one where the two species are competing for planktonic resources, i.e. there is not a predator/prey relationship. It is clear that each species in isolation (where there is no interaction) occupies a similar total area, but these areas have significantly different distributions. The overlap in species is mainly restricted to a coastal band from Hondeklip Bay to about 100 km east of Port Elizabeth. In this case study, GIS use is clearly forming a basis to further ecosystems observations and management.

Fig. 4.6 Degree of interaction between anchovy and sardine stocks in coastal waters around South Africa

4.5 Challenges Facing the Use of GIS for Fisheries Management or Research Since a recent publication (Meaden 2004), has set out in some detail the main challenges facing the use of GIS in fisheries related fields, here it will be appropriate to simply highlight a limited number of these. It is pertinent to

112

G. Meaden

mention that challenges for those working in this field cover not only factors directly related to the optimum functioning of GIS as a useful tool, but also to areas that are strongly associated with this working environment, e.g. adequate funding, training, modelling, database management, digital structures, etc. but space forbids a look at these. It is also pertinent to note that an important reason for examining challenges is that it is often scientists, or those doing fisheries research, who are made scapegoats when things go wrong in specific fisheries. ‘‘It is important that scientists demonstrate that they can supply sensible answers to the panoply of problems, and that the burden of responsibility (for the plight of fisheries) can be shifted to users and politicians where it should rest’’ (Meaden 2004; p. 14). A final introductory remark is that, although the challenges have been compartmentalised here, in reality many of them will be interlinked.

4.5.1 Intellectual and Theoretical Challenges Most GIS-based work in the terrestrial domain concentrates on the mapping and analysis of objects that are fixed in space, e.g. roads, buildings, forests, crime scenes, retail locations, etc. Obviously there may be temporal changes in the location of mapped features, but these are typically one-off, occasional movements. The marine milieu could hardly be more different. Here almost everything moves. This refers not only to the free-ranging objects that are being mapped, e.g. fish schools, fishing vessels, lobsters, whales, etc., but also to the marine milieu itself. Features related to this such as water temperature, salinity, turbidity, phytoplankton, etc. will all be in continual motion. Whilst some of these motions may be relatively predictable e.g. major ocean currents, others are highly erratic or chaotic and thus unpredictable, e.g. a plankton bloom or fish foraging movements. So, a challenge for the GI analyst is – How best to map moving objects? Clearly this is a challenge that has to be overcome because at the root of fisheries GIS work lies the need to know the spatial disposition of marine resources. There is now a considerable body of work being deployed into factors such as animal movements and fish migrations, and movement models are being developed. Additionally, GIS animations are being deployed so as to best analyse movements through the spatio-temporal domain. A second intellectual challenge relates to scale and resolution. Clearly fisheries science is concerned with processes operating at scales varying from the relatively micro to the very macro, and in both the time and space dimensions. This leads to challenges such as – What is the optimum scale or scales at which to be working? How can a single study integrate different processes occurring at multiple time and space scales? What space/time gap do I leave between data gathering points? It is important to be working at the correct scale because many distributional patterns can only be discerned at this appropriate scale. Estimating such a scale is

4

GIS in Fisheries Management and Research

113

made more difficult because water body movements (speed and direction) will rarely coincide with movements made by faunal species. Resolution is intimately associated with scale because it is a decision as to what measurement interval to use when gathering data. St. Martin (2004) offers some useful comments on the relationship between data collection and scale explaining in particular the range of complexities involved. The third intellectual challenge is posed by the fact that nearly all fisheries related GIS work takes place in 3 or 4 dimensions. This contrasts with terrestrial GIS which only has to consider 2 or 2.5 dimensions (the latter being associated with the fact that ground-based phenomena also have altitudinal attributes). Arguably it is the need to be working in at least 3D that has been the greatest barrier to the emergence of full marine GIS software. However, much work is going on in this sphere and slowly systems are emerging, e.g. Varma (2000). It is not only GIS that must function in these added dimensions, but of course there have to exist database structures, functions and storage capabilities that are designed to work efficiently in 3D. A fourth intellectual challenge is that associated with applying spatial statistics and modelling in such a way that these procedures can best be integrated to GIS. What is being considered here is the use of GIS as a software platform or activity surface on which numerical models, usually in the form of equations, may be conceived, evaluated or tested. For example, consider the spatial distribution of several species through a temperature gradient in water. Through this gradient it should be possible to establish an equation that best describes (models) this distribution. Of course, this is an extremely simple example; in real life the species will also be reacting to other distributions such as adjacency of predators, bottom sediment type, salinity levels, etc., all of which might need to be integrated to establish a holistic model. At the present time it is likely that most marine spatial modelling is performed via the use of specialist modelling software that can be integrated to GIS in some way, and some authors have noted that it could be sensible to stick with a ‘‘spatial analysis toolkit’’ as GIS-independent software having the ability to be integrated when required (Openshaw and Clarke 1996). Box 4.5 sets out some of the advantages of modelling within a GIS environment. Box 4.5 Advantages of Modelling within a GIS Environment The following (adapted from Meaden 2004) is an illustrative listing of some of the major modelling advantages that can be obtained whilst using GIS in the marine fisheries sphere:

 The raster data structure provides an ideal platform for various spatial procedures.

 Most GIS have built-in, or self selecting, formulas that can be used.  It is easy to add extra variables that might influence or improve the model.  A range of types of weightings can be added.

114

   

G. Meaden

Temporal iterations can be accomplished for dynamic modelling. Most GIS can be coupled to external modelling or statistical software. GIS is ideal for exploratory data analyses. GIS can easily accommodate scale, time and area changes.

A final intellectual challenge is that of optimising visualisation. Whilst being accurate, it is essential that output from the GIS is easily comprehensible to as wide an audience as possible. Unfortunately for the cartographer, humans all have variable perceptual preferences which clearly means that what may be easily comprehensible to some will not be so to others. To optimise visualisation as far as possible it is necessary to consider a wide range of factors associated with cartography, including font size, style and placement, colour combinations, quantitative classifications, data representation (symbology), etc. Additionally, many ‘‘map objects’’ may be classified as ‘‘fuzzy’’, meaning that they may be imprecise, or that different individuals may not translate terms in the same way. For instance, what is the continental shelf, or where exactly is the Gulf Stream? The ‘‘intertidal zone’’ will differ from month to month. Marine biomes will be interpreted differently by different authors, and the marine area is notorious for having indeterminate boundaries. A further challenge to visualisation is that of the meaningful mapping of extremely complex surfaces, often in 3D. Think here of the mapping of benthic communities or the large scale mapping of a coral reef. And consider the difficulties of mapping species ranges. These may be patchy, exhibit varying densities, will vary seasonally and all vary in the third dimension.

4.5.2 Practical and Organisational Challenges For those applying GIS to fisheries research or management one of the most pressing needs is for data, and this may be expensive, scarce and difficult to locate. If data can be located then invariably it may not cover the exact spatial area or time period needed; it may have been gathered at an incorrect resolution; it may be stored in an inconvenient structure; it may lack adequate metadata information; and it may have been gathered for a different purpose and thus have not exactly the desired parameters. If the desired data cannot be found then it has to be obtained as primary data. This typically involves large cost considerations and it may require specialised equipment, vessel charter time scheduling, reliance on weather conditions, etc. There is then the problem of statistical significance of sampled data. In a 3 or 4D moving marine milieu, how can any researcher estimate how significant his data is, or how large a sample might be required? Despite the huge increases in the availability of online data, challenges to data gathering will persist mainly because projects are likely to get ever more specialised hence requiring greater levels of detail and

4

GIS in Fisheries Management and Research

115

accuracy. One major way in which access to data can be better facilitated will be through the advent of enhanced interoperability, allowing for disparate data and computer systems to be linked for the purposes of data sharing. Considerable progress is being made on this front through the creation of OpenGIS via the OpenGIS Consortium. The advantages of this development will be greatly manifest in the widely scattered and fragmented fisheries research sectors. The problem of ‘‘subject organisation’’ is a further challenge to the use of GIS in the fisheries sphere. ‘‘Fisheries research’’ per se may be carried out in conjunction with a host of other subject areas, e.g. oceanography, marine ecosystems, climatology, biology as well as various areas of technology. And since the research itself is typically carried out in isolated and small scale establishments, then fragmentation will undoubtedly be manifest in the fact that the researcher will not have access to all requisite resources. It is likely that the computing, or associated hardware and software, will be inadequate in some aspect, and access to the appropriate guidance and support may not be easy. Attempts at overcoming subject fragmentation are emerging through workshops, conferences, the Internet, etc., but this author is aware that there is a large amount of duplication or ‘‘reinventing the wheel’’ still going on. A final practical challenge is that concerned with delivery of the information output. Although it might be relatively easy for the researcher to pass GIS-based information to relevant personnel within his organisation, actually reaching a wider audience can be very challenging. Though we have mentioned the symposia and some of the support and guidance media that are emerging, it is true to say that the vast majority of literature about GIS functionality is still to be seen in the relatively inaccessible grey literature. Although many papers appearing in better known fisheries journals may be reporting research that has utilised GIS, this is seldom highlighted, so the ‘‘methods’’ section of papers invariably say little on how a GIS has been used. The Internet is likely to slowly have an impact on disseminating GISbased fisheries information, and this is likely to be through providing easier access to data inputs as well as delivering the outputs from GIS-based research.

4.5.3 Economic, Social and Cultural Challenges Over the last two decades the world of computing has undergone a major transition in terms of cost allocation. Whereas computers plus their software, were major items of expenditure in the 1980s, relatively speaking this is no longer the case. It is now recognised that in the GIS world data costs account for approximately 80% of all systems costs, though in fisheries related GIS work these costs are likely to be an even higher proportion. This balance has

116

G. Meaden

shifted because whereas computers and software can be mass produced, this cannot be said for most data. Additionally, data requisites have become much more stringent in terms of detail, quantity and quality. It is likely that the high cost of data has been a major reason for the relatively slow growth of fisheries related GIS, and this is especially the case in most developing countries. Here the costs of GIS are still extremely difficult to justify, partly because cost:benefit analyses are difficult to perform and thus the advantages are not easily demonstrated. Also, developing countries usually pay ‘‘western’’ prices for their GIS-related needs, and this may include not only computing equipment and data but also training and other support. Most fisheries related GIS work in developing areas must rely almost completely on donor support, though there are now some areas where local expertise is accruing, e.g. Sri Lanka, Philippines and some Persian Gulf states. Social and cultural challenges to the adoption of GIS in fisheries research relate to the working ambience or to institutional norms and practices, factors that largely relate to the degree of development or to the dominant culture. What this means is that in an area such as fisheries, with its relative isolation and small scale of operation, many institutions will lack a ‘‘champion’’ for GIS. As Campbell and Masser (1993) demonstrated some time ago, without someone in an institution who is really willing to push GIS and to nurture it carefully, then GIS has little chance of being successfully adopted. Other challenges in this area are that there is frequently a lack of appreciation that problems in the fishery domain are rooted in spatial differentiation, and many managers or researchers do not appreciate the importance of the geographic perspective. Data from fishers may be hard to obtain because there is a deep distrust over what scientists or politicians will do with the data. Many countries have found it extremely difficult to implement data gathering systems such as fisheries log-books, yet these have the potential to be a valuable source of research data. Interestingly CEFAS in the UK, the main fisheries research institution, has recently implemented a scheme whereby they hire commercial fishing boats on short term contracts, in order to both gather data and to show the fishers that they are working in sympathy with them. This will hopefully engender better attitudes to ‘‘scientific management’’. Some final challenges in this area arise from a poor appreciation among international institutions and donor organisations as to the realities of working with sophisticated IT systems in developing country situations, i.e. where the technology has little real relevance to the cultural setting. And, in both developed and developing countries, researchers may lack ‘‘geographical cognition’’. This is an innate appreciation of geographical (spatial) relationships, and thus an inability to recognise geographic patterns or surface trends in terms of adjacency, ubiquity, contiguity and heterogeneity. These skills are vital for the success of GIS.

4

GIS in Fisheries Management and Research

117

4.6 Conclusions and Future In the decade since the first edition of this book, advances in GIS methods and applications have been unprecedented. A perusal of the equivalent chapter in this first edition revealed that it was essential to devote a third of the chapter simply to describing what GIS was and how it functioned (Meaden 1996). There is now no longer a need for this – GIS has become virtually ubiquitous in the world of science and research, if not yet in the world of fisheries or its management. It is also worth noting that almost all uses of GIS in the fisheries area, as described in the first edition, were related to more-or-less static mapping, e.g. creation of nautical charts, coastal zone management, optimum locations for aquaculture, etc., and it was only in an examination of the potential uses for GIS was mention made of anything more complex. Nowadays, there is almost no facet of the use of GIS for spatial analyses that cannot be attempted. GIS is now in a situation to ‘‘provide robust capabilities to even the most novice of users’’ (Battista and Monaco 2004; p. 202). This has greatly stimulated its use. There is also a greater realization of its potential and this itself increases as GIS improves its analytical functional range. The whole of the GIS operational infrastructure is now being well supported through investments in back-up, guidance, support and training. Internet data portals have greatly facilitated research by providing access to hugely more varied data sources. Battista and Monaco (2004) further surmise that the future of GIS in marine associated work will be determined by the synthesis of three interdependent factors: software enhancement, data access and user-derived statistical, spectral and spatial data exploratory techniques. This synthesis ‘‘will ensure that GIS continues to enable researchers to readily visualise complex spatial relationships among species, assemblages, habitats, and physical regimes of marine environments’’ (p. 202). Progress in GIS functionality will be further enhanced by developments in parallel technologies. Foremost amongst these will be data supply via improvements in remote sensing resolution and image delivery systems, plus advances in acoustic underwater data collection, especially allowing for the rapid in situ mapping of sub-surface topography and habitat classes. Improved databases will greatly enhance data querying capabilities, and data usage will grow as standardisation and interoperability among different systems is enhanced. A great spur to the further applications of GIS in fisheries management and research will come from the need for ecosystems, rather than single species, management (St. Martin 2004). This author highlights the extremely complex spatial disposition of the interlinked parts that make up any marine ecosystem. He also notes that fisheries themselves are moving away from management systems whereby actions (fishing decisions) are either based on the decisions of individuals or decisions made by remote fishery authorities (such as those made by the EC’s Common Fisheries Policy). In the future decisions are likely to be made at local community level. St. Martin (2004) further concludes that the

118

G. Meaden

problems associated with fisheries and their management need to be recast in spatial rather than numeric terms, e.g. instead of quantitative catch quotas being allocated, fishers are likely to have their fishing effort allocated to a defined spatial area. This changing management scenario will require the computational efficiency of GIS to manage the complexities of the marine and management space. With the demise in fish stocks seemingly proceeding unchecked, and with the new found appreciation of the importance of spatialbased decision making, then these are important times for the use of GIS in fisheries management and research.

References Afonso-Diaz M, Simoes J and Pinto C (2004) A dedicated GIS to estimate and map fishing effort and landings for the Portuguese crustacean trawl fleet. In Nishida T, Lailola PJ and Hollingworth CE (eds) GIS/Spatial Analyses in Fishery and Aquatic Sciences (vol 2), Fishery-Aquatic GIS Research Group, Saitama, Japan, pp 323–340 Afonso-Diaz M, Pinto C and Simoes J (2006) GeoCrust 2.0 – A computer application for monitoring the Portuguese crustacean trawl fishery using VMS, landings and logbooks data. Poster presented at ICES Annual Science Conference, ICES CM 2006/ N:19. 19–23 September, 2006. Maastricht, Netherlands Agardy T (1994) Advances in marine conservation: the role of marine protected areas. Trends in Ecology and Evolution 9: 267–270 Allison GW, Lubchenco J and Carr MH (1998) Marine reserves are necessary but not sufficient for marine conservation. Ecological Applications 8: S79–S92 Al-Kharusi L (2006) Analysis of space and time variation of emperor (Lethrinus) in Omani waters. Unpublished MSc thesis, Department of Geography, University of Leicester, UK Barkai A (2004) An electronic fishery data management system: A demonstration of a unique, wheelhouse, software solution for the collection, management and utilization of commercial fishing data. In Nishida T, Lailola PJ and Hollingworth CE (eds) GIS/Spatial Analyses in Fishery and Aquatic Sciences (vol 2), Fishery-Aquatic GIS Research Group, Saitama, Japan, pp 599–606 Battista TA and Monaco ME (2004) Geographic information systems applications in coastal marine fisheries. In Fisher WL and Rahel FJ (eds) Geographic Information Systems in Fisheries. American Fisheries Society, Bethesda, USA, pp 189–208 Caddy JF, Carocci F and Coppola S (1998) Have peak fishery production levels been passed in continental shelf areas? Some perspectives arising from historical trends in production per shelf area. Journal of Northwest Atlantic Fisheries Science 23: 191–219 Campbell H and Masser I (1993) Implementing GIS: The organisational dimension. Association for Geographic Information. Conference Papers for AGI93, Birmingham, UK Christensen V, Guenette S, Heymans JJ, Walters CJ, Watson R, Zeller D and Pauly D (2003) Hundred-year decline of North Atlantic predatory fishes. Fish and Fisheries 4: 1–24 de Graff G, Marttin F, Aguilar-Manjarrez J and Jenness J (2003) Geographic information systems in fisheries management and planning. FAO Fisheries Technical Paper No 449. FAO, Rome, Italy Drapeau L, Pecquerie L, Freon P and Shannon LJ (2004) Quantification and representation of potential spatial interactions in the southern Benguela ecosystem. Ecosystem Approaches to Fisheries in the Southern Benguela, African Journal of Marine Science 26: 141–159 Dugan JE and Davis GE (1993) Applications of marine refugia to coastal fisheries management. Canadian Journal of Fisheries and Aquatic Science 50: 2029–2042

4

GIS in Fisheries Management and Research

119

Eastwood PD, Meaden GJ and Grioche A (2001) Modelling spatial variations in spawning habitat suitability for the sole Solea solea using regression quantiles and GIS procedures. Marine Ecology Progress Series 224: 251–266 Eastwood PD and Meaden GJ (2004) Introducing greater ecological realism to fish habitat models. In Nishida T, Lailola PJ and Hollingworth CE (eds) GIS/Spatial Analyses in Fishery and Aquatic Sciences (vol 2), Fishery-Aquatic GIS Research Group, Saitama, Japan, pp 181–198 Environmental Simulation Laboratory Inc. (2004) Introduction to Marine Explorer. In Nishida T, Lailola PJ and Hollingworth CE (eds) GIS/Spatial Analyses in Fishery and Aquatic Sciences (vol 2), Fishery-Aquatic GIS Research Group, Saitama, Japan, pp 615–623 FAO (2004) State of World Fisheries and Aquaculture (SOFIA) 2004. Food and Agriculture Organisation of the UN, Rome, Italy Fisher WL and Rahel FJ (eds) (2004) Geographic Information Systems in Fisheries. American Fisheries Society, Bethesda, Maryland, USA Garcia SM and de Leiva Moreno I (2001) Global overview of marine fisheries. Proceedings of Reykjavik Conference on Responsible Fisheries in the Marine Ecosystem, Reykjavik, Iceland. 1–4th October 2001. pp 1–24 Gill TA, Monaco ME, Brown SK and Orlando SP (2001) Three GIS tools for assessing or predicting distributions of species, habitats, and impacts: CORA, HSM, and CA&DS. In Nishida T, Kailola PJ and Hollingworth CE (eds) Proceedings of the First International Symposium on GIS in Fishery Science, Seattle, Washington, USA. 2–4 March 1999. pp 404–415 Itoh K and Nishida T (2001) Marine Explorer: Marine GIS software for fisheries and oceanographic information. In Nishida T, Kailola PJ and Hollingworth CE (eds) Proceedings of the First International Symposium on GIS in Fishery Science, Seattle, Washington, USA. 2–4 March 1999. pp 427–437 Kapetsky JM and Aguilar-Manjarrez J (2005) GISFish: The FAO global gateway to GIS, remote sensing and mapping for aquaculture and inland fisheries. In Nishida T, Shiba Y and Tanaka M (eds) Program and Abstracts for the Third International Symposium on GIS/Spatial Analyses in Fishery and Aquatic Sciences. Shanghai Fisheries University, Shanghai, China. 22–26 August 2005. p 5 Kemp Z and Meaden GJ (2002) Visualization for fisheries management from a spatiotemporal perspective. ICES Journal of Marine Science 59: (Part 1): 190–202 Kruse GH, Bez N, Booth A, Dorn MW, Hills S, Lipcius RN, Pelletier D, Roy C, Smith SJ and Witherell D (eds) (2001) Spatial Processes and Management of Marine Populations. University of Alaska Sea Grant AK-SG-01-02. Fairbanks, Alaska, USA Lubchenco J, Palumbi SR, Gaines SD and Andelman S (2003) Plugging a hole in the ocean: the emerging science of marine reserves. Ecological Applications 13(1): S3–S7 Meaden GJ (1996) Potential for geographic information systems (GIS) in fisheries management. In Megrey BA and Moksness E (eds) Computers in Fisheries Research. Chapman & Hall, London. pp 41–79 Meaden GJ (2000) Applications of GIS to fisheries management. In Wright DJ and Bartlett DJ (eds) Marine and Coastal Geographical Information Systems. Taylor and Francis, London. pp 205–226 Meaden GJ (2001) GIS in fisheries science: Foundations for the new millennium. In Nishida T, Kailola PJ and Hollingworth CE (eds) Proceedings of the First International Symposium on GIS in Fishery Science. Seattle, Washington, USA. 2–4 March 1999. pp 3–29 Meaden GJ (2004) Challenges of using geographic information systems in aquatic environments. In Fisher WL and Rahel FJ (eds) Geographic Information Systems in Fisheries, American Fisheries Society, Bethesda, USA. pp 13–48 Meaden GJ and Kapetsky JM (1991) Geographical information systems and remote sensing in inland fisheries and aquaculture. FAO Fisheries Technical Paper No 318. FAO, Rome, Italy

120

G. Meaden

Meaden GJ and Do Chi T (1996) Geographical information systems: Applications to marine fisheries. FAO Fisheries Technical Paper No 356. FAO, Rome, Italy Miceli MFL and Scott PC (2005) Estimativa preliminar do estoque da holotu´ria Isostichopus badionotus no entorno da Ilha Grande, RJ apoiado em Sistemas de Informac¸a˜o Geogra´´ fica e Sensoriamento Remoto, Anais XII Simposio Brasileiro de Sensoriamento Remoto, Goiaˆnia, Brasil. 16–21 April 2005. pp 3659–3665 Mikol R (2004) Data collection methods and GIS uses to enhance catch and reduce bycatch in the north Pacific fisheries. In Nishida T, Lailola PJ and Hollingworth CE (eds) GIS/Spatial Analyses in Fishery and Aquatic Sciences (vol 2), Fishery-Aquatic GIS Research Group, Saitama, Japan. pp 607–614 Mullon C, Freon P and Cury P (2005) The dynamics of collapse in world fisheries. Fish and Fisheries 6: 111–120 Neis B, Schneider DC, Felt L, Haedrich RL, Fischer J and Hutchins JA (1999) Fisheries assessment: What can be learned from interviewing resource users. Canadian Journal of Fisheries and Aquatic Sciences 56: 1949–1963 Nishida T, Kailola PJ and Hollingworth CE (2001) Proceedings of the First International Symposium on GIS in Fishery Science. Seattle, Washington, USA. 2–4 March 1999 Openshaw S and Clarke G (1996) Developing spatial analysis functions relevant to GIS environments. In Masser I and Salge F (eds) Spatial Analytical Perspectives on GIS: Gisdata4. Taylor and Francis, London. pp 21–37 Pauly D, Christensen V, Guenette S, Pitcher TJ, Sumaila UR, Walters CJ, Watson R and Zeller D (2002) Towards sustainability in world fisheries. Nature 418: 689–695 Rahel FJ (2004) Introduction to geographic information systems in fisheries. In Fisher WL and Rahel FJ (eds) Geographic Information Systems in Fisheries, American Fisheries Society, Bethesda, USA. pp 1–12 Rubec PJ, Coyne MS, McMichael RH and Monaco ME (1998) Spatial methods being developed in Florida to determine essential fish habitat. Fisheries 23: 21–25 St. Martin K (2004) Geographic information systems in marine fisheries science and decision making. In Fisher WL and Rahel FJ (eds) Geographic Information Systems in Fisheries, American Fisheries Society, Bethesda, USA. pp 237–258 Tomlinson RF (1989) Presidential address: Geographic information systems and geographers in the 1990s. The Canadian Geographer 33(4): 290–98 Valavanis VD (2002) Geographic Information Systems in Oceanography and Fisheries. Taylor and Francis, London, UK Varma H (2000) Applying spatio-temporal concepts to correlative data analysis. In Wright DJ and Bartlett DJ (eds) Marine and Coastal Geographical Information Systems. Taylor and Francis, London, UK. pp 75–93 Wood BA (2004) Open-source and freely available geographic information system software and resources. In Nishida T, Lailola PJ and Hollingworth CE (eds) GIS/Spatial Analyses in Fishery and Aquatic Sciences (vol 2), Fishery-Aquatic GIS Research Group, Saitama, Japan. pp 625–640 Wood BA and Dunn A (2004) Visualisation of fisheries data using free scientific visualisation software. In Nishida T, Lailola PJ and Hollingworth CE (eds) GIS/Spatial Analyses in Fishery and Aquatic Sciences (vol 2), Fishery-Aquatic GIS Research Group, Saitama, Japan. pp 641–648 Wright DJ (2000) Down to the sea in ships: the emergence of marine GIS. In Wright DJ and Bartlett DJ (eds) Marine and Coastal Geographical Information Systems, Taylor and Francis, London, UK. pp 1–10

Chapter 5

Remote Sensing Olav Rune Godø and Eirik Tenningen

5.1 Introduction Management of the marine environment and its resources according to the ecosystem approach represents a tremendous increase in demand for quality data. This is needed for establishing correct understanding of ecosystem dynamics and is essential for precise assessment of state and development of ecosystem components, particularly those associated with human exploitation. Adequate information for these purposes is obtainable only if we can be present at the time when and location where processes take place. Such objectives strongly contrast those of current monitoring strategies. These are often based on snapshot pictures of the state at regular temporal distance as recorded by standardized surveys. Is it possible to adjust current approaches to satisfy future demands? What is the realism costwise? What are the alternatives? The motivation for extended use of remote sensing is often reduced cost and improved resolution (temporal and spatial). Expanded data collection with conventional methods becomes easily unrealistic due to the lack of capacity to effectuate apt sampling at the needed temporal and spatial resolution. Remote sensing methods can easily give fundamentally better data in that respect, but can it be used in operation monitoring? The fact is that such monitoring is often based on long time series of data. The dependency of trends in these time series for taking managements decision represents thus a halt for introduction of more modern sensing methods. Therefore, the strongest limitation to full utilization of modern remote sensing techniques is the lack of an operational framework for handling, treatment and synthesis of the information. Remote sensing techniques, particularly acoustics is important and will become even more important in the future. These methods are non-intrusive and non-destructive. New development promise categorization that helps the fishers to select species and size without expensive and destructive net catches. Remote sensing can either be effectuated by the sensors themselves or the sensors can be located at site/platform but operated through a remote connection. O.R. Godø (*) Institute of Marine Research, Nordnes, 5817 Bergen, Norway

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_5, Ó Springer ScienceþBusiness Media B.V. 2009

121

122

O.R. Godø and E. Tenningen

In this chapter we will give an overview of some of the most promising remote sensing technologies (sensors and platforms) for observing and monitoring of the marine environment and its resources. Applications spanning from observing ocean currents or fish movements, to measuring densities of whales or abundance of the smallest zooplankton organisms show that remote sensing offer cost efficient solutions to major challenges set by the ecosystem approach to fisheries assessment and management. Technology strengths and weaknesses will be discussed. In a larger perspective; what are the potential for remote sensing sensors to take over and in concert replace or enhance present data sources. Most important, the success of modern sensor technology is totally dependent of development of modelling and computational solution that efficiently utilizes the collected information. This will be theme of the concluding part of the chapter.

5.2 Sensors In this part we will describe some of the remote observation sensors we consider the most important and promising related to monitoring of the marine environment and its resources. We have categorized the sensors according to their operational wavelengths and frequencies used and associated them to the most common platforms carrier types and tasks (Fig. 5.1). In (Fig. 5.2) the search volume for some of the sensors is shown.

Fig. 5.1 Remote sensors with their most common carrier and tasks in the water column. The yellow lines indicate the sampling ranges and thus the limitations

5

Remote Sensing

123

Fig. 5.2 Search volume for some acoustic sensors and platforms compared to lidar from an airplane. Vessel echo sounder acoustics are limited at surface. This can be compensated by the horizontal pointing sonar. The limitation close to bottom (bottom dead zone) can be compensated through other platform, e.g. like shown her with an AUV or drifting buoy close to bottom. We can also improve coverage of the surface area through an anchored upward pointing acoustic system. This will also improve temporal resolution. Finally, lidar is a quick and efficient means of covering the surface zone and thus compensating for the surface blind zone and vessel avoidance

5.2.1 Acoustics Acoustics uses sound to detect properties of marine organisms and their environment. Vertical sounding is most common in fisheries science while in commercial fishing horizontal looking systems are crucial for efficient fishing of pelagic species. Also acoustics sensors are used to detect current direction and speed. However, for centuries before acoustics was developed, remote sensing and quantification of fish schools has been done with hand line (Fig. 5.3). Most important, today is that acoustics enable novel answers to the requirements set by the ecosystem approach; including observing behaviour and interactions on individuals level and species overlap on stock level. Also, acoustics sensors enable simultaneous sampling of interacting species and association between e.g. bottom topography, bottom type or ocean currents. In a late section we will show that details of the whole water column can be sampled when using various acoustic sensors from different platforms. When choosing acoustic system there is a number of compromises related to needed range, resolution and sampling

124

O.R. Godø and E. Tenningen

Fig. 5.3 Sketch of fisherman with hand line detecting and quantifying a herring school. An experienced user could precisely determine the size, distribution and depth of the school and was a key person for an efficient fishery. The hand line and lead weight is slightly manipulated on the drawing by Lauritz Haaland 1900 (with courtesy from Norwegian Fisheries Museum)

volume. Present trend in development is to increase band and beam widths to avoid most of these compromises. Low frequency reaches long but has lower resolution than higher frequencies. Wide beam searches large volume but gives low resolution. Low frequency transducers are generally larger and more expensive and difficult to handle and mount on small vessels. Search volume can be increased without loosing resolution by using multibeam system. There is thus a long list of possible choices and it is a challenge to select a cost-efficient solution for a specific task. Common for all is the increasing amount of data produced with advances in technology. Efficient utilization is computer demanding, as real time processing is essential, particularly in commercial fishing and operational monitoring surveys. Data handling and post processing is discussed in a later section of this chapter. In the following is described in more detail some of the most common sensors in fisheries and fisheries research. Modern research vessels are equipped with a multitude of acoustic instruments (see e.g. http://www.uib.no/gosars/index.html) and support abundance estimation, behavioural studies, bottom habitat mapping, water movements etc.

5

Remote Sensing

125

5.2.1.1 Echo Sounders Echo sounders have been used for detecting fish since the 1930s (Sund 1935), and for surveying since the 1950s (Cushing 1952). Improved quantification through echo counting (Cushing 1964) and echo integration (Dragesund and Olsen 1965) can be used directly in quantification of fish biomass. Single beam, dual beam and split beam systems are in use but split beam systems are presently the most common in stock assessment. Single beam sounders reflect echo abundance but gives limited information about individual targets. In dual and split beam systems the beams are composed of two and four sectors respectively and facilitate position of the target and thus more exact information about the individual’s acoustic properties based on depth separation and phase differences (see e.g. Simmonds and MacLennan (2005)). Further, the individually recorded fishes can be tracked over time and thus reveal behaviour patterns. Such patterns have already revealed new understanding of interaction between species (see e.g. Onsrud et al. 2005) and between the individual and the environment (unpublished material, IMR, Bergen) as well as fishing vessel-fish interaction (Handegard et al. 2003; Ona et al. 2007). Echo sounders are normally run from research vessels and are used for collecting abundance data at stock level. Such surveys often use trawl samples for identification of acoustic recordings (Toresen et al. 1998). The sampling for identification of species and size represent a major source of uncertainty and take a substantial part of available survey effort (see the section on Validation below). Consequently, there has been a strong focus in later years to improve remote size and species recognition by acoustics. Sizing of single fish detection by target strength (TS) measurements by dual or split beam echo sounders is a well established approach (see e.g. Brede et al. 1990; Clay and Horne 1994; Jorgensen 2003; Fig. 5.5). TS is related to the swimbladder dorsal area and it is assumed that this area is size dependent (Fig. 5.4). However, (Ona 1990) and (Horne 2003) show that many physiological factors affect this relationship. Further, fish natural behaviour influence orientation and thus the area facing towards the echo sounder transducer and hence the acoustic TS (Horne 2003; Huse and Ona 1996). Another approach to

Fig. 5.4 Target strengthlength relationship for valley polloch from in situ observation (Horne 2003)

126

O.R. Godø and E. Tenningen

Fig. 5.5 A sample echogram from the Vestfjorden survey in 1997 showing a large (approximately 250 000 tons) herring school. The main echogram is about 3.5-nmi long and 0–500 m vertically (50 m between the marker lines). Inserted to the right is an echogram from the probing transducer at 100-m depth in the middle of the school, resolving individual herring tracks. Typically, TS data would be collected 5–25 m from the transducer (range indicated by red lines) (Ona 2003)

sizing of swimbladdered fish is searching for the resonance frequency (Thompson and Love 1996; Holliday and Smith 1978; McClatchie et al. 1996). As the resonance frequency will vary with swimbladder size fish length can be deduced due to the relationship between fish and swimbladder sizes. This methodology is not yet extensively utilized but is expected to become important in the future both for operational monitoring surveys and for commercial fishing. Also, remote species identification is becoming an operational feature in modern acoustics by analysing data from several frequencies (Horne 2000; Korneliussen and Ona 2002). This method utilizes variation in acoustic properties at different frequencies to categorize the recordings. The method has proven an efficient tool to separate fish with and without swimbladder and is continuously under development. In a quota regulated fishery the economic loss for fishermen if e.g. mackerel is caught by mistake during the low prize season could easily be several hundred thousand Euros. Traditional echo sounders cover a very limited volume of water right under the vessel. Multibeam echo sounders have been developed to improve efficiency for bottom mapping. These sounders return reliable bottom depths from a swathe width of e.g. 3–6 times the bottom depth, transverse to the vessel track (see e.g. http://www.km.kongsberg.com, Fig. 5.6). These sounders have

5

Remote Sensing

127

Fig. 5.6 Multibeam system for efficient bottom and habitat mapping (http:// www.kongsberg.com)

maintained resolution in spite of increasing the search volume because a fan of narrow beams builds the extended beam width. Multibeam echo sounders are also used for bottom habitat typing (Roberts et al. 2005). The most recent development in multibeam echo sounder technology is the Simrad ME70, with 500 beams (Mazauric et al. 2006). This sounder expand the sampling volume through a 2520 beam system, which thus may include a complete school in the sampling volume as well as give detailed information about fish located close to bottom. 5.2.1.2 Sonars Vessel mounted echo sounders has a very limited sampling volume close to the vessel due to the narrow beam. Also, the surface zone, often called the blind zone, is not covered at all as the transducer is looking downward from a certain depth. Horizontal looking sonars are extensively used in commercial fishing both through the searching and catching phases (see e.g. Fernø and Olsen 1994). In recent years commercial fishing sonars have attained a technical level of interest for scientist and have become an important tools for studying abundance and behaviour of pelagic species (Brehmer et al. 2006). Sonars dramatically improve the sampling volume in shallow water but the fish density information is associated with additional uncertainty compared to echo sounders. The horizontal view of the fish gives a higher variability of acoustic target strength and if the vessel affects the fish, as often is the case for pelagic species, the data become difficult to analyse. Also, calibration has been a problem although some solution now appears feasible (see Brehmer et al. 2006 and references therein). Recently a new generation multibeam sonar, the Simrad MS70, was introduced. This

128

O.R. Godø and E. Tenningen

sonar, which has a 2520 beam system comparable to the ME70 multibeam echo sounder, is specified for pelagic fish stock abundance estimation (Ona et al. 2006).

5.2.1.3 Other Acoustic Sensors While multibeam sonars insonify a large observation volume with one ping, scanning sonars have a single beam that must scan the same volume over time with many pings and over time build a similar picture. When detecting a moving object this might be a problem but represent a cost efficient solution for many applications. Examples are detection and assessing fish in shallow water (Farmer et al. 1999), tracking fish tagged with acoustic transmitters (Harden Jones et al. 1977), monitoring trawls (Ona 1994); bottom habitat mapping etc. Acoustic current profilers are acoustic systems that utilise the Doppler shift to detect and measure particle movements in the sea, and hence direction and speed of water. Such instruments are mostly used in oceanography but are very useful in conjunction with other instruments to judge fish behaviour in relation to ocean currents (Zedel et al. 2005; Wilson and Boehlert 2004). Broadband acoustics may, similarly to multi frequency echo sounders, supply information to support discrimination between species. We expect future research to undertake the development of broadband technology and associated software for species identification.

Fig. 5.7 AUV Hugin with SAS transducer. Example SAS bathymetry from a rock outcrop (60  60 m). Depth range 196.5–198.5 m. The vertical axis is exaggerated by a factor of 2 in this display

5

Remote Sensing

129

Synthetic Aperture Sonar (SAS) uses a technique actively applied in radar (described in the SAR section). It is now also becoming realistic in marine acoustics when using autonomous underwater vehicle as platform (AUV). The AUV is stable enough to enable reception of echoes from several pings as if they came from one ping from a large transducer. (Hagen et al. 2006) demonstrate successful bottom mapping (Fig. 5.7) giving very high resolution at long ranges. The system is presently ideal for bottom habitat mapping and for detecting fisheries related effects on the bottom habitat. The method demand advanced computers and demanding signal processing. Also, this complexity complicates the presentation of moving objects like fish. Many marine organisms produce sound, thus, recording and analysing these sounds may identify the presence of sound producers; who they are and what they are doing (Rountree et al. 2006). Presently such sensors and techniques are under-utilised in fisheries and fisheries science. Particularly, such information could considerably help scientist to better understand species interaction and thus ecosystem processes. 5.2.1.4 Alternative Platforms Traditionally acoustic systems are operated from research or fishing vessels. Modern systems are getting smaller and more robust and thus allow distributed application. Installing advanced scientific echo sounders on fishing vessels is becoming an important approach to increase coverage in time and space of fish stocks. In recent years stationary bottom mounted or anchored systems has become more common (see e.g. Farmer et al. 1999; Godø et al. 2005; Brierley et al. 2006; Trevorrow 2005). Detailed observations on one location improve the information of temporal dynamics and may become an important data source for quantifying and modelling these processes. Good examples are diurnal dynamics, e.g. see vertical migration in Fig. 5.8; (Wilson 1998; Onsrud et al. 2004), species interactions (Traynor 1996; Onsrud et al. 2004) as well as density variation over time (Brierley et al. 2006). New platforms also are being designed to meet specific challenges e.g. an autonomous buoy that cover the acoustic dead zone of the vessels (Ona and Mitson 1996), or moored multisensor platform to collect simultaneous fish density and behaviour data along with physical environmental data (The FAD buoy, Brehmer et al. 2006). With the leaping progress in technology such platforms, designed for particular purposes, will appear and give scientist access to detailed information about the ecosystem of greatest importance to realise the ecosystem approach to stock assessment and management. 5.2.1.5 Computer Demands in Acoustics Echo sounders entered the digitised world in the end of the 1980s. Since then most acoustic equipment used in commercial fishing and in science is computer based. As described above, development goes towards increasing beam and

130

O.R. Godø and E. Tenningen

Fig. 5.8 Diurnal dynamics of mesopelagic fish at the Midatlantic Ridge as measured from an echo sounder from 950 m depth. Part of the biomass shows extensive vertical migration with time of day (UTC). Some is unaffected while some of the surface biomass apparently concentrate at daytime and disperse at night. Mid day is about 1430 UTC

bandwidth, and the sensors are installed in new platform types. As a result, the amount of data explodes and creates a critical demand for efficient software solutions to scrutinize the overwhelming amount of data from a multitude of sensors. The acoustic signals are analysed by instrument computers before displayed in e.g. echograms. Also, the raw data are stored for post processing. While post processing has been standardized procedures for research application of acoustic data (Foote et al. 1991), such processing is not common in fisheries. However, with the increasing complexity of the available information there is growing demand for processing to enable the skipper to make decision based on the essential information. The computer related tasks are mainly: 1. Data screening, selection and storage 2. Data processing and scrutiny to extract essential information 3. Data integration and production of results. The MS/ME70 systems (see e.g. Ona et al. 2006) with 500 beams clearly demonstrate an urgent computer demanding issue: Every ping produces

5

Remote Sensing

131

5–10 MB of data and thus in the order of 250 GB per day. Firstly, computer power is needed to handle this data stream. Also, storage capacity is normally limited. As the sonar data contain no essential information except when schools are recorded, data reduction can be achieved by simply discarding pings without school information. This and similar routines may help reducing the data before storage. Such computer tasks are becoming increasingly important with expanding beam and bandwidth and when more sensors and platforms are involved. An additional critical demand is the need to process the data in real or close to real time. To extract the essential information from complex data sets we need powerful post processing systems. Available systems, e.g. BEI, Movies, Echoview and SSS, are designed to handle echo sounder data and some also sonar data. The expanding amount of data from an increasing number of sensors will demand substantial development of such software in the coming years. In particular visualisation and statistical evaluation of the data is needed. The most important tasks will be allocation of backscattering to species or groups of species, behavioural analysis of single fish and schools and quantification of species interaction. We expect that such tools will become vital in future monitoring strategies for marine ecosystems because they may improve quality of the data as well as understanding of processes. Both are essential to understanding ecosystems – a basic need for efficient modelling. Data integration will be discussed in a later section.

5.2.2 Satellites When moving from the acoustic sensors described above into the field of radars and optical technology discussed in the next sections, the fundamental wave propagation changes from sound waves to electromagnetic waves. The frequencies used in acoustics are usually in the kHz region, whereas the frequencies associated with the sensors described in the following sections range from GHz to 1015 Hz. Radars operates at microwave wavelengths, lidar in the visible spectrum (400–700 nm) and thermal IR sensors just above the visible spectrum. Spaceborne instruments have the ability to cover large areas in a short time making them suitable to monitor large scale day-to-day and seasonal variations as well as climate changes. They provide valuable inputs to modelling of the physical environment. A synthetic aperture radar (SAR) can be used to monitor global surface wave fields and polar sea ice conditions. Microwave scatterometers can be used to measure surface roughness, wind speed and direction. The sea surface temperature (SST) can be measured by a scanning multichannel microwave radiometer (SMMR). Visible and infrared sensors can provide information about ocean colour and sea surface temperature. NASA’s Seasat, launched in 1978, is presented here as an example of a successful satellite mission that ran for 25 years. It was the first Earth orbiting satellite to carry four complementary microwave experiments. An overview of the sensors is

132

O.R. Godø and E. Tenningen

given below. For a more detailed description of the mission and the different sensors (see Evans et al. 2005 and references therein). 5.2.2.1 Synthetic Aperture Radar (SAR) Synthetic aperture radar (SAR) is a radar in which a sophisticated post-processing program is used to produce a very narrow effective beam. The moving antenna creates a synthetic aperture that is much larger than the physical dimensions of the actual antenna by combining the pulses along the track as if they were made simultaneously. The post-processing involves Fourier analysis that requires significant computer resources. The phase and polarisation of the return signal provides additional information about the target. SAR can be used to observe surface and internal waves, current boundaries, meso scale eddies (10–400 km), temperature fronts, oil slicks, bathymetry in shallow areas as well as wind speed and direction. In the Polar Regions, SAR produces fine resolution images used to create ice motion charts and the most detailed maps of Antarctica ever made resulted from the RADARSAT mission in 1997 and in 2000 (Jezek 1999; Jezek et al. 2003). Finally, SAR can be used to monitor the movement of the fishing fleet and illegal ship discharge. The Seasat SAR operates at 1.28 GHz making it insensitive to clouds and darkness. The spatial resolution is 2525 m (Jordan 1980). 5.2.2.2 Microwave Scatterometer The Seasat mission carried for the first time a microwave scatterometer designed specifically for wind observations. Accurate measurements of wind velocity could be obtained producing 100-km-resolution maps of synoptic surface winds over the ocean used in operational numerical weather prediction and marine forecasting. Today, scatterometers have 60% global coverage in 6 h and 90% in 12 h, providing essential information for resolving diurnal and locally varying winds that drive ocean mixing and transport processes (Liu 2003). The microwave scatterometer measure the capillary waves on the ocean surface generated by the friction velocity of the wind. The wavelength of these waves are in the order of centimeters and their amplitude reflects the local wind force. The wind direction is found from the two-dimensional wave spectrum (Moore and Fung 1979). The wind vector measurements are used in ocean circulation models, air-sea flux studies, weather forecasting and to measure the seasonal melt cycle in the Polar Regions (Liu 2002; Carsey 1985). The Seasat microwave scatterometers operates at 14.6 GHz providing data during day and night and nearly all weather conditions. 5.2.2.3 Microwave Radar Altimeter A microwave radar altimeter, e.g. Seasat Radar Altimeter (ALT), Geosat altimeter and the GEOS-3 altimeter, can provide high-resolution mapping of

5

Remote Sensing

133

the sea surface topography. It is possible to detect a 1 cm change in sea level at a regional scale and to monitor changes in the global mean sea level to about 2 mm/year (Chelton 2001). The data can be used to monitor ocean dynamics, significant wave height, swell propagation and to predict phenomenons like El Nin˜o. The radar altimeter is also an important tool in the monitoring of the polar ice and sea ice freeboard (the portion of sea ice that extends above sea level) (Laxon et al. 2003). 5.2.2.4 Scanning Multichannel Microwave Radiometer The Seasat Scanning multichannel microwave radiometer (SMMR) used 5 dual-polarized simultaneous measurement channels ranging from 6.6 to 37 GHz. The SMMR can measure a variety of parameters including sea surface temperature and surface wind speed. It also provides path length corrections and atmospheric attenuation corrections for the scatterometer instrumentation. Two satellites planned for launch in 2007 (Aquarius) and 2008 (SMOS) will use SMMR technology to measure ocean salinity with a 3-beam pushbroom radiometer (Evans et al. 2005). 5.2.2.5 Other Sensors Sensors in the visible part of the electromagnetic spectrum are used to monitor algal blooms and chlorophyll a concentration in the ocean surface layer. Infrared sensors provide sea surface temperature measurements. The major disadvantage of these sensors is their limitations in overcast weather. They operate at a very high frequency, which is unable to penetrate clouds.

5.2.2.6 The Satellite Platform NASA’s Seasat mission described above was designed to collect data on seasurface winds, sea-surface temperatures, wave heights, ocean topography, internal waves, atmospheric water, and sea ice properties. The Seasat satellite orbit altitude is 800 km, and in the region 435–1336 km for other Earth orbiting satellites (Evans et al. 2005). Computers onboard the satellites have to be reliable and well tested. A computer crash in space can be fatal. The radiation that pervades space can trigger glitches that destroy the data or in worst case the computer. When highspeed particles, such as cosmic rays, collide with the microscopic circuitry or computer chips they can cause errors. To ensure safe operation, most space missions use radiation-hardened computer chips that contain extra transistors that take more energy to switch on and off. The downsides of these chips are that they are expensive and as much as 10 times slower than an equivalent CPU. One of the most limited resources onboard a satellite is its bandwidth. The data collected are transferred to Earth-based stations using regular radio

134

O.R. Godø and E. Tenningen

communication and the transmission speed is even slower than old dial-up modems (http://www.nasa.gov/).

5.2.2.7 Satellite Data Used in Modelling In ocean circulation models satellite data such as sea surface temperature and surface wind speed are often used as input. An example of such a model is the hybrid coordinate ocean model (HYCOM) that was developed as an open ocean model and later evaluated in basin scale studies and in shelf sea (Bleck 2002; Chassignet et al. 2003; Winther and Evensen 2006). In Fig. 5.9, the modelled data based on satellite measurements are compared to CTD temperature data. CTD data are collected monthly along several transects (e.g. Shetland – Feie in the North Sea collected by the Institute of Marine Research, Norway). See Fig. 5.9. By understanding how fish react to variations in the physical environment, one can create migration models based on satellite data. Zagaglia et al. (2004) provide a good example of the utilization of remote sensing data from satellites in their study on yellowfin tuna. They show a relationship between catch per unit effort (CPUE) and wind direction, sea surface temperature, chlorophyll – a concentration and sea surface height.

Fig. 5.9 Satellite data extended to the deep between Shetland and Feie. CTD measurements (top) and modelled data (bottom) (Winther and Evensen 2006)

5

Remote Sensing

135

5.2.3 Optics 5.2.3.1 Lidar Lidar is an acronym for light detection and ranging. An airborne lidar uses laser pulses to detect fish and plankton in much the same way as an echosounder uses acoustic pulses. Laser light in the mid-visible portion of the spectrum (532 nm) is capable of penetrating the sea surface and propagate through the upper layer of the ocean down to 25–50 m depending on the conditions. Reflection of the laser radiation back to the receiver is measured to generate lidargrams similar to echograms used in fisheries acoustics. A typical lidar system is shown in Fig. 5.10. The laser is a linearly polarized frequency-doubled Nd:YAG operating in the green portion of the visible spectrum (532 nm) with a pulse energy of 130 mJ. The pulse repetition rate is typically 30 Hz. The laser beam is diverged to create a surface footprint with a diameter of 5 m when operated from an altitude of 300 m. The laser pulse penetrates clean air without any significant loss, but is heavily scattered by water vapour in foggy or rainy conditions. There is a large reflection from the sea surface, however the laser is tilted 15 degrees from vertical to reduce specular reflections. The portion of the light penetrating the air-sea interface reflects from objects in the sea making observations of fish and plankton possible. The reflected signal is collected by a telescope shown next to

Fig. 5.10 A typical airborne lidar system

136

O.R. Godø and E. Tenningen

the laser in Fig. 5.10. In front of the telescope there is a rotatable polarizer allowing to choose between the co-polarized or crosspolarized return. To reject unwanted background light the collected light passes through an interference narrowband filter. The resulting signal is converted to an electric current through a photomultiplier tube. A 50 load resistor converts the electric current to a voltage that is digitized in the A/D system at 400–1000 MHz with 8–12 bits of resolution. At 1 GHz the depth resolution is 0.11 m. The laser and receiver optics are fixed on a mounting frame that can easily be fitted in most aircrafts with a camera port. Also shown in the figure is the power supply and cooling unit for the laser next to the DC/AC converter. All the components are connected to the industry pc, from which the user can control the operation of the lidar. The raw data collected from a survey flight of 6 h is in the order of 5 GB. For a more detailed description of the lidar (see e.g. Churnside et al. 2001). Post-processing of lidar data is a challenging job due to the great variations in surface and propagation conditions. During a survey flight the conditions are likely to change many times as a large area is covered. Calm water close to the coast may have low visibility resulting in bad lidar depth penetration due to algal blooms etc., whereas clear open sea can have a rough surface reducing the portion of the light that passes the air-sea interface drastically. Rain and fog result in unusable data that has to be removed. After the initial manual filtering

Fig. 5.11 A lidargram showing a mackerel school. The colour scaling is a relative measure of the strength of the lidar return

5

Remote Sensing

137

of the data, a more systematic post-processing can start. First the data above the sea surface and the data below the lidar penetration depth are removed. This is quite straightforward as the surface can be detected by its strong echo, and the deeper data can be removed by a signal-to-noise (SNR) threshold. Then the background signal or clear water return has to be removed, and the lidar attenuation coefficient can be found. For simplicity, the optical properties of the water are assumed to be constant with depth. If fish are present at some depth, there is an extra contribution to the signal at that depth, which depends on the backscattering properties of the fish. The processed data can be used to generate lidargrams. Figure 5.11 shows an example of a mackerel school found in the Norwegian Sea during summer feeding. The major advantages of an airborne lidar system compared to traditional echosounders is the absence of vessel avoidance by fish, and its ability to cover large areas in short time creating a snapshot of the survey area. An airborne lidar can also be used to detect plankton, pelagic recruits and oil slicks. Tenningen et al. (2006) showed the potential of target classification using two receivers measuring the degree of depolarisation in the return lidar signal. 5.2.3.2 Video and Imaging Video is perhaps the most effective tool in observing behaviour and interactions in nature. Under water however its use is strongly limited by the lack of light. Great care has to be taken to provide enough illumination of the target. We have all seen fantastic footage of large whales, milling fish schools or hunting sharks. These shots are taken close to the surface where the sunlight is still present and very often from underneath to obtain contrast from the surrounding seawater. In underwater remote sensing, video is mainly used as a validation tool. It has been used to observe individual fish in acoustic target strength measurements and to study escapement from trawl (Kloser and Horne 2003; Ingolfsson and Jørgensen 2006). The use of video in deep water is troublesome due to the short observation distances and extensive use of artificial light is needed. Video has also been used on remotely operated vehicles (ROV) to monitor variability in natural behaviour and reactions to large predators (Lorance and Trenkel 2006). 5.2.3.3 Aircraft Platform Aircrafts are widely used for counting of marine mammals (e.g. Haug et al. 2006). They form a great platform for visual observations, video and photo. Towards the end of the 1990s new sensors were tested for marine research purposes, and today, flight surveys are carried out using lidar, SAR, IR and visual observations. A single sensor, e.g. lidar, can be carried by smaller twin engine aircrafts used for photo surveying, whereas more sensors requires more room and electric power.

138

O.R. Godø and E. Tenningen

An aircraft survey provides a quick coverage creating a snapshot of the situation in the survey area. The flight speed during lidar surveys is typically 180 knots resulting in a low cost per nautical mile. The major downsides of using aircrafts are its limitations in bad weather and the problem of data verification. Both visual observations and sensors in the visual spectrum are dependent on visibility.

5.3 Data Management, Fusion and Synthesis 5.3.1 Management Under data management we discuss storage, quality control and accessibility. An inherent problem of modern technology is that new sensors combined with enhanced computer capacity produce more data. These data has to be stored for later use. Fortunately, there seem to be a never-ending interaction between the increasing data availability and the development of improved storage media both with respect to capacity and quality. However, the increasing amount of information also challenge our ability organise and make the data accessible for the users. Fisheries management build on advice from scientific evaluations of the harvested stocks and their environment. These evaluations are normally founded on time series of data. Thus, storage, quality and accessibility are particularly critical in this field of science. The development of the ecosystem approach is demanding in this respect and progress computer technology is key in realising the approach through efficient collection and management of field information. In general the ecosystem approach to fisheries management increases the demand for information due to the need for data on both predators and preys of the ecosystem independent of their importance as a target for fisheries. Also, process information about their interaction is essential. In addition, associated data from the physical and biological environment is crucial for developing the model tools needed for future ecosystem advice. Traditional data such as trawl catch information will never increase to a troublesome level due to the high cost in collecting them. However, remote sensing systems over-sample as no additional costs are involved when collecting data autonomously. In particular acoustic sensors manifold the data amount. For example, the new MS/ME70 multibeam sonar/sounder produce orders of magnitude more data than traditional echo sounders. (GB per hour, see e.g. Ona et al. 2006). Should all these data be stored for later analysis? Technically this is possible, but this is a good example that storage of all data is not always a feasible approach. Many sensors in modern scientific instrumentation collect data continuously and often oversample. The multibeam systems are good examples. In most cases less than 5% of pings will contain interesting information while the rest is thrash. It is thus possible to make decision during sampling that directly or at a delayed stage select the useful information and discard the rest. Such procedures need

5

Remote Sensing

139

consideration in all modern data management. They will ease the data storage problem and simplify later analysis. The complexity of collecting marine data makes outsiders sometimes very susceptible to inappropriate treatment and application. However, in the future we expect scientists from different fields and institutions to become frequent users of the data for developing new models and running analysis. Therefore modern storage and distribution systems of marine data need to be coupled with thorough information and guidance to avoid unsuitable use. Thus computer-based information system has to become an integral part of advanced and open access databases for complex data to meet the requirements set by future demands in fisheries science.

5.3.2 Utilising Data on Heterogeneous Scales Multi-sensor data collection represents a challenge in fisheries research. A scientific approach exists in environmental sciences (see e.g. Johannessen et al. 2006). But repeated trials in such a simple thing as combining acoustic and bottom trawl data has till now not resulted in an accepted general scientific approach (see e.g. Godø 1994; Hjellvik et al. 2003). However, an increased utilisation of remote sensing data will increase the demand for harmonising heterogeneous data. Known approaches in handling satellite information include:

 Multi-sensor analysis: Often used in classification of various properties based on data from satellite sensors. In fisheries this could be trawl and acoustic data, or surface satellite information and biomass information from acoustics in the water column.  Multi scale analysis: Presently models are developed for combining information from satellite sensors with various geographical resolutions. For fisheries the same example as given above are valid.  Multi temporal data: New methodology is underway for handling and analysing time series of data, e.g. time series of vegetation pictures observed from satellites. Similar combination could be interesting in fisheries, e.g. combination of time series satellite sensor data and lidar or acoustic data. The collection of data on heterogeneous scales from various sensors would never be possible without computer power. The transformation of this information through various models into useful comparable information is particularly computer demanding. Many remote sensing sensors produce large amounts of data and the more complex and diverse, the more complex analysis is needed to remove the scaling problem.

5.3.3 Validation Without proper validation, enthusiastic and well-meant use of new sensor systems in models and assessment systems may result in later confusion and

140

O.R. Godø and E. Tenningen

frustration. However, appropriate validation of marine data is often difficult and thus strong assumptions might be associated with the applied methodology. Examples such as constant acoustic target strength and trawl catchability demonstrate that such assumptions might survive without discrediting the methodology. However, when combining information assumptions that affect the relationship between sensor data might invalidate any combination analysis. Thus, for applying and combining remote sensing data and relate them to traditional data, validation becomes essential. Examples could be comparison of lidar and acoustic biomass data in the upper 50 m and ground truthing of the recordings with fish and plankton sampling gears.

5.3.4 Synthesis Synthesis includes the process of taking the data after scaling and validation and analyse them with respect to well defined objectives. In our case this includes evaluation of stock and ecosystem status and prediction of their development. Also, it involves anthropogenic effects on the marine habitat, e.g. effect of trawling on the bottom habitat (Hiddink et al. 2006). The ecosystem approach to management of the marine environment and its resources demand more and better data with better temporal and geographic resolution. There is no other way to obtain this at realistic cost than through use of remote sensing data at heterogeneous scales. In this respect computer power and advanced data models will become an important foundation in future scientific based management. Obvious challenges are to develop methodology that extends air based information from surface layers to the ocean interior. In other words; use new sensor technologies combined with available computer capacities to improve cost efficiency through inexpensive data collection (Malanotte-Rizzoli 1996). Another challenge emphasised by the ecosystem approach is the need for better information on the dynamic properties of the ecosystem, e.g. migration, species interaction, environmentbiology interaction etc. The inclusion of remote sensing techniques combined with methodological/modelling development opens the possibility to produce this kind of information to a much greater extent than before. Synthesis also presents another demanding requirement for computers. The handling and computational challenges associated with the increasing amounts of data and information is easily under evaluated, particularly when operating with data on heterogeneous scales. Due to the high costs of collecting data computers and models are used to get the best of what is there. In the future we anticipate a reversed situation: that computer capacity and model quality will prepare the ground for predicting new sampling regimes and strategies that over time will substantially improve cost efficiency of the field sampling.

5

Remote Sensing

141

References Bleck R (2002) An Oceanic general circulation model framed in hybrid isopycnic-cartesian coordinates. Ocean Modelling 4: 55–88 Brede R, Kristensen FH, Solli H, and Ona E (1990) Target tracking with a split-beam echo sounder. Rapports et Proce´s-Verbaux des Re´unions Conseil International pour l’Exploration de la Mer 189: 254–263 Brehmer P, Lafont T, Georgakarakos S, Josse E, Gerlotto F, and Collet C (2006) Omnidirectional multibeam sonar monitoring: applications in fisheries science. Fish and Fisheries 7: 165–179 Brierley AS, Saunders RA, Bone DG, Murphy EJ, Enderlein P, Conti SG, and Demer DA (2006) Use of moored acoustic instruments to measure short-term variability in abundance of antarctic krill. Limnology and Oceanography-Methods 4: 18–29 Carsey FD (1985) Summer Arctic sea ice characteristics from satellite microwave data. Journal of Geophysical Research 90(NC3): 5015–5034 Chassignet E, Smith L, Halliwell G, and Bleck R (2003) North Atlantic simulation with the hybrid coordinate ocean model (HYCOM): impact of the vertical coordinate choice, reference density, and thermobaricity. Journal of Physical Oceanography 33: 2504–2526 Chelton DB (2001) Report of the High-Resolution Ocean Topography Science Working Group Meeting. Corvallis OR: College of Oceanic and Atmospheric Sciences Oregon State University Ref. 2001–4 Churnside JH, Wilson JJ, and Tatarskii VV (2001) Airborne lidar for fisheries application. Optical Engineering 40: 406–414 Clay CS and Horne JK (1994) Acoustic models of fish – the Atlantic cod (Gadus morhua). Journal of the Acoustical Society of America 96: 1661–1668 Cushing DH (1952) Echo surveys of fish. J Cons CIEM 18: 45–60 Cushing DH (1964) The counting of fish with an echo sounder. Rapp P.-V. Re´un CIEM 155: 190–195 Dragesund O, Olsen S (1965) On the possibility of estimating yearclass strength by measuring echo abundance of O-group fish. Fiskeridir Skr ser Havunders 13: 47–71 Evans DL, Alpers W, Cazenave A, Elachi C, Farr T, Glackin D, Holt B, Jones L, Liu WT, McCandless W, Menard Y, Moore R, and Njoku E (2005) Seasat – A 25-year legacy of success. Remote Sensing of Environment 94: 384–404 Farmer DM, Trevorrow MV, and Pedersen B (1999) Intermediate range fish detection with a 12-kHz sidescan sonar. Journal of the Acoustical Society of America 106(5): 2481–2490 Fernoe A and Olsen S (1994) Marine Fish Behaviour in Capture and Abundance Estimation. Fishing News Books, Oxford Foote KG, Knudsen HP, and Korneliussen RJ (1991) Postprocessing system for echo sounder data. The Journal of the Acoustical Society of America 90: 37–47 Godø OR (1994) Factors Affecting the Reliability of Groundfish Abundance Estimates from Bottom Trawl Surveys. Ferno¨ A and Olsen S, Fishing News Books. A division of Blackwell Science Ltd Godø OR, Patel R, Torkelsen T, and Vagle S (2005) Observatory Technology in Fish Resources Monitoring. Proceedings of the International Conference ‘‘Underwater Acoustic Measurements: Technologies & Results’’ 28th June – 1st July 2005. Heraklion, Crete, Greece Hagen PE, Hansen RE, and Langli B (2006) Interferometric Synthetic Aperture Sonar for the HUGIN 1000-MR AUV. UDT Pacific 2006, San Diego, CA, USA, November 2006: 1–8 Handegard NO, Michalsen K, and Tjostheim D (2003) Avoidance behaviour in cod (Gadus Morhua) to a bottom-trawling vessel. Aquatic Living Resources 16: 265–270

142

O.R. Godø and E. Tenningen

Harden Jones FR, Margetts AR, Greer WM, and Arnold GP (1977) The efficiency of the Granton otter trawl determined by sector-scanning sonar and acoustic transponding tags: a preliminary report. ICES Symposium on Acoustic Methods in Fishery Research 32: 1–15 Haug T, Stenson GB, Corkeron PJ, and Nilssen KT (2006) Estimation of harp seal (Pagophilus groenlandicus) pup production in the North Atlantic completed: results from survey in the Greenland Sea in 2002. ICES Journal of Marine Science 63: 95–104 Hiddink JG, Jennings S, and Kaiser MJ (2006) Indicators of the ecological impact of bottomtrawl disturbance on seabed communities. Ecosystems 9: 1190–1199 Hjellvik V, Michalsen K, Aglen A, and Nakken O (2003) An attempt at estimating the effective fishing height of the bottom trawl using acoustic survey recordings. ICES Journal of Marine Science 60: 967–979 Holliday DV and Smith PE (1978) Seasonal-changes in fish size distributions as detected by acoustic-resonance survey and by conventional sampling. Journal of the Acoustical Society of America 64: S95. Notes: Supplement: Suppl 1 Horne JK (2000) Acoustic approaches to remote species identification: a review. Fisheries Oceanography 9: 356–371 Horne JK (2003) The influence of ontogeny, physiology, and behaviour on the target strength of walleye pollock (Theragra chalcogramma). ICES Journal of Marine Science 60: 1063–1074 Huse I and Ona E (1996) Tilt angle distribution and swimming speed of overwintering Norwegian spring spawning herring. ICES Journal of Marine Science 53: 863–873 Ingolfsson OA and Jørgensen T (2006) Escapement of gadoid fish beneath a commercial bottom trawl: relevance to the overall trawl selectivity. Fisheries Research 79: 303–312 Jezek KC (1999) Glaciological properties of the Antarctic ice sheet from RADARSAT-1 synthetic aperture radar imagery. Annals of Glaciology 29: 286–290 Jezek KC, Farness K, Carande R, Wu X, and Labelle-Hamer N (2003) RADARSAT 1 synthetic aperture radar observations of Antarctica: modified Antarctic Mapping Mission 2000. Radio Science 38(4) (art no 8067) Johannessen JA, Le Traon PY, Robinson I, Nittis K, Bell MJ, Pinardi N, and Bahurel P (2006) Marine environment and security for the European area – toward operational oceanography. Bulletin of the American Meteorological Society 87: 1081-+ Jordan RL (1980) The Seasat-A synthetic aperture radar system. IEEE Journal of Oceanic Engineering OE- 5(2): 154–163 Jorgensen R (2003) The effects of swimbladder size, condition and gonads on the acoustic target strength of mature capelin. ICES Journal of Marine Science 60: 1056–1062 Kloser R and Horne JK (2003) Characterizing uncertainty in target-strength measurements of a deepwater fish: orange roughy (Hoplostethus atlanticus). ICES Journal of Marine Science 60: 516–523 Korneliussen RJ and Ona E (2002) An operational system for processing and visualizing multi-frequency acoustic data. ICES Journal of Marine Science 59: 291–313 Laxon S, Peacock N, and Smith D (2003) High interannual variability of sea ice thickness in the Arctic region. Nature 425(6961): 947–950 Liu WT (2002) Progress in scatterometer application. Journal of Oceanography 58: 121–136 Liu WT (2003) Scientific Opportunities Provided by Seawinds in Tandem. JPL Publication 03-12, Jet Propulsion Laboratory, Pasadena, 38 pp Lorance P and Trenkel VM (2006) Variability in natural behaviour, and observed reactions to an ROV, by mid-slope fish species. Journal of Experimental Marine Biology and Ecology 332: 106–119 Malanotte-Rizzoli, P (ed.), (1996) Modern Approaches to Data Assimilation in Ocean Modeling. Elsevier Oceanography Series, 455 pp Mazauric V, Berger L, and Trenkel V (2006) Biomass estimation using the new scientific multibeam echo sounder ME70. ICES CM 2006/I:18

5

Remote Sensing

143

McClatchie S, Alsop J, and Coombs RF (1996) A re-evaluation of relationships between fish size, acoustic frequency, and target strength. ICES Journal of Marine Science 53: 780–791 Moore RK, and Fung AK (1979) Radar determination of winds at sea. Proceedings of the IEEE 67: 1504–1521 Ona E (1990) Physiological factors causing natural variations in acoustic target strength of fish. Journal of Marine Biological Association of the United Kingdom 70: 107–127 Ona E (1994) Recent Developments of Acoustic Instrumentation in Connection with Fish Capture and Abundance Estimation. Ferno¨ A and Olsen S, Marine Fish Behaviour in Capture and Abundance Estimation 200–216. Oxford, Fishing News Books, Blackwell Science Ltd Ona E (2003) An expanded target-strength relationship for herring. ICES Journal of Marine Science 60: 493–499 Ona E and Mitson RB (1996) Acoustic sampling and signal processing near the seabed: the deadzone revisited. ICES Journal of Marine Science 53: 677–690 Ona E, Dalen J, Knudsen HP, Patel R, Andersen LN, and Berg S (2006) First data from sea trials with the new MS70 multibeam sonar. Journal Acoustic Society of America 120: 3017–3018 Ona E, Godo OR, Handegard NO, Hjellvik V, Patel R, Pedersen G (2007) Silent research vessels are not quiet. Journal of the Acoustical Society of America 121: EL145–EL150 Onsrud MSR, Kaartvedt S, Rostad A, and Klevjer TA (2004) Vertical distribution and feeding patterns in fish foraging on the krill Meganyctiphanes norvegica. ICES Journal of Marine Science 61: 1278–1290 Onsrud MSR, Kaartvedt S, and Breien MT (2005) In situ swimming speed and swimming behaviour of fish feeding on the krill Meganyctiphanes norvegica. Canadian Journal of Fisheries and Aquatic Sciences 62: 1822–1832 Roberts JM, Brown CJ, Long D, and Bates CR (2005) Acoustic mapping using a multibeam echosounder reveals cold-water coral reefs and surrounding habitats. Coral Reefs 24: 654–669 Rountree RA, Gilmore RG, Goudey CA, Hawkins AD, Luczkovich JJ, and Mann DA (2006) Listening to fish: applications of passive acoustics to fisheries science. Fisheries 31: 433-+ Simmonds J and MacLennan DN (2005) Fisheries Acoustics. Blackwell Science, Oxford Sund O (1935) Letters to the editor. Nature 135: 953 Tenningen E, Churnside JH, Slotte A, and Wilson JJ (2006) Lidar target-strength measurements on Northeast Atlantic mackerel (Scomber scombrus). ICES Journal of Marine Science 63: 677–682 Thompson CH and Love RH (1996) Determination of fish size distributions and areal densities using broadband low-frequency measurements. ICES Journal of Marine Science 53: 197–201 Toresen R, Gjøsæter H, and Barros de P (1998) The acoustic method as used in the abundance estimation of capelin (Mallotus villosus Mu¨ller) and herring (Clupea harengus Linne´) in the Barents Sea. Fisheries Research 34: 27–37 Traynor JJ (1996) Target-strength measurements of walleye pollock (Theragra chalcogramma) and Pacific whiting (Merluccius productus). ICES Journal of Marine Science 53: 253–258 Trevorrow MV (2005) The use of moored inverted echo sounders for monitoring meso-zooplankton and fish near the ocean surface. Canadian Journal of Fisheries and Aquatic Sciences 62: 1004–1018 Wilson CD (1998) Field trials using an acoustic buoy to measure fish response to vessel and trawl noise. The Journal of the Acoustical Society of America 103: 3036 Wilson CD and Boehlert GW (2004) interaction of ocean currents and resident micronekton at a seamount in the central North Pacific. Journal of Marine Systems 50: 39–60 Winther NG and Evensen G (2006) A hybrid coordinate model for shelf sea simulation. Ocean Modelling 13: 221–237

144

O.R. Godø and E. Tenningen

Zagaglia CR, Lorenzzetti JA, and Stech JL (2004) Remote sensing data and longline catches of yellowfin tuna (Thunnus albacares) in the equatorial Atlantic. Remote Sensing of Environment 93: 267–281 Zedel L, Patro R, and Knutsen T (2005) Fish behaviour and orientation-dependent backscatter in acoustic Doppler profiler data. Ices Journal of Marine Science 62: 1191–1201

Chapter 6

Quantitative Research Surveys of Fish Stocks Kenneth G. Foote

6.1 Introduction Fisheries research surveys are designed and conducted for a variety of purposes. Some examples are those of defining geographical limits of distribution, describing diurnal movements or seasonal migrations, and determining abundance. The main purpose considered here is that of determining abundance. Quantitative research surveys of fish stock abundance are carried out to avoid the well-known shortcomings of fishery-dependent methods. Five generic survey types are described: egg and larval survey, mark-recapture experiment, fish capture survey, acoustic survey, and optical survey. Methods of direct observation (Godø 1998) are emphasized. Common to each of these generic surveys, in addition to overall aim, is the use of computing resources. This varies from survey type to survey type, because of the kinds of operations that are performed and quantities of data to be treated. Typical operations involving computers are listed. (a) Data collection The primary data are often collected with instruments whose control and operation is rapid, repetitive, and complicated too. A dedicated processor or even specially built computer is commonly used. The scientific echo sounder is an example. (b) Data preprocessing The numbers of certain kinds of data, especially acoustic and optical data, may be so large that preprocessing in real time is necessary. (c) Data display In operating some common instruments in research surveys, such as trawl-monitoring instruments and echo sounders, real-time display of data is an end in itself. Important information is conveyed by the image. Presentation of this is generally controlled by a computer and effected on an electronic screen.

K.G. Foote (*) Woods Hole Oceanographic Institution, Woods Hole, MA 02543, USA

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_6, Ó Springer ScienceþBusiness Media B.V. 2009

145

146

K.G. Foote

(d) Data logging Survey data are often so numerous that the only rational way to store these during collection, if after preprocessing, is through the mass storage devices associated with digital computers. Retrieval of data, as for postprocessing or analysis, is similarly effected by means of digital computers. (e) Data postprocessing and visualization These tasks may be necessary for one or more reasons. Quality control of collected data prior to their analysis may be expedited by assembly of frequency distributions, computation of statistics, identification of data outliers, and visualization. Each of these operations is generally undertaken with a computer. (f) Data analysis Derivation of estimates of abundance may involve relatively simple operations with modeling, as in egg and larval surveys or markrecapture experiments, or more complicated operations, as in reducing measurements of fish density along line transects to yield an estimate of abundance over an area. Use of a digital computer undoubtedly simplifies conduct of the mathematical operations, but may also be necessary because of the data volume. In every case where spatial data are involved, geostatistical analysis may be useful (Cressie 1991, Petitgas 1993, Rivoirard 1994, Rivoirard et al. 2000). The structure function, e.g., covariance function or variogram, is first computed for the experimental data, then modeled to characterize the observed spatial distribution. This may be used for interpolation or extrapolation, as in mapping the distribution where it is not directly measured. It may also be used in estimating the variance of the abundance estimate. Both mapping of the spatial distribution and estimation of the variance are computationally intensive operations. A spectrum of demands on processor type and computing power is thus made by the five generic surveys. However, the digital computer is essential to each survey type. Data quantities Data rates vary with survey and season. In 1990, data rates associated with state-of-the-art acoustic instruments used in fish surveying, namely scientific echo sounders, were of order 1–10 kb/s. Now, data rates associated with scientific multibeam echo sounders are of order 1 MB/s. This is comparable to data rates associated with operational optical systems that collect data in the form of images. Processor capabilities are such that data quantities and rates are not generally limiting factors, but given that the ultimate aim is to extract information from data at an early time, preprocessing is being pursued vigorously to accelerate this process. Ensuring quality is a constant consideration in all processing operations, which may cause postponement to the postprocessing stage, when collateral data can aid in the interpretation of the primary data. Ecosystem context Contributions of fisheries research surveys to ecosystem understanding are also recognized in this work, with ultimate applications in ecosystem monitoring and management. Consequently, the survey types

6

Quantitative Research Surveys of Fish Stocks

147

include tools and methods that can be applied to other ecosystem players, e.g., zooplankton, shellfish, squid, and macroalgae, and to their habitats as well, including both water column and bottom. Throughout, reference is generally made simply to fish, but this can be understood much more generally as encompassing a range of aquatic organisms. Terminology It is noted at the outset that the term abundance is used somewhat differently within the fields of fisheries and plankton acoustics. In fisheries acoustics, and as intended here, abundance refers to a global measure of quantity, e.g., total number of individual fish in a stock, or number of fish in a particular year- or size-class of that stock, or corresponding biomass. In plankton acoustics, abundance usually refers to a local measure of quantity, e.g., number of organisms per unit volume, or number of organisms of a certain species and developmental stage per unit volume, or corresponding biomass per unit volume. In the terminology of fisheries acoustics, abundance as measured in plankton acoustics is a measure of numerical density of the target organism. There is a presumption that this measure is characteristic of a larger water mass, and when multiplied by the volume of this water mass yields a cumulative measure of abundance or biomass corresponding to the plankton stock in the geographical area of observation.

6.2 Egg and Larval Surveys Given sufficient information about reproductive biology and early life history, the abundance of the spawning component of a number of fish stocks may be estimated by egg and larval survey. This involves sampling by physical capture of eggs and larvae over the region of occurrence, performance of laboratory measurements to determine biological quantities related to adult fecundity inter alia, and simple estimation of spawning population size by substitution of measured quantities in an equation. Lo et al. (in press) have recently written a review of the subject. Sampling requirements for performing a survey are: (i) definition of the time of spawning and region of occurrence of eggs and larvae; (ii) adequate spatial and temporal sampling; (iii) quantitative knowledge of the volume associated with each individual sample. It is essential that eggs and larvae of the target species, including developmental stage, be identified when analyzing samples. It must be possible to determine egg mortality and incubation rates, or stage duration. Knowledge of adult fecundity is assumed. Allowance for advection must generally be made. Some representative studies aiming to elucidate critical factors for performing an egg and larval survey are presented in Ellertsen et al. (1989), Fossum and

148

K.G. Foote

Moksness (1993), Hempel and Hempel (1971), Munk and Christensen (1990), Sundby (1991), Sundby et al. (1989), Talbot (1977), and Westga˚rd (1989). In Fossum and Moksness (1993), for example, hatching period is estimated. The efficiency of gear in sampling larvae is specifically addressed in Munk (1988) with respect to herring (Clupea harengus) and in Suthers and Frank (1989) with respect to cod (Gadus morhua). A rather recent technological development is the Continuous Underway Fish Egg Sampler (CUFES) (Checkley et al. 1997). In this operational system, eggs and other small organisms are sampled by a high-volume submersible pump attached to the vessel hull. The organisms are concentrated on board and passed through an Optical Plankton Counter (OPC) (Herman 1988, 1992) [Section 6.6.1] for enumeration. Physical subsamples are collected downstream from the OPC to enable species composition to be determined and the output of the OPC to be interpreted. The Laser Optical Plankton Counter (LOPC) (Herman et al. 2004) [Section 6.6.2] can be used in place of the OPC, enabling imaging with automated classification to be performed. CUFES has been applied in early studies to map the distribution of Atlantic menhaden eggs off the coast of North Carolina and to assess the spawning stocks of northern sardine and anchovy off California and South Africa, respectively (Checkley et al. 1997). The eggs of northern anchovy (Engraulis mordax) and Pacific sardine (Sardinops sagax) have been the object of surveys off central and southern California (Checkley et al. 2000). Population size is computed according to one of three methods. These depend on determination of annual egg production, daily reduction in fecundity, and daily egg production, respectively. Corresponding equations are quite simple. Details on the methods, as well as examples of successful pelagic egg surveys and demersal egg surveys, are given in Gunderson (1993). A particular example of the daily-egg-production method is given for Cape anchovy (Engraulis capensis) in Hampton et al. (1990). Total production of fertilized eggs has been both measured and calculated for cod and plaice (Pleurnectes platessa), establishing the usefulness of plaice egg surveys for stock assessment (Heessen and Rijnsdorp 1989). Example of larval abundance surveys are given for herring in Anthony and Waring (1980), Burd and Holford (1971), Hempel and Schnack (1971), Lough et al. (1985), Saville (1971), and Stevenson et al. (1989). Data quantities are quite modest. However, as with other survey types involving spatial sampling, a geostatistical analysis (Petitgas 1993, Rivoirard et al. 2000) may be computationally intensive. Particular aims of such an analysis may be determination of structure through a generally anisotropic covariance function or variogram, and use of this in mapping the distribution of eggs and larvae and in estimating the variance of the mean abundance estimate due to coverage of the region of occurrence in relation to observed distributional properties.

6

Quantitative Research Surveys of Fish Stocks

149

6.3 Mark-Recapture Experiments Mark-recapture experiments on fish have a historical precedent in demographic statistics on humans, which is cited by Ricker (1975). In applications to fish, mark-recapture experiments are typically used to determine exploitation rate as well as total fish population. In essence, a number of fish are marked, as by tagging, and released. After allowing a suitable time for mixing, the subsequent fish catch is registered and number of recaptures identified. Under certain simplifying assumptions, the population abundance is the ratio of total catch to rate of exploitation, where this rate is estimated by the ratio of the number of recaptures to number of marked fish. This so-called Petersen method is described in detail by Ricker (1948) and Seber (1973), among others. Ricker and Seber each enumerates six assumptions that must be fulfilled for application of the mentioned, simpler formula. These are essentially the same, differing principally in the degree of explicitness of the assumptions of mortality (Ricker 1948) and equal sampling probabilities in first and second samples (Seber 1973). In accordance with the summary advice given in Gunderson (1993) on biological aspects of survey design, availability, sampling gear, and vulnerability and selectivity, these assumptions should be verified. For the Petersen method, studies would have to determine the following (Ricker 1975): (i) natural mortality due to marking; (ii) degree to which marks are lost; (iii) vulnerability of marked fish to catching; (iv) degree to which mixing of marked fish with the population of unmarked fish is uniformly random, or degree to which fishing effort is proportional to density; (v) degree to which marks are detected among recaptured fish; (vi) degree to which recruitment to the catchable population during the time of recapture is negligible. Effects of violations in the assumptions are considered by the cited authors, as are more general cases of inference about fish populations. In particular, the following generic cases are considered: constant and variable survival rates, closed populations with both single-mark release and multiple releasing, and open populations with mark release before or during the sampling period. Jones (1976) gives an accessible, detailed account of the use of mark-recapture data for abundance estimation, as well as for determination of survival rates, growth, and movement parameters. Practical examples on the role of mark-recapture experiments in fishstock abundance estimation are found in, for example, Aasen (1958), Dragesund and Jakobsson (1963), and Dragesund and Haraldsvik (1968). Exemplary studies on the tagging or marking process itself are found in Nakashima and Winters (1984), Parker et al. (1990), and Wetherall (1982), among others.

150

K.G. Foote

6.4 Fish Capture Surveys Fish capture surveys aim to estimate the abundance of a stock or a component of this. There are many ways to catch fish, but none that is unbiased (Godø 1998, Fre´on and Misund 1999). In the case of plankton, the bias may appear to be less because of the reduced mobility of such organisms, but this does not stand up to close inspection, as in the cases of krill (Kasatkina 1997) and smaller organisms, e.g., mesozooplankton, observed to be reacting to a small gauze recorder originally attached to the Video Plankton Recorder (Davis et al. 1992a) [Section 6.6.3]. The large number of physical sampling devices – Wiebe and Benfield (2003) list more than 150 – witnesses to a commitment by the community to improve the sampling characteristics. Notwithstanding admitted selectivity in catching, these devices have contributed significantly to knowledge of plankton distribution and abundance. Some of the longest time series derive from these, e.g., surveys with the Continuous Plankton Recorder (Hardy 1926, Warner and Hayes 1994, Planque and Batten 2000); zooplankton surveys about Iceland with Hensen net, opening diameter 73 cm, from 1961 to 1991, and WP-2 net, opening diameter 57 cm, in 1992 and 1993, both with 200-mm mesh (Astthorsen and Gislason 1995); and Marine Resources, Monitoring, Assessment, and Prediction (MARMAP) sampling program (Sherman et al. 1987, 1996, Kane 1993, 1997), using paired bongo nets of mesh sizes 333 and 505 mm. In the following, the emphasis is on research trawl surveys of fish, simply called trawl surveys. These aim to estimate abundance in a relative sense over a specific geographical area by means of a series of trawl hauls. Trawl locations may be distributed over the region according to a number of patterns. Basic examples are uniform or random grids, both with and without stratification. The major concern in nearly every trawl survey is that of representativity of sampling, alternatively, avoidance or minimization or understanding of bias (Fre´on and Misund 1999). Determining how representative trawl samples are involves auxiliary studies that often require more effort than that of the survey itself. Such studies examine the phenomena of size selectivity, vulnerability, and fish behavior, including avoidance reactions to vessel and gear. As these generally depend on species, size, biological state, season, depth, and light level, among other influences, quantifying the fundamental sampling process is indeed a formidable undertaking. Exemplary studies focusing on some of the mentioned problems for bottom trawls are described in Enga˚s and Godø (1986, 1989a,b), Godø and Enga˚s (1989), Godø and Sunnana˚ (1992), Godø and Walsh (1992), Harden Jones et al. (1977), and Ona and Godø (1990). The problems of changes in vertical and horizontal distributions are addressed in Godø and Wespestad (1993), for example. The cited study of Godø and Walsh (1992) has led to a change in trawl type, from rubber bobbin ground gear to heavy rockhopper ground gear. Under certain conditions, trawl surveys can yield a relative measure of abundance that may serve as an index. If the trawl performance can

6

Quantitative Research Surveys of Fish Stocks

151

additionally be quantified, so that the effective area swept by a bottom trawl or effective volume swept by a pelagic trawl is known for each individual haul, and the corresponding efficiency is known, then a trawl survey should be capable of yielding absolute estimates of abundance. Godø (1994) summarizes factors affecting the reliability of abundance estimates of bottom fish. Hjellvik et al. (2004) explain how the reliability can be improved by a close examination of the nature of variability in bottom trawl survey data. Vital data on trawl performance are provided by cable-free instrumentation, such as that of the SCANMAR System 400 and SIMRAD Integrated Trawl Instrumentation (ITI) system. Such instrumentation includes sensors that transmit data on gear depth, trawl velocity including sinking or rising rates, distance between trawl doors or wings, height of trawl opening, temperature at operating depth, and cod-end filling, among other things. The so-called trawleye is also part of the system; it monitors fish concentration in the vicinity of the trawl opening, as well as bottom contact, providing data for real-time display in echogram form on a color graphics monitor. Cable-free instruments operate on command to save battery power, using an acoustic link for reception of commands and transmission of data. Trawl surveys have traditionally been directed at bottom fish, but may apply equally well to pelagic fish, given suitable conditions. Certainly trawling of pelagic fish is important in acoustic surveys, hence the expressed concern about representativity in sampling by both kinds of trawling gear. In the case of pelagic trawls, the crucial issue of capture efficiency is described in Hylen et al. (1995). This is discussed further with respect to an acoustic index (Haug and Nakken 1977) and a catch-rate index (Randa 1984) and Nakken et al. (1995). Astthorsen et al. (1994) describe the derivation of abundance indices for juvenile cod in Icelandic waters by pelagic trawling and present results as a time series over the period 1970–1992. Just as the design of trawl surveys may vary from uniform sampling regimes to those seeking randomness, so do the precise analyses of trawl survey data vary. When direct use is made of the spatial location of trawls, by means of geostatistics, results may be expressed in terms of structure functions, maps of the distribution, and global estimate of abundance with variance estimate. When spatial information is ignored, computation of total abundance may merely involve summing catch numbers, possibly weighted by the respective stratum area or by the total survey area. During the International Workshop on Survey Trawl Mensuration held in St. John’s, Newfoundland, in 1991 (Walsh et al. 1993), 76 factors were listed as influencing survey trawl performance and fish capture efficiency. Should a sufficient number of the more important of these be quantified in the future and models for their compensation be developed, then this information might be used to improve local estimates of fish density. Abundance estimates over an area would be correspondingly improved, as illustrated in Hjellvik et al. (2004).

152

K.G. Foote

6.5 Acoustic Surveys Quantitative acoustic surveys are well-established, but improvements in electronics, transducers, and computing power are compelling qualitative changes in such surveys. Examples are due to increasing bandwidth of scientific echo sounders, development and use of multibeam sonars that provide a calibrated water-column signal, and use of sidescan sonar for benthic organisms. In addition to being quantitative, acoustic methods are generally non-invasive, remote, rapid, synoptic, and capable of sensing multiple scales. Depending on the particular sonar, they may be capable of observing the water column, bottom, and sub-bottom, including both biological organisms and habitat.

6.5.1 Generic Methods Prerequisites for conventional acoustic surveys are a transducer or sonar, electronics for controlling transmission and reception, a display device, a platform to beam these, namely a vessel or towed vehicle, and a digital computer. Essential functions of the instruments are those of echo sounding, data display, data processing, and data storage. The associated echo sounder or sonar system is generally operated in a calibrated state. Details on acoustic and data processing are included below. The survey area is spanned by a series of line transects. A major design aim in choosing transects is acquisition of maximal information about the spatial distribution of fish (Foote and Stefa´nnson 1993). Typical constraints are the known or suspected boundaries of the stock, possible large-scale movements of the fish during the course of the survey, diurnal patterns of fish movement, availability of ship time, possible need to conduct trawling or to perform hydrographic or other oceanographic measurements concurrently with the acoustic survey, among other considerations. Prior information about the fish stock is always valuable, as in allocation of sampling effort by stratification. Current practice is reviewed by Simmonds et al. (1992) and Rivoirard et al. (2000). In a particular case, that of Norwegian spring-spawning herring when wintering in a fjord system, the fish are generally surveyed by scientific echo sounder in two or more stages. Initially, the fjord areas are covered by a large-scale zigzag design (Fig. 6.1). The primary purpose is to locate significant quantities for further, finer-scale measurement. Identified sub-areas or significant strata are surveyed subsequently according to an ad hoc design that aims to maximize information about the particular concentrations. In December 1993, for example, the predominant concentration of herring was in central Ofotfjorden. Its second coverage was done with a regular, fine-scale grid (Fig. 6.2). The basic measurement is that of the echo signal recorded at a particular location. This contains information about the density of scatterers as a function of distance from the acoustic source and receiver, which are usually collocated, as assumed here. The intensity of the received signal is usually processed with some form of range compensation, then displayed. When successive or adjacent

6

Quantitative Research Surveys of Fish Stocks

153

Fig. 6.1 Large-scale zigzag survey designed applied initially in acoustic abundance surveys of Norwegian spring-spawning herring wintering in the Ofotfjorden - Tysfjorden system in the 1990s

Fig. 6.2 Fine-scale survey design composed of equally spaced north-south parallel transects and applied to the predominant herring concentration in central Ofotfjorden, December 1993

154

K.G. Foote

echo signals are aligned, as in a scientific echo sounder [Section 6.5.2.1] or multibeam sonar [Section 6.5.2.2], respectively, the resulting echogram presents an acoustic image of scatterers along or across the transect. This is illustrated for data from a scientific echo sounder in Fig. 6.3. Quantitative processing of the same information may specify the acoustic density of particular echogram features, or scatterers, as a function of location, including depth in the case of ordinary vertically oriented echo sounder beams or range in the case of sonar beams of general orientation. Because of the overall aim of stock abundance estimation, processing of echo sounder data is performed systematically. An important part of data postprocessing is allocation of the echo record to particular scatterer classes. Echo quantities are assigned to fish species and size classes. Auxiliary information useful in making the assignment includes composition of trawl hauls, data from oceanographic sensors, e.g., salinitytemperature-depth (STD) sonde [Section 6.8.1], acoustic Doppler velocimeter [Section 6.8.2], appearance of the echogram, and knowledge of fish biology. Results of this allocation are stored in retrievable form, usually in a database, together with metadata such as location and time of data collection. Based on knowledge of the fish species and size, as well as characteristics of the transducer and echo sounder system, the acoustic measures of fish density are converted to biological measures during data analysis. In this way, the numerical density of fish is computed, yielding the number of fish per unit volume or number of fish per unit area as projected onto a horizontal surface. In the case of the widely used, general method of echo integration, this is accomplished by dividing the area backscattering coefficient by the mean backscattering cross section (Foote 1993, Foote and Stanton 2000). Other operations may be necessary with this technique if, for example, the sampling volume is different from the nominal value (Foote 1991) or acoustic extinction is significant (Foote 1990, 1999, Foote et al. 1992). Once the fish density has been determined along the line transects spanning the survey area, the total abundance may be computed (Foote and Stefa´nsson 1993). This is the integral of numerical density over the entire surveyed area or volume. Since only particular information is generally available, gathered along the ship track, interpolation is generally necessary. This may however be accomplished implicitly, as by assuming that each measurement of density is equally representative of the mean as any other. The stock abundance is generally estimated from the density measurements by numerical integration. This number is often partitioned by age or size class, thus resulting in a series of abundance estimates that pertain to the several classes. An exposition of the general technique is given in Gerlotto and Ste´quert (1983). This is based on survey situations exemplifying high degrees of homogeneity or heterogeneity in spatial distribution. Johannesson and Mitson (1983) describe a wide range of survey types, while also presenting the basic methodology in heuristic fashion. A detailed example is worked for herring in Rivoirard et al. (2000). The outlined measurements, signal processing, data storage, retrieval, data postprocessing, and data analysis operations are computationally intensive.

6

Quantitative Research Surveys of Fish Stocks

155

Fig. 6.3 Exemplary echograms of Norwegian spring-spawning herring during wintering, derived from the Simrad EK500/38-kHz scientific echo sounder and displayed by the Bergen Echo Integrator. Top frame: Herring aggregation in Vinjefjord, December 1988, depth range 125–225 m, total sailed distance about 1.4 nautical miles. Bottom frame: Herring aggregation in Ofotfjorden, December 1992, depth range 0–500 m, total sailed distance 5 nautical miles

156

K.G. Foote

6.5.2 Sonar Systems Line-transect measurements of the numerical density of fish and other aquatic organisms have been made traditionally by scientific echo sounder. However, other acoustic systems have and are being used too, if still to a rather limited degree or proscribed range of applications, notwithstanding some noteworthy early applications of sonar to count fish schools (Hewitt et al. 1976). To understand these systems and their potential applications, some basic operating principles are described in this section.

6.5.2.1 Scientific Echo Sounder In its simplest form, an echo sounder is a box of electronics that controls transmission of an underwater acoustic signal by a transducer and reception of echoes by the same device. The primary purpose of an echo sounder is to determine the range to interesting scatterers or targets, for example, the bottom or fish. It generally does this through transmission of a short signal, called a ping, conversion of the echo pressure fluctuations to an electrical signal, which can then be amplified and filtered, and display of the electrical echo signal, as on a paper chart or electronic screen. When the echo amplitude is represented by brightness or color and each time series of amplitudes is displayed versus time or range, and aligned with the preceding time series, the display is called an echogram [Section 6.5.1] (Fig. 6.3). A scientific echo sounder performs the same function as an ordinary echo sounder, but with high sensitivity, large dynamic range, avoidance of receiver saturation, control of the signal throughout its passage through the electronics both in transmission and reception, and stable operation. Attached transducers are similarly sensitive and robust, with special directional characteristics. The result of this is provision of a high-quality quantitative echo signal, historically called a calibrated output signal. This signal makes possible repeatable quantitative use of echo information, as in echo integration, described further below. Design principles are documented in, for example, Bodholt et al. (1989), Furusawa (1991), Furusawa et al. (1993), and Andersen (2001). Calibration of scientific echo sounders is addressed in Section 6.5.4. The calibrated output signal usually incorporates a particular form of timevaried gain (TVG) applied as analog or electronic amplification during data collection (Mitson 1983, Medwin and Clay 1998) or range compensation applied digitally after data collection. In effect, the echo intensity is amplified according to the function r210ar/5, where r is the range, r=ct/2, c is the speed of sound, t is the echo time relative to the start of signal transmission, and a is the absorption coefficient expressed in decibels per unit distance. In the logarithmic domain, the TVG function is 20 log r + 2ar. This TVG function compensates for scattering by a layer, so that the echo strength is independent of the layer depth, the numerical density and other factors remaining constant.

6

Quantitative Research Surveys of Fish Stocks

157

Echo integration The value of echo intensity, following the above range compensation, is proportional to the volume backscattering coefficient (Medwin and Clay 1998); the constant of proportionality is generally determined by calibration [Section 6.5.4]. When the volume backscattering coefficient is integrated over a range interval, the resulting quantity is the area backscattering coefficient. This is often averaged over a number of pings corresponding to an interval of sailed distance. According to the fundamental equation of echo integration, the area backscattering coefficient is equal to the product of the numerical density of scatterers and their mean backscattering cross section. A second common form of TVG is that of r4102ar/10 in the intensity domain or 40 log r + 2ar in the logarithmic domain. This TVG function is useful for eliminating the range dependence in echoes from resolved scatterers. In the simplest form of a scientific echo sounder, the transducer is regarded as an integral device operating with a single beam. However, an echo sounder transducer generally consists of an array of rather small elements, which can be electrically divided into multiple sections allowing different beams to be formed in the receiver. Dual beams are sometimes formed by a transducer divided into a central circular core and surrounding, outer ring. By transmitting with the central core, and receiving echoes on both parts, it is possible to make a direct measurement of target strength without knowing the particular angular position of an isolated single target (Ehrenberg 1974, 1979). Split beams may be formed by dividing the transducer into quadrants and forming corresponding beams in the receiver. Four half-beams are formed in pairs: port and starboard, and fore and aft. The electrical phase differences between corresponding half-beams can be translated into physical angles from which the target strength can be determined, hence the beam pattern value in the target direction. The measured echo strength can thus be compensated for the beam pattern and the target strength determined directly without additional processing (Ehrenberg 1979, Foote et al. 1986). A special combined dual- and split-beam system was assembled for research purposes (Traynor and Ehrenberg 1990). The transducer array was divided into a total of five parts, including a small core and a large outer ring divided into quadrants. Transducers used in scientific echo sounding are usually quite directional. Typical beamwidths, which measure the one-way angular sensitivity of the central, most sensitive part of the transducer beam, are typically in the range from about 28 to 128, with 78 representing a common standard for many systems used in fish stock abundance surveys. Traditional scientific echo sounders are narrowband in operation, with a bandwidth expressed as a rather small percentage of the center, operating frequency. This is usually 10% or less. It was realized rather early that frequency diversity, or bandwidth, confers distinct benefits, potentially including classification (Holliday 1980), as illustrated in a four-frequency system with operating frequencies between 0.5 and 3 MHz (Holliday and Pieper 1980).

158

K.G. Foote

Scientific echo sounder systems currently operate up to six or more narrowband transducers simultaneously, enabling echo data to be collected at multiple frequencies, e.g., 18, 28, 30, 38, 50, 70, 88, 105, 120, 200, 208, 420, 714, and 720 kHz. This increase in bandwidth is being exploited in attempts to perform acoustic classification without reference to fish-capture or other data (Korneliussen and Ona 2002, Jech and Michaels 2006). The same scientific echo sounders can also operate transducers at the same frequency synchronously, but not simultaneously. In the case of a particular research system, the Multiple Acoustic Profiling System (MAPS) (Holliday et al. 1989), 21 narrowband transducers are excited. Their frequencies are logarithmically spaced over the total band 100 kHz – 10 MHz. In another research system, that of the Broadband Acoustic Scattering Signatures (BASS) system (Foote et al. 2005a), seven octave-bandwidth transducers continuously span the total band 25 kHz – 3.2 MHz. A third research system, that of the BIOlogical, Multi-frequency Acoustical, Physical and Environmental Recorder (BIOMAPER II) (Wiebe et al. 2002), consists of a mix of narrowband and broadband transducers spanning the approximate total range 24 kHz – 1.4 MHz. Examples of acoustic abundance surveys carried out with scientific echo sounders are described by Haug and Nakken (1977) for 0-group fish in the Barents Sea, including cod, saithe, haddock, polar cod, redfish, capelin, and herring; Johannesson and Robles (1977) for Peruvian anchoveta; Mais (1977) for northern anchovy along the California coast; Mathisen et al. (1977) for juvenile sockeye salmon in lakes; Midttun and Nakken (1977) for blue whiting northwest of the British Isles and capelin in the Barents Sea; Thorne (1977) for Pacific hake and herring in Puget Sound and for herring in Carr Inlet in southeast Alaska. The frequencies are mostly ultrasonic, an exception being the earlier surveys conducted by Mais (1977) at 12 kHz, later surveys being performed at 38 kHz. Some additional examples are described by Jakobsson (1983) for Icelandic summer spawning herring; Traynor and Nelson (1985) for walleye pollock (Theragra chalcogramma) in the eastern Bering Sea; Bailey and Simmonds (1990) for herring in the North Sea; Hampton et al. (1990) for spawning Cape anchovy off South Africa; Wespestad and Megrey (1990) for walleye pollock in the eastern Bering Sea and Gulf of Alaska; and herring in the Gulf of Maine and Georges Bank (Overholtz et al. 2006). Stocks of razor clam and surf clam along the coast of central Chile have also been assessed by scientific echo sounder (Tarifen˜o et al. 1990). Euphausiids have also been surveyed by scientific echo sounder. Species include Meganyctiphanes norvegica off Georges Bank (Greene et al. 1988) and Euphausia superba in the Southern Ocean (Hewitt and Demer 2000, Lawson et al. 2004), among other species. 6.5.2.2 Multibeam Sonar A principal disadvantage of scientific echo sounders with narrow-beam transducers is their rather small sampling volumes: echo statistics can be sparse, and it can be difficult to assess behavioral responses of fish, which may bias

6

Quantitative Research Surveys of Fish Stocks

159

measures of numerical density. With advances in electronics and computing power, it has become possible to form many beams spanning a broad angular sector, aiding spatial resolution of targets and assessment of behavioral effects at the same time that these can be quantified. The power of multibeam sonar is suggested by the nominal specifications of several multibeam sonars that provide the water-column signal. The Simrad SM2000 multibeam echo sounder forms 128 beams at each of two frequencies, 90 and 200 kHz, over an angular sector from 90 to 1508, depending on the particular transducer configuration. The RESON SeaBat 8101 forms 101 beams at 240 kHz spanning a 150-deg sector. The nominal beamwidth of an individual beam of either of these systems is 1.5  1.58. The bandwidth is rather narrow. Multibeam echo sounders can be calibrated [Section 6.5.4]. Multibeam sonars have seen early application to the quantification of fish schools (Misund et al. 1992) and to the observation of vessel-induced effects on fish behavior (Misund and Aglen 1992, Misund 1993, Gerlotto et al. 1999, Soria et al. 1996). Swimming speed of schools has been measured by multibeam sonar (Misund 1993, Hafsteinsson and Misund 1995). Behavior of pelagic fish during trawling has also been observed (Misund and Aglen 1992). Interaction effects of predators and prey have been observed (Nøttestad and Axelsen 1999, Axelsen et al. 2001, Benoit-Bird and Au 2003). Other measurements as well as limitations have been described (Reid 2000). Recently two multibeam sonar systems have been designed and built to operate over a broad bandwidth, nominally 70–120 kHz. The Simrad ME70 multibeam echo sounder (Trenkel et al. 2006) can form from 1 to 45 split beams of beamwidths in the approximate range 2–78 in a fan-shaped beam. The Simrad MS70 multibeam sonar (Ona et al. 2006) forms a two-dimensional array of 500 beams, 2520, covering a sector of 608 in the horizontal plane and 458 in the vertical plane. This second sonar enables the three-dimensional structure of fish aggregations to be imaged by a single ping. Design principles for both systems have been presented (Andersen et al. 2006). 6.5.2.3 Sidescan Sonar The acoustic essence of sidescan sonar is embodied in its transducer: a long linear array of elements. This is typically towed with its long axis in the horizontal plane. As a consequence of its geometry and orientation, a fanshaped beam is produced that is very narrow in the horizontal plane and quite broad in the vertical plane transverse to the longitudinal axis of the array. In the case of the Edgetech 272-TD dual-frequency analog towed vehicle, the transducer beamwidths at 105 kHz are 1.28 in the horizontal plane and 508 in the vertical plane. At 390 kHz, the respective beamwidths are 0.5 and 508. There are a total of four arrays, two each on port and starboard. Sidescan sonars are typically towed at some fixed height over the bottom. At 390 kHz, for example, with a working range of 25–50 m, the height would typically be in the range 5–10 m, with the axis of the vertical beam pattern

160

K.G. Foote

oriented below the horizontal by 208 if the target is the bottom or bottom structures, which is common for sidescan surveying. When excited, the array launches an outgoing pulse with the acoustic energy mostly contained within the described fan-shaped sector. It sweeps outwards across the bottom, producing an echo time series. While this may be dominated by the bottom, other surfaces, structures, and water-column scatterers that it intercepts will also give echoes. It has long been appreciated that sidescan sonars can detect fish schools and shellfish (Fish and Carr 1990, 2001), including beds of zebra mussels (Dreissena polymorpha). Individual salmon migrating in the Fraser River near Mission, British Columbia, in September 1995 were detected and counted with sidescan sonar at 100 kHz (Trevorrow 1998). The water depth was 4–12 m; the maximum detection range was 200 m. Herring schools in their shallow spawning habitat, 3–4 m deep, were imaged near Escuminac, New Brunswick, in May 1996 at 100 and 330 kHz. The maximum detection range was 150 m. Detection of such fish targets was reduced due to boat traffic and the presence of air bubbles entrained by breaking waves. Migrating salmon have also been observed with a towed sidescan sonar operating at 330 kHz in the Strait of Georgia near the mouth of the Fraser River near Vancouver, British Columbia, in September 1998 (Trevorrow 2001). The fan-shaped beam of the sidescan sonar, 1.8188, could be steered by mechanical rotation. The steerable transducer was mounted inside a towed vehicle that was towed at a depth of 18 m. Results were considered to agree reasonably with trawl and riverine escapement data. Rather recently sidescan sonar has been used to image and quantify benthic egg clusters of the squid Loligo opalescens in Monterey Bay, California (Foote et al. 2006). Data were collected along transects with 100% overlap of successive port or starboard swaths. The data were mosaicked, enabling egg clusters called mops and beds to be detected. By measuring the area of coverage and knowing, by direct measurement, the mean diameter of an egg capsule, 16 mm, and mean number of eggs per capsule, 150, it was possible to estimate the total number of eggs laid in a particular one-arc-second quadrat. The result was 36.5 million eggs, which represents the potential recruitment of this short-lived species. There is no other objective method to observe this biologically important quantity. Other quantitative imaging applications of sidescan sonar can be envisioned, but limitations in dynamic range will be an obstacle to this. In the case of benthic species, however, a particular advantage of sidescan sonar remains, namely that of habitat-mapping at the same time. A more sophisticated sidescan sonar exploits interferometry by means of multiple arrays at the same frequency and same side of the towed vehicle. These enable the phase of returning signals, hence angle of incidence, to be determined. Since the height over bottom and towing depth, hence depth at nadir, is known, the bottom depth can be determined. Interferometric sidescan sonar is also known as multibeam sidescan sonar.

6

Quantitative Research Surveys of Fish Stocks

161

The combination of bathymetric data and backscattering imagery is potentially powerful in studies of bottom fish in their benthic habitat. Sidescan sonar arrays are also carried on a number of autonomous underwater vehicles, e.g., the Remote Environmental Measuring Units (REMUS) (Allen et al. 1997, Purcell et al. 2000). The possibility of conducting simultaneous surveys of some benthic organisms and their habitat is noted. 6.5.2.4 Acoustic-Lens-Based Sonar Acoustic lenses are exactly analogous to optical lenses, the difference being primarily that between sound and light. A well-designed acoustic lens, like its optical counterpart, focuses the energy incident on it, more efficiently when the lens is large compared to the wavelength. The sound speed in the lens is necessarily different from that in the medium, a condition for refraction. For efficiency, the acoustic impedance, or product of mass density and longitudinalwave sound speed, of the lens is ideally very similar to that of the ambient medium, reducing losses due to refraction and extraneous scattering. If the sound speed of the lens material is less than that of water, then the lens shape will be convex. Many liquids possess this property, e.g., carbon tetrachloride (Clay and Medwin 1977), ethyl and other alcohols (Greenspan 1972), and fluorocarbon liquids (Wade et al. 1975), as does a silicone rubber, General Electric RTV-602, with refractive index of about 1.4 (Folds and Hanlin 1975). If the sound speed of the lens material is greater than that of water, then the lens shape will be concave. Polystyrene and syntactic foam are widely available materials that can be used for such a lens (Folds and Hanlin 1975). Like optical lenses, acoustic lenses are subject to aberrations (Jenkins and White 1957), of which spherical aberration is generally most significant because of use of the lens with narrowband signals. The traditional remedy is employed: use of multiple elements. The same remedy is employed to lessen effects due to differential changes in sound speed, hence refractive index, with temperature. Acoustic lenses are being used in active sonar systems designed for detection and imaging, because the focusing effects beamforming. This is instantaneous. In addition, the particular delay-and-sum method of beamforming (KamgarParsi et al. 1997) does not consume power. Acoustic-lens-based sonars are being applied to high-resolution underwater imaging (Kamgar-Parsi et al. 1997). The potential for surveying fish in rocky habitats may be unsurpassed. The Dual-Frequency Identification Sonar (DIDSON), with operating frequencies of 1.1 and 1.8 MHz, is particularly powerful. At 1.1 MHz it forms 48 beams spaced at the two-way beamwidth of 0.68; at 1.8 MHz, 96 beams spaced at 0.38. The total horizontal, or equatorial, field of view is 298 at both frequencies, with the same nominal two-way elevation, or transverse, beamwidth of 148. The photographic-like quality of its images has been applied to study fish behavior near a hydroelectric power station (Moursund et al. 2003) and to enumerate salmon in Alaskan rivers (Burwen et al. 2004). Typical ranges of operation are 1–30 m.

162

K.G. Foote

A longer-range version of DIDSON has been developed, with operating frequencies of 700 kHz and 1.2 MHz. Data derived from DIDSON in the form of images can be formidable in quantity. Automation is essential to assemble multiple images in a coherent, continuous record. Software has been developed to effect mosaicking of such data (Kim et al. 2005).

6.5.3 Data Visualization and Postprocessing Until about 1990, nearly every major marine research institute that surveyed fish stocks acoustically used its own digital computer system for the postprocesssing of acoustic and related survey data. This generally involved software that was developed for a particular configuration of computer hardware, both in terms of processor types and manufacture. Changing computer generally meant changing software too. The software was, moreover, often written in a low-level language and assembled in a monolithic program, requiring sequential execution of the entire program, if with options or branch points, no matter what the scope of the particular exercise. To remedy this situation and to achieve a much greater degree of flexibility in the postprocessing of survey data, a new system was developed at the Institute of Marine Research, Bergen, beginning in 1988. This system, called the Bergen Echo Integrator (BEI) (Foote et al. 1991), has spawned or inspired commercial versions with worldwide distribution and use. As the design principles of BEI are quite general, and since it may still serve as the model for other new systems, the basic development of BEI is described. It was planned as part of a data network on board a research vessel (Knudsen 1990) (Fig. 6.4).

Weather station

Log

PC datalogger

Satellite communication

Satellite navigator

Manual input of environmental data

Graphics worksation for postprocessing acoustic data

Database processor

PC terminal

LAN

Echo sounder and preprocessor

PC datalogger

Multi-beam sonar fish-school counter

PC datalogger

GPIB CTD profiler

Manual input of fish station data

Fig. 6.4 Elements of a research vessel data network (Knudsen 1990) (reproduced with permission, H. P. Knudsen)

6

Quantitative Research Surveys of Fish Stocks

163

6.5.3.1 User Requirements and Basic Needs At an early stage in system planning, users imposed the following requirements: (i) adequate capacity to process, display, and manipulate data from an echo sounder preprocessor; (ii) machine independence; (iii) high-level programming language; (iv) modular construction; (v) database; (vi) Ethernet local area network (LAN); (vii) full documentation; (viii) user friendliness. Several basic needs were thus identified. Firstly, the system should be user-specified, i.e., the user should make all important decisions. Secondly, the system should be easy to use; if it is not, it will not be used. Thirdly, the system should be capable of expansion, hence the architecture should be extendable and scalable. Finally, full documentation was considered to be essential, thus facilitating further expansion. 6.5.3.2 Computing Environment Hardware The system was designed to serve as a data logger for the Simrad EK500 echo sounder. The daily data rate was of order 100–400 MB. The system was dimensioned to store data continuously for 24 h before cleanup procedures had to be started. Data were displayed in echograms consisting of 650 values of backscattering strength per ping: 500 values from the water column and 150 values from the detected bottom. Echograms would be displayed within 10 s following transformations such as changing the noise threshold, and computations such as re-computing area backscattering coefficients. The raster graphics were to render the original echogram without compromising information. The graphics system had to include a frame buffer with 256 colors from a modifiable color palette. The estimated speed of the system was thus 10 million instructions per second (MIPS), 2 million floating point operations per second (MFLOPS), and input/output (I/O) bandwidth of 2–3 MB/s. Reduced instruction set computer (RISC) architecture was required at the outset of the project. The hardware technology that fulfilled these requirements in 1988 was that of the workstation class of machines. Software The requirement of an open architecture was satisfied by employing internal, nonproprietary standards. The operating system UNIX (Christian and Richter 1994) was chosen for portability, multitasking capacity, virtual memory, non-segmented memory, and networking. The programming language C (Kernighan and Ritchie 1988) was chosen for its support of both structured programming and efficient coding, with powerful interface to UNIX. A relational database INGRES (Date 1987a) was chosen, with standard structured-query language (SQL) (Date 1987b). A user-friendly interface was ensured by choosing that graphical user interface (GUI) that was both nonproprietary and portable, namely the X Window system (Young 1994). External communications Choice of a local area network (LAN) avoided the limitations inherent in the traditional monolithic system with peripherals attached directly to a central processing unit (CPU). The most widely used LAN interface was chosen, namely Ethernet and Transport Control Protocol/ Internet Protocol (TCP/IP). A subset of TCP/IP was selected for data

164

K.G. Foote

acquisition, namely User Datagram Protocol/Internet Protocol (UDP/IP). The advantage of this protocol is that data can be sent to the receiver even during times of temporary overflow, albeit with data loss, which is non-critical apropos of the quantities involved.

6.5.3.3 System Design Data flow High-volume data, such as the raw acoustic data from the echo sounder preprocessor, are stored directly in files, while information on their access is stored in the database. Postprocessed data and other low-volume data, such as navigation data and salinity-temperature-depth (STD) data, are stored directly in the database. They system displays and processes data that are accessed from both database and files. Upon completion of the postprocessing operations, report generators summarize the results. Configuration window Key data on the cruise are specified through the configuration window. These data typically include nation, vessel name, cruise number, cruise objectives, plan, sea area, and personnel. Channel depths for echo integration are included, as is the basic interval of sailed distance over which echo integration is performed. Exemplary layer depths for echo integration in the North Sea, Norwegian Sea, and Barents Sea are 50 m for pelagic fish, and 1, 2, 5, and 10 m for bottom fish as measured from the detected bottom. Typical intervals of sailed distance are 0.1, 0.5, 1, and 5 nm. Windows for data postprocessing (a) Survey grid window This displays the location of cruise data stations, where echo sounder, trawl, oceanographic, and other data are collected. (b) Echogram window This consists of four sub-windows. The main echogram sub-window displays the echogram in a form suitable for interactive postprocessing by the operator. The expanded bottom channel sub-window displays the echogram expanded about the detected bottom, with possibility of operator correction of this by redrawing the bottom line. The color map sub-window specifies the color palette. This may be defined by the operator, but is typically selected from among three choices: gray scale for observing shape, red-blue color scale for gauging signal strength, and dark red-light blue color scale as a compromise of the first two scales, combining the features of shape and signal discrimination. The zoom sub-window allows expansion of arbitrary, operator-selected parts of the main echogram sub-window or expanded bottom channel sub-window. (c) Interpretation window Results of echo integration are stored in the database by scatterer type according to operator delineation of the echogram. This window makes available names of scatterer classes, from database or by operator definition, for assignment of integration values. (d) Target strength window Target strength values of resolved single-target echoes may be selected, compiled, and displayed in a histogram for operatordelimited regions of the echogram.

6

Quantitative Research Surveys of Fish Stocks

165

(e) File selection window Acoustic data may be selected through this window, which provides an alternative means of selecting data, as when navigation data are lacking. (f) Fish station window Biological data derived by tracking or other means may be summarized and displayed in the form of a length distribution and number of measured specimens for each selected species that is represented in the catch data. The selection may be performed in the survey grid window. (g) STD window Profiles of salinity-temperature-depth data are displayed, as designated in the survey grid window.

6.5.3.4 System Implementation Integration techniques It should be possible to activate a subset of the system without activating the whole. Since the system is represented by a set of processes, each of which executes its own task independently of the others, communication among the processes is essential. This is achieved through the communication mechanisms available in the window system itself. Processes announce their entrance to and departure from other window-based processes. Performance issues (a) Input/output bandwidth If mechanisms for database input and output had sufficient bandwidth, all data could be stored in the database. Failing this, data are separated into two categories, according to their respective volume. As already mentioned, high-volume data are stored in files, while low-volume data are stored in the database. Refined data, such as results of echo integration, are stored in the database. (b) Pre-calculation Computationally intensive parts of the system are optimized by pre-calculating mathematical expressions and storing the results in index tables. An example is that of echo integration, in which a value of the volume backscattering strength, which is expressed as a logarithm to base two, is used as an index to a vector of pre-calculated values of area backscattering coefficient. 6.5.3.5 Summary of Special Features Five special features of the Bergen Echo Integrator are listed. (a) The user may delineate arbitrarily shaped regions of the echogram for echo integration, including non-constant-depth intervals, by means of interactive graphics. (b) Results of echo integration may be stored in the database with different degrees of resolution. In terms of sailed distance, these are typically 0.1, 0.5, 1, and 5 nm. In terms of depth, these are nearly arbitrary, but typically lie in the range from 0.1 to 500 m. (c) Errors made in preprocessing may be corrected. Examples include redefinition of the detected bottom and changing the noise threshold.

166

K.G. Foote

(d) Different color maps may be used to aid the extraction of information on shape and signal strength from the echogram. (e) Interconnections of graphical user interfaces, database, and data files are designed to optimize data flow and the exercise of system functionality.

6.5.4 Calibration Calibration may be effected according to a number of methods (Bobber 1970, Urick 1983), but the accepted, most widely used method in fisheries research is that of the standard target (Foote 1982, 1983, Foote et al. 1987, MacLennan 1990, Sawada and Furusawa 1993, Jech et al. 2005). In brief, a standard target is suspended at a known position in the transducer beam, and the echo due to a standard transmission is recorded, processed, and related to the acoustic backscattering cross section of the target. The overall system sensitivity is thus measured. Properties of directionality can be measured by repeating the measurements with the target in different parts of the transducer beam. For scientific echo sounders operating at ultrasonic frequencies, a solid elastic target is typically used. Its backscattering cross section, or target strength in the logarithmic domain, is known a priori by calculation (Faran 1951, Hickling 1962), with typographical errors in basic expressions noted and the given single-frequency formula generalized to an operational form recognizing the system bandwidth (Foote 1982). Broadband echo sounders can also be calibrated with standard targets (Foote et al. 1999, Foote 2000). The same is true of multibeam sonars (Foote et al. 2005b, Foote 2006) and parametric sonars, as well as other sonars operating in the mid-frequency band, 1–10 kHz (Foote et al. 2007).

6.6 Optical Surveys 6.6.1 Optical Plankton Counter Surveys The Optical Plankton Counter (OPC) is a self-contained operational system based on the principle of light blockage (Herman 1988, 1992). As a particle intercepts a beam of light, the amount of light falling on the detector drops, generating a pulse in the receiver voltage. This allows automatic counting of pulses, hence particles. The beam of light is due to an array of high-intensity light-emitting diodes active at 640 nm. In fulfillment of its design aims, the OPC can determine the numerical density, biomass, and distribution of mesozooplankton and macrozooplankton with size range 0.25–20 mm. It can resolve sizes associated with developmental stages such as those of the later calanoid stages of Calanus finmarchicus. It can resolve detected zooplankton at separation distances from millimeters to kilometers. The OPC is packaged robustly and can be hauled or towed at speeds 0.5–4 m/s to a maximum depth of 1000 m.

6

Quantitative Research Surveys of Fish Stocks

167

The OPC has been calibrated by passing beads of known size through the OPC towed in air and relating the measured, apparent size to the actual size. For non-spherical shapes, such as those of elongated euphausiids in contrast to the spherical eggs of zooplankton, the output of the OPC is given in terms of equivalent spherical diameter. In surveys of copepods, Herman et al. (1991) measured the concentration of copepods in vertical sections along the Banquereau and Halifax lines, respectively 250 and 220 km in length, across the Scotian shelf. Concentrations of the euphausiids Meganyciphanes norvegica and Thysanoessa inermis were quantified by the OPC in the Emerald and La Have Basins of the Scotian Shelf in autumn 1990 (Herman et al. 1993). The numerical density of Calanus finmarchicus fifth copepodite has been measured by the OPC on the continental shelf of the northwest Atlantic Ocean and compared to numerical densities determined by bongo tows and the Multiple Opening and Closing Net and Environmental Sensing System (MOCNESS) (Baumgartner 2003). The numerical density of a large diatom (Coscinodiscus wailesii) was also measured by OPC and compared to WP2 net samples (Woodd-Walker et al. 2000). The OPC has been used to characterize the size structure of zooplankton along the Atlantic Meridional Transect (AMT) between 508N and 508S (Gallienne and Robins 1998). The OPC results were compared with microscopic counts of zooplankton derived from concurrent samples. The OPC is also used on board vessels and in the laboratory for particlesizing and counting. It is an essential component of the Continuous Underway Fish Egg Sampler (CUFES) (Checkley et al. 1997) [Section 6.2]. The conspicuous weakness of the OPC is knowing the particles, or organisms, being detected and counted. The size distribution may be sufficient, given the patchiness of zooplankton distribution. However, independent physical capture data may be necessary to determine species composition. The nextgeneration OPC, the laser OPC, addresses the problem of automatic classification via shape determination (Herman et al. 2004) [Section 6.6.2].

6.6.2 Laser Optical Plankton Counter Surveys The Laser Optical Plankton Counter (LOPC) resembles the OPC but with use of a laser diode and lens to form a broad light beam, and an array of detectors that enable the shape of individual organisms, or other particles, to be sensed as they pass through the light beam (Herman et al. 2004). This assumes that light blockage by the organisms is registered on multiple elements of the detector array. Particles as small as 0.1 mm in equivalent spherical diameter can be detected by the LOPC. During postprocessing, shape profiles are recognized and object sizes are quantified. In an early application, the LOPC was applied to distinguish two groups of species: copepod eggs and nauplii in one group, and Oithona spp. and Oncaea spp. in the other group.

168

K.G. Foote

6.6.3 Video Plankton Recorder Surveys The Video Plankton Recorder (VPR) is an operational underwater video system with magnifying optics that aims to measure the numerical density of plankton and determine species composition automatically in near-real time during surveys at sea (Davis et al. 1996). The basic components of the VPR are high-quality images of individual plankton, an image processing and plankton registration system, and a data processing and visualization system. Particular design aims for the VPR are (i) data collection that is automatic and continuous over scales from millimeters to kilometers, with electronic data storage, (ii) resolution of individual particles that is sufficient for classification by major taxonomic group, e.g., copepods, chaetognaths, fish larvae, fish eggs, euphausiids, amphipods, pteropods, detritus including marine snow, inter alia, (iii) taxonomic analysis of data in near-real time, and (iv) acquisition of data at towing speeds 0.5–5 m/s (Davis et al. 1992a). An early design aim of physical capture of actual particles for calibration of the electronic data was addressed by placing a gauze recorder box, which was a modified Longhurst-Hardy Plankton Recorder (Haury and Wiebe 1982), behind the imaged volume. However, this induced avoidance reactions that affected the optical registration (Davis et al. 1992b), hence the gauze recorder box was subsequently removed. Imaging is accomplished by an underwater video system, with magnifying optics sufficient to resolve particle sizes in the range from 0.01 mm to about 20–30 mm. Two, three, or four video cameras are configured to view concentric volumes in either stereoscopic or monoscopic mode at multiple magnifications. The video cameras are synchronized at 60 frames per second to a red-filtered strobe. The exposure time is 1 ms. The distance between strobe and camera is 1 m, with intersection of the two optical axes at 0.5 m. Image processing is performed on the basis of live or recorded images. Operations that can be performed at the 60-Hz collection rate include field-grabbing, convolution, edge detection, and delivery of extracted-image coordinates in terms of pixels to the host computer. The presence or absence of particles is determined automatically, and particle edges are examined for sharpness in a focus-detection system. The region-of-interest surrounding a sharp, focused image is transmitted to a computer where it is stored in a file named according to the time code. Taxonomic composition is performed by means of a neural network trained on the basis of 2000 images. Additional characterization of detected objects, as in terms of shape and texture, is performed. By 1996, the data rate was 3600–10,800 in-focus objects per hour. Classified regions-of-interest can be sorted by species, and position data can be attached. The number of counted objects in a given class can be related to the common sampling volume, thereby measuring the numerical density of organisms. Averaging over time or space intervals may be performed, and the spatial distribution can be displayed graphically.

6

Quantitative Research Surveys of Fish Stocks

169

The VPR has been calibrated by means of comparison with samples collected with the Multiple Opening and Closing Net and Environmental Sensing System (MOCNESS). Observations with the VPR along a particular transect were compared with data from MOCNESS tows at each of five depths, and a range of organisms was identified, including copepods, pteropods, larvacea, amphipods, and euphausiids (Benfield et al. 1996). Mean numerical densities were established for the two gears. The respective distances were in general agreement for sufficiently high concentrations, but the difference in sampling volumes of the two gears is enormous, being 0.069 m3 for one of the VPR cameras and 150 m3 for a MOCNESS with opening 1-m2 and towing distance of 150 m. In an early application of the VPR, the patchiness of copepods, doliolids, and Trichodesmium were observed at scale sizes of the order of 1–100 m (Davis et al. 1992b). Later, Davis et al. (1996) compared the density distributions of the copepod Calanus finmarchicus relative to the distribution of fluorescence in the same volume on Georges Bank in May 1994. Gallager et al. (1996) observed distributions of a number of organisms together with the salinity and temperature fields in the Great South Channel of Georges Bank in May 1992. Norrbin et al. (1996) measured zooplankton density in different regions of Georges Bank and related this to mixing and stratification of water masses. Organisms included chaetognaths, hydroid colonies, Pseudocalanus sp., and foraminifera in a well-mixed area, and Pleurobrachia, Limacina pteropods, Calanus finmarchicus, and Oithona spp. in a stratified area. Benfield et al. (1996) measured numerical densities of copepods, pteropods, and larvacea distributions in a stratified region of Georges Bank. Ashjian et al. (2001) quantified zooplankton, including Calanus finmarchicus and Oithona spp. inter alia, across Georges Bank over distances from centimeters to hundreds of kilometers, associating species with the hydrography. Similar measurements and hydrographic associations have been made in the Japan East Sea (Ashjian et al. 2005). Much smaller, aggregated organisms have also been quantified by the VPR. These include colonial Radiolaria (Dennett et al. 2002) and mats of Rhizosolenia diatoms (Pilskaln et al. 2005) in the North Pacific, and colonial diazotrophic cyanobacteria of the genus Trichodesmium along a trans-Atlantic transect (Davis and McGillicuddy 2006). Improvements to the VPR are ongoing. The VPR has been configured for towing at 6 m/s (Davis et al. 2005), with tests performed at 200-m depth at 5 m/s (Thwaite and Davis 2002). Classification accuracy is also being studied, with performance of automated classification to major taxa and even to species being done in real time (Davis et al. 2004). Recently, a dual classification method has been described that promises to increase the accuracy of classification, as by reducing the false positive rates (Hu and Davis 2006). This is particularly useful for automatic quantification of organisms at relatively low concentrations. The method requires agreement of two different classification schemes, based respectively on (i) shape-based features classified by neural network, and (ii) texture-based features classified by a supportvector machine.

170

K.G. Foote

The VPR is being integrated with other sensors used in surveying watercolumn organisms because of its power of remote, high-speed classification. Examples are given in Section 6.7.6 for integration with a scientific echo sounder and in Section 6.7.7 for integration with MOCNESS.

6.6.4 Video Survey of Benthos Benthic organisms are being surveyed with a video system constructed at the University of Massachusetts School for Marine Science and Technology (SMAST), called the SMAST Video Survey Pyramid (Stokesbury 2002). The base is square, with 2.2-m side length. Structural members between apex and corners of the base are 2.5-m long. A video camera has been positioned below the apex at a height of 1.57 m over the base and oriented vertically downwards, with view of a quadrat of area 2.8 m2, which was determined by calibration. Lights are attached to the frame. This system is typically deployed from a fishing vessel by lowering it to the seafloor, then raising and moving it to the next sampling location either by drifting or motoring. Following each survey, videotapes of the seafloor are reviewed in the laboratory; the data quality is controlled; and image processing software is applied to automatically recognize and measure individual organisms, especially sea scallops (Placopecten magellanicus). Surveys of sea scallop abundance have been performed on Georges Bank (Stokesbury 2002) and off the northeast coast of the U.S. (Stokesbury et al. 2004) using a uniform sampling pattern. The same system has been used to quantify sea scallop shell growth (Harris and Stokesbury 2006) and to examine the effect of harvesting on the epibenthic community on Georges Bank (Stokesbury and Harris 2006). In this last-mentioned study, a large number of organisms were surveyed, including numerous species of starfish, bryozoans and hydrozoans, sponges, crabs, hermit crabs, flounder, hake, sculpins, skates, and other fishes.

6.6.5 Lidar Surveys The device called lidar, for light detection and ranging, is analogous to sonar. It typically consists of a laser beam, a telescope to focus backscattered light, and a detection system mounted in the focal or equivalent plane (Churnside and Hunter 1996). The laser light is usually in the blue-green band to coincide with the absorption minimum of sea water. The laser light is allowed to spread or is actively defocused to avoid potential retinal damage at the sea surface, observing standards for human vision that are also eye-safe for marine mammals including cetaceans and pinnepeds (Zorn et al. 2000). Lidar used in fish surveying is radiometric, as the transmit light beam and detection system can be calibrated, e.g., with a Kodak gray card serving as a standard target (Churnside et al. 2003). This allows the measurements to be expressed in physical units,

6

Quantitative Research Surveys of Fish Stocks

171

which is analogous to the situation with measurements performed with calibrated acoustic systems [Section 6.5.4]. Lidar has been used to detect fish since about 1980 (Squire and Krumboltz 1981). The potential for surveying a number of organisms quantitatively has been recognized (Hunter and Churnside 1995). Lidar has been used to count schools of anchovy in coastal waters off southern California (Lo et al. 2000), fish schools in the Gulf of Mexico (Churnside et al. 2003), and juveniles of a number of fish species in the coastal Atlantic water off southern Europe (Carrera et al. 2006). Survey data are often expressed through the so-called lidargram or lidar echogram. Understandably, depth of detection and species recognition are general problems in lidar surveying. However, in such a case as that of adult pink salmon (Oncorhynchus gorbuscha) in Alaska, individuals can be imaged (Churnside and Wilson 2004). Lidar lends itself to combined use with other instruments, for example, scientific echo sounders, given the possibility of achieving synoptic coverage with lidar and directing a research vessel to regions of high concentration. Combined surveys of the two instruments are described in Section 6.7.4.

6.6.6 Camera Surveys Cameras have a long history of use underwater, but with scant adaptation to surveying for reasons of attenuation of light and, in benthic studies, variable lighting over uneven terrain. Another limitation has been the compounding of images in a larger continuous image. Two recent advances in technology are overcoming some of these barriers to the survey use of cameras. The autonomous underwater vehicle Sea Benthic Exploration Device (SeaBED) (Roman et al. 2000, Singh et al. 2004a) has a hovering capability, establishing a stable platform that can hover or follow the bottom terrain at constant height. Photomosaicking enables relatively large swaths of seafloor to be imaged coherently (Pizarro and Singh 2003, Singh and Pizarro 2004, Singh et al. 2004b), presently for habitat definition, but with a clear potential for imaging and surveying demersal fish too.

6.7 Integrated Surveys Combination of different survey types can complicate the business of estimating abundance. However, it can enhance the comprehensiveness of particular survey types, as well as enabling or supporting data interpretation (McClatchie et al. 2000). If multiple tools are used to provide essentially independent estimates of abundance, these can be used for cross-validation.

172

K.G. Foote

6.7.1 Combined Acoustic and Optical Methods for Surveying Herring Populations of Pacific herring in Prince William Sound are assessed by scientific echo sounder surveys. These are being supplemented by the use of other acoustic and optical methods (Thorne et al. 2003). Sonars are used to delimit the geographical distribution of the herring. Underwater cameras with infrared lighting are being lowered into fish schools at night to identify the species, for example, to distinguish herring from walleye pollock. Infrared scanning of the sea surface is being used to observe night-time foraging by Steller sea lions, humpback whales, and seabirds. The high correlation of Steller sea lion counts and herring biomass is noted.

6.7.2 Combined Acoustic and Egg-Production Survey Acoustic and egg-production surveys have been combined in order to assess the spawning stock biomass of Cape anchovy (Engraulis capensis) (Hampton et al. 1990). Both methods have been applied at the same time but essentially independently. The echo integration method was applied over the targeted region of occurrence, the South African continental shelf. Eggs were counted at regular intervals along the acoustic transects, and the proportion of spawners was estimated by pelagic trawl. Estimates of spawning biomass were derived by the mean daily-egg-production method. Agreement of the two estimates for each of the reported years, 1985 and 1986, is excellent. Given the intrinsic difficulties associated with the execution of most surveys, the agreement adds considerable confidence to the abundance estimates at marginal additional cost.

6.7.3 Integrated Acoustic and Trawl Survey Physical capture is an essential part of most acoustic surveys of fish abundance. It is generally necessary for both species identification and determination of size or age composition of the acoustically observed fish (Thorne 1983, Traynor and Nelson 1985). In many studies, bottom fish that might otherwise be acoustically measured when off the bottom are inaccessible. The problem of the so-called bottom dead zone was first addressed by Thorne (1983) and Traynor and Nelson (1985). It is this realization, coupled with a wider appreciation of the problems of sampling by trawl that supported an integration of acoustic and trawl surveys, with joint performance of the two survey types (Godø and Wespestad 1993). The biomass of the mesopelagic community on the continental slope south of Tasmania has been estimated by both scientific echo sounder operating at

6

Quantitative Research Surveys of Fish Stocks

173

38 kHz and by midwater trawling (Koslow et al. 1997). The depth range is 0–900 m. The acoustic estimate of abundance was seven times larger than that based on the volume-swept trawling estimate. The acoustic estimates are consistent with estimates based on primary production and trophodynamics modeling.

6.7.4 Combined Lidar and Echo Sounder Surveys A necessary condition for many abundance surveys of fish and zooplankton is that the geographical area of occurrence be covered synoptically, that is, in a time that is small compared to large-scale movements of the species of interest. The contributions of lidar to such coverage have been recognized. For example, lidar observation from aircraft flying at 50–100 m/s can be used to direct survey vessels sailing at 5 m/s to areas of highest concentration of fish or zooplankton, where echo sounders can collect high-resolution data throughout the water column and perform trawling for physical identification (Churnside and Thorne 2005). Lidar can also be used to provide independent estimates of abundance (Churnside et al. 2001). In the summers of 1998 and 1999 lidar surveys were performed of epipelagic schools of juvenile anchovy, mackerel, and sardine, with depth of light penetration 25–30 m (Carrera et al. 2006). Acoustic surveys were performed with scientific echo sounders operating at 38 kHz, but essentially without limitation in depth of sound penetration. The correlation in observations was positive, but the problem of distinguishing plankton and fish optically was recognized. In December 2000, a similar method was applied to estimate the abundance of individual epipelagic fish schools, e.g, striped mullet (Mugil cephalus), off the west coast of Florida, but with 208-kHz scientific echo sounder (Churnside et al. 2003). Measures of acoustic backscattering were somewhat greater than those of optical backscattering. Lidar surveys were also combined with scientific echo sounder surveys at 420 kHz on the zooplankton Neocalanus spp. in Prince William Sound, Alaska, in May 2002 (Churnside and Thorne 2005). A correlation of 0.78 was obtained when an a posteriori threshold was applied to remove contributions from phytoplankton.

6.7.5 Integrated Camera and Net Survey An underwater camera with strobe flash has been used in conjunction with a 60-cm-diameter plankton net (Houde et al. 1989). The camera was mounted on the frame of the net and oriented to take pictures of plankton in a flow chamber immediately before the codend. Resulting silhouette photographs enabled fine details of the spatial distribution of bay anchovy (Anchoa mitchilli) eggs and larvae to be determined, thus supplementing net-derived measures of overall abundance.

174

K.G. Foote

6.7.6 Integrated Acoustic and VPR Surveys A number of zooplankton has been surveyed with both a scientific echo sounder operating at 420 kHz and Video Plankton Recorder (Benfield et al. 1998). The organisms included amphipods, chaetognaths, copepods, euphausiids, fish, and pteropods. The VPR provided the essential species identification data that enabled the acoustic record to be interpreted. Physonect siphonulae are important acoustic scatterers because of their gas-filled inclusions, called pneumatophores. They have been surveyed along track lines by a single-camera variant of the VPR and scientific echo sounders operating at 43, 120, 200, and 420 kHz (Benfield et al. 2003a). The VPR observations give confidence to the interpretation of the echo data. The VPR has been used in conjunction with acoustic surveys at 43, 120, 200, and 420 kHz of Antarctic krill (Euphausia superba) on the continental shelf west of the Western Antarctic Peninsula (Lawson et al. 2006). A particular result has been in situ determination of krill target strength, with improvements in parametrization of the acoustic scattering model.

6.7.7 Integrated MOCNESS and VPR Surveys Two studies have been conducted with the VPR mounted on the frame of the Multiple Opening and Closing Environmental Sensor System (MOCNESS) (Broughton and Lough 2006). In one, the opening area was 1 m2 and the mesh size, 333 mm; in the other, the opening area was 0.25 m2 and the mesh size, 64 mm. It is concluded that the two devices provide consistent results for some organisms, but act in a complementary manner for others: MOCNESS can capture low-abundance organims that might not be observed with the VPR, while the VPR can sample fragile organisms that would be damaged or destroyed by physical capture.

6.8 Auxiliary Instruments 6.8.1 Salinity-Temperature-Depth Sonde The salinity-temperature-depth (STD) sonde is a traditional profiling instrument that measures the temperature and salinity fields in the water column. This is usually achieved through direct measurements of conductivity, temperature, and pressure. Based on these, the sound speed profile cam be computed (Mackenzie 1981) and used in the range compensation function applied to echo sounder signals. The STD measurements are also used when interpreting echograms, because of the association of fish with particular water masses

6

Quantitative Research Surveys of Fish Stocks

175

and other oceanographic features. Examples of displays of information resulting from hydrographic surveys in the Norwegian Sea are given in Blindheim (1990, 2004).

6.8.2 Acoustic Doppler Velocimeter An acoustic Doppler velocimeter measures the Doppler shift in frequency due to movement of scatterers relative to the measuring device. This can be expressed as one or two components or the full three-dimensional vector of velocity depending on the configuration of the device. Some velocimeters make essentially point measurements in space; others make measurements over a range of distances. When such a device is oriented downwards in the water column, or upwards in the case of bottom-mounted devices, a vertical profile of current can be determined from backscattering, or reverberation, due to the myriad of small particles typically present in water. Since these velocimeters generally use multiple beams at oblique orientations, their surveying potential may be significant. A number of attempts have been made to exploit a particular acoustic Doppler velocimeter as a multiple-beam echo sounder. This is the acoustic Doppler current profiler (ADCP), which is manufactured by RD Instruments. The device has been used to measure zooplankton abundance both with modification (Flagg and Smith 1989, Cochrane et al. 1988, 1994) and without modification (Heywood et al. 1991, Roe and Griffiths 1993, Roe et al. 1996). It has been used to measure the velocity of swimming herring (Zedel and Knutsen 2000) and migrating salmon (Tollefsen and Zedel 2003), as well as the velocity of vertically migrating euphausiid and pteropod layers (Tarling et al. 2001). It has been used on a variety of platforms, including, for example, vessels with hull-mountings, and moorings. The frequencies of these devices have been in the ultrasonic range. Since a measurement of phase is sufficient to determine velocity, acoustic Doppler velocimeters are often restricted in dynamic range. Ultimately, a significant expansion of the electronic capability of these velocimeters would enable significant applications in fish surveying.

6.8.3 Acoustic Bottom-Classifier System Bottom fish show preferences for substrate type (Scott 1982, Orlowski 1989, Walsh 1992). According to the theory developed by Orlowski (1984), the ratio of energy contained in the bottom-surface-bottom echo with energy contained in the single bottom-bounce echo is related to bottom hardness and roughness. A commercial device, called RoxAnn, is used by both fishers and researchers (Burns et al. 1989). This device is essentially an echo sounder with special signal processing capability. Its potential to measure water-column scatterers and bottom habitat at the same time is evident.

176

K.G. Foote

6.8.4 Global Positioning System The Global Positioning System (GPS) or its more sensitive, hence more precise and accurate differential version (dGPS), have become ubiquitous. Use of GPS or dGPS in surveying operations has become routine. In particular, data are automatically labeled with their location and time of collection, enabling reconstruction, postprocessing, and analysis to proceed apace.

6.8.5 Geographical Information System A geographical information system (GIS) is a set of software that allows data to be retrieved and displayed on the basis of the position of data collection, namely latitude and longitude, as well as time in the general case. GISs allow superposition, or overlaying, of data of different kinds. This can aid data interpretation, as when the occurrence of fish has a particular association with water masses, currents, or habitat. GISs are invaluable when processing data because of the inherent reference of each datum to position and time.

6.9 Potential Survey Applications of New or Unique Acoustic Instruments or Techniques Several new or presently unique instruments or techniques for surveying fish are briefly described.

6.9.1 Long-range Forward Scattering by Fish Aggregations Large fluctuations have been observed in the received level of continuouswave low-frequency transmission signals at 1 kHz from 1.9 to 137 km in the Bristol Channel (Weston 1967, Weston et al. 1969). Signal amplitudes often change by 10 dB or more in the course of several minutes. Fluctuations associated with the sunrise and sunset periods were attributed to fish forming schools in the daytime, with better acoustic transmission conditions, and dispersing at night, with increasing acoustic absorption (Weston et al. 1969, Weston 1972). The effect was modeled as a consequence of acoustic extinction due to aggregation of fish bearing gas-filled sacs called swimbladders (Weston 1967). Diachok (2000) has formalized a method for observing the frequency or frequencies of swimbladder resonance, associated these with fish swimbladder sizes, hence fish sizes, and fish abundance. The method has been applied to

6

Quantitative Research Surveys of Fish Stocks

177

sardines (Sardina pilchardus) in the Mediterranean Sea (Diachok 1999, 2000). The transmission source is the parametric acoustic array (Westervelt 1963, Moffett and Konrad 1997).

6.9.2 Long-range Backscattering by Fish Aggregations This method, like that of fish swimbladder resonance spectroscopy [Section 6.9.1], involves sensing of large numbers of fish over rather long distances. Both methods exploit the powerful scattering due to excitation of the fish swimbladder at rather low frequencies, of order 0.4–10 kHz. Seminal observations of long-range backscattering by fish were made at least as early as 1958 (Weston and Revie 1971). The particular reported observations were made with a bottom-mounted 1-kHz sonar with horizontal transmit beamwidth of 158 oriented outwards in the Bristol Channel into water of depth 35–90 m, with near-colocated receiving array of beamwidth 48. Echoes due to moving targets were observed at ranges of 15–30 km. Collaboration with the Fisheries Laboratory, Lowestoft, revealed the nature of these targets: schools of Cornish pilchards, with mean fish length 23 cm. In an application of long-range backscattering, Rusby et al. (1973) made observations with a towed 6.4-kHz sidescan sonar in the Sea of the Hebrides in September 1971. This sidescan sonar was the Geological Long-Range Inclined Asdic (GLORIA). It was towed along 13-km base lines. Herring were detected at a greatest range of 15 km in water depths of 120–170 m. Over a 3-day period, a fishery area of 170 km2 was monitored. Recently, Makris et al. (2006) has adapted this method to detect and track large fish schools or layers, in addition to observing the short-time evolution of their shapes. Unlike the method used by Rusby et al. (1973) with collocated transmitter and receiver, a low-frequency sound source with total bandwidth 390–440 Hz is moored, and a vessel tows the receiving array in a dynamic bistatic mode of observation. In observations made near the edge of the continental shelf 200 km south of Long Island, New York, in May 2003, fish aggregations were detected and imaged at distances of 10–20 km.

6.9.3 Parametric Sonar Sonars based on the parametric acoustic array (Westervelt 1963, Moffett and Konrad 1997) are used in a number of applications because of certain exceptional properties. These include unusual directionality at low frequencies, with a main-lobe beamwidth that is disproportionately narrow compared to the transmitting transducer size and notable absence of sidelobes, and large bandwidths at low frequencies. The Simrad PS18 Parametric Sub-bottom Profiler (Dybedal 1993) transmits primary signals over the band 15–21 kHz, generating difference

178

K.G. Foote

frequencies in the range 0.5–6 kHz. While such directional low-frequency waves are very useful for observing the ocean sub-bottom, as in geophysical investigations and searching for hidden objects buried beneath the seafloor, the sonar also has a clear capacity to observe mid-water scatterers, for example fish in aggregations. Because of the low frequency, directional effects that are significant at ultrasonic frequencies will be much reduced, contributing to a greater stability in received echo signals. Recently, a standard target was specified for use in calibrating this parametric sonar (Foote et al. 2007). Applications of such a standard-target calibration method will enable measurements of fish to be expressed in physical units.

6.9.4 Passive Acoustic Survey Methods Passive acoustic methods are well developed for surveying a number of marine mammals, e.g., to determine their geographical limits of distribution and to determine their abundance. Bowhead whales have been surveyed acoustically during the spring migration past Point Barrow, Alaska (Clark and Ellison 1988). Sperm whales have been surveyed along transects in the Southern Ocean by means of towed hydrophones (Leaper et al. 2000). Humpback whales, pilot whales, and spinner porpoises have also been surveyed by passive acoustic methods. The potential of passive methods for surveying wintering pinnepeds has been recognized (Stirling et al. 1983). Fish species have also been surveyed by passive acoustic means. The spawning area of red drum, weakfish, spotted sea trout, and silver perch have been defined by a hydrophone suspended from a boat and ten timer-activated sonobuoys (Luczkovich and Sprague 2002). The croaker has been surveyed in the Gulf of Trieste in the northern Adriatic Sea (Bonacito et al. 2002). Vocalizations of coral reef fishes during spawning are being monitored, with recognition of the potential for surveying (Lobel 2002, Mann and Lobel 1995). The number of fish species that produce sound underwater exceeds 800 according to Kaatz (2002), revising the earlier estimate of some 600 species (Fish 1964). The potential for monitoring and surveying, as well as conducting behavior studies, is thus evident and is being pursued (Rountree et al. 2006).

6.9.5 Other Methods In addition to the potential survey applications of new or unique instruments, there are other sensors and systems that are or could be used for quantitative surveying in ecosystem investigations. Several of these are briefly mentioned. Two bathyphotometers are in use to measure bioluminescence: the HighIntake, Defined Excitation Bathyphotometer (HIDEX-BP) (Widder et al. 1993) and the Expendable Bathy-Photometer (XBP) (Fucile 2002). Two flow

6

Quantitative Research Surveys of Fish Stocks

179

cytometers are in use: the Flow Cytometer and Microscope (FlowCAM) (Sieracki et al. 1998) and FlowCytobot (Sosik et al. 2002), both typically mounted on stationary platforms. A number of optical imaging systems are operational. These include the Zooplankton Visualization System (ZOOVIS) (Benfield et al. 2003b), a deep-water video system for profiling marine snow and zooplankton (Gorsky et al. 1992, Picheral et al. 1998), a simplified video plankton recorder (Tiselius 1998), and a submersible microscope (Akiba and Kakui 2000). Laser line scan (LLS) systems are also being adapted for investigations of bottom organisms and habitat along line transects, with swathmapping analogous to multibeam-sonar swath-mapping (Reynolds et al. 2001). Several systems are being used, e.g., Shadowed Image Particle Profiling and Evaluation Recorder (SIPPER) (Samson et al. 2001), the Westinghouse SM2000 (Tracey et al. 1998), and an LLS being used for three-dimensional imaging (Caimi et al. 1993). The Polarized Larval Recorder (Gallager et al. 1989) detects larval molluscs, e.g., giant scallop larvae (Placoplecten magellanicus), due to birefringence in aragonite crystals. Acknowledgments M. Parmenter is thanked for a sustained, meticulous effort to cast the text and references in the official style.

References Aasen O (1958) Estimation of the stock strength of the Norwegian herring. J Cons Int Explor Mer 24:95–110 Akiba T, Kakui Y (2000) Design and testing of an underwater microscope and image processing system for the study of zooplankton distribution. IEEE J Oceanic Eng 25:97–104 Allen B, Stokey R, Austin T, Forrester N, Goldsborough R, Purcell M, von Alt C (1997) REMUS: a small, low cost AUV; system description, field trials and performance results. Proc MTS/IEEE Oceans 1997, pp 994–1000 Andersen LN (2001) The new Simrad EK60 scientific echo sounder system. J Acoust Soc Am 109:2336 Andersen LN, Berg S, Gammelsæter OB, Lunde EB (2006) New scientific multibeam systems (ME70 and MS70) for fishery research applications. J Acoust Soc Am 120:3017 Anthony VC, Waring G (1980) The assessment and management of the Georges Bank herring fishery. Rapp P-v Re´un Cons Int Explor Mer 177:72–111 Aoki Y, Sato T, Zeng P, Iida K (1991) Three-dimensional display technique for fish-finder with fan-shaped multiple beams. In: Lee H, Wade G (eds) Acoustical Imaging. Plenum Press, New York, Vol. 18, pp 491–499 Ashjian CA, Davis CS, Gallager SM, Alatalo P (2001) Distribution of plankton, particles, and hydrographic features across Georges Bank described using the Video Plankton Recorder. Deep-Sea Res II, 48:245–282 Ashjian CA, Davis CS, Gallager SM, Alatalo P (2005) Characterization of the zooplankton community, size composition, and distribution in relation to hydrography in the Japan/ East Sea. Deep-Sea Res II 52:1363–1392 Astthorsen OS, Gislason A, Gudmundsdottir A (1994) Distribution, abundance, and length of pelagic juvenile cod in Icelandic waters in relation to environmental conditions. ICES Marine Science Symposia 198:529–541

180

K.G. Foote

Astthorsen OS, Gislason A (1995) Long-term changes in zooplankton biomass in Icelandic waters in spring. ICES J Mar Sci 52:657–668 Axelsen BE, Anker-Nilssen T, Fossum P, Kvamme C, Nøttestad L (2001) Pretty patterns but a simple strategy: predator-prey interactions between juvenile herring and Atlantic puffins observed with multibeam sonar. Can J Zool 79:1586–1596 Bailey RS, Simmonds EJ (1990) The use of acoustic surveys in the assessment of the North Sea herring stock and a comparison with other methods. Rapp P-v Re´un Cons int Explor Mer 189:9–17 Baumgartner MF (2003) Comparisons of Calanus finmarchicus fifth copepodite abundance estimates from nets and an optical plankton counter. J Plank Res 25:855–868 Benfield MC, Davis CS, Wiebe PH, Gallager SM, Lough RG, Copley NJ (1996) Video plankton recorder estimates of copepod, pteropod and larvacean distributions from a stratified region of Georges Bank with comparative measurements from a MOCNESS sampler. Deep-Sea Res II 43:1925–1945 Benfield MC, Wiebe PH, Stanton TK, Davis CS, Gallager SM, Green CH (1998) Estimating the spatial distribution of zooplankton biomass by combining Video Plankton Recorder and single-frequency acoustic data. Deep-Sea Res II 45:1175–1199 Benfield MC, Lavery AC, Wiebe PH, Greene CH, Stanton TK, Copley NJ (2003a) Distributions of physonect siphonulae in the Gulf of Maine and their potential as important sources of acoustic scattering. Can J Fish Aquat Sci 60:759–772 Benfield MC, Schwehm CJ, Fredericks RG, Squyres G, Keenan SF, Trevorrow MV (2003b) Measurement of zooplankton distributions with a high-resolution digital camera system. In: Strutton P, Seuront L (eds) Handbook of Scaling Methods in Aquatic Ecology: Measurement, Analysis Simulation. CRC Press, Boca Raton, Florida, pp 17–30 Benoit-Bird K, Au W (2003) Hawaiian spinner dolphins aggregate midwater food resources through cooperative foraging. J Acoust Soc Am 114:2300 Blindheim J (1990) Arctic intermediate water in the Norwegian Sea. Deep-Sea Res 37:1475–1489 Blindheim J (2004) Oceanography and climate. In: Skjoldal HR (ed) The Norwegian Sea Ecosystem. Tapir Academic Press, Trondheim, pp 65–96 Bobber RJ (1970) Underwater electroacoustic measurements. Naval Research Laboratory, Washington, District of Columbia Bodholt H, Nes H, Solli H (1989) A new echo-sounder system. Proc Inst Acoust 11(3): 123–130 Bonacito C, Constantini M, Picciulin M, Ferrero EA, Hawkins AD (2002) Passive hydrophone census of Sciaena umbra (Sciaenidae) in the Gulf of Trieste (Northern Adriatic Sea, Italy). Bioacoustics 12:292–294 Broughton EA, Lough RG (2006) A direct comparison of MOCNESS and Video Plankton Recorder zooplankton abundance estimates: possible applications for augmenting net sampling with video systems. Deep-Sea Res II 53:2789–2807 Burd AC, Holford BH (1971) The decline in the abundance of Downs herring larvae. Rapp P-v Re´un Cons Int Explor Mer 160:99–100 Burns DR, Queen CB, Sisk H, Mullarkey W, Chivers RC (1989) Rapid and convenient acoustic sea-bed discrimination. Proc Inst Acoust 11(3): 169–178 Burwen D, Maxwell S, Pfisterer C (2004) Investigations into the application of a new sonar system for assessing fish passage in Alaskan rivers. J Acoust Soc Am 115:2547 Caimi FM, Blatt JH, Grossman BG, Smith D, Hooker J, Kocak DM, Gonzalez F (1993) Advanced underwater laser systems for ranging size estimation, and profiling. Mar Technol Soc J 27(1): 31–41 Carrera P, Churnside JH, Boyra G, Marques V, Scalabrin C, Uriarte (2006) Comparison of airborne lidar with echosounders: a case study in the coastal Atlantic waters of southern Europe. ICES J Mar Sci 63:1736–1750 Checkley Jr DM, Ortner PB, Settle LR, Cummings SR (1997) A continuous, underway fish egg sampler. Fish Oceanogr 6:58–73

6

Quantitative Research Surveys of Fish Stocks

181

Checkley Jr DM, Dotson RC, Griffith DA (2000) Continuous, underway sampling of eggs of Northern anchovy (Engraulis mordax) and Pacific sardine (Sardinops sagax) in Spring 1996 and 1997 off Southern and Central California. Deep-Sea Res II 47:1139–1155 Christian K, Richter S (1994) The UNIX operating system. 3rd edn. Wiley, New York Churnside JH, Hunter JR (1996) Laser remote sensing of epipelagic fishes. SPIE 2964:38–53 Churnside JH, Wilson JJ, Tatarskii VV (2001) Airborne lidar for fisheries applications. Opt Eng 40:406–414 Churnside JH, Demer DA, Mahmoudi B (2003) A comparison of lidar and echosounder measurements of fish schools in the Gulf of Mexico. ICES J Mar Sci 60:147–154 Churnside JH, Wilson JJ (2004) Airborne lidar imaging of salmon. Appl Opt 43:1416–1424 Churnside JH, Thorne RE (2005) Comparison of airborne lidar measurements with 420 kHz echo-sounder measurements of zooplankton. Appl Opt 44:5504–5511 Clark CW, Ellison WT (1988) Numbers and distributions of bowhead whales, Balaena mysticetus, based on the 1985 acoustic study off Pt. Barrow, Alaska. Rep Int Whal Comm 38:365–370 Clay CS, Medwin H (1977) Acoustical oceanography: principles and applications. Wiley, New York Cochrane NA, Whitman JWE, Belliveau D (1988) Doppler current profilers. Can Tech Rep Fish Aquat Sci, (1641), 89–92 Cochrane NA, Sameoto DD, Belliveau DJ (1994) Temporal variability of euphausiid concentrations in a Nova Scotia shelf basin using a bottom-mounted acoustic Doppler current profiler. Mar Ecol Prog Ser 107:55–66 Cressie NAC (1991) Statistics for spatial data. Wiley, New York Date CJ (1987a) A guide to INGRES. Addison-Wesley, Reading, Massa-chussets Date CJ (1987b) A guide to SQL standard. Addison-Wesley, Reading, Massachussetts Davis CS, Gallager SM, Berman MS, Haury LR, Strickler JR (1992a) The Video Plankton Recorder (VPR): design and initial results. Arch Hydrobiol Beih Ergeb Limnol 36:67–81 Davis CS, Gallager SM, Solow AR (1992b) Microaggregations of oceanic plankton observed by towed video microscopy. Science 257:230–232 Davis CS, Gallager SM, Marra M, Stewart WK (1996) Rapid visualization of plankton abundance and taxonomic composition using the video plankton recorder. Deep-Sea Res II 43:1947–1970 Davis CS, Hu Q, Gallager SM, Tang X, Ashjian CJ (2004) Real-time observation of taxa-specific plankton distributions: an optical sampling method. Mar Ecol Prog Ser 284:77–96 Davis CS, Thwaites FT, Gallager SM, Hu Q (2005) A three-axis fast-tow digital Video Plankton Recorder for rapid surveys of plankton taxa and hydrography. Limnol Oceangr: Methods 3:59–74 Davis CS, McGillicuddy DJ (2006) Transatlantic abundance of the N2-fixing colonial cyanobacterium Trichodesmium. Science 312:1517–1520 Dennett, MR, Caron DA, Michaels AE, Church M, Gallager SM, Davis CS (2002) Video Plankton Recorder reveals high abundance of colonial radiolaria in surface waters of the central north Pacific. J Plankt Res 24:797–805 Diachok O (1999) Effects of absorptivity due to fish on transmission loss in shallow water. J Acoust Soc Am 105:2107–2128 Diachok O (2000) Absorption spectroscopy: a new approach to estimation of biomass. Fish Res 47:231–244 Dragesund O, Jakobsson J (1963) Stock strengths and rates of mortality of the Norwegian spring spawners as indicated by tagging experiments in Icelandic waters. Rapp P-v Re´un Cons int Explor Mer 154:83–90 Dragesund O, Haraldsvik S (1968) Norwegian tagging experiments in the north-eastern North Sea and Skagerak, 1964 and 1965. Fiskeridirektoratets Skrifter Serie Havundersøkelser 14:98–120

182

K.G. Foote

Dybedal J (1993) TOPAS: parametric end-fire array used in offshore applications. In: Hobaek H (ed) Advances in Nonlinear Acoustics. World Scientific, Singapore, pp 264–275 Ehrenberg JE (1974) Two applications for a dual-beam transducer in hydroacoustic fish assessment systems. Proc IEEE Conf Eng Ocean Environ 1:152–154 Ehrenberg JE (1979) A comparative analysis of in situ methods for directly measuring the acoustic target strength of indivudual fish. IEEE J Oceanic Eng 4:141–152 Ellertsen B, Fossum P, Solemdal P, Sundby S (1989) Relations between temperature and survival of eggs and first feeding larvae of the Arcto-Norwegian cod (Gadus morhua L.). Rapp P-v Re´un Cons Int Explor Mer 191:209–210 Enga˚s A, Godø OR (1986) Influence of trawl geometry and vertical distribution of fish on sampling with bottom trawl. J Northw Atl Fish Sci 7:35–42 Enga˚s A, Godø OR (1989a) The effect of different sweep lengths on the length composition of bottom-sampling trawl catches. J Cons Int Explor Mer 45:263–268 Enga˚s A, Godø OR (1989b) Escape of fish under the fishing line of a Norwegian sampling trawl and its influence on survey results. J Cons Int Explor Mer 45:269–276 Faran Jr JJ (1951) Sound scattering by solid cylinders and spheres. J Acoust Soc Am 23:405–418 Fish MP (1964) Biological sources of sustained ambient sea noise. In: Tavolga WN (ed) Marine Bio-Acoustics. Pergamon Press, Oxford, pp 175–194 Fish JP, Carr HA (1990) Sound underwater images, a guide to the generation and interpretation of side scan sonar data. 2nd edn. Lower Cape Publishing, Orleans Fish JP, Carr HA (2001) Sound reflections, advanced applications of side scan sonar. Lower Cape Publishing, Orleans Flagg CN, Smith SL (1989) On the use of the acoustic Doppler current profiler to measure zooplankton abundance. Deep-Sea Res 36:455–474 Folds DL, Hanlin J (1975) Focusing properties of a solid four-element ultrasonic lens. J Acoust Soc Am 58:72–77 Foote KG (1982) Optimizing copper spheres for precision calibration of hydroacoustic equipment. J Acoust Soc Am 71:742–747 Foote KG (1983) Maintaining precision calibrations with optimal copper spheres. J Acoust Soc Am 73:1054–1063 Foote KG (1990) Correcting acoustic measurements of scatterer density for extinction. J Acoust Soc Am 88:1543–1546 Foote KG (1991) Acoustic sampling volume. J Acoust Soc Am 90:959–964 Foote KG (1993) Application of acoustics in fisheries, with particular reference to signal processing. In: Moura JMF, Lourtie IMG (eds) Acoustic Signal Processing for Ocean Exploration. North-Holland, Dordrecht, pp 381–390 Foote KG (1999) Extinction cross section of Norwegian spring-spawning herring. ICES J Mar Sci 56:606–612 Foote KG (2000) Standard-target calibration of broadband sonars. J Acoust Soc Am 108:2484 Foote KG (2006) Optimizing two targets for calibrating a broadband multibeam sonar. Proc MTS/IEEE Oceans 2006, electronic doc 060401-09, 4 pp Foote KG, Aglen A, Nakken O (1986) Measurement of fish target strength with a split-beam echo sounder. J Acoust Soc Am 80:612–621 Foote KG, Knudsen HP, Vestnes G, MacLennan DN, Simmonds EJ (1987) Calibration of acoustic instruments for fish density estimation: a practical guide. ICES Coop Res Rep 144:1–69 Foote KG, Knudsen HP, Korneliussen JR, Nordbø PE, Røang K (1991) Postprocessing system for echo sounder data. J Acoust Soc Am 90:38–47 Foote KG, Ona E, Toresen R (1992) Determining the extinction cross section of aggregating fish. J Acoust Soc Am 91:1983–1989 Foote KG, Stefa´nsson G (1993) Definition of the problem of estimating fish abundance over an area from acoustic line-transect measurements of density. ICES J Mar Sci 50:369–381

6

Quantitative Research Surveys of Fish Stocks

183

Foote KG, Atkins PR, Bongiovanni C, Francis DTI, Eriksen PK, Larsen M, Mortensen T (1999) Measuring the frequency response function of a seven-octave-bandwidth echo sounder. Proc Inst Acoust 21(1): 88–95 Foote KG, Stanton TK (2000) Acoustical methods. In: Harris RP, Wiebe PH, Lenz J, Skjoldal HR, Huntley M (eds) ICES Zooplankton Methodology Manual. Academic Press, London, pp 223–258 Foote KG, Atkins PR, Francis DTI, Knutsen T (2005a) Measuring echo spectra of marine organisms over a wide bandwidth. In: Papadakis JS, Bjørnø L (eds) Proc Int Conf Underwater Acoustic Measurements: Technologies and Results. Heraklion, Crete, Greece, 28 June–1 July 2005, pp 501–508 Foote KG, Chu D, Hammar TR, Baldwin KC, Mayer LA, Hufnagle Jr LC, Jech JM (2005b) Protocols for calibrating multibeam sonar. J Acoust Soc Am 117:2013–2027 Foote KG, Hanlon RT, Iampietro PJ, Kvitek RG (2006) Acoustic detection and quantification of benthic egg beds of the squid Loligo opalescens in Monterey Bay, California. J Acoust Soc Am 119:844–856 Foote KG, Francis DTI, Atkins PR (2007) Calibration sphere for low-frequency parametric sonars. J Acoust Soc Am 121:1482–1490 Fossum P, Moksness E (1993) A study of spring- and autumn-spawned herring (Clupea harengus L.) larvae in the Norwegian Coastal Current during spring 1990. Fish Oceanogr 2:73–81 Fre´on P, Misund OA (1999) Dynamics of pelagic fish distribution and behaviour: effects on fisheries and stock assessment. Fishing News Books, Blackwell Science, Osney Mead, Oxford Fucile PD (2002) An expendable bioluminescence measuring bathy-photometer. Proc MTS/IEEE Oceans 2002, pp 1716–1721 Furusawa M (1991) Designing quantitative echo sounders. J Acoust Soc Am 90:26–36 Furusawa M, Takao Y, Sawada K, Odubo T, Yamatani K (1993) Versatile echo sounding system using dual beam. Nippon Suisan Gakkaishi 59:967–980 Gallager SM, Bidwell JP, Kuzirian AM (1989) Strontium is required in artificial seawater for embryonic shell formation in two species of bivalve molluscs. In: Crick RE (ed) Origin, Evolution, and Modern Aspects of Biomineralization in Plants and Animals. Plenum Press, New York, pp 349–366 Gallager SM, Davis CS, Epstein AW, Solow A, Beardsley RC (1996) High-resolution observations of plankton spatial distributions correlated with hydrography in the Great South Channel, Georges Bank. Deep-Sea Res II 43:1627–1663 Gallienne CP, Robins DB (1998) Trans-oceanic characterization of zooplankton community size structure using an optical plankton counter. Fish Oceanogr 7:147–158 Gerlotto F, Ste´quert B (1983) Une me´thode de simulation pour e´tudier la disribution des densite´s en poissons: application a` deux cas re´els. FAO Fish Rep 300:278–292 Gerlotto F, Soria M, Fre´on P (1999) From 2D to 3D: the use of multi-beam sonar for a new approach in fisheries acoustics. Can J Fish Aquat Sci 56:6–12 Godø OR (1994) Factors affecting the reliability of groundfish abundance estimates from bottom trawl surveys. In: Ferno¨ A, Olsen S (eds) Marine Fish Behaviour in Capture and Abundance Estimation. Fishing News Books, Oxford, pp 166–199 Godø OR (1998) What can technology offer the future fisheries scientist – possibilities for obtaining better estimates of stock abundance by direct observations. J Northw Atl Fish Sci 23:105–131 Godø OR, Enga˚s A (1989) Swept area variation with depth and its influence on abundance indices of groundfish from trawl surveys. J Northw Atl Fish Sci 9:133–139 Godø OR, Sunnana˚ K (1992) Size selection during trawl sampling of cod and haddock and its effect on abundance indices at age. Fish Res 13:93–310 Godø OR, Walsh SJ (1992) Escapement of fish during bottom trawl sampling – implications for resource assessment. Fish Res 13:281–292

184

K.G. Foote

Godø OR, Wespestad VG (1993) Monitoring changes in abundance of gadoids with varying availability to trawl and acoustic surveys. ICES J Mar Sci 50:39–51 Gorsky G, Aldorf C, Kage M, Picheral M, Garcia Y, Favole J (1992) Vertical distribution of suspended aggregates determined by a new underwater video profiler. Annales de l’Institut oceanographique, Paris, 68:275–280 Greene CH, Wiebe PH, Burczynski J, Youngbluth MJ (1988) Acoustical detection of highdensity krill demersal layers in the submarine canyons off Georges Bank. Science 241:359–361 Greenspan M (1972) Acoustic properties of liquids. In: Gray DE (ed) American Institute of Physics handbook. McGraw-Hill, New York, pp 3-86–3-98 Gunderson DR (1993) Surveys of fisheries resources. Wiley, New York Hafsteinsson MT, Misund OA (1995) Recording the migration behavior of fish schools by multibeam sonar during conventional acoustic surveys. ICES J Mar Sci 52:915–924 Hampton I, Armstrong MJ, Jolly GM, Shelton PA (1990) Assessment of anchovy spawner biomass off South Africa through combined acoustic and egg-production surveys. Rapp P-v Re´un Cons Int Explor Mer 189:18–32 Harden Jones FR, Margetts AR, Greer Walker M, Arnold GP (1977) The efficiency of the Granton otter trawl determined by sector-scanning sonar and acoustic transponding tags. Rapp P-v Re´un Cons Int Explor Mer 170:45–51 Hardy AC (1926) A new method of plankton research. Nature 118:630 Harris BP, Stokesbury KDE (2006) Shell growth of sea scallops (Placopecten magellanicus) in the southern and northern Great South Channel, USA. ICES J Mar Sci 63:811–821 Haug A, Nakken O (1977) Echo abundance indices of 0-group fish in the Barents Sea, 1965–1972. Rapp P-v Re´un Cons Int Explor Mer 170:259–264 Haury LR, Wiebe PH (1982) Fine-scale multi-species aggregations of oceanic zooplankton. Deep-Sea Res 29:915–921 Heessen JJL, Rijnsdorp AD (1989) Investigations on egg production and mortality of cod (Gadus morhua L.) and plaice (Pleuronectes platessa L.) in the southern and eastern North Sea in 1987 and 1988. Rapp P-v Re´un Cons Int Explor Mer 191:15–20 Hempel I, Hempel G (1971) An estimate of mortality in eggs of North Sea herring (Clupea harengus L.). Rapp P-v Re´un Cons Int Explor Mer 160:24–26 Hempel G, Schnack D (1971) Larval abundance on spawning grounds of Banks and Downs herring. Rapp P-v Re´un Cons Int Explor Mer 160:94–98 Herman AW (1988) Simultaneous measurement of zooplankton and light attenuance with a new optical plankton counter. Cont Shelf Res 8:205–221 Herman AW (1992) Design and calibration of a new optical plankton counter capable of sizing small zooplankton. Deep-Sea Res 39:395–415 Herman AW, Sameoto DD, Shunnian C, Mitchell MR, Petrie B, Cochrane N (1991) Sources of zooplankton on the Nova Scotia Shelf and their aggregations within deep-shelf basins. Cont Shelf Res 11:211–238 Herman AW, Cochrane NA, Sameoto DD (1993) Detection and abundance estimation of euphausiids using an optical plankton counter. Mar Ecol Prog Ser 94:165–173 Herman AW, Beanlands B, Phillips EF (2004) The next generation of Optical Plankton Counter: the Laser-OPC. J Plank Res 26:1135–1145 Hewitt RP, Smith PE, Brown JC (1976) Development and use of sonar mapping for pelagic stock assessments in the California current. Fish Bull 74:281–300 Hewitt RP, Demer DA (2000) The use of acoustic sampling to estimate the dispersion and abundance of euphausiids, with an emphasis on Antarctic krill, Euphausia superba. Fish Res 47:215–229 Heywood KJ, Scrope-Howe S, Barton ED (1991) Estimation of zooplankton abundance from shipborne ADCP backscatter. Deep-Sea Res 38:677–691 Hickling R (1962) Analysis of echoes from a solid elastic sphere in water. J Acoust Soc Am 34:1582–1592

6

Quantitative Research Surveys of Fish Stocks

185

Hjellvik V, Godø OR, Tjøstheim D (2004) Decomposing and explaining the variability of bottom trawl survey data from the Barents Sea. Sarsia 89:196–210 Holliday DV (1980) Use of acoustic frequency diversity for marine biological measurements. In: Diemer FP, Vernberg FJ, Mirkes DZ (eds) Advanced Concepts in Ocean Measurements for Marine Biology. University of South Carolina, Columbia, South Carolina, pp 423–460 Holliday DV, Pieper RE (1980) Volume scattering strengths and zooplankton distributions at acoustic frequencies between 0.5 and 3 MHz. J Acoust Soc Amer 67:135–146 Holliday DV, Pieper RE, Kleppel GS (1989) Determination of zooplankton size and distribution with multi-frequency acoustic technology. J Cons Int Explor Mer 46:52–61 Houde ED, Ortner PB, Lubbers III L, Cummings SR (1989) Test of a camera-net system to determine abundance and heterogeneity in anchovy egg distributions. Rapp P-v Re´un Cons Int Explor Mer 191:112–118 Hu Q, Davis C (2006) Accurate automatic quantification of taxa-specific plankton abundance using dual classification with correction. Mar Ecol Prog Ser 306:51–61 Hunter JR, Churnside JH (eds) (1995) Airborne fishery assessment technology – a National Oceanic and Atmospheric Administration workshop report. SWFSC Administrative Rep No LJ-95-02, Southwest Fisheries Science Center, La Jolla, California Hylen A, Korsbrekke K, Nakken O, Ona E (1995) Comparison of the capture efficiency of 0-group fish in pelagic trawls. In: Hylen A (ed) Precision and relevance of pre-recruit studies for fishery management related to fish stocks in the Barents Sea and adjacent waters. Proc 6th Russian-Norwegian Symp, Bergen, 14–17 June 1994, Institute of Marine Research, Bergen, Norway Jakobsson J (1983) Echo surveying of the Icelandig summer spawning herring 1973–1982. FAO Fish Rep 300:240–248 Jech JM, Foote KG, Chu D, Hufnagle Jr LC (2005) Comparing two 38-kHz scientific echosounders. ICES J Mar Sci 62:1168–1179 Jech JM, Michaels WL (2006) A multifrequency method to classify and evaluate fisheries acoustics data. Can J Fish Aquat Sci 63:2225–2235 Jenkins FA, White HE (1957) Fundamentals of optics. McGraw-Hill, New York Johannesson KA, Robles AN (1977) Echo surveys of Peruvian anchoveta. Rapp P-v Re´un Cons Int Explor Mer 170:237–244 Johannesson KA, Mitson RB (1983) Fisheries acoustics. A practical manual for aquatic biomass estimation. FAO Fish Tech Pap 240:1–249 Jones R (1976) The use of marking data in fish population analysis. FAO Fish Tech Pap 153:1–42 Kaatz IM (2002) Multiple sound-producing mechanisms in teleost fishes and hypotheses regarding their behavioural significance. Bioacoustics 12:230–233 Kamgar-Parsi B, Johnson B, Folds DL, Belcher EO (1997) High-resolution underwater acoustic imaging with lens-based systems. Int J Imag Syst Tech 8:377–385 Kane J (1993) Variability of zooplankton biomass and dominant species abundance on Georges Bank, 1977–1986. Fish Bull 91:464–474 Kane J (1997) Persistent spatial and temporal abundance patterns for late-stage copepodites of Centropages hamatus (Copepoda: Calanoida) in the U.S. northeast continental shelf ecosystem. Fish Bull 95:85–98 Kasatkina SM (1997) Selectivity of commercial and research trawls in relation to krill. CCAMLR Sci 4:161–169 Kernighan BW, Ritchie DM (1988) The programming language. 2nd edn. Prentice Hall, Englewood Cliffs, New Jersey Kim K, Neretti N, Intrator N (2005) Mosaicing of acoustic camera images. IEE Proc-Radar Sonar Navig 152:263–270 Knudsen HP (1990) The Bergen Echo Integrator: and introduction. J Cons Int Explor Mer 47:167–174

186

K.G. Foote

Korneliussen RJ, Ona E (2002) An operational system for processing and visualizing multi-frequency acoustic data. ICES J Mar Sci 59:293–313 Koslow JA, Kloser RJ, Williams A (1997) Pelagic biomass and community structure over the mid-continental slope off southeastern Australia based upon acoustic and midwater trawl sampling. Mar Ecol Prog Ser 146:21–35 Lawson GL, Wiebe PH, Ashjian CJ, Gallager SM, Davis CS, Warren JD (2004) Acousticallyinferred zooplankton distribution in relation to hydrography west of the Antarctic Peninsula. Deep-Sea Res II 51:2041–2072 Lawson GL, Wiebe PH, Ashjian CJ, Chu D, Stanton TK (2006) Improved parametrization of krill target strength models. J Acoust Soc Am 119:232–242 Leaper R, Gillespie D, Papastavrou V (2000) Results of passive acoustic surveys for odontocetes in the Southern Ocean. J Cetacean Res Manage 2:187–196 Lo NCH, Hunter JR, Churnside JH (2000) Modeling statistical performance of an airborne lidar survey system for anchovy. Fish Bull 98:264–282 Lo NCH, Smith PE, Takahashi M (in press) Egg, larvae and juvenile surveys. In: Jakobsen T, Fogarty M, Megrey MA, Moksness E (eds) Fish Reproductive Biology and its implications for Assessment and Management. Blackwell, Oxford Lobel PS (2002) Diversity of fish spawning sounds and the application of passive acoustic monitoring. Bioacoustics 12:286–289 Lough RG, Bolz GR, Pennington, M, Grosslein MD (1985) Larval abundance and mortality of Atlantic herring (Clupea harengus L.) spawned in the Georges Bank and Nantucket Shoals areas, 1971–1978 seasons, in relation to spawning stock size. J Northw Atl Fish Sci 6:21–35 Luczkovich JJ, Sprague MW (2002) Using passive acoustics to monitor estuarine fish populations. Bioacoustics 12:289–291 Mackenzie KV (1981) Nine-term equation for sound speed in the oceans. J Acoust Soc Am 70:807–812 MacLennan DN (1990) Acoustical measurement of fish abundance. J Acoust Soc Am 87:1–15 Mais KF (1977) Acoustic surveys of northern anchovies in the California Current system, 1966–1972. Rapp P-v Re´un Cons int Explor Mer 170:287–295 Makris NC, Ratilal P, Symonds DT, Jagannathan S, Lee S, Nero RW (2006) Fish population and behavior revealed by instantaneous continental shelf-scale imaging. Science 311:660–663 Mann DA, Lobel PS (1995) Passive acoustic detection of sounds produced by the damselfish, Dascyllus albisella (Pomacentridae). Bioacoustics 6:199–213 Mathisen OA, Croker TR, Nunnallee EP (1977) Acoustic estimation of juvenile sockeye salmon. Rapp P-v Re´un Cons Int Explor Mer 170:279–286 McClatchie S, Thorne RE, Grimes P, Hanchet S (2000) Ground truth and target identification for fisheries acoustics. Fish Res 47:173–191 Medwin H, Clay CS (1998) Fundamentals of acoustical oceanography. Academic Press, Boston Midttun L, Nakken O (1977) Some results of abundance estimation studies iwth echo integrators. Rapp P-v Re´un Cons int Explor Mer 170:253–258 Misund OA (1993) Abundance estimation of fish schools based on a relationship between school area and school biomass. Aquat Liv Resourc 6:235–241 Misund OA, Aglen A (1992) Swimming behaviour of fish schools in the North Sea during acoustic surveying and pelagic trawl sampling. ICES J Mar Sci 49:325–334 Misund OA, Aglen A, Beltestad AK, Dalen J (1992) Relationships between the geometric dimensions and biomass of schools. ICES J Mar Sci 49:305–315 Mitson RB (1983) Acoustic detection and estimation of fish near the sea-bed and surface. FAO Fish Rep 300:27–34 Moffett MB, Konrad WL (1997) Nonlinear sources and receivers. In: Crocker MJ (ed) Encyclopedia of Acoustics. Wiley, New York, Vol. 1, pp 607–617

6

Quantitative Research Surveys of Fish Stocks

187

Moursund RA, Carlson TJ, Peters RD (2003) A fisheries application of a dual-frequency indentification sonar acoustic camera. ICES J Mar Sci 60:678–683 Munk P (1988) Catching large herring larvae: gear applicability and larval distribution. J Cons Int Explor Mer 45:97–104 Munk P, Christensen V (1990) Larval growth and the separation of herring spawning groups in the North Sea. J Fish Biol 37:135–148 Nakashima BS, Winters GH (1984) Selection of external tags for marking Atlantic herring (Clupea harengus harengus). Can J Fish Aquat Sci 41:1341–1348 Nakken O, Hylen A, Ona E (1995) Acoustic estimates of 0-group fish abundance in the Barents Sea and adjacent waters in 1992 and 1993. In: Hylen A (ed) Precision and Relevance of Pre-recruit Studies for Fishery Management Related to Fish Stocks in the Barents Sea and Adjacent Waters. Proc 6th Russian-Norwegian Symp, Bergen, 14–17 June 1994, Institute of Marine Research, Bergen, Norway Norrbin MF, Davis CS, Gallager SM (1996) Differences in fine-scale structure and composition of zooplankton between mixed and stratified regions of Georges Bank. Deep-Sea Res II 43:1905–1924 Nøttestad L, Axelsen BE (1999) Herring schooling manoeuvres in response to killer whale attacks. Can J Zool 77:1540–1546 Ona E, Godø OR (1990) Fish reaction to trawling noise: the significance for trawl sampling. Rapp P-v Re´un Cons Int Explor Mer 189:159–166 Ona E, Dalen J, Knudsen HP, Patel R, Andersen LN, Berg S (2006) First data from sea trials with the new MS70 multibeam sonar. J Acoust Soc Am 120:3017 Orlowski A (1984) Application of multiple echoes energy measurements for evaluation of sea bottom type. Oceanologia 19:61–78 Orlowski A (1989) Application of acoustic methods to correlaøtion of fish density distribution and the type of sea bottom. Proc Inst Acoust 11:179–185 Overholtz WJ, Jech JM, Michaels WL, Jacobson LD (2006) Empirical comparisons of survey designs in acoustic surveys of Gulf of Maine-Georges Bank Atlantic herring. J Northw Atl Fish Sci 36:127–144 Parker NC, Giorgi AE, Heidinger RC, Jester DB Jr, Prince ED, Winans GA (eds) (1990) FishMarking Techniques, American Fisheries Society, Bethesda, Maryland, USA Petitgas P (1993) Geostatistics for fish stock assessment: a review and an acoustic application. ICES J Mar Sci 50:285–298 Picheral M, Grisoni J-M, Stemmann L, Gorsky G (1998) Underwater video profiler for the ‘‘in situ’’ study of suspended particulate matter. Proc MTS/IEEE Oceans 1998, pp 171–173 Pilskaln C, Villareal T, Dennett M, Darkangelo-Wood C, Meadows G (2005) High concentrations of marine snow and diatom algal mats in the North Pacific Subtropical Gyre: implications for carbon and nitrogen cycles in the oligotrophic ocean. Deep-Sea Res I 52:2315–2332 Pizarro O, Singh H (2003) Toward large-area mosaicing for underwater scientific applications. IEEE J Ocean Eng 28:651–672 Planque B, Batten SD (2000) Calanus finmarchicus in the North Atlantic: the year of Calanus in the context of interdecadal change. ICES J Mar Sci 57:1528–1535 Purcell M, von Alt C, Allen B, Austin T, Forrester N, Goldsborough R, Stokey R (2000) New capabilities of the REMUS autonomous underwater vehicle. Proc MTS/IEEE Oceans 2000, pp 147–151 Randa K (1984) Abundance and distribution of 0-group Arcto-Norwegian cod and haddock 1965–1982. In: Godø OR, Tilseth S (eds) Reproduction and recruitment of Arctic cod, Proc 1st Russian–Norwegian Symp, Leningrad, 26–30 September 1983, Institute of Marine Research, Bergen, Norway Reid DG (ed) (2000) Report on echo trace classification. ICES Coop Res Rep 238 Reynolds JR, Highsmith RC, Konar B, Wheat CG, Doudna D (2001) Fisheries and fisheries habitat investigations using undersea technology. Proc MTS/IEEE Oceans 2001, pp 812–820

188

K.G. Foote

Ricker WE (1948) Methods of estimating vital statistics of fish populations. Indiana University Publications Science Series 15:1–101 Ricker WE (1975) Computation and interpretation of biological statistics of fish populations. Bull Fish Res Board Can 191:1–382 Rivoirard J (1994) Introduction to disjunctive kriging and non-linear geostatistics. Oxford University Press, New York Rivoirard J, Simmonds J, Foote KG, Fernandes P, Bez N (2000) Geostatistics for estimating fish abundance. Blackwell Science, Oxford Roe HSJ, Griffiths G (1993) Biological information from an Acoustic Doppler Current Profiler. Mar Biol 115:339–346 Roe HSJ, Griffiths G, Hartman M, Crisp N (1996) Variability in biological distributions and hydrography from concurrent Acoustic Doppler Current Profiler and SeaSSoar surveys. ICES J Mar Sci 53:131–138 Roman C, Pizarro O, Eustice R, Singh H (2000) A new autonomous underwater vehicle for imaging research. Proc MTS/IEEE Oceans 2000, pp 153–156 Rountree RA, Gilmore RG, Goudey CA, Hawkins AD, Luczkovich JJ, Mann DA (2006) Listening to fish: applications of passive acoustics to fisheries science. Fisheries 31:433–446 Rusby JSM, Somers ML, Revie J, McCartney BS, Stubbs AR (1973) An experimental survey of a herring fishery by long-range sonar. Mar Biol 22:271–292 Samson S, Hopkins T, Remsen A, Langebrake L, Sutton T, Patten J (2001) A system for highresolution zooplankton imaging. IEEE J Ocean Eng 26:671–676 Saville A (1971) The distribution and abundance of herring larvae in the northern North Sea, changes in recent years. Rapp P-v Re´un Cons Int Explor Mer 160:87–93 Sawada K, Furusawa M (1993) Precision calibration of echo sounder by integration of standard sphere echoes. J Acoust Soc Jap (E) 14:243–249 Scott JS (1982) Selection of bottom type by groundfishes of the Scotian Shelf. Can J Fish Aquat Sci 39:943–947 Seber GAF (1973) The estimation of animal abundance and related parameters. Griffin, London Sherman K, Smith WG, Green JR, Cohen EB, Berman MS, Marti KA, Goulet JR (1987) Zooplankton production and the fisheries of the northeast shelf. In: Backus RH (ed) Georges Bank. MIT Press, Cambridge, Massachusetts, pp 268–282 Sherman K, Grosslein M, Mountain D, Busch D, O’Reilly J, Theroux R (1996) The Northeast Shelf Ecosystem: an initial perspective. In: Sherman K, Jaworsk NA, Smayda, TJ (eds) The Northeast Shelf Ecosystem: Assessment, Sustainability, and Management. Blackwell Science, Cambridge, Massachusetts, pp 103–126 Sieracki, CK, Seiracki ME, Yentsch CS (1998) An imaging-in-flow system for automated analysis of marine microplankton. Mar Ecol Prog Ser 168:285–296 Simmonds EJ, Williamson NJ, Gerlotto F, Aglen A (1992) Acoustic survey design and analysis prodecure: a comprehensive review of current practice. ICES Coop Res Rep 187:1–127 Singh H, Pizarro O (2004) Advances in large-area mosaicking underwater. IEEE J Ocean Eng 29:872–886 Singh H, Can A, Eustice R, Lerner S, McPhee N, Pizarro O, Roman C (2004a) Seabed AUV offers new platform for high-resolution imaging. Eos 85:289, 294–295 Singh H, Armstrong R, Gilbes F, Eustice R, Roman C, Pizarro O, Torres J (2004b) Imaging coral I: imaging coral habitats with the SeaBED AUV. Subsurf Sens Technol Applic 5:25–42 Squire JL, Krumboltz H (1981) Profiling pelagic fish schools using airborne optical lasers and other remote sensing techniques. Mar Technol Soc J 15:27–30 Soria M, Fre´on P, Gerlotto F (1996) Analysis of vessel influence on spatial behaviour of fish schools using a multi-beam sonar and consequences for biomass estimates by echo sounder. ICES J Mar Sci 53:453–458

6

Quantitative Research Surveys of Fish Stocks

189

Sosik, HM, Olson RJ, Neubert MG, Shalapyonok A, Solow AR (2002) Time series observations of a phytoplankton community monitored with a new submersible flow cytometer. Proc Ocean Opt XVI, 12 pp Stevenson DK, Sherman KM, Graham JJ (1989) Abundance and population dynamics of the 1986 year class of herring along the Maine coast. Rapp P-v Re´un Cons Int Explor Mer 191:345–350 Stirling I, Calvert W, Cleator H (1983) Underwater vocalizations as a tool for studying the distribution and relative abundance of wintering pinnipeds in the high Arctic. Arctic 36:262–274 Stokesbury KDE (2002) Estimation of sea scallop abundance in closed areas of Georges Bank, USA. Trans Am Fish Soc 131:1081–1092 Stokesbury KDE, Harris BP, Marino II MC, Nogueira JI (2004) Estimation of sea scallop abundance using a video survey in off-shore US waters. J Shellfish Res 23:33–40 Stokesbury KDE, Harris BP (2006) Impact of limited short-term sea scallop fishery on epibenthic community of Georges Bank closed areas. Mar Ecol Prog Ser 307:85–100 Sundby S (1991) Factors affecting the vertical distribution of eggs. ICES Mar Sci Symp 192:33–38 Sundby S, Bjørke H, Soldal AV, Olsen S (1989) Mortality rates during the early life stages and year-class strength of northeast Arctic cod (Gadus morhua L.). Rapp P-v Re´un Cons Int Explor Mer 191:351–358 Suthers IM, Frank KT (1989) Inter-annual distributions of larval and pelagic juvenile cod (Gadus morhua) in southwestern Nova Scotia determined with two different gear types. Can J Fish Aquat Sci 46:591–602 Talbot JW (1977) The dispersal of plaice eggs and larvae in the Southern Bight of the North Sea. J Cons Int Explor Mer 37:221–248 Tarifen˜o E, Andrade Y, Montesinos J (1990) An echo-acoustic method for assessing clam populations on a sandy bottom. Rapp P-v Re´un Cons Int Explor Mer 189:95–100 Tarling GA, Matthews JBL, David P, Guerin O, Buchholz F (2001) The swarm dynamics of northern krill (Meganyctiphanes norvegica) and pteropods (Cavolinia inflexa) during vertical migration in the Ligurian Sea observed by an acoustic Doppler current profiler. Deep-Sea Res I 48:1671–1686 Thorne RE (1977) Acoustic assessment of Pacific hake and herring stocks in Puget Sound, Washington and southeastern Alaska. Rapp P-v Re´un Cons Int Explor Mer 170:265–278 Thorne RE (1983) Application of hydroacoustic assessment techniques to three lakes with contrasting fish distributions. FAO Fish Rep 00:269–277 Thorne RE, Thomas GL, Foster MB (2003) Application of combined optical and acoustic technologies for fisheries and marine mammal research in Prince William Sound and Kodiak, Alaska. Proc MTS/IEEE Oceans 2003, electronic doc, 5 pp Thwaite FT, Davis CS (2002) Development of a towed, flyable fish for the Video Plankton Recorder. Proc MTS/IEEE Oceans 2002, pp 1730–1736 Tiselius P (1998) An in situ video camera for plankton studies: design and preliminary observations. Mar Ecol Prog Ser 164:293–299 Tollefsen CDS, Zedel L (2003) Evaluation of a Doppler sonar system for fisheries applications. ICES J Mar Sci 60:692–699 Tracey GA, Saade E, Stevens B, Selvitelli P, Scott J (1998) Laser line scan survey of crab habitats in Alaskan waters. J Shellfish Res 17:1483–1486 Traynor JJ, Nelson MO (1985) Methods of the U.S. hydroacoustic (echo integrator-midwater trawl) survey. Int N Pac Fish Commiss Bull 44:30–38 Traynor JJ, Ehrenberg JE (1990) Fish and standard sphere measurements obtained with a dual-beam and split-beam echo-sounding system. Rapp P-v Reun Cons Int Explor Mer 189:325–333 Trenkel V, Mazauric V, Berger L (2006) First results with the new scientific multibeam echosounder ME70. J Acoust Soc Am 120:3017

190

K.G. Foote

Trevorrow MV (1998) Salmon and herring school detection in shallow waters using sidescan sonars. Fish Res 35:5–14 Trevorrow MV (2001) An evaluation of a steerable sidescan sonar for surveys of near-surface fish. Fish Res 50:221–234 Urick RJ (1983) Principles of underwater sound. 3rd edn. McGraw-Hill, New York Wade G, Coelle-Vera A, Schlussler L, Pei SC (1975) Acoustic lenses and low-velocity fluids for improving Bragg-diffraction images. Acoust Hologr 6:345–362 Walsh SJ (1992) Factors influencing distribution of juvenile yellowtail flounder (Limanda ferruginea) on the Grand Bank of Newfoundland. Netherlands J Sea Res 29:193–203 Walsh SJ, Koeller PA, McKone WD (eds) (1993) Proceedings of the international workshop on survey trawl mensuration, Northwest Atlantic Fisheries Centre, St. John’s, Newfoundland, March 18–19, 1991. Can Tech Rep Fish Aquat Sci 1911:1–114 Warner AJ, Hayes GC (1994) Sampling by the Continuous Plankton Recorder survey. Prog Oceanogr 34:237–256 Wespestad VG, Megrey BA (1990) Assessment of walleye pollock stocks in the eastern North Pacific Ocean: an integrated analysis using research survey and commercial fisheries data. Rapp P-v Re´un Cons Int Explor Mer 189:33–49 Westervelt PJ (1963) Parametric acoustic array. J Acoust Soc Am 35:535–537 Westga˚rd T (1989) Two models of the vertical distribution of pelagic fish eggs in the turbulent upper layer of the ocean. Rapp P-v Re´un Cons Int Explor Mer 191:195–200 Weston DE (1967) Sound propagation in the presence of bladder fish. In: Albers VM (ed) Underwater Acoustics. Plenum, New York, pp 55–88 Weston DE (1972) Fisheries significance of the attenuation due to fish. J Cons Int Explor Mer 34:306–308 Weston DE, Horrigan AA, Thomas SJL, Revie J (1969) Studies of sound transmission fluctuations in shallow coastal waters. Phil Trans Roy Soc London 265:567–607 Weston DE, Revie J (1971) Fish echoes on a long range sonar display. J Sound Vib 17:105–112 Wetherall JA (1982) Analysis of double-tagging experiments. Fish Bull 80:687–701 Widder EA, Case JF, Bernstein SA, MacIntyre S, Lowenstine MR, Bowlby MR, Cook DP (1993) A new large volume bioluminescence bathyphotometer with defined turbulence excitation. Deep-Sea Res II 40:607–627 Wiebe PH, Stanton TK, Greene CH, Benfield MC, Sosik HM, Austin T, Warren JD, Hammar T (2002) Biomapper-II: an integrated instrument platform for coupled biological and physical measurements in coastal and oceanic regimes. IEEE J Ocean Eng 27:700–716 Wiebe PH, Benfield MC (2003) From the Hensen net toward four-dimensional biological oceanography. Prog Oceanogr 56:7–136 Woodd-Walker RS, Gallienne CP, Robins DB (2000) A test model for optical plankton counter (OPC) coincidence and a comparison of OPC-derived and conventional measures of plankton abundance. J Plank Res 22:473–483 Young DA (1994) The X window system: programming and applications with Xt. 2nd edn. Prentice-Hall, Englewood Cliffs, New Jersey Zedel L, Knutsen T (2000) Measurement of fish velocity using Doppler sonar. Proc MTS/IEEE Oceans 2000, pp 1951–1956 Zorn HM, Churnside JH, Oliver CW (2000) Laser safety thresholds for cetaceans and pinnepeds. Mar Mamm Sci 16:186–200

Chapter 7

Geostatistics and Their Applications to Fisheries Survey Data: A History of Ideas, 1990–2007 Pierre Petitgas

7.1 Introduction Geostatistical applications to fisheries survey data were first developed to estimate population abundance and its precision for survey designs such as the systematic design where sample points are not independent from each others (Gohin 1985, Conan 1985, Petitgas 1993a, 1996). The International Council for the Exploration of the Sea (ICES) and its Fish Technology Committee (FTC) played a major role in promoting the application of geostatistics to fisheries by organising workshops (ICES 1989, 1992, 1993). Centre de Geostatistique (Fontainebleau) was also active and organised in 1992 a course for fisheries scientists which was advertised by ICES (Armstrong et al. 1992). The course explained geostatistical modeling and provided illustrative fisheries case studies in acoustic and egg surveys. Petitgas and Lafont (1997) produced a software specifically dedicated to the geostatistical estimation of global fish abundance and its precision for a variety of survey designs. Based on the experience gained, Rivoirard et al. (2000) presented the theory, a variety of demonstrative fisheries applications to acoustic and trawl surveys and provided much guidelines in the application of geostatistics to fisheries survey data. Petitgas (2001) reviewed concepts in geostatistics and statistics as well as tools for estimating population abundance with different survey designs. Geostatistical applications have now flourished not only in the field of fisheries survey-based abundance estimation but also in marine science in general (e.g., method assumptions and ecological characteristics: Rossi et al. 1992; distribution of invertebrates: Rufino et al. 2006; variograms inside schools: Gerlotto et al. 2006; interpolation in a predator-prey space: Bulgakova et al. 2001). Methods and tools are now widely documented, available and used. It seems useful at this time, to assemble the past history of ideas and formulate challenging questions for the future. Theoretical foundations of geostatistics can be found in Matheron (1971), Journel and Huijbregts (1978) and more recently in Chile`s and Delfiner (1999). P. Petitgas (*) IFREMER, Department Ecology and Models for Fisheries, BP. 21105, 44311 cdx 9, Nantes, France

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_7, Ó Springer ScienceþBusiness Media B.V. 2009

191

192

P. Petitgas

Rivoirard et al. (2000) provides a useful guide to the geostatistical text book literature. The purpose of Geostatistics (Matheron 1971) is to model spatial variability in a variable of interest and use this model for the estimation in space of that variable or a function of it. The objective of the estimation can be mapping the variable (interpolation), estimating its mean over an area, performing a change of scale, or mapping the probability of passing a threshold. The methodology provides the estimation variance (i.e., the variance of the estimation error) for any type of survey design. Further, the geostatistical spatial structural model contains ecological information characterising aggregation patterns. Since the structural analysis is the corner stone of geostatistics, the method is of interest both to the fisheries assessment scientist and the marine ecologist. Geostatistical applications for the evaluation of marine resources have developed to consider more complex structural models than the variogram and more complex survey designs than random or systematic designs. They have also considered solutions to outliers, as well as the use of geostatistical simulations. Ecological applications have mainly dealt with ways to characterise spatial fish aggregation patterns. The spatial variographic structure has been used to characterise schooling aggregative patterns, density dependence in spatial organisation, or border effects or geometry of the area. Other tools than the variogram have been used to characterise the spatial pattern and its link to co-variates, including the inertio-gram, the D2-variogram or a point process approach. Also indicators of spatial patterns have been developed to monitor spatial distributions of fish stocks as an element of the ecosystem approach to fisheries. Each topic in this chapter will be reviewed and comments are offered that focus on the concepts. For full theoretical descriptions, the reader is referred to the literature and in particular to Rivoirard et al. (2000) or Petitgas (1996, 2001).

7.2 Abundance Estimation and Mapping 7.2.1 Geostastistical Concepts and Basic Geostatistics 7.2.1.1 Random Functions Geostatistics is applied in two steps (Matheron 1971). The first step is the structural analysis in which a model is chosen and fitted that interprets the underlying spatial continuity in the data. The second phase is that of estimation, which involves using the model to derive estimates of the variable and their estimation variances. The mathematical framework (Matheron 1971) is that of random functions: the sampled values are interpreted as the outcome of one realisation of a random function Z (Z(x1), Z(x2), . . .Z(xn),. . .) within a defined domain. The structural model (e.g., the variogram) applies to the the random function Z, not the particular realisation sampled. Inference is possible by

7

Geostatistics and Their Applications to Fisheries Survey Data

193

making stationary assumptions, which can be made at various spatial scales, e.g., for all distances in the domain (strict stationarity) or only at short distances (quasi-stationarity). The random function model is then: Z(x) = E[Z(x)] + Y(x), where the expectation E is taken over all realisations. E[Z(x)] is the drift (or trend), which due to its definition is not necessarily a smooth surface in space. Y(x) are the residuals, which have some degree of stationarity in space. The random function is a mathematical representation. Thus the practicioner will usually prefer to estimate quantities of the sampled realisation (at locations unsampled or means over blocks) rather than of the random function (drift values). Geostatistics will allow the estimation of quantities of the sampled realisation as well as the drift. In contrast, classical statistical theory (e.g., linear models) will only allow the estimation of the drift values, i.e., that of the random function (Matheron 1989, Petitgas 2001). The variogram  is the structural tool. It is defined as the half variance of increments of Z between pairs of points separated by vector distance h (Matheron 1971): g(h) = 0.5 E [(Z(x) – Z(x + h))2]. The expectation E is taken over all realisations of the random function. The intrinsic model of random function to which the variogram applies is more general than the strict stationary model: the increments Z(x)–Z(x+h) are supposed to have zero mean and a stationary semivariance (the variogram) depending on h only. The inference for the variogram is possible by assuming a certain degree of stationarity in space which allows one to estimate the expectation over different realisations by spatial averaging. Quasistationarity is sufficient in general, which in practice applies to distances smaller than a third of the sampled domain (Journel and Huijbregts 1978).

7.2.1.2 Variances When estimating the mean over a domain v by the simple average of point n P values Z(xi): Zv ¼ 1n Zðxi Þ, the geostatistical measure of precision is the i¼1

estimation variance. It is written as a function of the variogram only (Matheron  ðv; xÞ   ðv; vÞ   ðx; xÞ, where  ðv; vÞ is the 1971): 2E ¼ var½Zv  Zv  ¼ 2 mean variogram value for all distances in v (model dispersion variance in v),  ðx; xÞ is the mean variogram value for all sample points x used in the estimation (sample dispersion variance) and  ðv; xÞ is the average variogram value for all distances between each point sample x and all points of the domain v. The estimation variance depends on the geometry of the domain, the position of the samples relative to each other, and the position of the samples relative to the domain. The more continuous the spatial structure and the tighter the sampling, the smaller the estimation variance. It is noteworthy that because the estimation variance depends on the variogram and the configuration of the sampling, sampling schemes can be compared and optimised based on the variogram.

194

P. Petitgas

Matheron (1971) defined two types of variances, the dispersion variance and the estimation variance. The dispersion variance over domain v is the variance of the random function over the domain and equals  ðv; vÞ. It can be estimated by using the variogram,  ðv; vÞ, or alternatively with n random samples, P 1 ðzi  zÞ2 . The dispersion variance is the classical variance but the spatial n1 i

domain on which it applies is explicitly defined. The estimation variance is the variance of the error between the (true) value of the random function and its estimate,  2E ¼ var½Zv  Zv , where the variance is taken over all realisations of the random function. The estimation variance differs conceptually from the variance of the estimate, var½Zv , which is the classical measure of precision in statistics. Petitgas (2001) further discussed differences in estimation quantities when using geostatistics and generalised linear models (GLM: McCullagh and Nelder 1995) as GLMs have been used for estimation purposes in fisheries. Geostatistical estimates of variance are model-based. In contrast to sampling theory (e.g., Cochran 1977) where variance estimates are design-based, which requires randomising the sample locations, here the sample locations may be fixed. Matheron (1989) further discussed the passage from randomisation of samples to the use of random functions (see also Petitgas 2001). When sample points are not positioned independently from each other and when the population sampled is spatially structured the estimation of any variance requires a model of the spatial correlation in the population (Cochran 1977, Matheron 1971). Thus geostatistics solves the problem of the estimation of variance for survey designs that are not random, in particular grids of points as in ichyoplankton surveys, or parallel or zig-zag transects as in acoustic surveys (ICES 1993).

7.2.1.3 Kriging Kriging is a linear estimation procedure (Matheron 1971) that is unbiased and of minimum variance. By kriging one can estimate either point values (point kriging), mean values over blocks (block kriging) or the mean value over the entire domain (kriging the mean). Kriging not only provides estimates but also their estimation variance. Suppose we want to estimate the mean over block v centred Pon x0 by a linear combination of sample values known at points x: Zv ¼ i Zðxi Þ.  is the neighbourhood in which n samples are considered. The i2 P PP i j ðxi ; xj Þ. estimation variance writes: 2E ¼ 2 i  ðv; xi Þ   ðv; vÞ  i

i

j

The kriging weights are those that minimise the estimation variance (named kriging variance at the minimum). The minimisation can be done under the P constraint that the kriging weights sum to unity: i ¼ 1, which will ensure the i2

estimate to be unbiased. The estimation will be done with only those linear combinations of samples that filter a constant mean which can stay unknown.

7

Geostatistics and Their Applications to Fisheries Survey Data

195

In practice the constraint will result in keeping the estimate close to the neighbourhood mean of the samples. Such estimation procedure is called ordinary block kriging with a moving neighbourhood and is widely used for mapping purposes. A quasi-stationary variogram model is sufficient in which stationarity of the mean and variance applies for only those distances involved in the neighbourhood. 7.2.1.4 Comparing and Optimising Survey Designs A software tool (EVA: Petitgas and Lafont 1997) was specifically designed to calculate the estimation variance of the global mean estimate for a variety of sampling schemes used in fisheries surveys. Considering that other survey designs than the one performed would provide a similar variogram, different sampling schemes can be compared based on the estimation variance formula of the global mean estimate. On an illustrative example, Petitgas (1996) showed that a regular sampling design would have performed as precisely as the uneven design performed. Doray et al. (2008) compared different star acoustic surveys around a Fish Aggregative Device and defined the appropriate number of branches to the star. In acoustic surveys, a compromise in survey time allocation must be achieved between acoustic transect sampling for measuring fish density and trawl haul sampling for measuring fish length. Simmonds (1995) (also in Simmonds and McLennan 2005, Chap. 8) analysed the effort allocation between the number and spacing of acoustic transects and the number of trawl hauls and found that a fine tuning between more acoustic transects or more trawl hauls was not necessary for the Scottish acoustic surveys on North Sea herring.

7.2.2 Variography 7.2.2.1 Inference and Model Choice Three types of variograms can be distinguished (Matheron 1989): the regional, experimental and model variograms. The regional variogram is that of the sampled realisation if all values were known at all points. The experimental variogram is the estimate of the regional variogram based on the data samples. The model variogram is that of the underlying random function. When fitting a variogram model, one passes from the experimental variogram of one realisation to the model variogram of the random function. Matheron (1989) supplies theoretical proof that the variability between regional variograms of the same random function is small for short distances, thus it is possible in practice to infer the variogram model from one realisation of the random function (i.e., one data set). He also supplies experimental proof that models with similar parameters give similar kriging estimates and estimation variances. Variogram models are mathematically appropriate functions ensuring that calculated

196

Table 7.1 Common 2D variogram functions and their physical characteristics: C is sill, r is range, h is distance Model name Model formula Behaviour at origin h ! 0 Sill Modeled irregularity  Spherical Linear Yes Medium 3 3 Cð1:5 h=r  0:5 h =r Þ if 0  h  r C if h  r Exponential Gaussian Power

C(1 – exp(h/r)) C(1 – exp(h2/r2)) ha with 0 < a < 1 ha with a=1 ha with 1 < a < 2

Nugget



C0 if h40 0 if h ¼ 0

Linear Parabolic (horizontal tangent) Increasingly vertical tangent as a ! 0 Linear Increasingly horizontal tangent as a ! 2 Discontinuous

Asymptotic Asymptotic No

Medium Very smooth Very irregular

No No

Medium Smooth

Yes

Purely random

P. Petitgas

7

Geostatistics and Their Applications to Fisheries Survey Data

197

variances are positive. The behaviour at the origin of the variogram model (e.g., for short distances smaller than the grid mesh size) has a major influence on the estimation variance (Matheron 1971, Petitgas 2001). Table 7.1 lists different model functions often used in fisheries applications. The choice of the model function corresponds to a physical interpretation of the spatial regularity of the underlying process from which the data have been sampled (constitutive assumption: Matheron 1989). Model fitting then has two steps: first the choice of the model function then the fit of that function to estimate its parameters. 7.2.2.2 Model Validation Adequacy of the model to represent the spatial variability in the data can be checked in various ways. Goodness of fit criteria of the model to the experimental variogram can be helpful in fitting the variogram model (e.g., Fernandes and Rivoirard 1999, Rivoirard et al. 2000). Comparing the dispersion variance in the model to that of the data is also advised (Matheron 1971) as the model should contain as much variance as there is in the data: the model dispersion variance over the entire domain  ðV; VÞ should be close to that of the data variance (if the variogram model is fitted for all distances in V: strict stationarity) or alternatively the model dispersion variance over a small block  ðv; vÞshould be close to the data variance within such block size (if the variogram model is fitted for only small distances: quasi-stationarity). Last, crossvalidation of data values by kriging provides a way to measure how the model and the kriging procedure reproduce the data (e.g., Journel and Huijbregts 1978). 7.2.2.3 Variogram Characteristics Important characteristics of the variogram are the nugget, sill, range and anisotropy. The nugget is a discontinuity of amplitude C0 at the origin of the variogram (Table 7.1). It has three interpretations which cannot be distinguished in practice. These are a purely random component, a measurement error, and spatial structures with range smaller than the grid mesh size. The sill is the variance of the random function. Because the (dispersion) variance is a function of the domain on which it is computed (  ðv; vÞ), the variogram sill and the data variance need not coincice. In general, the variogram sill will be greater than the data variance. They will be close in value when the variogram range is short relative to the domain studied (stationary case). The range is the distance at which correlation vanishes. It relates to the average dimension of patches of either low or high values. The anisotropy models directional differences in the spatial variation. In a geometric anisotropic variogram model, all directions have the same sill but the range varies elliptically with direction (e.g., fish aggregations are elliptical rather than circular). In zonal anisotropy, the sill varies with direction meaning that the spatial distribution is more heterogeneous is certain directions (e.g., an in-shore off-shore gradient in fish density

198

P. Petitgas

can be modelled with zonal anisotropy). A variogram model can be the sum of various variogram functions when the data show different nested structures (e.g., nugget + isotropic spherical variogram + linear directional variogram). A variety of case studies in modelling experimental variograms can be found in Journel and Huijbregts (1978) and Chile`s and Delfiner (1999). Petitgas (1996) also documents the variety of variogram models of interest in fisheries applications. 7.2.2.4 Variogram Estimates The classical experimental variogram (Matheron 1971) is: P 1 ðzðxÞ  zðyÞÞ2 , where n(u, h) is the number of pairs of   ðu; hÞ ¼ 2 nðu;hÞ jxyjh

points (x, y) separated by distance class h in direction class u and z(x) the sample value at location x. Distance and direction classes are defined depending on the sampling scheme configuration to ensure a sufficient number of pairs in each class. It is recommended to compute the variogram for distances not exceeding half of the maximum dimension of the domain studied (Journel and Huijbregts 1978). In fisheries applications, sample point coordinates are usually in navigational units of degrees of latitude and longitude. To compute distances between sample points a transformation of the coordinates to geographic units is necessary. Because the classical estimate of the variogram involves a square difference between sample values, the experimental variogram may be erratic. Thus capturing the spatial structure with that estimate may be difficult although there is spatial structure in the data. Alternative estimates of the variogram have been proposed for dealing with the effect of zeroes, or high values, or inhomogeneity in the sampling locations (Rivoirard et al. 2000). The irregularity in the sampling design (e.g., clusters of points in particular areas) can affect the experimental variogram. Spatial weights can be given to sample points (e.g., area of influence) leading to a weighted P P wx wy ðzðxÞ  zðyÞÞ2 = wx wy , where wx is variogram estimate: 0:5 jxyjh

jxyjh

the spatial weight of point x. Rivoirard et al. (2000) have used such estimate in the case of zig-zag acoustic surveys. Also, in acoustic surveys the interaction between spatial and temporal variability may be such that the along transect variogram can be more easy to interpret than the 2D variogram calculated along and across transects. Another situation is when the spatial distribution shows many zeroes with occasional patches of positive values. Then a variogram estimate based on the non-centred covariance can be better adapted to P P 1 zðxÞzðyÞ, where N is the reveal the spatial structure: N12 zðxi Þ2  nðhÞ i

jxyjh

total point number. This covariance estimate of the variogram may underestimate the sill due to departure from strict stationarity thus care should be taken in comparing the model dispersion variance with the data variance. For dealing with high values, transformations of the variable have been suggested

7

Geostatistics and Their Applications to Fisheries Survey Data

199

based on distributional assumptions. Cressie (1991) proposed a variogram estimate robust to outliers from a Gaussian distribution. Guiblin et al. (1995) (also in Rivoirard et al. 2000) suggested to log-transform the data Z into L (L=Ln(1+Z/b)), estimate the variogram of the log-transform L using the classical estimate, model the variogram of the log-transform  L and back-transform this variogram model to obtain h i the variogram model of 2 the original data : ðhÞ ¼ ðm þ bÞ þ varZ 1  expðð2 L ðhÞ=varLÞÞ h i with 2 ¼ Ln 1 þ varZ=ðm þ bÞ2 and m=E[Z]. Such a procedure has been successfully used on Northern North Sea herring (Rivoirard et al. 2000) and its robustness tested by simulations. 7.2.2.5 Automated Fitting Procedures Once the variogram model function is chosen, its parameters can be fitted by eye or by an automated algorithm using a least squares procedure. Cressie (1991) and Chile`s and Delfiner (1999) document a variety of statistical fitting procedures. Fernandes and Rivoirard (1999) (also in Rivoirard et al. 2000) used weighted least squares to estimate variogram model parameters and compared models with a goodness of fit criteria. The function to be minimized was: P qðbÞ ¼ wj   ðhj Þ  ðhi ; bÞ where b is the set of model variogram parameters j

and * denotes the variogram estimate. The weights wj can be proportional to the number of pairs of points in distance class j or an inverse power of the distance hj. This second possibility will ensure aPgood variogram fit for small distances. wj ½  ðhj Þðhi ;bmin Þ j P  The goodness of fit criteria was: gof ¼ where bmin is the set of w  ðh Þ j

j

j

fitted variogram model parameters. Monitoring yearly fisheries surveys result in a time series of spatial data in which the spatial structure of a given species shows both variability and some consistency across years. One would therefore like to model a variogram in each year as well as coherently across all years. Bellier et al. (2007) considered that all years had a similar variogram model function (e.g., spherical) but that the parameters varied between years. They fitted a spherical variogram model to a set of yearly experimental variograms using non-linear mixed-effects regression (Pinheiro and Bates 2000). They used fixed effects for the range parameter and random effects for the sill and nugget, which resulted in estimating a constant range across all years and yearly variable sill and nugget.

7.2.3 Multivariate Approaches The correlation information between a target fish species and explanatory covariates may be used to improve the estimation of the target species.

200

P. Petitgas

Multivariate geostatistics comprise a diversity of methods adapted to different situations such as e.g., particular covariation between covariate and target, or differences in the sampling configurations between covariate and target or when a drift surface is considered. A comprehensive presentation of multivariate geostatistics and related topics can be found in Wackernagel (1995) or Chile`s and Delfiner (1999). Rivoirard (1994) provides a simple introduction to cokriging and its relationship with non-linear geostatistics. Here we shall review the different situations that have been encountered so far in fisheries applications when considering an ancillary and a target variable. Two classes of covariates should be distinguished, spatial and non spatial. Non spatial covariates are explanatory variables controlling fish behaviour (e.g., time of day) and therefore variation in fish density is independent of space. Spatial covariates are explanatory variables (e.g., bottom depth, river plume) that covary in space with fish density and therefore can explain the fish spatial distribution. 7.2.3.1 Universal Kriging Large scale drifts in the data can result from the response of fish concentration to explanatory environmental parameters (e.g., a gradient in fish density from coast to off-shore depending on bottom depth). In the Universal kriging model the drift is separated from the residuals and explicitly modelled at large scale. Matheron (1971) showed that drift and residuals could not be estimated together using only one realisation of the random function (i.e., one data set from 1 year): the variability in the (estimated) residuals that result from the spatial smoothing procedure in estimating the drift is an underestimate of the (true) residual process variability. Rivoirard and Guiblin (1997) advocate considering a bias term coming from the estimation of the drift which is needed when estimating the estimation variance of the mean estimate. Ancillary covariates have been used to estimate the drift by regression. Sullivan (1991) was confronted with a gradient in demersal fish density with bottom depth. He took advantage of the fact that the drift in the fish density developed in one geographical direction only, across the isobaths. The variogram of the residuals was then estimated along the isobaths and applied in all directions. The drift was estimated across the isobaths using a regression with depth. A Universal kriging procedure was then used for mapping. When the drift is consistent in time, repeated surveys have been used to estimate the drift directly as the mean over diffferent realisations of the process. Using repeated surveys on the same grid of stations, Petitgas (1997) estimated the dome-shape drift in sole egg distributions on a spawning ground by averaging egg density in time. The model developed was multiplicative as the residual variance was proportional to the drift. Doray et al. (2008) worked on repeated surveys over a tuna aggregation around a FAD (fish aggregative device). The drift was first estimated by time averaging then modelled using an advection diffusion equation. It formulated the balance between oriented movements toward the aggregation centre and dispersive non oriented movements and resulted in modeling the decrease of fish

7

Geostatistics and Their Applications to Fisheries Survey Data

201

concentration from the FAD’s head to the border of the aggregation. Residuals from the advection diffusion model were estimated. The variogram of the residuals was used to estimate the mean density around the FAD by kriging. The variogram was also used to optimise the star acoustic survey design. 7.2.3.2 External Drift With the external drift procedure, the kriged estimates are constrained to conform to the shape of an ‘external’ variable in time or space. This is achieved by imposing constraints on the kriging weights. Constraints on the kriging weights can be extended to filter other functions than a constant mean. Suppose the drift is linearly related to an explanatory variable f(x): E [Z(x)]P =af(x) + b. i ¼ 1 and Imposing supplementary conditions on the kriging weights: i2 P i fðxi Þ ¼ fðx0 Þ will result in constructing the estimate with only those linear i2

combinations of the samples that filter the drift whatever the values of a and b. The result is that the kriged estimates will conform to the shape of the variable f(x). The price to pay is that the kriging variance will be increased as the number of constraints is increased. The values of f(x) need to be known at all sample points as well as at all points or blocks to be estimated. The variogram to be used is that of the residuals Y(x). The external drift procedure allows one to adjust the drift to conform to a particular functional relationship with a covariate, which plays the role of a guiding variable. The procedure is helpful when the target variable is undersampled in comparison to the covariate or for dealing with a drift that has a functional relationship with a covariate. Rivoirard et al. (2000) (also in Guiblin et al. 1996) and Petitgas et al. (2003b) used an external drift procedure to map fish length while accounting for a linear relationship between fish length and bottom depth. Rivoirard and Wieland (2001) accounted for the effect of time of day on trawl haul catch using an external drift procedure and mapped the fish spatial distribution at a given time using both day and night samples. Bouleau et al. (2004) investigated how densily sampled acoustic data between trawl haul stations could help the mapping of the sparcely sampled trawl haul data (heterotopic and undersampled configuration). An external drift procedure was used to map the trawl haul data while following the spatial distribution of the acoustically sampled fish density. The map of the acoustic data was first estimated by ordinary kriging and represented the shape surface to conform to. Its values were available at all trawl stations as well as all points to be estimated. The external drift procedure was then used to guide the mapping of the trawl hauls with the map of the acoustic data. 7.2.3.3 Co-kriging In this section, the cross-variogram is the structural tool. It is an extension of the variogram for multivariate random functions: ij ðhÞ ¼ 0:5E½ðZi ðxÞ  Zi ðx þ hÞÞ ðZj ðxÞ  Zj ðx þ hÞÞ, where i and j are indices for different co-varying variables.

202

P. Petitgas

The cross-variogram is a symmetrical function of h. Also both variables play a similar role. The coherent fitting of variograms and cross-variograms is not always an easy task (Wackernagel 1995) and simplifications of co-kriging have been developed depending on the variograms and cross-variograms structures. There are simplifying cases: factorisation of covariates and intrinsic correlation. Until now only intinsic correlation has been used in fisheries applications. Intrinsic correlation applies when cross-variograms and variograms are all proportional to a same variogram: ij ðhÞ ¼ 2ij ðhÞ. Co-kriging is helpful for improving the estimate of the target by using correlated covariates known at more locations than the target (undersampled configuration) or for ensuring coherence in the estimated values when variables are functionally related (the co-kriging estimate will comform to the relationship). In their analysis of acoustic and trawl data and in addition to using an external drift procedure, Bouleau et al. (2004) also used an intrinsic co-kriging model to map trawl haul data using both acoustic and trawl data. The authors compared maps and estimation variances obtained by co-kriging and external drift to that obtained by kriging the trawl data. Co-kriging and External drift methods made use of the acoustic data between trawl hauls while the kriging of the trawl hauls did not make use of the acoustic information. Co-kriging or external drift provided similar maps. They had more details than the univariate kriged map, in particular they were less smooth and areas of high abundance were more restricted. The external drift had slightly increased estimation variance in comparison to the co-kriging, as is expected by theory. Petitgas (1991) proposed a co-kriging model accounting for the aggregation/disaggregation of fish schools between day and night. The day and night samples were functionally related considering that the night fish density at point x was equal R to the average of the day values over area v centred on x: Znight ðxÞ ¼ 1v Zday ðx þ uÞdu. The relationship v

allowed one to specify coherently day and night variograms as well as derive the cross-variogram model between day and night values. The co-kriging pocedure allowed one to estimate a day (or night) map using both day and night samples. In all, multivariate approaches allowed one to make full and coherent use of the available information which resulted in more realistic maps. But the estimation variances were not dramatically decreased in comparison to the univariate case. The reason why is perhaps that the mathematical dimension of the estimation problem was increased in the multivariate situation in comparison to the univariate case as more sources of variations in the target variable were considered.

7.2.4 Simulations The interest in simulations is to generate maps that contain all the variability that is in the data. In contrast to kriging which results in a smoothed interpolated surface, a simulated field shows all the variability in the process while

7

Geostatistics and Their Applications to Fisheries Survey Data

203

respecting the data variogram and the histogram. A non conditional simulation is one realisation of the random function model. A conditional simulation is one realisation that conforms to the sample point values. Simulations are an appropriate tool for evaluating the impact of spatial uncertainty on the result of complex procedures. They are useful when estimating derived variables that require variabilty in the surveyed variables (e.g., estimating the length of the sea floor to an island for designing a cable), or computing the estimation variance of estimates that are themselves a non-linear combination of the surveyed variables (e.g., combining fish length and acoustic backscatter for estimating fish abundance in acoustic surveys), or testing estimation procedures and survey design (e.g., testing different rules for adding samples in adaptive sampling). Chile`s and Delfiner (1999) and Lantue´joul (2002) document comprehensively the many methods and algorithms for simulating spatially structured random functions with defined histogram and covariance, along a line (1D), on the plane (2D) or in 3D, as well as the conditioning to the data values. Among the variety of methods, the turning bands method due to Matheron (1973) is practical and efficient. The different steps for constructing a conditional simulation with the turning bands method are assembled as a flow chart on Fig. 7.1. Raw data

1 : Gaussian transform

Gaussian data

2 : Joint variography Variogram of Gaussian

3 : Turning bands

Non conditional simulation of Gaussian 4 : Conditioning by kriging

Conditional simulation of raw data

5 : Back-transform

Conditional simulation of Gaussian

Fig. 7.1 Flow chart showing the steps for constructing a geostatistical conditional simulation that match the histogram, the variogram and the data samples, using the turning bands method and conditioning by kriging. Non conditional simulation will match histogram and variogram. A conditional simulation will match histogram, variogram and data values. In Step 1 when transforming the raw data into a Gaussian, the use of a Gibbs sampler can be helpful for dealing with many zeroe values. Step 2 is in fact a joint analysis where variograms of raw and transformed data are modelled coherently. Steps 3 and 4 can be obtained directly by other methods, e.g., sequential gaussian simulation, that requires bi-gaussian assumption of the transformed data [adapted from Chile`s and Delfiner (1999)]

204

P. Petitgas

The anamorphosis and back-transform steps result in matching the simulated values to the histogram of the data. The turning bands method results in matching the covariance of the simulated values to that of the data. Last, kriging is used to match the simulated values to the data values resulting in a conditional simulation.

7.2.4.1 Gaussian Anamorphosis Because the methods simulate Gaussian random functions, it is useful to Gaussian transform the data. A Gaussian anamorphosis is used where the Gaussian value y is associated to the data value z that has the same cumulative probability: P(Z(x)yc using an invertable monotone anamorphosis function, leading to a truncated gaussian variable at cut-off yc (P[Z(x) = 0] = P[Y(x) < yc]). How to assign Gaussian values to the zeroes in order to match the covariance structure and knowing the Gaussian values Y(x)>yc? It is useful here to use a Gibbs sampler otherwise the assignment of Gaussian values Y(x)
7.2.4.2 The Turning Bands Method Chile`s and Delfiner (1999) and Lantue´joul (2002) provide full and up-to-date documentation of the method. The turning bands method constructs a simulation in is the projection of point x of
on the line of direction d. Using a large number of lines n, the central limit theorem implies that Y is Gaussian. The directions d can be random or following a quasi-random sequence (e.g., van der Corput sequence) that is more efficient in filling space evenly as n increases. In 3D the isotropic covariance C3(r) of Y is simply related to the 1D covariance C1(r) of X1 as: C1 ðrÞ ¼ drd ½rC3 ðrÞ. For each covariance model classically used (e.g., Table 7.1)

7

Geostatistics and Their Applications to Fisheries Survey Data

205

the associated 1D covariance is known. The simulation of a random process along a line with a specified covariance C1 can be constructed with a variety of methods, e.g., using autoregressive or dilution algorithms. Note that because the relationship between the 1D and the 2D covariances is more complicated than that between the 1D and the 3D covariances, it is easier to simulate in the 2D plane as if it was a section of the 3D space. In comparison to other simulation methods, the turning bands method offers flexibility in the implementation as well as control on the simulated covariance structure. Also, it allows one to simulate a large number of points in the simulated fields with computational efficiency. 7.2.4.3 Conditioning by Kriging Some methods based on more assumptions than the Turning bands method (e.g., sequential gaussian simulation) allow one to simulate conditionally to the data directly but the Turning bands method does not. Conditioning is obtained by simulating non conditionnally a kriging error. The steps are as follows: (i) perform a non conditional simulation of Z at the nodes x of the simulation grid: Snc(x) and at the data points x: Snc(x); (ii) perform kriging at the grid nodes x using the data Z(x): Z k(x); (iii) perform kriging at the grid nodes x using the simulated values at the data points Snc(x): Snck(x). The conditional simulation at the grid nodes Sc(x) is constructed as: k ðxÞ. Since kriging is an exact interpolator, the Sc ðxÞ ¼ Z k ðxÞ þ ½Snc ðxÞ  Snc conditional simulation conforms to the data values. 7.2.4.4 Testing Homogeneous Survey Designs and Variogram Estimators Since transects in acoustic surveys are being sampled continuously, it has been suggested to sum the recorded fish density along the transect lines and estimate fish stocks using a one-dimensional procedure (Petitgas 1993a). Following on this idea, Simmonds and Fryer (1996) (also in Rivoirard et al. 2000, Chap. 5) tested a variety of survey designs applied to a variety of simulated onedimensional processes. Designs considered were parallel transects that were randomly or regularly spaced or randomly spaced within strata and zig-zag transects. The 1D simulations were non conditional and used an autoregressive method. Simulated processes varied in their degree of correlation (nugget, range) as well as in the incorporation of a trend (linear). Designs were ranked considering precision on the mean estimate, bias, and precision on the variance. The conclusion was that the systematic design (regularly spaced transects) was the best strategy for estimating abundance with highest precision and negligible bias. Random stratified design with two or one transect per strata were ranked as best strategies (but close to the systematic design) when the aim was both the estimate of the mean and that of the variance. The interest in using a zig-zag transects design in comparison to a parallel transects design depended on the correlation range and transects spacing. Rivoirard et al.

206

P. Petitgas

(2000, Chap. 5) also used simulations to compare variogram estimators. They used similar sets of simulations (nugget, range, linear trend) than previously in addition to varying skewness in the histogram. Three variogram estimators (classical, log back transform, non-centred covariance) were tested for their ability to infer the simulated variogram structure for a large range of skew in the data. Relative bias in variogram parameters was less than 5% with all variogram estimates. 7.2.4.5 Investigating Adaptive Survey Designs The sampling design is homogeneous when sample points cover the surveyed domain independently from the underlying spatial distribution being sampled. The position of sample points can be random, stratified random, or on a regular or irregular grid. Geostatistics by modelling the spatial covariance and weighting samples by kriging gives some flexibility in the design. When the design incorporates additional sample points in areas of higher abundance with the idea to gain information in those areas, the design is heterogeneous or adaptive. Adaptive designs result in targetting the sampling effort in positive areas rather than dispersing the effort in all areas including empty ones. Simmonds and MacLennan (2005) describe various adaptive rules for fisheries acoustic surveys. Typically, an adaptive design is a two stage sampling procedure. Level 1 samples are located according to a homogeneous sampling scheme. Then level 2 samples are added in the vicinity of level 1 samples conditionally on the values observed at those level 1 samples. Adaptive sampling suffers the risk of bias in the design. A rich block will contain low and rich values. When sampling the rich block, if the level 1 sample is low, no additional sample will be added and the block will be considered low in abundance. In contrast, if the level 1 sample has a high value, additional samples will be added and lower values will be sampled. The result is systematic underestimation of rich areas. Clearly, the bias depends on the rule adopted to allocate additional samples. And simulations have been used to evaluate the bias associated to particular rules. With a design-based approach Thompson and Seber (1996) proposed an unbias adaptive sampling rule (adaptive cluster sampling) with corresponding estimators. The cluster around the rich value that triggers the addition of samples must be sampled entirely. Such design and corresponding mean estimates were applied by Lo et al. (1997) for a larval survey. Conners and Schwager (2002) simulated non conditionally 2D fields of Gaussian values with various cases of spatial correlation to test adaptive cluster sampling against homogeneous designs. Gaussian simulated values were exponentiated. Correlation in the simulated values was obtained with an ‘ad hoc’ procedure (not allowing complete control of the simulated variogram). The adaptive cluster design performed with no more bias than homogeneous designs and had higher precision in cases of patchy distributions. A geostatistical approach to adaptive sampling is to model the relationship between point values and that of block

7

Geostatistics and Their Applications to Fisheries Survey Data

207

means or to model the position of high values relative to lower ones. Non-linear geostatistics can be helpful in analysing data sampled with an adaptive rule. Petitgas (1997b) post-stratified ichtyoplankton survey data based on the correlation structure between high and low abundance areas. Petitgas (2004) used simulations to test bias in the design depending on the adaptive rule. 2D fields were non conditionally simulated using the Turning bands method and the Gaussian values were exponentiated. The level 1 samples were positioned on a systematic grid of points with a mesh size equal to the variogram range. The rule for adding level 2 samples was based on the mean of 3 consecutive level 1 samples. That rule allowed one to obtain higher precision on the mean than a systematic design that had more points while the bias in the mean stayed lower than 3%.

7.2.4.6 Variance Estimate of Combined Variables The abundance of a given species derived from acoustic surveys is a combination of the species length and the acoustic backscatter assigned to that species, where each variable comes from a specific sampling process (e.g., Simmonds and McLennan 2005). The fish length is measured by sampling fish schools with appropriate gear using (quasi) point sampling with pelagic trawl hauls or purse seine sets. The acoustic backscatter is (quasi) continuously recorded by the echosounder along the sailed transect line and integrated over depth and a unit sailed distance (one nautical mile). Fish density is then estimated by a nonsAx linear combination of the two variables: ZðxÞ ¼ 2 b=10 where sA is the acouslx 10 tic (calibrated) nautical area scattering coefficient (sensu Simmonds and McLennan 2005), l 2 is the mean fish length squared and b is the (known) coefficient of the species target strength to length relationship. Fish density Z(x) can be further assigned to ages using an age-length key giving the proportion of ages at any given length p(a,l): Z(a,x)=p(a,l)Z(x). In his study on optimising effort allocation between acoustic transects and trawl hauls, Simmonds (1995) considered variance terms for the different sampling processes but the correct variance of the mean estimate of Z was not established. The estimation variance of the mean fish density over the surveyed area is best obtained by geostatistical conditional simulations, allowing for the combination of the many possible maps (realisations) of length and acoustic backscatter that contain the correct spatial variability in each variable. Working with the Scottish acoustic surveys for North Sea herring, Gimona and Fernandes (2003) and Woillez et al. (2006a) estimated the error variance in the mean fish abundance by conditional simulations. Gimona and Fernandes (2003) used sequential Gaussian simulations and Woillez et al. (2006a) used the turning bands method. Conditional simulations of the fish length and the acoustic backscatter sA were performed on the same grid of points. Then the maps were combined to estimate the map of the fish density using the above acoustic formula:

208

P. Petitgas

Zði; rÞ ¼ fðsAi;r ; li;r 2 Þ where i is the index for the grid points and r corresponds to the realisations. The maps of Z(i,r) were averaged over the domain V to obtain the mean fish abundance in each realisation ZV (r). For a large number of realisations, the histogram of ZV (r), its mean E [ZV (r)] and variance Var[ZV (r)] were estimated. The estimation variance of the mean estimate was Var[ZV (r)]. The bias in the simulations was: E ½ZV ðrÞ  ZV , where ZV was the mean survey estimate. Simulating fish length caused no problem because structure was well defined and the variable is close to Gaussian. In contrast, the simulation of the acoustic backscatter was more difficult because of the large amount of zero values in the data. To solve that problem Woillez et al. (2006a) used a Gibbs sampler to coherently assign Gaussian values to the zero samples knowing the surrounding positive values and the covariance structure of the raw data. The resulting relative estimation error was close to 15% for the different surveys analysed. This is the order of magnitude that can be expected in acoustic surveys when the assignment of echo-traces to species is considered without error, as was the case in the present example.

7.3 Ecological Considerations In contrast to the previous part in which spatial variation was modelled to serve the estimation process, in this part we focus on different ways by which geostatistical structural analysis tools can be used to reveal ecologically meaningful characteristics in the spatial variation and enhance ecological understanding.

7.3.1 Spatial Relationships Between Variables There can be many ways by which two variables are related and therefore spatial relationships between variables can be investigated in several ways. A variety of tools have been suggested to characterise collocation, covariation and conditional variation. Collocation characterises the point-to-point agreement of two spatial distributions while covariation captures how the change in values as a function of distance (e.g., gradients) is correlated between two maps. Conditional variation looks at the variation of one variable relative to the other. 7.3.1.1 Overlap The Global index of collocation (GIC: Bez and Rivoirard 2000a, Table 7.2) is a measure of how closely collocated two spatial distributions are. The index provides a measure of overlap between two spatial distributions and can serve as a simple distance between maps allowing for the classification of maps.

i¼pðaÞ

Attribute

Index name

Index description

Occupation

Positive area

Area of non null values

Formula X si Izi 40 PA ¼

Aggregation

Spreading area

Spatial concentration of abundance relative to Z1 QðaÞ a homogeneous distribution Þda SA ¼ 2 ð1  Q

Equivalent area

Integral range of the relative covariogram, also EA ¼ Q2 =gð0Þ the inverse probability for two random individuals to be at same location Z Weighted average of sample positions zðxÞ CG ¼ x dx Q rule: rank z in decreasing order, start Number of Patches as defined using a distance computing CG of richest values; if y too threshold distant from CG of previous values, consider new patch; continue Z zðxÞ Weighted variance of sample positions around dx I ¼ ðx  CGÞ2 a gravity centre Q ffi Ratio of inertia for directions carrying minimal A ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Imax =Imin and maximal inertia

Reference Woillez et al. (2007)

i

Woillez et al. (2007)

0

Location

Gravity Center (CG) Number of Patches

Dispersion

Inertia (I) Anisptropy

Correlation

Decrease of correlation at short distance

Range

Distance beyond which correlation vanishes

gð0Þ  gðh0 Þ gð0Þ First u for which g(u)=0

Ratio of distance between gravity centres and random individuals

GIC ¼ 1 

Overlap Global index of between two collocation distributions

MI ¼

CG2 CG2 þ I1 þ I2

Bez et al. (2001) Woillez et al. (2007)

Bez et al. (2001) Woillez et al. (2007) Woillez et al. (2007) Matheron (1971) Bez et al. (2000a)

209

Microstructure index

Bez et al. (2001)

Geostatistics and Their Applications to Fisheries Survey Data

i¼pðaÞ

distance between nearest sample neighbours)

7

Table 7.2 Attributes of spatial distributions and indices to characterise them. Notations are as follows. i: index of samples; z(x): fish density at x zi > 0 and 0 otherwise; Q: total abundance as defined by the spatial (counts of fish perRn.m.2); si: area of influence of sample i; Izi: equals 1 if sample R integral of z:Q ¼ zðxÞdx; g: geostatistical transitive covariogram:gðhÞ ¼ zðxÞzðx þ hÞdx; Q(a): summed abundance from richest values that stand N N P P on area a:QðaÞ ¼ si zi where a is a proportion of total domain: a ¼ si /A; CG: centre of gravity; h0: a lag distance chosen as appropriate (mean

210

P. Petitgas

The GIC can be used to investigate inter-annual changes in the spatial distribution of the same fish or between ages to characterise differences in habitats across the life cycle (Woillez et al. 2007). It could also be helpful as a measure of spatial overlap between predator and prey. For investigating more complexe relationships between spatial distributions other tools are helpful.

7.3.1.2 Inertiogram To investigate whether a fish distribution is constrained to particular domains of the distribution of an explanatory variable, Bez and Rivoirard (2000b) suggested the use of the inertiogram. The concept was illustrated with fish egg spatial distributions and temperature fields. The control of the fish egg spatial distribution by that of temperature was tested by translating the temperature field relative to the egg distribution. Temperature was weighted by egg abundance occurring at the same location leading one to estimate the mean temperature per individual egg as well as the temperature variance per egg. The inertiogram was the temperature variance per egg as a function of the vector translation distance. The inertiogram map was constructed by considering translations in different directions. The inertiogram maps allowed one to visualise whether a particular temperature range or spatial domain controlled the fish egg distribution. If so, the inertiogram showed low values in those areas. In the previous examples, the structural tools were based on the geostatistical transitive methodology (Matheron 1971, Petitgas 1993a, Bez 2002). The transitive method deals simply with the zeroe values without the need to delineate the domain of presence. It is appropriate for case studies where many zeroes occur in the data. The structure in the transitive covariogram then characterises both the intrinsic spatial structure of the variable (that characterised by the variogram) as well as the influence of the geometry of the domain (e.g., lower values near the borders). In intrinsic geostatistics (usually named geostatistics: Matheron 1971), the domain of study is assumed to be known with no influence on the spatial distribution of the variable of interest.

7.3.1.3 Cross-Variogram The cross-variogram characterises the spatial covariation between two continuous variables. It is symetrical as both variables play the same role in the analysis. It is the structural tool in multivariate linear geostatistics (Wackernagel 1995). Barange et al. (2005) used covariograms to analyse how sardine and anchovy were spatially organised one relative to the other for different years with different total abundances. The analysis revealed that sardine and anchovy spatially alternated in years of low abundance while they co-occurred in years of high abundance.

7

Geostatistics and Their Applications to Fisheries Survey Data

211

7.3.1.4 Ratio of Cross and Simple Indicator Variograms The cross-variogram is also a structural tool in non-linear geostatistics (Rivoirard 1994) as a non linear model is a multivariate model for all indicator cut offs of the variable (the indicator Iz(x)>c equals 1 when the value z(x) is greater than the cut-off c and 0 otherwise). Building from the properties of the non-linear model (e.g., diffusive or not), cross-variograms of indicators are helpful in ecology to investigate whether transitions are progressive or sharp when crossing a border. Indicators of the target variable spatially define geometrical sets or domains. The cross-variogram of two indicators (Iz(x)c1 and Iz(x)c2 with c1  c2) divided by the variogram of the indicator of the low cut-off c1 represents the probability as a function of distance h to encounter values higher than the high cut-off c2 when in the domain of the low cut-off c1 (one extremity of the vector h is outside the domain of c1 and the other inside that of c2): Ic1 xIc2 ðhÞ I ðhÞ

¼ Prob½Zðx þ hÞ  c2 = ðZðxÞ5c1; Zðx þ hÞ  c1Þ. Petitgas (1993b)

c1

analysed the spatial setting of high herring densities relatively to low and medium densities. The analysis revealed that high densities could occur anywhere within the domain of medium values, which was a large domain, meaning that high densities were difficult to predict. One can also envisage to use these tools when an explanatory variable is given as a polygon. For instance, polygons can represent a characteristic of the environment, e.g., the area of a river plume, the area where gyres develop, etc. The cross variogram between the polygon indicator and the target variable will allow one to investigate how the target variable responds spatially to the polygon.

7.3.1.5 Constrained Variogram In a similar way than above and to demonstrate change in the spatial continuity of a target variable when crossing the border of geometrical sets defined by the spatial setting of an ancillary variable, the variogram can be computed with a selected set of samples resulting in constrained variograms. The selection can be for those pairs of points that are inside or outside defined spatial sets or that each stand on one side of the border limit. The difference between the constrained variogram and the overall variogram computed on all pairs of points irrespective of the border limits will serve as a test to demonstrate the impact of a particular explanatory variable on the spatial structure of the target variable. Rivoirard et al. (2000, Chap. 4) investigated the influence of the shelfbreak contour on the variogram of blue whiting. They computed a ‘constrained’ variogram with only those pairs of points that fell close to the shelfbreak contour. The constrained variogram was lower than the overall variogram, meaning that the shelf break was associated with a difference in variance between along and across the contour.

212

P. Petitgas

7.3.1.6 D2-Variogram At sample point x the variable of interest may not always be one value z(x) but a p-component vector (v1(x), v2(x), . . .,vp(x)). For instance, when characterising schools with many parameters, one may construct a p-component vector at the spatial scale of a few nautical miles to characterise the acoustic image (echogram). What is then the spatial structure of the acoustic images? Petitgas (2003)  2 P 2 1 VkðxÞ  Vk ðx þ hÞ . It showed suggested the D2-variogram: D ðhÞ ¼ 2 E k

good spatial structure of acoustic images for the Bay of Biscay echograms.

7.3.2 Indices of Spatial Pattern Indices have been developed at various spatial scales to summarise various aspects of spatial distributions that cannot be explicitly characterised with the variogram. Such indices can then be used in a monitoring approach of spatial distributions. Indices have been developed to characterise the spatial distribution of fish density values over a few nautical miles, the schooling pattern and the clustering pattern of schools. The multi-scale organisation of spatial distributions will be discussed using density-dependence as making the link between scales of spatial organisation. 7.3.2.1 Indices for Density Values Until now we have used tools that mainly characterise correlation. But correlation is only one aspect of spatial distributions. How then can one characterise a spatial distribution in its many aspects? For a full characterisation of the many properties of spatial distributions a variety of geostatistical indices have been developed. Woillez et al. (2007) proposed a list of 10 indices (Table 7.2) to characterise occupation, aggregation, location, dispersion, correlation and overlap. These notions are somewhat related (e.g., aggregation, dispersion and occupation) and formal relationships exist between indices (Woillez et al. 2007). The centre of gravity of a population with a measure of dispersion around it had been proposed already (Swain and Sinclair 1994, Atkinson et al. 1997, Bez and Rivoirard 2001). The occupation and aggregation indices are not truely spatial in the sense that they are sensitive to the histogram and not to the spatial location of values. Various indices to characterise aggregation have been suggested (area coverage: Swain and Sinclair 1994, Gini index: Myers and Cadigan 1995, spatial selectivity index: Petitgas 1998) which all relate to the area associated with the highest values. But the spreading index is more general in the sense that the amount of zeroes do not affect this index. Therefore in calculation of the spreading index the delineation of the positive data domain is not necessary. Spatial indices are useful in characterising the spatial organisation of the life cycle. It can be demonstrated that young immatures, young

7

Geostatistics and Their Applications to Fisheries Survey Data

213

matures and older matures differ in some aspects of their spatial distributions, in particular for location, aggregation and dispersion (Woillez et al. 2007). Also, these indices have the potential to be used in a monitoring system so as to detect changes in the spatial distributions, which could be helpful in the context of climate change or habitat conservation for fish stocks. 7.3.2.2 Indices of Schooling Patterns The previous spatial indices characterise how density values (numbers of fish per unit area) are organised geographically. The area unit of such samples is in general a few nautical miles. At a smaller spatial scale, fish are organised in schools or shoals or aggregations. For pelagic fish, a set of indices have been proposed to characterise the schools (ICES 2000) which relate to school geometry, internal density and vertical position. These parameters can be estimated by analyzing by school the digital acoustic records (echogram) using image analysis software. School size and abundance are in general related on a logscale (Fre´on and Misund 1999, Chap. 4). The bivariate plot can be summarised by summing the schools acoustic backscatter by ascending order of school size. Such (spectrum) curves indicate how biomass is distributed in classes of school size and their curvature indicates the higher contribution of particular school sizes. Curvature was characterised by geostatistical (schooling) indices (Petitgas 2000, p. 29: spectrum indices 1 and 2) which are an extension of the spatial selectivity index where occupation area is replaced by school size. Two indices were considered. Index spectrum 1 was defined as the area difference between the observed curve and the diagonal and characterised the curvature. It is sensitive to the skew in the distribution of fish biomass as a function of school size as well as to the skew in the distribution of school size. Index spectrum 2 was defined as the area difference between the observed curve and the curve obtained considering that all schools had equal density (equal to the ratio of the summed school biomass over the summed school sizes). This second index characterised the skew in the distribution of biomass as a function of school size irrespective of the distribution of school size. Inter-annual variation in the schooling pattern could be characterised using these indices: in the Bay of Biscay school biomass varied but not the distribution of school size while for northern North Sea herring schools, school biomass varied with school size distribution. 7.3.2.3 Indices of Clustering Patterns of Schools At a higher spatial level of organisation, schools occupy habitats with particular spatial distributions as they generally occur in clusters of the schools (Fre´on and Misund 1999, Chap. 4). Digital acoustic records can be replayed by school using image analysis softwares (ICES 2000) providing data sets of georeference school parameters. In such data, the nearest school neighbour distance is in general skewed, which demonstrates that schools are aggregated in clusters of

214

P. Petitgas

schools. Swartzman (1997) defined school clusters by grouping schools together based on a chosen distance threshold applied to the nearest school neighbour along the sailed acoustic transects. Petitgas (2003) suggested a procedure to define the threshold based on a spatial point process approach which maximised homogeneity of the school distribution within the clusters. School cluster indices were estimated to characterise the clustering pattern: number of clusters, number of solitary schools, dimension of clusters, number of schools per unit cluster length, skewness in the nearest neighbour distance within clusters. Resulting clusters of schools showed spatial scales of a few nautical miles (3–5 n.m.), while larger spatial components of tens of miles could also be present in the data sets (regional or meso scale structures). The pair correlation function, an analog of the variogram but for point process (Stoyan and Stoyan 1994), was a helpful structural tool to demonstrate scale in the schools spatial distribution. 7.3.2.4 Multiscale Organisation and Density-Dependence How do we relate the different spatial scales at which fish populations are organised? Statistical relationships have been examined for between global characteristics of the population and particular spatial indices. Relationships between population abundance and area occupied have been observed either in the form of a relationship between global abundance and local density (Myers and Stokes 1989, Fisher and Frank 2004) or between global abundance and spatial indices of occupation and dispersion (Swain and Sinclair 1994, Atkinson et al. 1997, Woillez et al. 2007). But abundance may not always vary with occupancy (Swain and Morin 1996). Petitgas et al. (2001) attempted a multiscale analysis of the variation in the spatial distribution with global abundance, including schooling and clusters of schools. They found no relationship between global abundance and indices of schooling and clustering but found a relationship between total school number and clustering parameters. A variety of situations seem possible perhaps because the scales and controls in the population spatial organisation are difficult to identify clearly using fisheries survey data. Four scenarios (Fig. 7.2) have been suggested with a geostatistical procedure for testing them on data (Swain and Sinclair 1994, Petitgas 1998): proportionality between global abundance and local density; change in habitat occupancy with no change in average local density; and intermediate cases where lower density areas or specific sites are replenished first when global abundance increases. Based on ecological theory, an underlying mechanism has been proposed to explain abundance – occupancy relationships. It is that of the density dependent suitability of habitats (MacCall 1990: ‘basin model’) which balances potential suitability of habitats with intra specific competition. This ‘basin model’ falls into one of the intermediate scenarios. The way by which the spatial organisation of a fish population can vary as an ensemble in all its organisational scales (schools, clusters, regional) cannot be predicted as yet as no multi-scale integrative model of spatial distribution exists. Such development would not only require a behavioural spatial mechanism to pass from one

7

Geostatistics and Their Applications to Fisheries Survey Data

215

Fig. 7.2 Four scenarios of spatial distribution change with global abundance. The lines marked (1) and (2) represent the density surface when global abundance increases from year (1) to year (2). Abscissa x represents space. Ordinate Z(x,t) represents fish density at location x in year t [after Petitgas (1998)]

scale to the other but also a relation between external forces (fishing, environment, predation) and habitat suitability and fish behaviour. Comprehensive data analyses using spatial indices at various spatial scales in different stock situations and other forcing parameters are needed for knowledge to progress. Further, the spatial distribution does not seem to relate to population abundance only: Woillez et al. (2006b) also showed correlation between spatial indices and population dynamics parameters such as mortality or recruitment, making spatial indices good candidates for indicator based monitoring of fish stocks.

7.3.3 Variation in the Spatial Structure The way by which fish aggregate and occupy their habitats is the expression of fish behaviour and can therefore be expected to depend on a variety of ecological factors. Thus the spatial structure as characterised by the variogram can be expected to depend at least on particular constraints (e.g., light, habitat geometry) and vary in time (e.g., day and night, seasonally, inter-annually) as well as with fish length or total abundance. Also, because sampling across space requires a certain amount of time, the space-time interaction within the survey data potentially affects the variographic structure and this has been investigated as well. 7.3.3.1 Aggregative Behaviour The size and anisotropy of the domain over which fish distribute constrain their spatial organisation. Giannoulaki et al. (2006) compared the variographic structure of sardine and anchovy in different areas and seasons and reported that the range of the variogram varied with the size of the areas overwhich fish distributed. Also variograms varied seasonally concomittantly with seasonal variation in fish length. Time of day is also a factor affecting the aggregative

216

P. Petitgas

behaviour of fish. Pelagic fish tend to aggregate in schools during day time and disagregated at night (Fre´on and Misund 1999). Rivoirard et al. (2000) reported day/night variographic difference on the Norwegian spring spawning herring when wintering in the fjord complex at Lofoten. Night variograms had longer range than day variograms, meaning that the fish aggregations were shorter during day than during night. The aggregative behaviour also varies seasonally (e.g., feeding vs. spawning or migratory schooling behaviour: Fre´on and Misund 1999). Mello and Rose (2003) reported different types of variographic structure depending on seasonally varying types of aggregating behaviour. They analysed which seasonal spatial structure had the best survey precision for different types of survey designs. 7.3.3.2 Inter-Annual Variation Unless total abundance dramatically changes across years, the variographic structure shows good consistency across years, the range being the most consistent while sill and nugget show more variability. Analysing a large number of years and different species, Fernandes and Rivoirard (1999) (also in Rivoirard et al. 2000) used an automated procedure to fit a variogram model for each year and species. Different models (e.g., spherical, exponential, linear) were fitted in each year and a goodness of fit criteria was used to finally select the model. Nearly all years shared similar variographic models and range parameters except for a few years which where atypical and dominated by a few very high values. For these years, the average variogram across all years was used. In a similar situation with egg survey data, Bellier et al. (2007) estimated variogram model parameters in each year by fitting all variograms of all years simulatneously using non linear mixed effects regression (Pinheiro and Bates 2000). Considering that all years shared a similar underlying variogram, all years were fitted with a spherical model with a constant range but varying nugget and sill across years. Fixed effects in the range, sill and nugget and random effects in the sill and nugget were estimated by the mixed effects regression procedure, which provided variogram parameters in each year. 7.3.3.3 Density-Dependence Stock collapse is often associated with reduction in the spatial occupation and this has been reported using indices of spatial occupation (see above). Variographic structure is also expected to show density-dependence at particular abundance levels. Warren (1997) reported changes in the Northen cod spatial structure associated with the stock collapse: variogram range decreased and nugget increased with collapsing stock abundance. Barange et al. (2005) reported little change in the indicator variograms for low and high cut-offs in contrasting years of high and low total abundance. But the variographic structure for intermediate cut-offs varied with total abundance. The range increased with abundance, meaning that when abundance increased, the intermediate

7

Geostatistics and Their Applications to Fisheries Survey Data

217

values occupied larger areas. This spatial behaviour is coherent to that reported using other but related tools, e.g., by Swain and Sinclair (1994) using occupied areas at different cut-offs, by MacCall (1990) using the ‘basin model’ based on density-dependent habitat selection theory, by Petitgas et al. (2001) using clustering indices, or by Petitgas (1998) using geostatistical aggregation curves. 7.3.3.4 Multi-Survey Spatial Structural Models Annual monitoring fisheries surveys result in a time series of spatial data. To model the space-time spatial structure in all years coherently, and make best use of all the available information, 3D variogram models have been considered. In acoustic fisheries surveys, the number of trawl hauls are scarce in each year but when surveys are repeated on an annual basis with similar sampling design the number of trawl hauls available across years is larger. When the spatial distribution and variogram structure is consistent across years, the map in any given year could be more precise when using all samples from all years rather than just the samples of the current year. For that purpose Guiblin et al. (1996) (also in Rivoirard et al. 2000, Chap. 4) inferred a multi-year spatial structure. They worked on mapping herring length in the northern North Sea. The spatial structure of fish length was consistent across the years. Temporal variation was collapsed as either ‘in the same year’ or ‘in different years’ irrespective of the year lag. The space-time variogram model was addtitive: g(h, t) = gspa (h) + (t), where was a nugget effect in time. The variogram in space irrespective of time gspa (h) was estimated as the variogram for pairs of points belonging to the same year and averaged across all years. The 3D variogram was that for pairs of points separated in space by distance h and belonging to two different years. When kriging at a particular point in a particular year using a 3D neighbourhood, sample points of the given year will be considered as well as sample points from other years. A temporal nugget effect is thus added in the kriging system for those samples not belonging to the year considered. It should be noted that the full model contained a trend surface guided by depth and that the space-time spatial structure applied to the residuals. A similar model was fitted to anchovy length in the Bay of Biscay although the structure was slightly different (Petitgas et al. 2003b): g(h, t) = It = 0 gspa (h) +It>0 . In this case study, interannual variations were such that the location of patches changed between years resulting in a pure nugget structure between sample points belonging to two different years. Here, in the 3D neighbourhoods, there is a switch in the model (structure or nugget) depending on whether the sample points belong to the same year for which kriging is performed or not. The multi-year modelling approach used in both case studies could have a generic interest for analysing fisheries survey data in each year while using the multi-year information. Space time models are also of interest when estimating total annual egg abundance using repeated egg surveys over the spawning season. The estimation problem can be solved in two steps, first the estimation of egg abundance at each time by spatial

218

P. Petitgas

integration and second the integration under the egg abundance ogive to estimate total annual egg abundance. Petitgas (1997a) fitted a multiplicative space-time model on sole egg distributions and derived the estimation variance for the annual egg abundance with that model. In that model, the temporal structure was in the spatial mean while the spatial structure was constant in time.

7.3.3.5 Space-Time Interaction Within a Survey It is usual practice to consider survey data as synoptic and correlation in the data as spatial only. But because of fish movement, variation in aggregating behaviour and because sampling across space takes a certain amount of time, survey data will contain space-time interactions. The importance of the interaction will depend on biological variation, survey design, and boat speed. The influence of time cannot reduce to the addition of a further dimension because sampling is not performed in 3D space but in a changing 2D space (e.g., Petitgas 2001). Rivoirard (1998) proposed space-time variogram models for different types of fish movements. For instance, in the case of brownian fish movement and isotropic spatial structure, the space-time covariance can be written as a convolution of the underlying spatial covariance and diffusion due to movement. The effect of different fish movements (random, cyclical, migration) on the variogram was investigated on northern North Sea herring using simulations (Rivoirard et al. 2000, Chap. 5). An underlying average spatial distribution was estimated over the time series of surveys that represented the probability map of the fish distribution. A large number of patches of fish were considered for motion, which were located initially using the probability map. In the case of the random motion of patches, a constraint on the probability of motion was imposed so as to conform at any time and location to the underlying probability map of the fish distribution. The fish distribution was dynamically simulated while acoustic survey transects were simulated that sampled the fish. Variograms were estimated using the simulated data for the different types of fish movements. The result was that the influence of random and tidal motion had little influence on the variographic structure. In contrast the influence of migration was a concern when the survey transects were in the direction of the migration. If the survey transects crossed the migration direction, the effect was less important.

7.4 A Word on Software A large variety of geostatistical computer software and libraries are available. Most include variography and kriging on a grid. Some offer a comprehensive list of geostatistical tools (e.g., isatis, gslib) and others are dedicated to particular aspects of geostatistics (e.g., Variowin, Eva). Updated information on

7

Geostatistics and Their Applications to Fisheries Survey Data

219

computer software is available on the internet. The ai-geostats home page (http://www.ai-geostats.org/) provides, among other information about geostatistics, an inventory of software with descriptions of their functionalities and links to their home page. The ai-geostats web resource is also a forum for discussion on geostatistics. Rivoirard et al. (2000) provides an appendix with guidance on geostatistical books and description of software tools. Some of this information is updated here. Only softwares that are the most comprehensive or address particular aspects that others do not and that are of interest in fisheries science are presented here (Table 7.3). The present selection also covers the range of different computer platforms. Isatis and Gslib are the most comprehensive packages. Isatis is a complete geostatistical package offering all geostatistical methods (linear, non-linear, stationary, non-stationary, monovariate, multi-variate, simulations). It is a commercial software package that runs under Unix or an emulated PC. Gslib (Deutsch and Journel 1992) is a suite of FORTRAN routines that also covers a wide range of geostatistical methods though perhaps less complete for nonlinear geostatistics. The code of PC executables is free and dowloadable from the internet. An interface for running Gslib (WinGslib) exists as a commercial product for Windows. The MATLAB kriging tool box is a collection of MATLAB routines for kriging and co-kriging based on Gslib routines. The code is free and accessible from the internet. Gstat (Pebesma 2004) is an S-plus library as well as an R package, which covers multivariate geostatistics and simulations (the R package may offer slightly less functionalities than the S library). The code is free and is downloadable from the internet. Variowin (Pannatier 1996) is a PC software tool dedicated to the estimation and fitting of variograms. The code is not open source. The book and executable software can be downloaded from the internet. None of the previous softwares considers the

Table 7.3 Selected geostatistical softwares. An inventory of softwares is available at http:// www.ai-geostats.org/software/ Name Access Code Internet Reference Isatis Gslib

Commercial Free: on internet

No Yes

http://www.geovariances.fr/ http://www.gslib.com/

Matlab tool box

Free: on internet

Yes

Gstat Variowin

Free: on internet Free: on internet

Yes No

Eva

Free: from authors

No

http:// www.globec.whoi.edu/ software/ http://www.gstat.org/ http://www.sst.unil.ch/ research/variowin/ index.html contact: [email protected]

Deutsch and Journel (1992)

Pebesma (2004) Pannatier (1996) Petitgas and Lafont (1997)

220

P. Petitgas

question of global estimation over a domain. EVA (Petitgas and Lafont 1997) is a PC software tool that specifically targets global abundance and its estimation variance for a large variety of survey designs including adaptive design. It is the only software available that allows one to estimate the global estimation variance. The code is not open source but the software executable is freely available from the authors.

7.5 Future Challenges Geostatistics formalizes the relationship between sample locations, correlation stucture in the population and precision in the estimates. Geostatistics provides model-based variance estimates of global abundance as well as mapping by kriging. Therefore and in contrast to a design-based approach, geostatistics allows one to separate between data analysis and survey design, giving more flexibility in the design. These have been the early contributions of Geostatistics to fisheries science that resolved the problem of estimating the precision of global abundance over a domain for sampling designs in which the samples were not taken independently from each other. Because the corner stone of geostatistics is the modelling of a spatial structure, the methodology also offers tools to characterise population aggregation patterns and address the key biological question of changes in the spatial organisation of fish populations under the controls of density dependence, environment and behaviour. The need to understand the spatio-temporal variability in fish stocks and its controls as well as develop multiscale models of fish populations spatial organisations remain. In the classical univariate geostatistical approach, variability in the data is interpreted as spatial variation only. But when using spatiotemporal and multivariate approaches, mathematical dimensionality is increased in order to properly consider the variability in the data. To solve that paradox and because biological variability originates from biological processes occurring at different scales, multiscale data collection schemes would be helpful. Multivariate geostatistics has the potential to assemble multiscale information as well as combine stochastic and deterministic approaches. Multiscale sampling can be achieved in various ways, by adaptive sampling designs or by combining surveys performed at different scales, e.g., from the large scale fisheries surveys to the fine scale survey on aggregation behaviour over time. Multivariate geostatistical approaches represent a large field with challenging fisheries applications, that in addition to linear multivariate geostatistics also includes non-stationary and non-linear geostatistics as well as conditional simulations. Software tools are now widely available that will allow for the development of many future applications using these more complexe approaches. Fisheries management issues have expanded to include population conservation issues as well as an ecosystem approach to fish stock diagnostics. Thus

7

Geostatistics and Their Applications to Fisheries Survey Data

221

there is a need to relate fish population dynamics with population distribution and the occupation of essential habitats, trophic interactions and climate forcing on the habitats. In all these topics it is key to monitor the spatial distribution of a range of target fish species as well as their environment. Geostatistical spatial indices have shown potential in elaborating indicator based diagnostics of fish stocks. This approach can be extended to a variety of indices in the different components of the ecosystem. Also, an atlas of geostatistically derived fish distribution maps can be helpful in establishing regions of essential fish habitat on a multispecies and multi-year basis. Geostatistics will be useful in developing spatial indices to serve as indicators for management as well as develop space-time methods to analyse large numbers of distribution maps with functional relationships at different scales.

References Armstrong M, Renard D, Rivoirard J, Petitgas P (1992) Geostatistics for fish survey data. Course C-148, Centre de Ge´ostatistique, Fontainableau, France Atkinson D, Rose G, Murphy E, Bishop C (1997) Distribution changes and abundance of northern cod (Gadus morhua), 1981–1993. Canadian Journal of Fisheries and Aquatic Sciences 54(Suppl. 1): 132–138 Barange M, Coetzee J, Twatwa N (2005) Strategies of space occupation by anchovy and sardine in the southern Benguela: the role of stock size and intra-specific competition. ICES Journal of Marine Science 62: 645–654 Bellier E, Planque B, Petitgas P (2007) Historical fluctuations in spawning location of anchovy (Engraulis encrasicolus) and sardine (Sardina pilchardus) in the Bay of Biscay during 1967–1973 and 2000–2004. Fisheries Oceanography 16: 1–15 Bez N (2002). Global fish abundance estimation from regular sampling: the geostatistical transitive method. Canadian Journal of Fisheries and Aquatic Sciences 59: 1921–1931 Bez N, Rivoirard J (2000a) Indices of collocation between populations. In: Chekley D, Hunter J, Motos L, van der Lingen C (eds) Report of a workshop on the use of Continuous Underway Fish Egg Sampler (CUFES) for mapping spawning habitat of pelagic fish. GLOBEC Report 14 Bez N, Rivoirard J (2000b) On the role of sea surface temperature on the spatial distribution of early stages of mackerel using inertiograms. ICES Journal of Marine Science 57: 383–392 Bez N, Rivoirard J (2001) Transitive geostatistics to characterise spatial aggregations with diffuse limits: an application on mackerel ichtyoplankton. Fisheries Research 50: 41–58 Bouleau M, Bez N, Reid D, Godo O, Gerritsen H (2004) Testing various geostatistical models to combine bottom trawl stations and acoustic data. ICES CM 2004/R:28 Bulgakova T, Vasilyev D, Daan N (2001) Weighting and smoothing of stomach content data as input for MSVPA with particular reference to the Barents Sea. ICES Journal of Marine Science 58: 1208–1218 Chile`s JP, Delfiner P (1999) Geostatistics: modelling spatial uncertainty. Wiley, New York Cochran W (1977) Sampling techniques. Wiley, New York Conan, G (1985) Assessment of shellfish stocks by geostatistical techniques. ICES CM 1985/ K:30 Conners E, Schwager S (2002) The use of adaptive cluster sampling for hydroacoustic surveys. ICES Journal of Marine Science 59: 1314–1325 Cressie N (1991) Statistics for spatial data. Wiley, New York

222

P. Petitgas

Deutsch C, Journel A (1992) Geostatistical software library and user’s guide. Oxford University Press, Oxford Doray M, Petitgas P, Josse E (2008) A geostatistical method for assessing biomass of tuna aggregations around Fish Aggregation Devices with star acoustic surveys. Canadian Journal of Fisheries and Aquatic Sciences 65: 1193–1205 Fernandes P, Rivoirard J (1999) A geostatistical analysis of the spatial distribution and abundance of cod, haddock and whiting in North Scotland. In: Gomez-Hernandez J, Soares A, Froideveaux R (eds) GeoENV II – Geostatistics for Environmental Applications. Kluwer Academic Press, Dordrecht. pp 201–212 Fisher J, Frank K (2004) Abundance distribution relationships and conservation of exploited marine fishes. Marine Ecology Progress Series 279: 201–213 Fre´on P, Misund O (1999) Dynamics of pelagic fish distribution and behaviour: effects on fisheries and stock assessment. Blackwell Science, Oxford Gerlotto F, Bertrand S, Bez N, Gutierrez M (2006) Waves of agitation inside anchovy schools observed with multi beam sonar: a way to transmit information in response to predation. ICES Journal of Marine Science 63: 1405–1417 Giannoulaki M, Machias A, Koutsikopoulos C, Somarakis S (2006) The effect of coastal topography on the spatial structure of anchovy and sardine. ICES Journal of Marine Science 63: 650–662 Gimona A, Fernandes P (2003) A conditional simulation fo acoustic survey data: advantages and pitfalls. Aquatic Living Resources 16: 123–129 Gohin F (1985) Planification des expe´riences et interpre´tation par la the´orie des variables re´gionalise´es: application a` l’estimation de la biomasse d’une plage. ICES CM 1985/ D:03 Guiblin P, Rivoirard J, Simmonds J (1995) Analyse structurale de donne´es a` distribution dissyme´trique: exemple du hareng e´cossais. Cahiers de Ge´ostatistique 5: 137–159 Guiblin P, Rivoirard J, Simmonds J (1996) Spatial distribution of length and age for OrkneyShetland herring. ICES CM 1996/D:14 ICES (1989) Report of the workshop on spatial statistical techniques. ICES CM 1989/K:38 ICES (1992) Acoustic survey design and analysis procedure: a comprehensive review of current practice. ICES Cooperative Research Report 187 ICES (1993) Report of the workshop on the applicability of spatial statistical techniques to acoustic survey data. ICES Cooperative Research Report 195 ICES (2000) Report on Echotrace Classification. ICES Cooperative Research Report 238 Journel A, Huijbregts Ch (1978) Mining geostatistics. Academic Press, London Lantue´joul C (2002 Geostatistical simulations: models and algorithms. Springer-Verlag, Berlin Lo N, Griffith D, Hunter J (1997) Using a restricted adaptive cluster sampling design to estimate hake larval abundance. CalCOFI report 38: 103–113 MacCall A (1990) Dynamic geography of marine fish populations. University of Washington Press, Seattle Matheron G (1971) The theory of regionalised variables and their applications. Les Cahiers du Centre de Morphologie Mathe´matiques, Fascicule 5. Centre de Ge´ostatistique, Fontainebleau Matheron G (1973) The intrinsic random functions and their applications. Advances in Applied Probability 5: 439–468 Matheron G (1989) Estimating and choosing: an essay on probability in practice. SpringerVerlag, Berlin McCullagh P, Nelder J (1995) Generalised linear models. Chapman and Hall, London Mello L, Rose G (2003) Using geostatistics to quantify seasonal distribution and aggregation patterns of fishes: an example of Atlantic cod (Gadus morhua). Canadian Journal of Fisheries and Aquatic Sciences. 62: 659–670 Myers R, Cadigan N (1995) Was an increase in natural mortality responsible for the collapse of northern cod? Canadian Journal of Fisheries and Aquatic Sciences 52: 1274–1285

7

Geostatistics and Their Applications to Fisheries Survey Data

223

Myers R, Stokes K (1989) Density dependent habitat utilization of groundfish and the improvement of research surveys. ICES CM 1989/D:15 Pannatier Y (1996) Variowin: software for spatial data analysis in 2D. Springer-Verlag, Berlin Pebesma E (2004) Multivariate geostatistics in S: the gstat package. Computers and Geosciences 30: 683–691 Petitgas P (1991) Un mode`le de co-re´gionalisation pour les poissons pe´lagiques formant des bancs le jour et se dispersant la nuit. Note N33/91/G, Centre de Ge´ostatistique, Fontainebleau Petitgas P (1993a) Geostatistics for fish stock assessments: a review and an acoustic application. ICES Journal of Marine Science 50: 285–298 Petitgas P (1993b) Use of disjunctive kriging to model areas of high pelagic fish density in acoustic fisheries surveys. Aquatic Living Resources 6: 201–209 Petitgas P (1996) Geostatistics and their applications to fisheries survey data. In: Megrey B, Moksness E (eds) Computers in Fisheries Research. Chapman and Hall, London. pp 113–142 Petitgas P (1997a) Sole egg distributions in space and time characterized by a geostatistical model and its estimation variance. ICES Journal of Marine Science 54: 213–225 Petitgas P (1997b) Use of disjunctive kriging to analyse an adpative survey design for anchovy eggs in Biscay. Ozeanografika 2: 121–132 Petitgas P (1998) Biomass dependent dynamics of fish spatial distributions characterized by geostatistical aggregation curves. ICES Journal of Marine Science 55: 443–453 Petitgas P (ed) (2000) Cluster: Aggregation patterns of commercial fish species under different stock situations and their impact on exploitation and assessment. Final report to the European Commission, contract FAIR-CT-96.1799. European Commission, DG-Fish, Brussels Petitgas P (2001) Geostatistics in fisheries survey design and stock assessment: models, variances and applications. Fish and Fisheries 2: 231–249 Petitgas P (2003) A method for the identification and characterization of clusters of schools along the transects lines of fisheries acoustic surveys. ICES Journal of Marine Science 60: 872–884 Petitgas P (2004) About non-linear geostatistics and adaptive sampling. In: Report of the Workshop on Survey Design and Data Analysis (WKSAD). ICES CM 2004/B:07. Working Document 11 Petitgas P, Lafont T (1997) EVA2: Estimation variance version 2, a geostatistical software for the precision of fish stock assessment surveys. ICES CM 1997/Y:22 Petitgas P, Masse´ J, Beillois P, Lebarbier E,. Le Cann A (2003a) Sampling variance of species identification in fisheries acoustic surveys based on automated procedures associating acoustic images and trawl hauls. ICES Journal of Marine Scienc 60: 437–445 Petitgas P, Masse´ J, Grellier P, Beillois P (2003b) Variation in the spatial distribution of fish length: a multi-annual geostatistics approach on anchovy in Biscay, 1983–2002. ICES CM 2003/Q:15 Petitgas P, Reid D, Carrera P, Iglesias M, Georgakarakos S, Liorzou B, Masse´ J (2001) On the relation between schools, clusters of schools, and abundance in pelagic fish. ICES Journal of Marine Science 58: 1150–1160 Pinheiro J, Bates D (2000) Mixed effects models in S and Splus. Springer-Verlag, Berlin Rivoirard J (1994) Introduction to disjunctive kriging and non-linear geostatistics. Clarendon, Oxford Rivoirard J (1998) Quelques mode`les spatio-temporels de bancs de poissons. Note N12/98/G. Centre de Ge´ostatistique, Fontainebleau Rivoirard J, Guiblin P (1997) Global estimation variance in presence of conditioning parameters. In: Baafi E, Schofield N (eds) Geostatistics Wollongon ’96, Volume I. Kluwer Academic Publishers, The Netherlands. pp 246–257 Rivoirard J, Simmonds J, Foote K, Fernandes P, Bez N (2000) Geostatistics for estimating fish abundance. Blackwell Science, Oxford

224

P. Petitgas

Rivoirard J, Wieland K (2001) Correcting for the effect of daylight in abundance estimation of juvenile haddock (Melanogrammus aeglefinus) in the North sea: an application of kriging with external drift. ICES Journal of Marine Science 58: 1272–1285 Rossi R, Mulla D, Journel A, Franz E (1992) Geostatistical tools for modeling and interpreting ecological spatial dependence. Ecological Monographs 62: 277–314 Rufino M, Maynou F, Abello P, Yule, A (2006) Small-scale non-linear geostatistical analysis of Liocarcinus depurator (Crustacea: Brachyura) abundance and size structure in a western Mediterranean population. Marine Ecology Progress Series 276: 223–235 Simmonds J (1995) Survey design and effort allocation: a synthesis of choices and decisions for an acoustic survey. North Sea herring is used as an example. ICES CM 1995/B:09 Simmonds J, Fryer R (1996) Which are better, random or systematic acoustic surveys? A simulation using North Sea herring as an example. ICES Journal of Marine Science 53: 39–50 Simmonds J, McLennan D (2005) Fisheries acoustics: theory and practice. Blackwell Science, Oxford Stoyan D, Stoyan H (1994) Franctals, random shapes and point field. Wiley, New York Sullivan P (1991) Stock abundance estimation using depth-dependent trends and spatially correlated variation. Canadian Journal of Fisheries and Aquatic Sciences 48: 1691–1703 Swain D, Morin R (1996) Relationships between geographic distribution and abundance of American plaice (Hippoglossoides platessoides) in the southern gulf of St. Lawrence. Canadian Journal of Fisheries and Aquatic Sciences 53: 106–119 Swain D, Sinclair A (1994) Fish distribution and catchability: what is the appropriate measure of distribution? Canadian Journal of Fisheries and Aquatic Sciences 51: 1046–1054 Swartzman G (1997) Analysis of the summer distribution of fish schools in the Pacific Boundary Current. ICES Journal of Marine Science 54: 105–116 Thompson S, Seber, G (1996) Adaptive sampling. Wiley, New York Wackernagel H (1995) Multivariate geostatistics: an introduction with applications. SpringerVerlag, Berlin Warren W (1997) Changes in the within-survey spatio-temporal structure of the northern cod (Gadus morhua) population, 1985–1992. Canadian Journal of Fisheries and Aquatic Sciences 54(Suppl. 1): 139–148 Woillez M, Poulard JC, Rivoirard J, Petitgas P, Bez N (2007) Indices for capturing spatial patterns and their evolution in time with an application on European hake (Merluccius merluccius) in the Bay of Biscay. ICES Journal of Marine Science 64: 537–550 Woillez M, Rivoirard J, Fernandes P (2006a) Evaluating the uncertainty of abundance estimates from acoustic surveys using geostatistical conditional simulations. ICES CM 2006/I:15 Woillez M, Petitgas P, Rivoirard J, Fernandes P, terHofstede R, Korsbrekke K, Orlowski A, Spedicato MT, Politou CY (2006b) Relationships between population spatial occupation and population dynamics. ICES CM 2006/O:05

Chapter 8

Ecosystem Modelling Using the Ecopath with Ecosim Approach Marta Coll, Alida Bundy and Lynne J. Shannon

8.1 Introduction Marine ecosystems are dynamic and complex, with interactions, feedback loops and environmental effects occurring concurrently. Fishing activities impact on their structure and functioning, modifying their features and affecting the interactions established between their biological components. The task of making predictions of future states of the ecosystem and understanding marine resource dynamics might seem Herculean or madness, even with reductionist modelling approaches. It is a large task which has been simplified and made tractable with the development of the ecosystem modelling software system, Ecopath with Ecosim (Polovina 1984, Walters et al. 1997, Pauly et al. 2000). In recent years, it has become an ecosystem modelling tool that is used globally for static analyses of marine ecosystems and tropho-dynamic and spatial simulations. This chapter briefly reviews the history and development of Ecopath with Ecosim (EwE) and describes the theory and assumptions on which is it based. Then it uses case studies to illustrate EwE utility and the insights it can bring to understand ecosystem structure and functioning, ecosystem changes and to examine the likely consequences/benefits of different management options at the ecosystem level.

8.1.1 History and Development Fisheries management efforts have largely failed given that at least 75% of the world’s major fisheries resources are either fully exploited (‘‘mature’’) or overexploited, with clear signs of declines in catches (FAO 2005). Improvement of our fisheries management requires new approaches and techniques. These need to accommodate the net effects of alternative fishing strategies on the ecosystem as a whole, through taking into account the effects on ecosystem M. Coll (*) Institute of Marine Science (ICM-CSIC), Passeig Marı´ tim de la Barceloneta, 37-49, 08003 Barcelona, Spain

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_8, Ó Springer ScienceþBusiness Media B.V. 2009

225

226

M. Coll et al.

structure and functioning as well as the effects on the biology and dynamics of the targeted stocks. Single-species fisheries models are unable to capture interactions between species (especially trophic interactions), and do not accommodate spatial aspects of fish stocks and their prey and predators. Thus they are limited in the breadth and scope of management objectives and strategies that can be explored for future management. Careful consideration and quantification of trophic interactions is important when addressing management objectives to exploit or conserve species that may be interacting strongly with a particular species targeted by fisheries. Information on the spatial distribution and overlap between species is also needed if closed areas are to be implemented as effective management tools. In both respects, the dynamic ecosystem modelling approach of EwE (through Ecosim and Ecospace) offers a means to incorporate interactions and spatial constraints into an approach in which fisheries effects can be explored and carefully examined, providing a useful tool for ecosystembased fisheries management. The Ecopath and Ecosim modelling tool (EwE) is composed of a core mass balance model (Ecopath, which stands for Ecological Pathways Model) (Polovina 1984, Pauly et al. 2000; Christensen and Walters 2004a, Christensen et al. 2005) from which temporal and spatial dynamic simulations can be developed (Walters et al. 1997, 1999, Christensen and Walters 2004a). This tool has been widely used to quantitatively describe aquatic systems and the ecosystem impacts of fishing (Christensen and Pauly 1993, Christensen and Walters 2004a). Ecopath with Ecosim has its roots in classic ecology. Food chains are considered to be based on trophic flows between discrete trophic levels (Lindeman 1942) and thus species are allocated to distinct trophic levels and positions in a food chain or food web. Based on this theory and using the concept of mass balance and energy conservation, Polovina (1984) developed the first Ecopath model for the French Frigate Shoals in the Northwestern Hawaiian Islands. Christensen and Pauly (1992) further developed the model to include fractional trophic levels to take into account species that feed across a range of trophic levels. The latter forms the basis of network analysis and the current Ecopath modelling approach (e.g. Wulff et al. 1989, Pauly et al. 2000). Since the mid-1990s, with the coalescence of increased computing power and new ideas, the scope of Ecopath has exploded: the trophodynamic simulation model Ecosim (Walters et al. 1997, Christensen et al. 2005) has introduced the capability to conduct multispecies simulations to explore ecosystem structure and functioning, the impact of fishing, policy exploration and more; a year later, the development of Ecospace, a spatially explicit simulation model, began (Walters et al. 1999, Christensen et al. 2005). This immediately addressed questions relating to marine protected areas and spatial management, in addition to exploring aspects related to spatial distribution of organisms, and behaviour and the role of water movement. Ecopath with Ecosim is thus the first ecosystem level simulation model that is widely accessible. There are 3681 registered users in 150 different countries from

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

227

November 2003 to 8 March, 2007 (www.ecopath.org, 14 September, 2006) and over 200 publications making Ecopath with Ecosim the default modelling approach to explore ecosystem related questions in the world of fisheries science. A total of 325 EwE models have been constructed to date, 42% to describe ecosystem structure, 30% to address fisheries management problems, 9% to address policy issues, 6% for marine protective areas and 11% to explore questions in theoretical ecology (Morissette 2007). In this chapter, we briefly describe the theory and assumptions on which EwE is based (see Section 8.2). We use a selection of case studies to illustrate the scope of EwE and the insights it can bring to understanding ecosystem structure and functioning, ecosystem changes, the impacts of fishing and policy exploration (see Section 8.3). Finally, we take a step back and discuss the limits of EwE, give a critical perspective and finish with a discussion of future directions (Section 8.4).

8.2 Ecopath with Ecosim 8.2.1 Fundamental Theory and Equations of Ecopath 8.2.1.1 Mass Balance Modelling An Ecopath model provides a quantitative representation of the studied ecosystem, or a snapshot, in terms of trophic flows and biomasses for a defined time period. The ecosystem is represented by functional groups, which can be composed of species, groups of species with ecological similarities or ontogenetic fractions of a species. The key principle of Ecopath is mass balance: for each group represented in the model, the energy removed from that group, for example by predation or fishing, must be balanced by the energy consumed, i.e. consumption. Two linear equations represent the energy balance within a group and the energy balance among groups. The production (P) of each functional group (i) in the ecosystem is divided into predation mortality (M2ij) caused by the biomass of the other predators (Bj); exports from the system both from fishing activity (Yi) and other exports (Ei); biomass accumulation in the ecosystem (BAi); and baseline mortality or other mortality (1-EEi), where EE is the ecotrophic efficiency of the group within the system, or the proportion of the production of (i) that is exported out of the ecosystem (i.e. by fishing activity) or consumed by predators within it. Pi ¼

X

Bj  M2ij þ Yi þ Ei þ BAi þ Pi  ð1  EEi Þ

(8:1)

j

Equation (8.1) can be re-expressed as: B

  X     P Q P ¼ Bj   DCij þ Yi þ Ei þ BAj þ Bi  ð1  EEi Þ B i B B j i j

(8:2)

228

M. Coll et al.

where (P/B)i indicates the production of (i) per unit of biomass and is equivalent to total mortality, or Z, under steady-state conditions (Allen 1971); (Q/B)i is the consumption of (i) per unit of biomass; and DCij indicates the proportion of (i) that is in the diet of predator (j) in terms of volume or weight units. Ecopath parameterizes the model by describing a system of linear equations for all the functional groups in the model. For each functional group, three of the basic parameters: Bi, (P/B)i, (Q/B)i or EEi have to be known in addition to the fisheries yield (Yi) and the diet composition. The energy balance within each group is ensured when consumption by group (i) equals production by (i), respiration by (i) and food that is unassimilated by (i). The units of the model are expressed in terms of nutrient or energy related currency by unit of surface (frequently expressed as tkm–2 yr1). 8.2.1.2 Definition of Functional Groups and Balancing Procedure An ecological model includes different functional groups spanning the whole ecosystem, from lower to higher trophic levels (i.e. from primary producers to top predators) and detritus groups (natural detritus and detritus generated by discarding during fishing operations). Definition of these groups is based on similarities between species in their ecological and biological features (e.g. feeding, habitat, mortality), their ecological role and on their importance as harvestable resources. In very speciose systems, multivariate statistical methods can be used to define mixed groups composed of various species by applying a systematic analysis of available ecological information. For example, Factorial Correspondence Analysis (FCA) and Hierarchical Cluster Analysis have been applied to stomach-content data in the Mediterranean Sea (Pinnegar 2000, Coll et al. 2006a, 2007). To encompass ontogenetic changes in feeding, behaviour and habitat preference, a multiple stanza representation has been incorporated into the model (Christensen and Walters 2004a), succeeding an earlier two-stanza version (Walters et al. 2000). The multiple stanza model enables the representation of all life-stages and ensures consistency between ontogenetic groups. Values of (P/B)i and diet composition have to be provided for all multiple stanza groups, while Bi and (Q/B)i need to be introduced for the leading stanza group only. A key stage in the development of a mass balance model is the process of assembling the data from different components of an ecosystem into one coherent picture, with flows that meet the mass balance criteria. It should be an exercise where information is gained about the ecosystem, since a single species view of the ecosystem (from which much of the input data is derived) will often not reflect the demands and constraints of the multispecies world these species inhabit. Preconceived ideas may have to be revisited for these models to balance. Thus, after parameterization, the model is considered balanced when the results show consistent values for the following: (1) estimates of EE <1; (2) values of total production/total consumption (P/Q) for most functional groups

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

229

between 0.1 and 0.35; and (3) values of total respiration/total system biomass (R/B) consistent with the group’s activities with high values for small organisms and top predators (Christensen et al. 2005). Solutions should first be explored manually by iteratively changing parameters (beginning with the most uncertain) within their range of uncertainty. If a solution cannot be found, or to explore alternative mass balance solutions, an Automatic Mass Balance Procedure (AMBP, Kavanagh et al. 2004) can be applied. The autobalance procedure randomly selects the initial input parameters from a pre-defined range of values using the Pedigree routine, see below. Thus each run starts with a different set of conditions, and the routine searches for the combination that will produce a balanced model. This is useful for exploring the range of feasible solutions for groups with poor parameter estimates. Only the biomass and diet parameters can be directly perturbed in the current autobalance routine, thus it is not a full perturbation analysis. The Pedigree routine can be used to describe the origin and quality of the data and of the model and to derive confidence intervals for the input data for use during the balancing procedure (Christensen and Walters 2004a, Christensen et al. 2005). The uncertainty associated with the model parameters (biomass, P/B, Q/B, catch and diet input) is defined using percent ranges. With this routine, the overall Pedigree of the model can be compared with other models, see Christensen et al. (2000). It differentiates between models that are based on local empirical data (higher quality models) and those lower quality models where parameters may be borrowed from other models or are estimates. Although this is not a direct estimate of the uncertainty and its affects on model estimates, it is a step towards exploring such effects.

8.2.1.3 Model Analysis Ecological analyses integrated in EwE can be used to examine ecosystem features based on trophic flows, thermodynamic concepts, information theory and trophodynamic indicators (Mu¨ller 1997, Christensen and Walters 2004a, Christensen et al. 2005, Cury et al. 2005). Trophic flows in an ecosystem are quantified in terms of consumption, production, respiration, exports and imports, and flow to detritus (tkm2 yr1). The sum of all these flows is the Total System Throughput (TST) and is considered to be an indirect indicator of the size of the food web (Christensen and Pauly 1993). Trophic Levels (TL) by functional groups are also provided by the model. The TL identifies the position of organisms within food webs and was first defined as an integer (Lindeman 1942) and later modified to a fractional value (Odum and Heald 1975). By convention, primary producers and detritus have TL = 1; while values for consumer groups are calculated from the weighted average TL of their prey, determined using mass-balance models, gut content analysis or isotope data (Stergiou and Karpouzi 2002). The TL is formulated as follows:

230

M. Coll et al.

TLj ¼ 1 þ

n X

DCji  TLi

(8:3)

i¼1

where j is the predator of prey i, DCji is the fraction of prey i in the diet of predator j and TLi is the trophic level of prey i. The average trophic level of the community (TLco) reflects the structure of the community and can be calculated easily as the weighted average of the TL of all the species within the ecosystem (Rochet and Trenkel 2003). From trophic flows and TLs, the Transfer Efficiency (TE) is calculated, which summarizes all the inefficiencies of the food web (due to respiration, excretion, egestion and natural mortality) at each step of the trophic chain (Lindeman 1942). The TE is an important emergent property of the system that is difficult to obtain from field work. It indicates how energy is transfered from low to high trophic levels and in Ecopath is estimated from the ratio of production at a given TL to production at the preceding TL (Lalli and Parsons 1993, Pauly and Christensen 1995). Flows, TLs and TE can be visualized using the flow diagram tool (Fig. 8.1) and in the form of a Lindeman Spine (Lindeman 1942, Ulanowicz 1986, Wulff et al. 1989, Libralato et al. 2002) (Fig. 8.2).

TL V Pelagic habitat

Demersal habitat

Anglerfish

Dolphins Dolphins

Conger

Large pelagics

IV

Adult hake hake Adult

Bonito

Dem. fishes (3) Fin whale Mackerel

Benthopelagic fishes

Anchovy

III

Various Shrimps pelagic fishes

Sardine

Juv. hake

Blue Blue whiting whiting Audouin gull

Horse Horse mackerel mackerel

Dem. sharks

Bentop. Cephalop.

Poor Poor cod cod Dem. fishes (2)

Dem. fishes (1)

Crabs

Macrozooplankton

Jellyfish

Red mullets mullets

Benthic cephalop.

Flat fishes

Norway Norway lobster lobster

Marine turtles

II

Sea birds

Micro- & & meso meso zooplankton zooplankton

Suprabenthos

Benthic invertebrates

Polichaetes

Detritus

I

Phytoplankton Phytoplankton

Discard (1)

Discard (2)

Fig. 8.1 Flow diagram of the South Catalan Sea organised by functional groups and fractionated trophic levels (TL) and divided between demersal and pelagic habitats. Boxes arranged along vertical axis by trophic level. Flows leave boxes in the upper half and enter in the lower half (only main flows shown for clarity). Adapted from Coll et al. 2008a

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

231

Fig. 8.2 Trophic flows by integer trophic levels (TL) in the form of Lindeman Spine for a representative sea grass meadow habitat of the Venice Lagoon ecosystem. Primary producers and detritus are separated to clarify the representation (both with TL = I) (tkm2 yr1). Reprinted from Libralato et al. 2002. Comparison Between the Energy Flow Networks of Two Habitats in the Venice Lagoon. Libralato et al. (2002), with permission to reproduce

Ecological models also provide various analyses by functional group in terms of food intake, niche overlap, partition of mortality, consumption and respiration (Christensen et al. 2005) that are useful for analyzing the ecological role of the different groups within the system. In particular, Mixed Trophic Impact (MTI) analysis enables the quantification of the combined effects of direct and indirect trophic interactions among groups. It measures the relative impact of a change in the biomass of one component on other components of the ecosystem (Ulanowicz and Puccia 1990). It is based on an input–output method for assessing direct and indirect economic interactions (Leontief 1951). In this way, positive effects that a prey has on its predator, negative effects that a predator has on its prey, and positive and negative effects that a group may have on another group through interactions with other groups (i.e. indirect effects) are all quantified. Therefore, matrices of the relative, net impacts (scaled between –1 and 1) of each group on all other groups are constructed (Fig. 8.3). Results from the model can also be related to ecosystem development theory sensu Odum (Margalef 1968, Odum 1969, Christensen 1995a). It has been found that when ecosystems develop, their biomass, information and complexity tends to increase, whereas when they are perturbed, e.g. by fishing, they show the opposite tendency. Therefore, to put ecosystem development into context, indicators can be compared across different modelled systems, or compared within the same system for different time periods. Such indicators include various coefficients of flows and biomasses (such as total primary production/total respiration and total biomass/total production), nutrient recycling indexes (such as Finn’s cycling index and predatory cycling index), the System Omnivory Index (SOI) and indices related to the information theory of Ulanowicz (1986).

232

M. Coll et al.

Fig. 8.3 Mixed Trophic Impact (MTI) analysis from the Gulf of California, Mexico (Arreguı´ nSa´nchez et al. 2002). Impacted groups are placed along the horizontal axis and impacting groups are placed along the vertical axis. The bars indicate relative impact (between 0 and 1) where positive impacts are above the zero line and negative impacts are below. Reprinted from Arreguı´ n-Sa´nchez et al. 2002. Flow of biomass and structure in an exploited benthic ecosystem in the Gulf of California, Mexico. Arreguı´ n-Sa´nchez et al. (2002), with permission from Elsevier

Moreover, various results from EwE can be analyzed in terms of direct fishing impacts. These are, for example, the exploitation rates (F/Z) by functional group, the gross efficiency of the fishery (GEf = total catch/primary production) and the mean trophic level of the catch (TLc). Values of F/Z by functional group are calculated based on the partitioning of mortality and increase with fishing pressure. Values of GEf are higher for systems with a high fishing impact, they are generally much lower than 1.0 and the weighted global average was described to be about 0.0002 (Christensen et al. 2005). The TLc reflects the overall strategy of a fishery and is calculated by weighting the proportions of each type of organisms from the catch by their respective TLs (Pauly et al. 1998). TLco and TLc decrease as fishing impacts increase in the ecosystem since fishing tends to first remove the high trophic level organisms. Thus these indices can reflect structural changes due to fishing and the effect of fishing down the food web can be identified (Pauly et al. 1998, Pinnegar et al. 2002, Jennings et al. 2002). The model-derived indicators presented above are illustrated for some ecosystems in the next section (see Section 8.3), whilst other indicators indirectly derived from ecological models are presented in Section 8.4 (see Section 8.4.3).

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

233

8.2.2 The Trophodynamic Simulation Module Ecosim Ecosim (Walters et al. 1997, 2000) re-expresses the linear equations of Ecopath as difference and differential equations that dynamically respond to changes in fishing mortality and biomass, enabling dynamic simulations at the ecosystem level from the initial parameters of a baseline Ecopath model (Fig. 8.4): dBi ¼ dt

  X X P  Qji  Qij þ Ii ðMi þ Fi þ ei Þ  Bi Q i

(8:4)

where dBi/dt is the growth rate during the time interval dt of (i) in terms of Bi; (P/Q)i is the gross efficiency; Mi is the non-predation natural mortality rate; Fi is the fishing mortality rate; ei is the emigration rate; Ii is the immigration rate; and ei  Bi–Ii is the net migration rate. Consumption rates (Q) are calculated based upon the ‘‘foraging arena’’ theory where the biomass of (i) is divided into a vulnerable and a non-vulnerable fraction and the transfer rate () between the two fractions is what determines the flow control (Fig. 8.5, Walters et al. 1997; Christensen and Walters 2004a): Qij ¼

aij  vij  Bi  Bj  Ti  Tj  Sij  Mij=Dj vij þ vij  Ti  Mij þ aij  Mij  Bj  Sij  Tj=Dj

(8:5)

where aij is the rate of effective search for i by j, Ti represents prey relative feeding time, Tj is the predator relative feeding time, Sij is the user-defined seasonal or long term forcing effects, Mij is the mediation forcing effects, and Dj represents effects of handling time as a limit to consumption rate. Default values of  represent mixed flow control ( = 2), whilst these values can be modified to represent bottom–up flow control ( = 1; i.e. donor-driven or prey control) and top–down flow control ( >> 1; i.e. Lotka-Volterra dynamics or predator control). In addition, a vulnerability value assigned to a given predator-prey interaction represents the factor by which a large increase in the predator biomass will cause predation mortality exerted by the predator on the prey to increase. For example, for an interaction assigned a high  (e.g.  =100), a doubling in the predator biomass would cause an approximately two-fold increase in the predation mortality inflicted upon the prey. Conversely, for a predator-prey interaction characterised by a low vulnerability (e.g.  close to 1), a large increase in predator biomass would have an unnoticeable effect on the predation mortality exerted by that predator on the prey in question (V. Christensen, pers. comm.).

8.2.2.1 Tuning the Model to Real Data There are many variables that can affect Ecosim simulations, including assumptions about flow control (Bundy 1997, Walters et al. 1997, Bundy

234

M. Coll et al.

Fig. 8.4 Example of Ecosim gaming simulations applied to the Southern Benguela ecosystem: Using EwE to simulate the potential effects on biomass of increased fishing mortality (F; lower panel in each scenario) under three types of flow control assumptions: top–down, bottom–up and mixed flow control: (a) fourfold increase in F of small pelagic fish (anchovy, sardine, round herring) from year 10–50; (b) pulsed fourfold increase in F of small pelagics from year 10–15 and (c) pulsed fourfold increase in F of hake from year 10–15. Biomass plotted relative to original biomass. Species groups as follows: 1 round herring; 2 hake; 3 sardine; 4 anchovy; 5 cephalopods; 6 other small pelagic fish; 7 chub mackerel; 8 seals; 9 large pelagic fish; 10 seabirds; 11 cetaceans; 12 horse mackerel; 13 chondrichthyans; 14 mesopelagic fish. Reprinted from Shannon et al. 2000. Modelling effects of fishing in the Southern Benguela ecosystem. Shannon et al. (2000), with permission from Elsevier

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

235

Biomass of predator (Bj)

aijViBj Non-vulnerable biomass of the prey

(Bi-Vi)

v(Bi-Vi)

v’(Vi)

Vulnerable biomass of the prey (Vi)

Fig. 8.5 Graphic representation of the foraging arena theory where the prey population is split into a vulnerable (Vi) and invulnerable (Bi–Vi) group. There is a steady flow of biomass between the two groups, and with the assumption that v = vij = vji. Prey from the vulnerable group is removed using the standard Lotka-Volterra equation, aijViBj, where Bj is the biomass of the predator and aij is the instantaneous rate of search. Adapted from Walters et al. 1997

2001, Shannon et al. 2004b). To this end, dynamic simulations with Ecosim can be carried out to test hypotheses and to calibrate the model to time series data. Ecosim simulations can be tuned to a time series of biomass and catch data as references, along with estimates of how fishing impacts have changed over a period of time (e.g. total or fishing mortality by functional group or fishing effort by fleet, Fig. 8.6). This enables the estimation of a statistical measure of goodness-of-fit to these data each time dynamic simulations are performed, comparing predicted model results to available (observed) trajectories. This goodness-of-fit measure is a weighted sum of squared deviations (SS) of log biomasses and catches from log predicted biomasses and catches (Christensen and Walters 2004a, Walters and Martell 2004). 8.2.2.2 Ecosim Properties and Routines Based on SS measures, three types of analyses are available and depend on a nonlinear SS minimization procedure. A non parametric procedure can be used to determine the sensitivity of SS to the vulnerability of functional groups () by changing each one slightly and then re-running the model to see how much SS is changed. After that, the best vulnerability values by functional groups can be estimated to give the better ‘‘fits’’ of Ecosim to the time series data (giving a reduction of the SS). In addition, an automatic procedure can be implemented to search for time series values of forcing functions (e.g. annual relative primary productivity) that represent productivity changes impacting biomasses throughout the ecosystem. A forcing function is sketched over time and applied to user-defined interactions only, usually to primary production. If applied to a primary producer, the forcing function alters the P/B of the producer, whereas in the case of a consumer-prey interaction, the rate of consumption of a prey

236

Fig. 8.6 (continued)

M. Coll et al.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

237

group by the predator is altered. Thus, when calibrating the model with available time series data, the roles of internal ecosystem forcing factors (e.g. flow controls determining trophic interactions) and external ones (e.g. fishing activities and environmental forcing) in driving the dynamics of marine resources over time can be explored (Christensen and Walters 2004a). Searching for a forcing function that produces an EwE model better fitted to observed time series data corresponds to searching for ‘‘nuisance parameter’’ estimates of the ‘‘process errors’’ in single-species assessment (Hilborn and Walters 1992). Christensen and Walters (2004b) introduced a new routine within Ecosim to search for alternative exploitation patterns setting different sustainability objectives. The ‘‘optimum policy search’’ routine is used to evaluate what changes in fishing effort or fishing mortality over time would maximize the performance of a particular measure for management. Implementing this routine facilitates searching for fishing rates that would maximize a combination of social, economical and ecological criteria (for further details see Section 8.3.2).

8.2.3 The Spatial Dynamic Module Ecospace Ecospace is a spatially explicit version of Ecosim that represents biomass dynamics over 2-D space (Walters et al. 1999), removing the assumption of homogenous spatial distribution and behaviour implicit in Ecopath and Ecosim. It employs the same set of differential equations as Ecosim (see Equations (8.4) and (8.5)), but additionally takes into account habitat preferences, movement due to advection and migration, the spatial behaviour of fishing fleets as well as trophic interactions and population dynamics.

Fig. 8.6 (continued) Example of fitting of the model to data for the southern Benguela: Abundance (A. biomass) and catch (B) time-series estimated by EwE (lines) and from timeseries data (dots) for the period 1978–2002. Biomass time series data are treated as relative and are scaled to match the EwE model series. Catches are scaled similarly (t km2 yr1) for both model and data. (A). (a) Estimated total biomass of anchovy compared to EwE anchovy biomass; (b) Sardine spawner biomass (surveys) compared to EwE sardine biomass; (c) Pairs of breeding gannets off South Africa compared to EwE seabird biomass; (d) Seal pups in the southern Benguela (census data) compared to EwEl seal biomass; (e) Exploitable biomass (single species model) of M. paradoxus on the West coast (R. Rademeyer, UCT, pers. comm.) compared to EwE large M. paradoxus biomass; (f) Exploitable biomass (single species model) of M. capensis on the South coast (R. Rademeyer, UCT, pers. comm.) compared to EwE large M. capensis biomass; (g) Estimated combined biomass of small and large M. capensis on both the West and South coasts (from surveys) compared to EwE large M. capensis biomass (Ecosim); (h) Estimated combined biomass of small and large M. paradoxus on both the West and South coasts (from surveys) compared to EwE large M. paradoxus biomass (Ecosim). (B).(a) Anchovy, (b) Sardine, (c) Chokka Squid, (d) Large horse mackerel, (e) Snoek, (f) Large M. capensis, (g) Large M. paradoxus, (h) Small M. paradoxus. Adapted from Shannon et al. (2004) with permission of NISC

238

M. Coll et al.

The model area is defined by a grid of cells, representing up to 8 different habitat types. In each habitat type, the cells share common attributes which affect movement, feeding rate and survival. Commonly these habitats will support a particular sub-food web, although they usually include the entire water column, from the benthos to the pelagic zone. Species that transcend habitat types (e.g. marine mammals, plankton) connect the sub-food webs together. The user sketches the different habitat types (e.g. shelf, coastal, deep, reef) on a base map, and then assigns which are the preferred habitats for the various functional groups in the model. An Eulerian approach is used in Ecospace to explicitly model movement, where the movement or the flow of biomass occurs with respect to fixed reference points, the grid of cells. Minimally this approach approximates the changing centre of distribution of the biomass (Christensen and Walters 2004a). Movement occurs across each of the four faces of the grid cell, unless the cell is on the boundary, in which case it is assumed that emigration out of the boundary cell is equivalent to immigration into the boundary cell. Movement between cells is determined by several factors which account for (1) movements of each functional group due to dispersal (mi) and advection (Vi), (2) predation risk and food availability and (3) fishing effort (Walters et al. 1999). Initial estimates of dispersal rates are measured as average swimming speed (km yr1) and advection is estimated from current fields (see below). The emigration term in Equation (8.4) is the sum of the flows across each cell face, represented by (mi+Vi)Bi (Martell et al. 2005). Immigration is also represented by (mi+Vi)Bi input rates, in this case proportional to the biomass in the adjacent cells. The instantaneous movement rates, mi, vary with pool type, habitat type in the source cell and response to predation risk and feeding conditions (risk ratio). Note that mi is not a directed rate, but should be considered as dispersal. A ‘‘habitat gradient function’’ enables a more realistic representation of movement where organisms respond to gradients in the environment (e.g. depth, salinity, temperature) and intentionally move towards cells with favourable habitat types. Martell et al. (2005) have explored this further and investigated the effects of different fitness dispersal rates (see Section 8.3). Two other forms of movement can also be explicitly modelled in Ecospace: migration and advection (advances since the first published model in 1999, see Christensen and Walters 2004a, Walters et al. 2004). Migration is modelled by defining a monthly series of preferred positions for the migrating species and associated concentration parameters (the spatial spread of the migrating fish around these preferred cells). Advection is a critical oceanic process for the distribution of larvae, nutrients and productivity in general. Ecospace models consider advection by first importing user defined current patterns or other types of physical forcing to define surface movement in the model. Using a series of linearised pressure field and velocity equations, which include sea surface anomalies, bottom friction force, the Coriolis force, down-welling/ up-welling rate and acceleration due to sea surface slope, Ecospace estimates equilibrium flow fields (horizontal and upwelling/down-welling) across each

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

239

cell face in the Ecospace grid (Walters et al. 2004). The flow fields maintain water mass balance and Coriolis force. Once defined, the user specifies which functional groups are subject to advection. A great advantage of spatial modelling is the ability to better represent the spatially aggregated patterns of fishing: fishermen generally know the best places to fish! Furthermore, open and closed areas are important tools for fisheries management and their efficacy can be explored using Ecospace (Fig. 8.7). Ecospace uses a ‘‘gravity model’’ to distribute fishing effort spatially. The distribution of effort is based on the habitat type, whether the area is open or closed, biomass of species of interest, price of fish and cost of fishing. This is essentially a spatial economic model where it is assumed that fishing fleets will operate in areas which are most cost effective providing they are accessible. Once an Ecospace model has been parameterized, with the biomass of the functional groups assigned to preferred habitats, the user-set preferences described above will determine the movement of the biomass pools. At this stage, an iterative approach is recommended where the results from Ecospace are compared with the Ecopath model for consistency (Walters et al. 1999). The predicted distribution maps of species or groups of species can be used to validate model results.

Fishing zone Island

No fishing zone

Fig. 8.7 A simple diagrammatic representation of the potential effects of a fisheries exclusion zone on pepino (S. fuscus) biomass at the end of a 10-year Ecospace simulation at a hypothetical Gala´pagos island. Darker areas represent high biomasses and lighter areas represent low biomasses. Catchable emigration of pepinos can be seen as dark shading outside the dotted lines that demarcate the boundaries of the hypothetical fisheries exclusion zone. Pepinos still decline to a biomass lower than present, but the no-fishing zone prevents the intense fishery from extirpating them. Reprinted from Okey et al. 2004. A trophic model of a Galapagos subtidal rocky reef. Ecological Modelling, 172: 383–401, with permission from Elsevier

240

M. Coll et al.

8.2.4 Data Requirements, Sources and Shortcuts During the parameterization of an Ecopath model, three of the four basic parameters in Equation (8.2): Bi, (P/B)i, (Q/B)i or EEi, need to be provided per functional group, in addition to catch data and trophic behaviour by functional group. Biomass of functional groups can be obtained from different sources of information: e.g. swept area method, egg production method, acoustic surveys and visual census. Ideally production/biomass (P/B) and consumption/biomass ratios (Q/B) should be calculated empirically, but are usually calculated by the application of empirical equations using length, weight and growth data (Nilsson and Nilsson 1976, Pauly 1980, Innes et al. 1987, Pauly et al. 1990, Christensen et al. 2005). Diet composition can be estimated from stomach contents analysis. Data on total catches need to be included in the model by functional group and fishing fleet, considering official landings statistics, discards and estimates of illegal, unregulated or unreported (IUU) landings. The Ecosim module uses initial parameters inherited from the baseline Ecopath model, therefore essential data requirements do not increase substantially when performing temporal simulations. However, time series of data for fishing effort by fleet and total or fishing mortality and time series of biomass and catches for various functional groups are required to tune the model to data (Christensen and Walters 2004a, see Section 8.2.2). The latter step of fitting an Ecosim model to time series data (see Section 8.3.2), although a time-consuming and demanding process, is advisable before detailed exploratory simulations are performed. Ecospace also uses the initial parameters from Ecopath. Additional data requirements (see Section 8.2.2) include the identification of habitat types, sketched on to base map and assignation of the functional groups to these habitats. In its most simple implementation, the only further input required by Ecospace are the dispersal rates for each functional group, movement rates in different types of habitat (good, bad) and identification of the location of fishing activity for each fleet. Other optional input data include importing GIS maps for the basemap, setting up MPAs, importing nutrient data and current patterns or surface currents forcing patterns to set up advection patterns (e.g. Martell et al. 2005, see below).

8.2.5 Allowing for Uncertainty Mass balance models are deterministic and require many input parameters, some of which may be poorly known, or adapted from other ecosystems or Ecopath models. This introduces a high level of uncertainty to the results of the model estimates. Therefore the uncertainty associated with model output should always be explored. To this end, EwE includes a data pedigree routine

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

241

% change in model estimate

1.2 1 0.8 0.6 0.4 0.2 0 –0.2

–50% –40% –30% –20% –10% 0%

10% 20% 30% 40% 50%

% change in input parameter

–0.4 Zooplankton

Phytoplankton

Sergestid Shrimp

Fig. 8.8 Simple sensitivity analysis of model estimates of biomass to the sergestid shrimp production/biomass and ecotrophic efficiency values in a model of San Miguel Bay, the Philippines. Adapted from Bundy 1997

(see Section 8.2.3), a simple sensitivity analysis, the Ecoranger utility, mixed trophic impact assessment and the autobalance routine. The simple sensitivity analysis in Ecopath quantifies the effects of increasing and decreasing each of the basic input parameters (B, P/B, Q/B and EE) in steps of 10%, by up to 50% of its original value (Fig. 8.8). Model output is in the form of a table of the difference between the new estimated output and its original value, as a proportion of the original value (Christensen and Walters 2004a). Mixed trophic impact assessment (MTI, see Section 8.2.1, Fig. 8.3) assumes that trophic structure is constant. This means that the technique cannot be used for predictive purposes, but should rather be considered as a simple form of sensitivity analysis. It is an indicator of which groups have negligible effects on others within the system, and for which there is likely to be little gained from an effort to collect additional data to refine estimates. On the other hand, it identifies groups having large trophic impacts on others, and for which it would be useful to refine estimates. Ecoranger is a Monte Carlo approach within Ecopath (Christensen and Pauly 1995, 1996, Pauly et al. 2000), enabling the incorporation of variability around values for the basic input parameters: B, Q/B, P/B, EE and diet composition for all groups. The mean or mode and range of these parameters can be entered and a frequency distribution (uniform, normal, log-normal or triangular) defined from which random samples are drawn to generate distributions for output variables. To put Ecoranger into a semi-Bayesian context, a ‘‘sampling/importance resampling’’ procedure based upon that of McAllister et al. (1994) is used (Christensen and Pauly 1996). Each possible model output is evaluated and of all the runs, the best-fit model is selected using a least square method. The best-fit model is that giving the smallest residuals (based on mean/mode of each selected parameter

242

M. Coll et al.

and the squared deviations of each), or the smallest deviation from three other user-defined criteria, namely maximum system biomass, maximum throughput or maximum ascendancy. Ecoranger can also be used in the Ecosim dynamic module using the confidence intervals from the model pedigree to run repeated temporal simulations. Each simulation begins by selecting random input combinations from normal distributions centered on the initial inputs of the baseline model, then the resulting model (if balanced) is used to develop dynamic simulations. This result can be seen as bands of uncertainty giving an idea of how sensitive Ecosim results are to input parameters (Christensen and Walters 2004a). The Autobalance routine is another means to explore uncertainty. However, only the biomass and diet parameters can be directly changed in the current Autobalance routine, thus it is not a full sensitivity analysis. Bundy (2005) used the Autobalance routine as a perturbation analysis of balanced Ecopath models to explore the effects of uncertainty on model results. For tractability reasons, only thirty Autobalance model runs were completed, but these provided estimates of 95% confidence limits for all input and output parameters. When comparing two models, a Mann-Whitney U, two independent samples test was used to test whether differences between the models were significant, or an artefact created by the uncertainty of the input parameters. Though this approach used only 30 replicates, it does provide some rigour for this type of comparative analysis. Model estimates can also be compared with alternative estimates of certain input parameters. For example, an alternative estimate of biomass of a poorly known yet important group such as gelatinous zooplankton was available and thus tested for the northern Benguela ecosystem model of the period 1980–1989 (Shannon and Jarre-Teichmann 1999a,b). Generally, these tools are basic and more rigorous sensitivity analyses are needed to formally analyze the propagation of uncertainty of input variables on the value of outputs, providing ranges of output values (Aydin and Friday 2001, Bundy 2005, Gaichas 2006).

8.2.6 User Beware – A Guide to Common Pitfalls and Advice for Avoidance EwE carries several of the same risks as single-species models, such as uncertainty associated with estimates of biomass, misinterpretation of trends in data series, and problems pertaining to disentangling the often confounding effects of environmental changes and the effects of fishing (Christensen and Walters 2004a). Users should make careful choices when selecting periods to be modelled. The period over which simulations can be considered reliable is often limited by data quality (Walters et al. 1997). On the other hand, focusing on such short time periods runs the real risk that important long-term effects may be missed (Mackinson et al. 1997, Walters et al. 1997). Use of erroneous or poor input data and parameters will obviously severely reduce the value of an EwE model and can lead to model outputs that are unrealistic and incorrect.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

243

However, these kinds of erroneous model outputs normally arise from errors in a few key input parameters rather than from general uncertainties in the model as a whole (Christensen and Walters 2000). Christensen and Walters (2000) list five main pitfalls of EwE: (i) prey that is rarely found in the diet of a predator may be omitted from diet composition estimates, leading to inaccuracies in the modelled effects of the predator on these prey, and visa-versa; (ii) trophic mediation effects, though they can be included in EwE, they may be overlooked (these are the indirect effects that the behaviour or presence of a third group may have on a predator-prey interaction); (iii) predation vulnerabilities are often underestimated, lessening the modelled impacts of predation; (iv) predators share foraging arenas; if abundance of one predator decreases, another may fill its place so that prey species do not benefit; (v) temporal variation in factors affecting species-specific habitats is not incorporated. In addition, non-trophic mediation effects (such as the effect of habitat type, the presence of refuges or other behavioural effects) can also be modelled in EwE, although they have not been frequently included in the models. These pitfalls can be overcome with good quality data and a notable knowledge of the system to model. Plaga´nyi and Butterworth (2004) provide several suggestions for EwE users to avoid common pitfalls. Users are warned against adopting default parameter settings without questioning their meaning or examining their effects on the results they obtain. A further warning has been sounded that care should be taken when focussing EwE applications on marine mammals and seabirds, given their very different life history traits to the traditional fish groups for which the EwE modelling approach was specifically developed. The authors advise that data quality should guide decisions on which functional groups to include in an EwE model and that effort be made to ensure that time-specific and spatially-specific diet compositions are used wherever possible. Regarding analysis and presentation of EwE model results, they emphasize the need for recognition of model complexity and uncertainty through presentation of model outputs as a range of likely scenarios. Sensitivity of EwE model results to the choice of vulnerability settings in Ecosim has been shown to be a major factor in interpretation of model outputs for fisheries management advice. In response to this, Plaga´nyi and Butterworth (2004) have proposed some guiding steps for EwE users, including searching for group-specific values for vulnerabilities rather than adopting default settings across all groups, and in particular, using available time series to fit Ecosim models by searching for ‘‘best’’ fit vulnerabilities (Christensen et al. 2000). It is advisable to follow the lead set by Arreguı´ n-Sa´nchez (2000) and Bundy (2001), for example, who presented fisheries management scenarios for a range of flow control (vulnerability) assumptions.

244

M. Coll et al.

8.3 Case Studies Ecopath models have been used to examine the trophic structure and functioning of a host of aquatic ecosystems, including lakes, aquaculture systems, estuaries, small bays, coastal systems and coral reefs, shelf systems, upwelling systems, and open seas (see examples in Table 8.1). This section will discuss a selection of the numerous available Ecopath models. In many cases, these models have served as the basis for Ecosim and Ecospace simulations to explore the trophic functioning and the ecosystem effects of fishing in these systems.

Table 8.1 Examples of Ecopath, Ecosim and Ecospace case studies System modelled Application (use and/or model output) References A. Ecopath models 1. Energy budgets, trophic structures and network analyses Tiahura reefs, Moorea Examined ecosystem structure and Island (French functioning of the fringing and barrier Polynesia) reefs North and Central Highlighted key role of benthic-pelagic Adriatic Sea coupling, small pelagics and jellyfish (Mediterranean) and the importance of the microbial food web. The ecosystem was described as highly impacted by fishing South Catalan Sea Showed the ecosystem was pelagically (NW Mediterranean) dominated, important pelagic-benthic coupling and importance of detritus and detritivors. The ecosystem showed high fishing intensity with large ecosystem impacts Cantabrian Sea Represented the Cantabrian Shelf (Bay (Bay of Biscay) of Biscay, Spain) in 1994 Strong relationships between pelagic, demersal and benthic compartments were identified Southern Plateau Described a low productive ecosystem (New Zealand) with importance in terms of feeding seabirds, seals and fish and of commercial fishing Kuosheng Bay, Taiwan Investigated effect of power plant on bay ecosystem Explored the definition of the Pribilof archipelago boundaries of open marine (Southeast Bering ecosystems by comparing the area of Sea) maximum energy balance by means of a mass balanced model from the 1990s with estimates of the foraging rage of the northern fur seals

Arias-Gonza´lez et al. 1997 Coll et al. 2007

Coll et al. 2006a

Sa´nchez and Olaso 2004

Bradford-Grieve et al. 2003

Lin et al. 2004 Ciannelli et al. 2005

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

Table 8.1 (continued) System modelled Central Pacific Ocean and northern Gulf of Mexico

Tongoy Bay (Northern Chile)

245

Application (use and/or model output)

References

The maximum biomass of top predators possible to accommodate within the models excluding fishing and non changed primary production to investigate the carrying capacity of ecosystems Examined ecosystem structure of a suspended scallop culture system.

Christensen and Pauly 1998

2. Placing fisheries within their ecosystem context San Miguel Bay (Pacific To examine the ecosystem effects of Coast of Southeast large-scale and small-scale fishing Luzon, Philippines) gears Gulf of California To describe the trophic web of the (Mexico) California Gulf during the late 1970s and study the shrimp trawling exploitation to understand the role of by-catch Black Sea (Eastern Models applied to the Black Sea Mediterranean) ecosystem to study the outburst of the gelatinous Mnemiopsis leidyi and the decline of small pelagic fish A mass balance model was used to South Humboldt describe the system in 1992 and assess Upwelling the impacts of fishing activities (Central Chile) Investigated the consequences of fishing Northern and Southern on small pelagic fish: market effects Benguela, Southern both on the higher and lower trophic Humboldt and the levels of the food web, causing Mediterranean Sea decrease of predators, the proliferation of other species and the disruption of energy flows. 3. Comparing ecosystems through time Southern Benguela Similar trophic functioning in 1980s and (South Africa) 1990s despite changes in abundance of species groups; fishing-induced changes from pristine ecosystem state Northern Benguela Large changes in ecosystem structure (Namibia) and trophic functioning since 1970s; shift from pelagic to demersaldominated ecosystem South Humboldt Compared models of 1992 and 1998. (Central Chile) Showed importance of predation mortality on fish production, quantified strong fishing effects, higher biomass supported in 1998 than in 1992 but smaller catches

Wolff 1994

Bundy and Pauly 2001 Arreguı´ n-Sa´nchez et al. 2002

Gucu 2002

Neira and Arancibia 2004 Shannon et al. in press

Shannon et al. 2003; Watermeyer 2007, and Watermeyer et al. 2008 Heymans et al. 2004, Roux and Shannon 2004 Neira et al. 2004

246 Table 8.1 (continued) System modelled Eastern Scotian shelf (North Atlantic)

Eastern Scotian shelf (North Atlantic)

Venice lagoon (NE Italy)

Orbetello lagoon (Central western Italy)

M. Coll et al.

Application (use and/or model output)

References

Models of before and after collapse of Atlantic cod explored changes in ecosystem structure and demonstrated a shift from benthic-feeder dominance to pelagic-feeder dominance and an increase in piscivory Models of before and after collapse of Atlantic cod explored reasons for the non-recovery of Atlantic cod and concluded that competition for food from the large biomass of pelagic fish contributes to the non-recovery Two models representing the lagoon from 1988–1991 and 1998 were compared to analyze the fishing impacts of Manila clam dredging in the area, which was developed from the middle 1980s Two models described the lagoon in 1995 and 1996 to analyze the effects of management activities developed in the area to control eutrophication.

Bundy 2005

4. Comparing studies across ecosystems Upwelling ecosystems Four models representing the Northern and Southern Humboldt and the Northern and Southern Benguela models were standardized and compared A model from representing the South Upwelling ecosystems Catalan Sea (NW Mediterranean in and NW 1994, Coll et al. 2006a) was compared Mediterranean Sea with the four models reported in Moloney et al. (2005) to assess ecosystem effects of fishing taking into account differences on ecosystem features Different habitats of the Comparison of two models representing two different habitats of the Venice Venice lagoon lagoon: the seagrass meadows and (NE Italy) Manila clam (Tapes philippinarum) fishing grounds. B. Ecosim models 1. Exploring impacts of fishing and management simulations Gulf of Thailand Two mass-balance models were used to reproduce changes on the ecosystem from 1963 to the 1980s by developing dynamic simulations of increasing or decreasing fishing activity

Bundy and Fanning 2005

Pranovi et al. 2003

Brando et al. 2004

Moloney et al. 2005

Coll et al. 2006b

Libralato et al. 2002

Christensen 1998

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

Table 8.1 (continued) System modelled Upwelling ecosystems

Different habitats of Tongoy Bay (Chile)

Gulf of Mexico

South Catalan Sea (NW Mediterranean)

Northern Colombian Caribbean Sea

South Catalan Sea (NW Mediterranean)

Central North Pacific ecosystem

247

Application (use and/or model output)

References

Exploration of ecosystem effects of fishing on small pelagic fish from three upwelling ecosystems: Peru, Venezuela and Monterey Bay Exploration of different fishing options in four models representing different benthic habitats of Tongoy Bay, Chile: seagrass, sand-gravel, sand and total ecosystem Dynamic simulations were performed using a mass-balance model of two ecosystems of the Gulf of Mexico to evaluate the ecological role of snappers and the impact of their exploitation Showed low resistance of the ecosystem to fisheries, an increase in fishing effort resulted in decreased catches under all scenarios analysed Explored the ecosystem impacts of reducing fishing mortality due to by-catch reduction devices applied to a tropical industrial shrimp fishery Available field data from bottomtrawling selectivity by applying sorting grids and square meshes have been put into ecosystem context simulating the consequent reduction of fishing mortality on target species in the area Assessed the ecological and economic impacts of alternative fishing methods to reduce the by-catch of marlin.

Mackinson et al. 1997

2. Examining energy flow controls Southern Benguela Exploring the effects of fishing on small (South Africa) pelagic fish and hake were explore under different scenarios of top–down and bottom–up flow control Southern Benguela EwE was used to simulate changes in the (South Africa) Southern Benguela ecosystem from an anchovy-dominated system to a sardine-dominated one NewfoundlandExploration of effects of fishing and Labrador, (Canada) predation on the ecosystem from 1980s to see if they could reproduce ecosystem changes observed from the early 1990s

Ortiz and Wolff 2002a

Arreguı´ n-Sa´nchez and ManickchandHeileman 1998

Coll et al. 2006a

Criales-Hernandez et al. 2006

Coll et al. 2008a

Kitchell et al. 2002

Shannon et al. 2000

Shannon et al. 2004a

Bundy 2001

248 Table 8.1 (continued) System modelled South Brazil Bight coastal ecosystem

Barents Sea

San Miguel Bay (Philippines)

M. Coll et al.

Application (use and/or model output)

References

Exploration of the effects of changing fishing strategies in terms of increasing squid’s catches and of livebaitfish for sardines A mass-balance model was used to explored different functional response hypotheses of minke whales, their prey and theirfisheries Application of an adaptive management approach to explore management options under different flow assumptions.

Gasalla and RossiWongtschowski 2004

3. Analyzing environmental forcing of ecosystem dynamics Black Sea Exploration of fishing and (Mediterranean) eutrophication on the Black Sea ecosystem and contrast of results with biological time series Study of how the effects of el Nin˜o – Pelagic system (eastern Southern Oscillation (ENSO) might tropical Pacific affect different organisms at middle Ocean) and high trophic levels. 4. Fitting models to data Southern Benguela (South Africa)

Northern Benguela (Namibia)

South Catalan Sea (Mediterranean)

Southern Humboldt upwelling (Chile)

A mass-balance model was fitted to available time series data for a 25-year period from 1978 to 2002, exploring how fish stock dynamics may be determined by feeding interaction patterns (flow controls), fishing strategies and environmental changes Temporal dynamics affecting the ecosystem were explored by fitting the model to time series of data over 30 years from 1970s and assuming wasp–waist flow control by small pelagic fish A model was fitted to available time series of data from 1978 to 2003 explaining 78% of data variability taking into account trophic control (67%), fishing (7%), and environmental factors (4%) The model fitted from 1970 to 2004 showed that fishing mortality explained 28% of the variability in the times series, vulnerability parameters explained 21%, and a forcing function affecting primary production explained a further 11–16% of the observed variability

Mackinson et al. 2003

Bundy 2004a

Daskalov 2002

Watters et al. (2003)

Shannon et al. 2004b.

Heymans 2004

Coll et al. 2008b

Neira et al. in prep.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

Table 8.1 (continued) System modelled

Application (use and/or model output)

Eleven ecosystems from different areas

Exploration of results of fitting eleven mass-balance models to time series of data and of ecosystem effects of harvesting species to their single species maximum sustainable catch. 5. Policy optimization and management scenarios Southern Benguela The optimization routine was applied to (South Africa) this upwelling system. Results showed that extreme optimal fishing scenarios forced the model parameters beyond their likely ranges producing highly unrealistic outcomes Prince Willinam Sound The optimization routine was used (Alaska) to analyze management options for within the context of rebuilding pinnipeds populations Gulf of Thailand Searching for alternative exploitation patterns setting different sustainability objectives optimizing for profit, value and conservation Application of the optimization routine La Paz Bay, Baja to the artisanal fisheries based on California Sur hook-and-line and on gillnets coexist, (Mexico) in conjunction with a shrimp fishery. 6. Back to the future simulations Reconstructing past systems using Strait of Georgia in modelling and traditional or local British Columbia, knowledge, historical documentation, Newfoundland, and archaeological data to explore (Northern British future policy goals. Columbia) 7. Pollution studies Prince William Sound, (Alaska)

Faroe Islands ecosystem

Exploration of impacts produced by Exxon Valdez oil spill by performing simulations changing mortalities of different functional groups Methyl mercury concentration on food web and marine mammals was modelled to explore the implications of human diet on cod and pilot whales.

C. Ecospace models 1. Modelling spatial dynamics Tongoy Bay, Chile Explored policies for sustainable exploitation of four benthic species in different benthic habitats, including exploitation exclusively in one habitat to exploitation across all habitats.

249

References Walters et al. 2005

Shannon 2002

Okey and Wright 2004

Christensen and Walters 2004b

Arreguı´ n-Sa´nchez et al. 2004

Dalsgaard et al. 1998 Pitcher et al. 2002 Ainsworth et al. 2002

Okey and Pauly 1999, Okey 2004

Booth and Zeller 2005

Ortiz and Wolff 2002b

250 Table 8.1 (continued) System modelled

M. Coll et al.

Application (use and/or model output)

2. Establishment and assessment of MPAs Brunei Darussalam, Introduction of Ecospace as an Southeast Asia important exploratory tool for MPA definition and function and policy exploration. Authors discuss the affects of harvesting on MPA boundaries, trophic cascades and density dependent effects and concluded that few large MPAs are more effective than more small MPAs Explored the consequences of alternative Gwaii Haana National MPA zoning policies. Authors Marine Conservation concluded that, coupled with a Area, British reduction in harvest pressure, a large Columbia, Canada MPA with a buffer area around it leads to the greatest increase in biomass Galapagos subtidal Authors demonstrate how the rocky reef functional extinction of sea cucumbers could be avoided by protecting some of the reef from fishing Central North Pacific Exploration of the relative importance of different assumptions about dispersal and advection under different fishing policy scenarios with respect to marine protected areas and concluded that MPAs for large pelagics need to be large Hong Kong Spatial policy exploration such as tradeoffs between compliance with fishery regulations and conservation in the Hong Kong artificial reef system, for example where fishing was permitted in one artificial reef, assuming that this would lead to greater support for the artificial reef scheme and self-enforcement.

References Walters et al. 1999

Solomon et al. 2002

Okey et al. 2004

Martell et al. 2005

Pitcher et al. 200b

8.3.1 ECOPATH 8.3.1.1 Energy Budget, Trophic Structure and Network Analysis A fundamental use of Ecopath models is the estimation of energy budgets, trophic flow and structure. Arias-Gonzalez et al. (1997) modelled two Tiahura reefs (Moorea Island, French Polynesia) (Arias-Gonzalez et al. 1997) highlighting the high proportion of primary productivity processed and recycled within both systems and the importance of detritus and microbially mediated food web.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

251

A model of the North and Central Adriatic Sea, representing the widest continental shelf of the Mediterranean Sea, highlighted the key role of benthic-pelagic coupling, small pelagics and jellyfish in this exploited Mediterranean shelf ecosystem (Coll et al. 2007). In addition, results indirectly underlined the importance of the microbial food web in the Adriatic Sea. Another model representing the exploited ecosystem of the South Catalan Sea (NW Mediterranean) in 1994 also showed that the ecosystem was pelagically dominated but that pelagic-benthic coupling, by means of flows from the pelagic food webs to detritus, and the abundance of detritivores, were important in the system (Coll et al. 2006a) (Fig. 8.1). Similarly, Sa´nchez and Olaso (2004) modelled the Cantabrian Shelf (Bay of Biscay, Spain) demonstrating a strong relationship between the pelagic, demersal and benthic compartments in the ecosystem. Table 8.2 presents several results from mass balance models of the South Catalan Sea, North and Central Adriatic Sea and Cantabrian Sea with respect to global statistics, network flow indices and information indices. Ecopath has also been applied to food webs with very limited primary production. For example, the Southern Plateau of New Zealand (BradfordGrieve et al. 2003) is characterized by low levels of phytoplankton biomass and was described by means of an ecosystem model that highlighted the importance of the microbial loop. The system was dominated in terms of trophic flows by the pelagic compartment, mainly retaining 69% of the biomass and 99% of the production. The mean transfer efficiency of the ecosystem between trophic levels II and IV was very high (23%), underlining the energy limitation in that ecosystem. The site of a coastal nuclear power plant in Kuosheng Bay, Taiwan, is one of the more unique applications of the Ecopath modelling approach (Lin et al. 2004). The Ecopath model was used to explore whether the impingement and entrainment of organisms during the intake of vast quantities of water for cooling of the power plant, and the expelling of warm water into the bay, impacted the coastal ecosystem. The total ecosystem attributes (total biomass, total system throughput, etc.) suggested that the bay system behaved like a normal coastal ecosystem in terms of structure and functioning. At the bay scale, the power plant was not having large impacts; the effects were likely to be localized to the warm water radius of the plume of expelled water. However, the Kuosheng Bay ecosystem was found to be more detritus-dependent than other coastal ecosystems to which it was compared, related to the rapid turnover rates of phytoplankton feeding into the detritus box. The boundaries of open marine ecosystems were explored by Ciannelli et al. (2005) using an Ecopath model of the Pribilof archipelago (Southeast Bering Sea) and comparing the area of maximum energy balance with estimates of the foraging range of the northern fur seals. Considering foraging theory, an ecosystem boundary should include the foraging range of the species that live within it for a part of their life cycle; considering ecosystem energetics an ecosystem should be the area within which the predatory demand is in balance

252

M. Coll et al.

Table 8.2 Global statistics, network flow indices and information indices from mass balance models of the South Catalan Sea (Coll et al. 2006a), North and Central Adriatic Sea (Coll et al. 2007) and Cantabrian Sea (Sa´nchez and Olaso 2004) Catalan Adriatic Cantabric Units sea(1) sea(2) sea(3) Statistics and flows Sum of all consumptions Sum of all exports Sum of all respiratory flows Sum of all flows into detritus Total system throughput Total primary production/total respiration Net system production Total primary production/total biomass Total biomass/total throughput Total biomass (excluding detritus) Total respiration/total biomass Ecopath Pedigree index Gross efficiency of the fishery Mean trophic level of the catch Mean trophic level of the community Network flow indices Predatory cycling index (% of throughput w/o detritus Finn’s cycling index (% of total throughput) Finn’s mean path length System Omnivory Index Information indices Ascendency (%) Capacity (Total) (1) Coll et al. 2006a; (2) Coll et al. 2007; (3) Sanchez and Olaso 2004.

851.73 1251.89 326.86 1607.52 4038.0 4.83

1305.04 730.75 421.09 1388.07 3845.0 2.73

2528.35 1075.86 950.88 1513.15 6068.0 2.13

t/km2/yr t/km2/yr t/km2/yr t/km2/yr t/km2/yr

1250.14 26.74

729.37 8.83

1074.12 10.60

t/km2/yr

0.02 58.97 5.54 0.670 0.003 3.12 1.50

0.03 130.30 3.23 0.665 0.002 3.07 1.39

0.03 191.00 4.98 0.669 0.006 3.76 2.31

3.33

3.97

3.55

%

6.77

14.69

4.89

%

2.56 0.22

3.34 0.19

2.99 0.27

35.08 12738.9

27.0 15409.6

25.9 29577.2

t/km2

% Flowbits

with the prey production. This work examined the limitations of the current definition of an open ocean ecosystem in a spatial context. Christensen and Pauly (1998) used Ecopath to estimate the maximum biomass of top predators that it was possible to accommodate in two different ecosystems, under two scenarios, one excluding fishing and one with no change in primary production. They observed that in both models, the Central Pacific Ocean and the northern Gulf of Mexico, top predators biomass were able to be increased by an order of magnitude and changes in food web structure were in agreement with the theory of ecosystem development sensu Odum (1969). Based on these results they proposed a functional definition of the carrying

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

253

capacity: ‘‘the upper limit of biomass that can be supported by a set primary production and within a variable food web structure is reached when the total system respiration equals the sum of primary production and detritus import’’. Aquaculture systems have also been analyzed using ecosystem models. In Northern Chile, for example, the suspended scallop culture located in Tongoy Bay has been studied using a 17 compartment model (Wolff 1994). Benthic invertebrates dominated the system in terms of total food intake and total biomass in the water column. The system was described to be of low maturity and to have high capacity to withstand ecological perturbations.

8.3.1.2 Placing Fisheries Within an Ecosystem Context Fisheries can be considered as top predators in ecosystems, exerting strong top– down control and often causing cascading effects down the food web. Fisheries may compete with natural top predators and with other fisheries within an ecosystem; the ways that one fishery impacts various ecosystem components will have impacts on other fisheries operating in the same ecosystem. The Ecopath models of the South Catalan Sea and the North and Central Adriatic Sea, Mediterranean (Coll et al. 2006a, 2007) showed that these ecosystems were heavily fished during the 1990s. Similarly, in the Cantabrian Sea model (Sa´nchez and Olaso 2004), the ecosystem effects of different fleets operating in the area were analysed. Results suggested that fishing activity in the area was comparable to the most intensively exploited temperate shelves of the world. In both the Mediterranean and the Cantabrian Sea, the trawling fleet was identified to be the gear having the strongest impacts in the ecosystem. Bundy and Pauly (2001) modelled the San Miguel Bay fisheries (Pacific Coast of Southeast Luzon, Philippines) to examine the ecosystem effects of large-scale and small-scale fishing gears. Results highlighted that both fishing sectors had high impacts on the ecosystem, but that the cumulative impact of the range of small-scale fishing gears was greater and more diverse than that of the large-scale fishery sector. While the large scale sector fished across most trophic levels, the small-scale sector as a whole caught a greater number of species across an even wider trophic range. These results demonstrated the complexity of the interactions that occur amongst the effects of fishing mortality, predation mortality and competition. Arreguı´ n-Sa´nchez et al. (2002) described the trophic web of the California Gulf (Mexico) during the late 1970s to examine the ecosystem effects of shrimp trawling and to understand the impacts of by-catch (Fig. 8.3). They concluded that maintaining by-catch would be beneficial to maximize shrimp yields due to the fact that some important predators of shrimps were captured by trawling in notable proportions. Therefore, reducing by-catch would lead to an increase in the biomass of species preying on shrimps. However, the analysis highlighted that if the fisheries management objective was to maximize overall production

254

M. Coll et al.

of the ecosystem or reduce the impact of fishing on other species, a reduction of by-catch would be compulsory. Gucu (2002) studied the outburst of the gelatinous Mnemiopsis leidyi in the 1990s and the decline of small pelagic fish in the Black Sea ecosystem using Ecopath. Results suggested that gelatinous species only played a small role in the decline of small pelagic fish, while overfishing was singled out as the primary cause. Overfishing, however, did play a crucial role in the successful development of gelatinous species by vacating the ecological niche occupied by small pelagics, and enabling gelatinous organisms to proliferate, taking advantage of the increase in plankton productivity associated with eutrophication during the 1980s. Similarly, modelling studies of the upwelling system of Central Chile in 1992 suggested that fishing activities removed about 15% of total calculated system primary production (Neira and Arancibia 2004), exploiting organisms with an intermediate to low trophic level. Although natural mortality by predation was found to be high off Chile, for example predation mortalities of small pelagic fish were estimated to have between 0.7 and 1.4 yr1 (Shannon et al. in press), the fishery also removed a large proportion of commercial species like anchovy, horse mackerel and Chilean hake production. Shannon et al. in press compared the consequences of fishing on small pelagic fish in the ecosystems of Northern and Southern Benguela, Southern Humboldt and the Mediterranean Sea. They underlined that fishing mortality rates for small pelagics were high in the Mediterranean Sea, Northern Benguela and Humboldt. In addition, models showed that a decrease in small pelagic fish abundance would have market effects both on the higher and lower trophic levels of the food web, causing decrease of predators, the proliferation of other species that are prey or competitors (e.g. gelatinous zooplankton and benthopelagic fish) and, generally, the disruption of energy flows. Together, these could result in an increase in flows to detritus, increasing importance of demersal processes in the system and often reduced summed impacts of the demersal compartment on the degraded pelagic food web.

8.3.1.3 Comparative Studies: Examining Changes in an Ecosystem Over Time Much can be learned from comparative modelling studies where Ecopath models of the same system are developed for different time periods; changes in ecosystem structure and trophic relationships due to fishing, environmental perturbations or a combination of both can be quantitatively assessed. Given the uncertainty associated with Ecopath estimates, and recent advances in estimating the effects of uncertainty, it is critical to include confidence limits when comparing the results of models (see above) when possible. Several models have been constructed for upwelling regions (e.g. JarreTeichmann et al. 1998, Neira et al. 2004, Heymans et al. 2004, Roux and Shannon 2004, Shannon 2001, Shannon et al. 2003, 2004b, Moloney et al.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

255

2005). Upwelling ecosystems are characterized by large fluctuations in abundance of dominant species (anchovy and sardine regime shifts), respond fairly vigorously to environmental events or changes, and have been shown to undergo major changes in their food web structure and sometimes also their trophic functioning (e.g. Lluch-Belda et al. 1989, 1992a, 1992b, Schwartzlose et al. 1999). Thus it is not surprising that most of the examples of Ecopath models developed for the same system in more than one period are from upwelling ecosystems, some of which have been selected for discussion in this section. For example, Ecopath models of the southern Benguela ecosystem (Fig. 8.9) have been constructed for the period 1980–1989, when anchovy was the dominant small pelagic fish species and for 1990–1997, when sardine began to increase (Shannon et al. 2003), for the period 2000–2004, when anchovy and sardine were both present at high levels of abundance (Shannon et al. in press, Shannon in prep.), and for three earlier periods selected to mark the undisturbed/pristine, pre-industrial (1652–1910), industrial (1960s) periods (Watermeyer et al. 2008). Comparing steady-state model outputs of the 1980s and 1990–1997 models, biomass per trophic level, transfer efficiencies, mixed trophic impacts and whole system properties suggested that the system functioned in a similar way trophically in these two contrasted periods (Shannon et al. 2003). However, there is some suggestion that the ecosystem was more mature in the 1990s than in the 1980s according to ecosystem-level attributes extending E.P. Odum’s ecosystem development theory (Odum 1969), such as the smaller total primary production/total respiration, net system production, total primary production/total biomass, residence time and relative ascendancy in the 1990s model, and the larger flows to/from detritus, connectance and total respiration/total biomass. In the 1990s, smaller catches were made whereas model zooplankton and small pelagic fish biomasses were larger, leading to the conclusion that the southern Benguela ecosystem was less tightly constrained by predators, fishing and food availability in the 1990s than the 1980s (Shannon 2001, Shannon et al. 2003). Pending completion of the three earlier models, the effects of man’s intervention on the Benguela ecosystem will be quantified (Watermeyer et al. 2008). Industrial fisheries sequentially exploited and depleted sardine (1960s–1970s), Cape hake (1970s–1980s) and horse mackerel (1980s–1990s) in the northern Benguela ecosystem off Namibia. Heymans et al. (2004) presented Ecopath models of these three periods, highlighting major changes in fishing, the food web structure and possibly also the trophic functioning of the northern Benguela ecosystem (Fig. 8.9). In particular, there has been a large increase in jellyfish off Namibia since the 1970s (Venter 1988, Fearon et al. 1992), and top predators switched from preying predominantly on sardine and anchovy to pelagic goby (Sufflogobius bibarbatus), following the likely dramatic increase in goby biomass since the late 1960s (Crawford et al. 1985) and the reduction in abundance of anchovy and sardine. By the 1990s, energy was flowing through few pathways and trophic efficiency was lower than in the 1980s. The trophic level of the catch increased in the 1980s, when hake catches were large, and

256

M. Coll et al.

Fig. 8.9 Diagramatic summary of the changes in ecosystem structure and relative abundance of dominant species (small pelagic fish, hake, goby, jellyfish, Cape gannet, Cape fur seal; number of individuals represents approximate abundance) in the northern and southern Benguela between the 1970s (left) and the early 2000s (right). Reprinted from van der Lingen et al. 2006, Benguela: Predicting a Large Marine Ecosystem, with permission from Elsevier

decreased again in the early 1990s, before environmental forcing in the form of the Benguela Nin˜o caused huge declines in the pelagic stocks. Two Ecopath models were also constructed for the Humboldt ecosystem off central Chile for 1992 and 1998 (Neira et al. 2004). These models showed the importance of predation mortality on fish production in the system, especially

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

257

of juvenile fish, and quantified the strong impact of fishing mortality exerted on Chilean hake, common sardine, horse mackerel and anchovy. Although biomass was 1.5 times larger in 1998 than in 1992, catches were 20% smaller (Neira et al. 2004), suggesting that the Chilean ecosystem was less tightly constrained later in the decade. However, the distribution of biomass and throughput across trophic levels was similar in the 2 years, suggesting that the ecosystem structure had not changed dramatically (Neira et al. 2004). Comparative studies have been conducted for other shelf and coastal ecosystems which undergo temporal changes in ecosystem structure. In the North West Atlantic, several Atlantic cod populations collapsed in the early 1990s. Bundy (2005) constructed two models of the eastern Scotian shelf, one before and one after the collapse of the Atlantic cod stock to explore changes in the structure and functioning of the ecosystem and to explore theories for the nonrecovery of the Atlantic cod stock. Despite similar total biomass and productivity before and after the cod collapse, there were marked differences in the trophic structure and energy flows through the system. Piscivory increased as a result of the increase in small pelagic fish abundance, and the system switched from a demersally-dominated system to a pelagically-dominated system (the pelagic:demersal ratio increased from 0.3 to 3.0.) indicating a shift in trophic flow from the demersal to the pelagic part of the food web. The pelagic:demersal ratio is an indicator of the negative effects of fishing (Zwanenburg 2000, Rochet and Trenkel 2003). The rationale is that as longer-lived large demersal predators are removed by fishing, the abundance of small, short-lived pelagics increase, due to a release from predation pressure. An analysis of the trophic interactions of Atlantic cod within the eastern Scotian Shelf ecosystem suggested that the lack of recovery of cod after their collapse in the early 1990s could be explained by trophic factors (Fig. 8.10), at least for small cod (Bundy and Fanning 2005). Their low biomass makes them vulnerable to both predation, and to increased competition for prey. Small cod compete for their prey with highly abundant forage fish competitors, and this likely leads to food limitation. Pranovi et al. (2003) constructed two models comparing the Venice lagoon (Adriatic Sea, Mediterranean) in 1998 and during 1988–1991, a period prior to the spread of the Manila clam and its intensive exploitation in the area by mechanical dredges that pose high impacts to the system. Manila clam (Tapes philippinarum) was introduced in the Venice lagoon in 1983 and their exploitation became the first exploitation activity due to high economic value of manila clam. However, until the end of 1980s, biological resources in the area were exploited only by the artisanal fishery. These models enabled the detailed study of the complex effects of clam harvesting in the area and the ‘‘Tapes paradox’’ (Libralato et al. 2002): the Manila clam population was apparently enhanced by dredging due to the nutritional advantages that this species was gaining from the re-suspended organic matter. The model also demonstrated the indirect negative impacts, mediated through the food web, of the clam fishery on the artisanal fishery.

258

M. Coll et al. 0.4 (a)

0.3

0.2

0.1

00.0 Seals L.Cod Hadd Skates Squid Dogfish Pollock L.Hal SDFs SDPisc LDPisc Cetea S.Cod SLDF L.Shak Birds LDF Predator 0.04 (b)

0.03

0.02

0.01

00.00 Seals

Dogfish

LDPisc

Predator

Fig. 8.10 Comparison of predation mortality on (a) small cod and (b) large cod on the eastern Scotian Shelf, Canada in 1980–1985 and 1995–2000. Horizontal lines represent the median, the boxes represent 50% of the values, and the whiskers extend to the highest and lowest levels (excluding outliers). L.cod=large cod (>40 cm), Hadd=haddock, L.Hal=large halibut, SDFs=small demersal feeders, SDPisc=small demersal piscivores, LDPisc=large demersal piscivores, Cetea=ceteacea, S.cod=small cod (40 cm), SLDF=small large demersal feeders, L.Shak=large silver hake (>30 cm), LDF=large demersal feeders. Adapted from Bundy (2005)

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

259

On the west coast of Italy, two models were constructed to describe the shallow water coastal system of Orbetello lagoon (Central Western Italy) for 1995 and 1996 (Brando et al. 2004) and to analyze management activities developed in the area to control eutrophication. This lagoon is characterized by limited exchange with the sea and a high availability of nutrients, showing increasing eutrophication from 1975 to 1993. From 1993 a series of management activities were carried out to reduce eutrophication by reducing nutrient loading, increasing water circulation and selective harvesting of macroalgae. Network analysis differentiated between the models for 1995 and 1996, revealed the first effects of algal harvesting and indicated that the ecosystem was more mature and stable in 1996. 8.3.1.4 Comparative Studies: Examining Different Ecosystems Comparative modelling across ecosystems using standardized models provides a context for the interpretation of a single ecosystem’s attributes and a relative measure of the intensity and impacts that fishing may be having in an ecosystem. Furthermore, it enables the comparison of emergent properties and the evaluation of potential generic ecosystem properties. EwE is a particularly useful tool for comparative studies as it enables a standardized approach (e.g. by using the same number of functional groups in each model with similar species in ecological terms), allowing the separation of biological features from modelling artefacts, and alleviating problems and biases that may result from the way in which groups have been aggregated in a model, or ecosystem attributes that relate to discrepancies in life history parameter estimates. For example, Moloney et al. (2005) compared four different upwelling ecosystems representing different areas and periods: models represented the Southern Humboldt (Chilean) upwelling ecosystem in 1992, the northerncentral Humboldt (Peruvian) upwelling ecosystem in 1973–1981, the southern Benguela (South African) ecosystem in 1980–1989 and the northern Benguela (Namibian) ecosystem in 1995–2000. After the standardization process, the four models shared similar structures based on 27 groups but differed in terms of representation of some species. A comparison of indicators revealed differences between the Humboldt and Benguela systems, while indicators based on integrated biomass, total production and total consumption were able to differentiate the Namibian model (where some exploited resources have been severely depleted) from the others. Coll et al. (2006b) standardized an ecological model representing a NW Mediterranean exploited ecosystem and compared it with the four standardized models from coastal upwelling ecosystems describe above (Moloney et al. 2005) (Fig. 8.11). A comparison of biomasses, flows and trophic levels indicated important expected differences between the ecosystems, mainly caused by differences in primary production, which was lowest in the Mediterranean model. In addition, fishing pressure was high relative to the low primary production in the Mediterranean ecosystem. Comparisons of %PPR (the proportion of primary

260

M. Coll et al.

a)

b) 4.0

0.8

TLcatch TLcommunity

Peru Chile South Africa Namibia NW Med

Catch/Biomass ratio

3.6

TL

3.2

2.8

0.6

0.4

0.2

2.4

0

2.0 Chile

Peru

S.Africa Namibia NW Med

2.0

2.5

3.0

3.5 TL

4.0

4.5

5.0

Fig. 8.11 Results on comparing studies for examining different ecosystems using standardized models representing NW Mediterranean exploited ecosystem in 1994 and the upwelling ecosystems of the Southern Humboldt (Chilean) in 1992, the northern-central Humboldt (Peruvian) in 1973–1981, the southern Benguela (South African) in 1980–1989 and the northern Benguela (Namibian) in 1995–2000. (a) Trophic level of the community (TLco) excluding TL = 1 and total catches TLc; (b) Trophic spectra of catch:biomass rations. Adapted from Coll et al. (2006b)

production required to support the fishery, see Section 8.4.3), the trophic level of the community (TLco), the biomass of consumers and exploitation ratios (F/Z) captured the ecosystem effects of fishing; stronger in the NW Mediterranean, Namibian and Peruvian models, weaker in the Southern Humboldt. The importance of pelagic-demersal coupling and of gelatinous zooplankton in the consumption of production in the Namibian and Mediterranean case studies was in remarkable contrast to the other ecosystems. These identified similarities were related to ecosystem effects of fishing. Libralato et al. (2002) also compared standardized models from two different habitats of the Venice lagoon: the seagrass meadows (Fig. 8.2) and Manila clam fishing grounds. The seagrass meadows showed higher primary production, species diversity and complexity and appeared to represent an ecosystem at a higher stage of ecological succession. The Manila clam exploited ecosystem was dominated by consumption and respiration flows and most of the energy were stored within the detritus compartment.

8.3.2 ECOSIM The dynamic simulation modelling tool Ecosim has broadened immensely the capabilities of Ecopath for exploring the temporal impacts of fishing and environmental factors. Ecosim allows users to change fishing mortality or fishing effort over time, enabling the exploration of fishing options and changes in

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

261

ecosystem functioning. In this section, some examples of Ecosim implementation are reviewed (Table 8.1). 8.3.2.1 Exploring the Impacts of Fishing and Management Simulations In the early 1960s, fishing in the Gulf of Thailand was mainly artisanal and restricted to coastal areas. Subsequently an intense trawl fishery was developed in the area and changes in the community were well documented from the early 1960s to the 1980s. Christensen (1998) used two mass-balance models of the area representing the ecosystem in 1963 (early trawling) and the 1980s (highly exploited) to reproduce changes over time, developing temporal simulations and increasing fishing activity in accordance with field data. Observed depletion of several demersal species was well reproduced from 1963 to the 1980s. The biomass of cephalopods, however, increased with increased fishing, as did shrimps and scads. The study also explored the possible effects of reducing fishing pressure from the 1980s to analyse whether ecosystem changes would be similar to those characterizing the 1963 scenario. Simulations predicted an increase in biomass to past levels within a few years if fishing effort were to have been reduced. However, this result should be interpreted with caution since other studies have shown that reducing fishing effort can lead to an increase in modelled biomass that is not observed in the field (e.g. Bundy 2001), and the reality of the non-reversibility of fishing impacts should be recognized (e.g. Scheffer et al. 2001, Bundy and Fanning 2005). Mackinson et al. (1997) used a dynamic mass-balance model to compare the impacts of different exploitation options on small pelagic fish within the three different upwelling ecosystems of Peru, Venezuela and Monterey Bay. Exploratory simulations showed that by intensively fishing small pelagic fish, positive effects were seen in both prey and competitors in all ecosystems, while predators located at higher trophic levels had the longest recovery times. Moreover, fishing mortalities corresponding to the maximum sustainable catches predicted by the models were higher than those obtained from single species models. Ortiz and Wolff (2002a) used Ecosim to analyze different management scenarios from five Ecopath models representing different benthic communities of Tongoy Bay (Chile) in order to explore strategies of sustainable use. Different simulations were explored by modifying fishing mortality and energy control (bottom–up, mixed control and top–down). Management options explored the exploitation of scallops Argopecten purpuratus, their principal predator (the sea star Meyenaster gelatinosus) and the snail Xantochorus cassidiformis. The results indicated that the estimated maximum sustainable catch was lower than when estimated from single species approaches. Furthermore, simulating the exploitation of the clam Mulinia sp. showed the strong impact of this group on other functional groups, indicating that this species was a keystone element of the ecosystem. The internal stability of the system was measured in each simulation by means of the System Recovery Time.

262

M. Coll et al.

The ecological role of snappers and the impact of their exploitation in two ecosystems in the Gulf of Mexico (Western Gulf of Mexico and the Continental Shelf of Yucatan) was evaluated using Ecosim (Arreguı´ n-Sa´nchez and Manickchand-Heileman 1998). The impacts of fishing on snappers were analyzed in terms of individual functional groups and in terms of the stability of the ecosystem using measures of persistence, recovery time, magnitude of change and resilience of groups. Snappers occupied a top predator role in both systems and when exploited had low values of persistence in the ecosystem. However, they exhibited different dynamics under different fishing scenarios and the authors suggested that the snapper stocks in the two areas should be managed independently due to these differences. Coll et al. (2006a) used Ecosim simulations to support the theory that the Mediterranean system has a low resistance and is thus highly susceptible to fishing and environmental impacts. Their results demonstrated that since 1994, an increase in fishing effort always resulted in decreased catches. Combining different scenarios of moderate increase of fishing effort, flow control and an environmental forcing affecting the availability of prey groups to small and medium-sized pelagic fish under wasp–waist control, the decline in observed catch and biomass was consistently reproduced by the model. Ecosim has also been used to put selectivity measures applied to trawl fisheries into the ecosystem context. Criales-Hernandez et al. (2006) used an Ecosim model of northern Colombian system (Caribbean Sea) and data from the Gulf of Mexico to explore the reduction of fishing mortality due to bycatch reduction devices applied to a tropical industrial shrimp fishery. Coll et al. 2008a used field data from bottom-trawling selectivity by applying sorting grids and square meshes and simulated the consequent reduction of fishing mortality on target species in a North-Western Mediterranean system. Kitchell et al. (2002) assessed the ecological and economic impacts of alternative fishing methods to reduce the by-catch of marlin in the central north pacific ecosystem. In the three cases, the potential benefits of the implementation of selectivity management options within an ecosystem are highlighted. 8.3.2.2 Examining Energy Flow Controls Early use of Ecosim underscored the importance of assumptions about energy flow control on the dynamics of the ecosystem (Bundy 1997, Walters et al. 1997). Shannon et al. (2000) explored the ecosystem dynamics of the Southern Benguela system in the 1980s using EwE under different flow control scenarios: bottom–up control of predators by their zooplankton prey; wasp–waist flow control of small pelagic fish (both top–down control of zooplankton and bottom–up control of predators by small pelagic fish) and mixed control (neither bottom–up nor top–down control), Fig. 8.4. Results obtained from simulations were very different under different flow controls and highlighted the importance of considering trophic flow control while assessing the effects of fishing. Under bottom–up flow control, effects of fishing were smaller than

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

263

under a wasp–waist scenario, where they vigorously propagated thought the ecosystem. Mixed control scenario showed intermediate results. EwE was then used to simulate changes in the Southern Benguela ecosystem from an anchovy-dominated system to a sardine-dominated one (Shannon et al. 2004a). Two hypotheses of mechanisms that may have caused the observed ecosystem changes were tested: fishing and environmentally-induced changes in the structure of the zooplankton community. Scenarios of altered fishing mortality on sardine, anchovy and horse mackerel were examined using Ecosim, and ‘‘forcing functions’’ (see Section 8.2.2) were applied to mesozooplankton were used to test the effects of altered prey availability to anchovy and sardine, which by virtue of their different feeding behaviours are better suited to different sized zooplankton prey (Van der Lingen 1999). Simulations suggest that it is unlikely that observed changes in pelagic fish catches between the two decades examined played a large role in driving the changes in abundance of anchovy and sardine. Rather, shifts between anchovy- and sardine-dominated periods may have been caused by environmentally-mediated changes in the availability of mesozooplankton prey to anchovy and sardine. Ecosim was used to explore whether ecosystem changes in the NewfoundlandLabrador ecosystem could be explained by considering changes in fishing and predation mortality under different trophic controls (Bundy 2001). Atlantic cod and other ground fish species collapsed or seriously decreased in the early 1990s causing enormous economic, social and ecological impacts. This study replicated the collapse (and posterior non recovery) of cod under top–down flow control situations and increasing fishing mortality. Predicted results from the model also suggested an increase in seal populations and shrimps. Simulations showed that an increase in the seal population would have negative effects on the recovery of cod under top–down and mixed flow control. The study supported the hypothesis that the Newfoundland cod collapse was due to overfishing and that, under the depleted cod stock situation, seal population increases could retard its recovery. Gasalla and Rossi-Wongtschowski (2004) explored the effects of changing fishing strategies in the South Brazil Bight coastal area in terms of increasing catches of squid and live-baitfish for sardines. They found that ecosystem effects of altered squid fishing were more pronounced under the assumption of top–down control, which led them to propose that a precautionary fisheries management measure should refer to simulations assuming top–down control. Fishing on live-baitfish for sardines had little impact in the area with the exception of impacts on sharks and rays. Mackinson et al. (2003) explored different functional response hypotheses using an Ecosim model of minke whales, their prey and fisheries in the Barents Sea. Results showed clear patterns in the response of the ecosystems irrespective of the functional response used. For example, impacts of intense fishing on whale prey had longer lasting impacts on whale biomass than direct exploitation of whales. However results also showed that simulations with different vulnerability settings led to different feeding and biomass dynamics of minke whales.

264

M. Coll et al.

Uncertainty concerning the nature of flow control in ecosystems, and the relative ecosystem impacts of fishing need not result in management inaction. Bundy (2004a) used an adaptive management approach to explore management options under different flow assumptions for San Miguel Bay, the Philippines and concluded that, with respect to the management options under consideration, there was no value in learning more about the uncertainty or distinguishing between the different resource models. She reached this conclusion by exploring the effect of five management options on four models of the resource, using six performance criteria to evaluate their impact. For each of the performance criterion, there was a robust policy for all models. Furthermore, the results demonstrated that top–down control assumptions led to more precautionary management. 8.3.2.3 Analyzing Environmental Forcing of Ecosystem Dynamics The effects of fishing and eutrophication on the Black Sea ecosystem were investigated using Ecosim and a biological time series of data to explore trends and correlations (Daskalov 2002). Thirty year simulations were run, where fishing mortality was changed over time and a forcing function was included to consider the effects of increased primary production. Results showed the occurrence of a trophic cascade where a decrease in predators led to an increase in forage fish, a decrease in zooplankton and an increase in phytoplankton. The trophic cascade was related to overfishing together with eutrophication. Watters et al. (2003) studied how the El Nin˜o – Southern Oscillation (ENSO) might affect different organisms at mid- and high trophic levels. A mass-balance model of the pelagic system of the eastern tropical Pacific Ocean was dynamically explored by forcing it with two environmental functions affecting phytoplankton biomass and predator recruitment. Results showed that environmental effects applied to recruitment of predators may be the dominant source of interannual variability in the ecosystem and that top–down flow control posed by fishing may be dampened by these effects (example in Fig. 8.12). 8.3.2.4 Fitting Models to Data Using Ecosim, a mass-balance model of the Southern Benguela ecosystem was fitted to available time series data for a 25-year period from 1978–2002, exploring how fish stock dynamics may be determined by feeding interaction patterns (flow controls), fishing strategies and environmental changes (Shannon et al. 2004b). In this model, it was estimated that fishing patterns explained about 5% of the variability in the times series, an estimated productivity forcing pattern applied to phytoplankton explained 11% of the variability and assumptions about the vulnerability of prey to predators (flow control patterns) explained around 33% of the variability. When flow control was assumed to be wasp– waist around small pelagic fish (exerting top–down control on their prey and

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach (a)

(b)

2.5

0.95

265

2.0 1.5

0.85

1.0

0.75

(d)

Relative biomass (Bt / B0)

(c)

1.00

4 3

0.90

2 1

0.80

(e)

(f)

1.6

0.95

1.2

0.85

0.8

0.75

(g)

(h)

1.05

1.0

1.00

0.8

0.95

0.6 0.4

0.90 0

20

40

60

80

100

0

20

40

60

80

100

Years since start

Fig. 8.12 Simulated indirect effects of global warming (under modeled bottom–up forcing of large phytoplankton biomass) on middle and upper trophic levels: components in the axis of (a) large marlins, (b) spotted dolphins, (c) large sharks, (d) small dorado, (e) large yellowfin, (f) Auxis, (g) flying fishes, and (h) large phytoplankton). The simulations were forced with winter mean sea-surface temperature anomalies predicted by the Max Planck global climate model. Solid lines are from simulations with F = average F during 1993–1997, and broken lines are from simulations with F = 0. A horizontal broken line is drawn for reference at Bt/B0 = 1.0. Reprinted from Watters et al. 2003. Physical forcing and the dynamics of the pelagic ecosystem in the eastern tropical Pacific: simulations with ENSO-scale and global-warming climate drivers. Watters et al. (2003), with permission to reproduce

bottom–up control on their predators), the model was best fit to the data series examined. The aim of studies such as these is to prepare models with improved parameterization and enhanced credibility, which may be useful in testing alternative fishing scenarios with a view to providing fisheries management advice in an ecosystem context. Figure 8.6 is an example of a Southern Benguela ecosystem model calibrated with time series data (Shannon et al. 2004b), and currently in the process of being updated (Shannon et al. in prep.). Heymans (2004), studying the Northern Benguela upwelling system, fitted a mass-balance model of the ecosystem to time series data from the 1970s to the present under different scenarios of trophic flow control. Sixty-five percentage of total variability of data was explain by considering internal (e.g. flow controls) and external (fishing and environmental factors) factors within the simulations. Environmental factors were included by using the model to calculate an

266

M. Coll et al.

environmental anomaly to increase the goodness of fit of the model. This environmental anomaly was then correlated to environmental variables and significant correlations were found with sea surface temperature and wind under wasp–waist flow control by small pelagic fish. This model is currently being updated with expanded data series (Heymans in prep.). Coll et al. 2008b presented results on fitting a model to available data from the temperate shelf ecosystem of the South Catalan Sea (North-Western Mediterranean) from 1978 to 2003. This work highlighted that flow control patterns explained up to 37–53% of the total variability of time series, while fishing factors and the environmental variability explained 14% and 6–16% of time series dynamics, respectively. This also showed that small pelagic fish in the area were involved in wasp–waist and bottom–up flow control situations and is also currently in the process of being updated. Neira et al. (in prep.) fitted a model of the Southern Humboldt upwelling system in 1970 to available time series of relative biomass, catch and fishing mortality for the period 1970–2004, finding that fishing mortality explained 28% of the variability in the times series, vulnerability (v) parameters explained an additional 21%, and a forcing function affecting primary production explained a further 11–16% of the observed variability. The model fitted primary productivity anomaly compared favourably to an independent time series of sea surface temperature and an upwelling index available from 1970–2000. Walters et al. (2005) reported the results of fitting eleven mass-balance models to time series of data and demonstrated the suitability of this methodology to reproduce past dynamics of exploited ecosystems. Models were fitted by using available data and were analyzed and discussed in terms of methodological procedures (e.g. trophic interactions, mediation effects). Moreover, ecosystem effects of harvesting species to their single species maximum sustainable catch was evaluated and shown to cause notable deterioration in ecosystem structure (e.g. loss of biomass from top predators). Mackinson et al. (2008) compared a series of fitted models to explore the contribution of fishing and environmental forcing as drivers of marine resources. 8.3.2.5 Policy Optimization and Management Scenarios At an exploratory workshop held in Vancouver (2001), a set of studies explored the impacts of different fishing strategies to meet broadly defined ecological, social (employment) and economic objectives in several ecosystems (Pitcher and Cochrane 2002). In most cases, fishing strategies that optimized economic or employment objectives were relatively easily understood, whereas the ‘‘optimal’’ fishing strategies were usually ecologically unrealistic, forcing the model ecosystems to unrealistic extremes. For example, Shannon (2002) found that extreme ‘‘optimal’’ fishing scenarios forced the model parameters (biomass, diet composition, consumption and production rates) beyond their likely ranges, producing highly unrealistic outcomes in the Southern Benguela ecosystem. Nonetheless, simulations such as these are still useful measures of the economic

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

267

and social value of careful fisheries management, and highlight the importance of interactions between species (i.e. extending beyond the traditional singlespecies approaches) in determining the ecological, social and economic impacts of various fishing policies under consideration. The workshop and subsequent studies using the EwE fishing policy optimization routine have certainly served to emphasize, in the realm of Ecosystem Approaches to Fisheries, the necessity of carefully defining policy objectives within an ecosystem context. Okey and Wright (2004) applied the optimization routine to analyse management options for Prince William Sound (Alaska) within the context of rebuilding pinniped populations. Again, maximizing economic and social criteria resulted in scenarios where predators were reduced to maximize prey production profitable for fisheries. When ecological criteria were emphasized, predators and their prey increased. Competition between fisheries and predators was evident since predators increased with decreases in fishing. This study also suggested that a 20% increase in pinniped biomass could be achieved with a modest reduction of fishing activity (Fig. 8.13). The policy optimization routine was also applied to the Gulf of Thailand and results highlighted that optimizing for profit led to an ecosystem were the emphasis was focused on maintaining productive stocks of profitable species while decreasing competitors and predators in the ecosystem (Christensen and Walters 2004b). Optimizing exploitation for economic profit produced an intense increase in effort of specialized fleets and a catch dominated by trash fish and shrimps, impacting larger fish and diversity of fishing activity.

Fig. 8.13 Catch levels by three fishing categories at the beginning and end of the 20-yr simulations relative to the actual 1994–1996 levels in Prince Williams Sound ecosystem, Alaska. The beginning and ending levels were selected by the nonlinear search procedure as optimized solutions to the objectives specified for each run. In the ‘‘Straight compromise’’ case, the economic, employment, and ecological objectives were weighted equally. Reprinted from Okey and Wright (2004). Toward ecosystem-based extraction policies for Prince William Sound, Alaska: integrating conflicting objectives and rebuilding pinnipeds. Okey and Wright (2004), with permission to reproduce

268

M. Coll et al.

Optimizing for ecological considerations implied a notable reduction in effort and resulted in an increase of biomass of several functional groups. In global terms, the optimization for profit was compatible with conservation measures, while optimizing for value was in conflict with both profit and ecology. In a case study of La Paz Bay, South Baja California (Mexico) by Arreguı´ nSa´nchez et al. (2004), economic, social and ecological criteria were examined for two artisanal fisheries and a shrimp fishery. Results showed how the optimization of economic and social criteria, and aiming for the maximum sustainable yield (MSY), resulted in depletion of some stocks and in unrealistic increases in fishing effort. However, by combining economic and social criteria with ecological ones, model simulations were developed in which stock depletion was avoided.

8.3.2.6 Back to the Future Simulations The back to the future approach (Pitcher 2001, 2005) reconstructs past systems using modelling, together with other forms of knowledge such as traditional or local knowledge, historical documentation and archaeological data, to explore future policy goals (Fig. 8.14). The ultimate goal is to progress towards ecosystem rebuilding of a known state that can be measured in terms of economic, social and ecological utility. When past models have been constructed, dynamic modelling is used to predict future situations of the ecosystem implementing different rebuilding measures and results are compared with past ones. This methodology has been applied to different ecosystems such as the Strait of Georgia in British Columbia (Dalsgaard et al. 1998), Newfoundland (Pitcher et al. 2002a) and Northern British Columbia (Ainsworth et al. 2002).

8.3.2.7 Studies Considering Pollution After the Exxon Valdez oil Spill in Prince William Sound and adjacent areas in 1989, research into the biological components of the ecosystem increased and an Ecopath model was constructed (Okey and Pauly 1999). Model simulations were used to explore the consequences of fishing and other anthropogenic disturbances (e.g. oil spill), restoration and resource planning of the ecosystem. Increased mortalities due to the oil spill were integrated in the model and temporal simulations were performed to evaluate the effects. Results indicated that the oil spill severely disturbed the ecosystem and had important impacts on various functional groups, considerably reducing their biomass. In global terms, simulations also demonstrated that the impacts produced by the oil spill could produce a shift on the marine community to an alternate state (Okey 2004). Methyl mercury concentration in food webs and marine mammals was modelled for the Faroe Islands ecosystem by Booth and Zeller (2005) to explore the implications of cod and pilot whales for human consumption.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

269

Fig. 8.14 Diagram illustrating the ‘‘Back-to-the-Future’’ concept for the adoption of past ecosystems as future policy goals. Triangles at the left represent a time series of historical ecosystem models, constructed at appropriate past times before the present (thick grey vertical line), where the vertex angle is inversely related and the height directly related to biodiversity and internal connectance. Time lines of some representative species in the ecosystems are indicated; size of the boxes represents relative abundance and solid circles represent local extinctions (= extirpations). Sources of information for constructing and tuning the ecosystem models are illustrated by symbols for historical documents (paper sheet symbol), data archives (tall data table symbol), archaeological data (trowel), the traditional environmental knowledge of Indigenous Peoples (open balloons) and local environmental knowledge of coastal communities (solid balloons). At right are alternative future ecosystems, representing further depletion, the status quo, or restoration to ‘‘Lost Valleys’’ that may be used as alternative policy goals. Restored ‘‘Lost Valleys’’ may be fished with sustainable, responsible fisheries designed according to specified criteria, and aiming at Optimal Restorable Biomasses determined using objective quantitative policy searches. Final choice of BTF policy goals are made by comparing trade-offs, cost and benefits among possible futures using socio-economic and ecological objectives agreed among industrial and small-scale fishers, government, conservation, coastal communities and other stakeholders in order to maximize compliance. Diagram does not show evaluation of risks from climate fluctuations and model parameter uncertainty. Modified from Pitcher 2007 and Pitcher and Ainsworth 2007

During this study, Ecotracer, an Ecosim routine that predicts the movement and accumulation of tracers within a food web (Christensen and Walters 2004a) was used. Mercury was predicted to increase in the ecosystem under present conditions and also during climate change scenarios. Results showed that the highest levels of mercury in human diet originated from whale meat consumption and that climate change would exacerbate this situation. The study also predicted that inflow rates of mercury to the ecosystem should be reduced by 50% to ensure secure levels of intake under current levels of consumption of marine resources.

270

M. Coll et al.

8.3.3 ECOSPACE Models which are spatially explicit are a high priority for advancement of fisheries science and management (e.g. Walters et al. 1999, Salomon et al. 2002, Martell et al. 2005), particularly in the light of the move from single species fisheries management to the ecosystem approach to fisheries. However, most EwE applications, including those fitting models to time series data, have not simultaneously accounted for spatial considerations such as the spatial behaviour of organisms, the spatial overlaps or co-existence of species (predator-prey; competing species, etc.), the spatial behaviour of fisheries, fish migration and spatial management approaches. Ecospace provides the methodology to pursue these questions, although currently there are relatively few primary publications using Ecospace. In a complete EwE study of Tongo Bay, Ortiz and Wolff (2002b) continued their analysis using Ecospace to explore spatial management options, distinguishing 4 habitat types, seagrass meadows (0–4 m), sand and gravel (4–10 m), sand flats (10–14 m) and mudflats (>14 m). Five different scenarios were explored where fishing on the main species, a red alga, a scallop, a gastropod and a crab, occurred in either seagrass, sand-gravel or sand habitats exclusively, on both seagrass and sand-gravel habitats, or on all habitats. On the basis of biomass changes, the authors concluded that the sand-gravel habitat was the most resistant to fishing, and that fishing 2–3 habitats simultaneously had the greatest negative effect on the ecosystem. They further concluded that a rotational harvest policy should be recommended for this Bay. Most Ecospace applications have focused on Marine Protected areas, and indeed, Walters et al. (1999) identified Ecospace as an important exploratory tool for MPA definition and function and policy exploration. Using a simple model of Brunei Darussalam, they concluded that biomass gradients would exist along the edges of MPAs, resulting from fishing activity along the MPA boundary on the increased density of predators within the MPA, causing a decrease in the density of predators within the MPA (fishing would also lower immigration to and emigration from the MPA). At the same time, trophic cascades are likely to occur within the reserve due to the increase in density of large predators, resulting in a decrease of small fish. Density dependent effects would lead to the emigration of large predators, resulting in reduced predator biomass within the MPA. They further conclude that a few large MPAs are more effective than more small MPAs. This is largely due to the fact that a few large MPAs have a smaller boundary (where fishing takes place) than many small MPAs (for the same total area). Thus the more boundaries there are the more dispersion can take place, and the greater the effects of fishing on the boundaries. These results have been confirmed by other Ecospace studies (e.g. Pitcher et al. 2002b, Salomon et al. 2002, Martell et al. 2005) (Fig. 8.15). Indeed Salomon et al. (2002), who explored various options for marine protected area zoning policies in the Gwaii Haana National Marine Conservation Area, Canada, concluded

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

271

% Change in biomass from baseline

8 7

Single Large MPA Several Small MPAs

6 5 4 3 2 1 0 Avian Preds

Herring

Lingcod

Rockfish

Shallow infauna

Fig. 8.15 The Single Large or Several Small (SLOSS) debate. The percent change in ecosystem component biomass from a baseline simulation of no protection as a consequence of splitting one large Marine Protected Area (MPA) into three small MPAs. Simulations were compared once they had run for 10 years. Reprinted from Salomon et al. 2004. Modelling the trophic effects of marine protected area zoning policies: a case study, Salomon et al. (2002), with permission from Springer Science and Business Media

that buffer zones were effective tools against edge effects and biomass density gradients but that there is no optimum reserve design since the design depends on the goal. Okey et al. (2004) used an Ecospace model to demonstrate how the functional extinction of sea cucumbers in a Galapagos subtidal rocky reef could be avoided by protecting some of the reef from fishing (Fig. 8.7). Martell et al. (2005) explored the relative importance of different assumptions about dispersal and advection for Ecospace predictions under different fishing policy scenarios for the central North Pacific with respect to marine protected areas and concluded that MPAs for large pelagics need to be large. They split the central North Pacific into two habitats, warm water and cool water and developed 3 movement models: the default advection-diffusion movement model where dispersal rates were random and non-directional and two models where movement responded to a fitness measure, where fitness was the difference between productivity and predation. Movement increased in areas of low fitness and decreased in areas of high fitness. In the variable emigration model, any movement in response to fitness was random; in the directed movement model, movement was directed towards cells with higher fitness. The authors imported monthly current information from sea surface topography to calculate advection fields. Primary production in each grid cell was both dynamic and assumed to be proportional to the rate of upwelling (dynamic models) or static with long term average primary productivities. The three scenarios represented different degrees of closed areas and protected species policy, including the status quo. The authors concluded the static versus dynamic models results were sufficiently different, regardless of the movement model, to indicate that the temporal differences in surface currents should be explicitly considered when exploring and developing closed area policies

272

M. Coll et al. No surface currents

Surface currents

Surface currents shift

MPA Boundary

Relative primary production

MPA Boundary

U

D

U

D

Adult & juvenile predator (F= 0) Adault & juvenile predator (F > 0)

Adevtion-diffusion movement

Relative forage density

D

MPA Boundary

MPA Boundary

Juvenile

MPA Boundary

Adult

Fishing effort

Fishing effort

MPA Boundary

Fishing effort

MPA Boundary

Adault & juvenile predator (F> 0)

Adult & juvenile predator (F= 0)

MPA Boundary

Fitness driven dispersal

MPA Boundary

Fishing effort

Fishing effort

Fishing effort

Fig. 8.16 A representation of the main Ecospace findings for closed area policies, with alternative movement models representing different hypotheses about animal behaviour and static forced distributions of production (first column) versus dynamic forced distributions of production, where monthly surface current information us used to predict spatial variability in primary production (second and third columns). The second row represents the spatial distribution of forage, were surface currents advect forage in the second and third columns. Each diagram represents a cross section or transect of the spatial distribution of abundance (and fishing effort) across an Ecospace map. Fishing effort distribution for top predators is represented by as the dotted line. Shaded polygons represent distributions of biomass along the transect at equilibrium; the area of each polygon is proportional to biomass. Vertical dashed lines represent marine protected area boundaries; arrows represent current directions; U, upwelling; D, downwelling or convergence zones. Reprinted from Martell et al. 2005. Interactions of productivity, predation risk, and fishing effort in the efficacy of marine protected areas for the central Pacific. Canadian Journal of Fisheries and Aquatic Sciences 60: 1320–1336, with permission from NRC Research Press

(Fig. 8.16). In particular, they were concerned that interannual variability in oceanic processes can affect the efficacy of MPAs since the position of convergence zones, where species aggregate, including tuna, can change from year to year, potentially lying beyond the MPA boundary. In this case, protected species would be subject to high fishing mortality. In general, the results were robust to the three movement models, suggesting that further research into the movement of large pelagics is not required.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

273

Like Ecosim, Ecospace can also be used for policy exploration, such as tradeoffs between compliance with fishery regulations and conservation in the Hong Kong artificial reef system (Pitcher et al. 2002b). In the latter, a scenario was explored in where fishing was permitted in one artificial reef, assuming that this would lead to greater support for the artificial reef scheme and selfenforcement.

8.4 What Ecopath with Ecosim Can Do for You (and What It Can’t) 8.4.1 Ecopath with Ecosim as a Diagnostic Tool Ecopath models have proved useful in identifying data gaps and sensitive interactions, thus guiding research (e.g. Halfon and Schito 1993, Bundy 2004b). In addition, they are useful in refining parameter estimates for poorly known groups according to the constraints of the interactions defined within the Ecopath models (Okey and Pauly 1999). By construction of an Ecopath model, the biological and ecological data available from an ecosystem is identified, analyzed, contextualized and evaluated. Ecopath models can also be used to test low quality data, for example the biomass of benthopelagic species or suprabenthos, which are difficult to estimate but represent an important proportion of the diet of many species within marine ecosystems. These models can calculate the minimum biomass necessary in the ecosystem to sustain total mortality of these groups if predation and fishing mortality are well characterized (e.g. Lam and Pauly 2005). Another example can be found in stomach contents data, where soft preys can be underestimated with respect to species with hard-body parts (like fish and crustaceans). Thus, Ecopath can be useful in the correction of trophic data given biomass of predators and prey, and estimates of consumption.

8.4.2 Dynamic Simulations as Management Tools for an Ecosystem Approach to Fisheries It is readily accepted that multispecies approaches to fisheries management should not and cannot fulfil the role of single-species fisheries management approaches, but that they should rather be considered as complementary approaches in a model ‘‘toolbox’’ from which management advice can be drawn (Starfield et al. 1988, Whipple et al. 2000). In the long-term, multispecies approaches can produce totally different management advice to the more traditional single species modelling approaches (e.g. Magnu´sson 1995, Stokes 1992), yet in the short term, advice may be similar (Christensen 1996). Cox et al.

274

M. Coll et al.

(2002) showed that EwE was better able to represent and explain species recoveries after severe fishing of apex predators than single species models. The lack of any formal process to reconcile (or check whether it is feasible to reconcile) the management strategies of single-species management approaches to attain different goals, has led to conflicting management advice being provided at the single-species level, and thus highlights the need for a multispecies or ecosystem approach (Murawski 1991). Dynamic Ecosim simulations can shed light on the possible ecosystem effects of different fishing strategies, although the model assumptions, such as flow control parameter settings, need to be carefully acknowledged and sensitivities of simulation results explored. The more recent development of the ability to fit EwE models to time series data, which includes refinement of vulnerability settings describing flow controls, has increased confidence in model predictions. This creates a more robust basis for testing hypothetical fishing scenarios, and leads to greater confidence in the information that these may provide for fisheries management. The fishing policy search and optimization routine provides an additional means of exploring the dynamic responses of the ecosystem to hypothetical fishing strategies which may optimize one or a combination of policy objectives, and provides managers with guidelines as to the likely trade-offs that are involved in prioritising one objective over another, or in trying to optimize several simultaneously.

8.4.3 Examining Emergent Properties Through Ecosystem Indicators The construction of EwE models facilitates the estimation of trophodynamic indicators and ecological analyses that can be useful tools for an ecosystem approach to fisheries. Trophodynamic indicators measure the strength of interactions between species or species groups within an ecosystem, and the structural and functional changes that occur in an ecosystem as a result of fishing (Cury et al. 2005). Of the 46 trophic indicators identified in the literature, Cury et al. (2005) selected six for closer examination using data and EwE models of the northern and southern Benguela ecosystems (Fig. 8.17), namely catch or biomass ratios, production or consumption ratios and predation mortality, primary production required to produce catch (PPR), trophic level of the catch (TLc), fishing in balance (FIB) index and mixed trophic impact (MTI), see above. The PPR expresses the catch in terms of equivalent flows of primary producers and detritus and can be normalized per unit of catch relative to the primary production and detritus of the ecosystem (%PPR). This measure is used as an indicator of the footprint of the fishery and can be employed as an indicator of fishing intensity (Pauly and Christensen 1995).

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

Catch (’000 tons)

NORTHERN BENGUELA Hake

Sardine

2 000

Other

Large horse mackerel

1 500

Other small pelagics

(a)

275

SOUTHERN BENGUELA 900

Hake

Sardine

Other

700

Anchovy

Horse mackerel

500

1 000

300

500

100

(b)

Demersal catch ratio

0.9

0.4

0.7

0.3

0.5

0.2

0.3

0.1

0.1

TL of catch

(c) 3.9 3.7 3.5 3.3 3.1 2.9 2.7

3.8 3.7 3.6 3.5 3.4 3.3

(d) 0.7 0.6 0.5 0.4 0.3 0.2 0.1

FiB index

2.5 2 1.5 1 0.5

TL of catch

1955 3.9 3.7 3.5 3.3 3.1 2.9 2.7

1965

1975

1985

1995

1993

1974 1968

1961

1950

4.5

5.5

6.5

Log (catch)

1965

1975

(e)

2000 1986

1991

1955

7.5

3.8 3.7 3.6 3.5 3.4 3.3

1985

1995

1991 1956

1996

1998 2000 1976

1950

1961 1959

5.5

6

6.5

Log (catch)

Fig. 8.17 Example of selected ecosystem indicators estimated using EwE models for the northern and southern Benguela ecosystems. Adapted from Cury et al. (2005) and used by permission of Elsevier

The Fishing in Balance (FIB) index can be easily calculated from catch data and TLc over time (Christensen 2000, Pauly et al. 2000) and measures whether a change in the trophic level of the catch in a given ecosystem is matched by concurrent changes in productivity (i.e. lower trophic level of the catch, higher productivity and FIB = 0). Overfishing is evident when the trophic level of the fishery decreases, but is not matched by increased productivity (FIB < 0).

276

M. Coll et al.

Alternatively, FIB can indicate whether there is an expansion of the fishery (FIB > 0), whether bottom–up effects are occurring (FIB > 0) or whether discarding is not considered in the analysis of impacts of fisheries on the ecosystem and is so high that its functioning is impaired (FIB < 0). The above trophodynamic indicators have been widely applied to different ecosystems. However, Cury et al. (2005) noted that these indicators were relatively conservative as they are slow to respond to large structural ecosystem changes. For example, the mean trophic level (TL) of the Namibian catch failed to reveal the sequential depletion of Namibian commercial fish stocks because the ecosystem had shifted productivity to non-exploited species that were not reflected in the catch data. On the other hand, the FIB index, viewed in conjunction with plots of TL of catches against catches, was shown to reflect the historical development and status of fisheries in South Africa and Namibia more successfully than other indicators that can be derived from catch statistics (Fig. 8.17). Ecopath mass-balance models can also be used to calculate standardized size spectra of the ecosystem, i.e. the distribution of biomass according to size of individuals (Pauly and Christensen 2004), which can be then compared between ecosystems. Size spectra is used to characterize the structure of a system and the fishing intensity over time due to the fact that the slope of size spectra plots reflects the exploitation level, being steeper when exploitation is high (Bianchi et al. 2000). Results from Ecopath and Ecosim can also be used to track functional changes of the ecosystems. Examples of emergent properties are the transfer efficiency, the flow to detritus and production ratios presented in Section 8.2.1.3. From MTI analysis, Libralato et al. (2006) developed and applied a method for identifying keystone species (or groups of species) in different ecosystems. Keystone species are those that are present at relatively low biomass levels but have a structuring role in the ecosystem (Power et al. 1996). Therefore they can be identified when the relative overall effect and the ‘‘keystoneness’’ are plotted against one another. The index is high when species or groups of species have both low biomass proportions within the ecosystem and high overall effects. Changes in keystone species can be analysed when different trophic models of an ecosystem are available, e.g. the importance of cetaceans as keystone groups was seen to decrease over time in various ecosystems (Libralato et al. 2006) (Fig. 8.18). Comparisons of ecosystem indicators from the same ecosystem in different periods of time or from different ecosystems when using standardized models can be very useful tools as discussed in the previous case studies section. However, trophodynamic indicators are mostly still descriptive at this stage and it remains for reference points to be clearly identified. Cury et al. (2005) advise that a suite of indicators be used to monitor and quantify ecosystem changes as a result of fishing. To define quantitative reference levels to analyze fishing impacts on ecosystems, a new composite index (integrating PPR, TLc and transfer efficiency TE was defined by Libralato et al. 2008): L index. This index represents the theoretical depletion in secondary production due to fishing and is formulated as a proxy for quantifying ecosystem effects of fishing. The

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach 0.5

0.5

Newfoundland 0.0 13 15 14 1918 17 20 2221 24 2523 26 27 28 29 30 31 32 33 34 36 35

–0.5

–1.0

37 38 39 –1.5

–2.0

40 41

–2.5 42 43 44

Keystoness

277

–3.0

1198 710 12

6

Estern Tropic Pacific

1 4 5

3

2

0.0

98 1110 12 15

13

65 4 7 14

21

3

16

16

18

–0.5

17

19 20 2221 23 24

–1.0

1 Cetaceans 2 Capelin 3 Harp seals 4 Hooded seals 5 L zooplankton 6 G halibut 7 Bivalves 8 Pollock 9 G halibut juv. 10 Phytoplankton 11 Small mesopelagics 12 S zooplankton 13 Piscivorous birds 14 Polychaetes 15 Large crabs

16 Echinoderms 17 Herring 18 Shrimp 19 Juv dem pisc 20 S plank dem feed 21 Small pelagics 22 Squid 23 Sandlance 24 Redfish 25 O benthic inver 26 Arctic cod 27 L plankt dem feed 28 Skates 29 Cod + 40cm 30 Aplaice<=35cm

25 26 27 28 –1.5

31 O s demersals 32 Transient mackerel 33 Dogfish 34 L dem pisc 35 Flounders 36 Aplaice +35cm 37 Small crabs 38 Cod <40 cm 39 Transient pelagics 40 Salmon 41 Lobster 42 Lumpfish 43 Ducks 44 Planktivorous birds

29 30

1 Lg sharks 2 Toothed whales 3 Sm sharks 4 Lg bigeye 5 Secondary consum 6 Primary consum 7 Cephalopods 8 Sm marlins 9 Lg yellowfin 10 Flying fish 11 Lg marlins 12 Producers

–2.0 31 32 33 34 –2.5 35 36 –3.0

0.5

13 Sm yellowfin 14 Misc epi fish 15 Misc meso fish 16 Auxis 17 Skipjack 18 Misc pisc 19 Crabs 20 Sm wahoo 21 Sm mahimahi 22 Lg wahoo 23 Sm bigeye 24 Spotted dolphin

25 Meso dolphin 26 Lg sailfish 27 Pursuit birds 28 Lg mahimahi 29 Bluefin 30 Sm sailfish 31 Sea turtles 32 Lg swordfish 33 Baleen whales 34 Rays 35 Sm swordfish 36 Grazing birds

0.5

1

Chesapeake Bay 0.0

7 11108 9

645 12

3 2

6 10

14 13 16 15 17 1918 20 21

–0.5

Bolinao reef 0.0

13 15 1412 17 1816

–0.5

22 23 24 25 26 –1.0

5

4 3

2 1 7

8 9 11

19 20 21 22

–1.0

27 28

23

29 –1.5

–1.5 24 25

–2.0 30

31

–2.5

32

–3.0 33

–3.5

–2.0 1 Zooplankton 2 Sea nettle 3 Ctenophore 4 Blue crab 5 Sed bacteria 6 Oysters 7 Bay anchovy 8 Microzooplank 9 Other polychae 10 Meiofauna 11 Spot

12 Phytoplankton 13 Macoma spp 14 Heteromicroflag 15 Free bacteria 16 Menhaden 17 Nereis 18 Summer flounder 19 Blue fish 20 Striped bass 21 Weak fish 22 Benthic prod

23 Catfish 24 Crus. dep. feed 25 Mya 26 White perch 27 Hog choker 28 Other sus. feed 29 Sus. bacteria 30 Croaker 31 Ale. blue herr. 32 Shad 33 Fish larvae

–2.5

–3.0

–3.5

1 Zooplankton 2 Crustaceans 3 Damselfishes 4 Wrasse 5 Cardinalfishes 6 Sea uchins 7 Seaweeds 8 Squid

9 Coral cons 10 Other inverteb 11 Seagrass 12 Other omniv f 13 Coral prod 14 Siganus fusc. 15 Parrotfish 16 Molluscs

17 Other pisciv f 18 Groupers (e.m.) 19 Sea cucumber 20 Siganus spinus 21 Gobies 22 Moray 23 Phytoplankton 24 Other planktiv f 25 Other herbiv f

Relative overall effect

Fig. 8.18 Keystoneness for the functional groups of four marine trophic webs. For each functional group, the keystoneness index (y axis) is reported against overall effect (x axis). Overall effects are relative to the maximum effect measured in each trophic web, thus for x axis the scale is always between 0 and 1. Within each trophic web the species are ordered by decreasing keystoneness, therefore the keystone functional groups are those ranking between the first groups. Reprinted from Libralato et al. 2006. A method for identifying keystone species in food web models Libralato et al. (2006), with permission from Elsevier

index was associated with the probability of a system being sustainable fished or overfished sensu Murawski (2000), providing a general basis for defining the level of disruption for ecosystems subjected to different fishing pressures. Catch and biomass ratios, as well as production, consumption and predation mortality as a result of EwE dynamic models can be used to analyse how species or species groups within the ecosystem have changed over time due to fishing, environmental and trophic interactions. They can be used, for example, to analyse predicted changes in mortalities and prey preference over time (Shannon et al. 2004b) or to analyse predicted proliferation of species biomass and catches (Bundy 2001). Dynamic simulations have also been shown to be useful tools to

278

M. Coll et al.

test emergent patters of exploited ecosystems such as trophic cascades and regime shifts as discussed within the Case study section (e.g. Daskalov 2002, Shannon et al. 2004a). In addition, dynamic EwE simulations are ideal for testing the impacts of fishery collapses or ecosystem changes using indicators. For example, Shannon and Cury (2003) quantified theoretical concepts of species interactions by means of indices of interaction strength, functional impact and trophic similarity. Exploring ecosystem effects through the use of such indices calculated from EwE simulation outputs facilitates the potential outcomes of hypothetical management options to be quantitatively assessed without confounding effects such as environmental perturbations necessarily being simultaneously considered. Thus, indices such as these can be a useful basis from which to set about monitoring fishing effects and ecosystem changes over time within real ecosystems. For example, the index of interaction strength measures the relative impact of a collapse of one stock on the biomass of all other groups in the model ecosystem, and is comparable across ecosystems provided that models being used have been standardized (see Section 8.3.1). As such, the index provides a measure of the relative importance of a species group within its ecosystem, and thus the likely consequences of altered fishing on that group.

8.4.4 Limitations, Caveats and Criticisms of EwE Pimm et al. (1991) considered data availability to be a major obstacle to progressing food web research, although they viewed the lack of standards in food web methodology (observing and reporting) to be an even larger problem. EwE rises to this occasion by providing a standardised modelling approach for food web analysis, which facilitates meaningful comparisons to be made between ecosystems (see Section 8.3.1). In many instances, EwE operates at a sufficiently low level of complexity to facilitate parameter estimation and simulation, although Magnusson (1999) warns that Ecosim is not statistically based and the empirical basis for estimation of some of the parameters is weak. Aydin and Friday (2001) supported this notion when they highlighted the need for a statistical framework to summarize the confidence levels surrounding EwE model results or predictions if they are to be used in a formal multispecies fisheries management context. Walters et al. (1997) list three major disadvantages of Ecosim compared to other more detailed (and thus often less applicable) multispecies simulation approaches: (i) switching and satiation in predators are not well represented. There is some debate as to the theory underlying the vulnerabilities of prey to their predators, which can be user-specified in Ecosim but may contain some inconsistencies. For example, Plaga´nyi and Butterworth (2004) note that

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

279

the part of a prey population vulnerable to one predator is distinct from that vulnerable to another, whereas the whole prey population is available to fishing. Further, despite a term describing prey handling time, EwE has been criticized for predator consumption rates that in effect increase linearly with increased prey abundance (Plaga´nyi and Butterworth 2004), (ii) complex but smooth changes in predation rates associated with rapid changes in size structure are not well represented by average interaction rates defined for individual groups in Ecosim, and (iii) Ecosim is dependent on the mass-balance (equilibrium) assumption of Ecopath. The latter infers the risk of encountering errors when attempting to interpret Ecosim results when these are extrapolated far beyond the equilibrium for which Ecopath data are available (Mackinson et al. 1997, Walters et al. 1997). Ecosim assumes that changes in the fisheries occur within a given ‘‘regime’’ or period, and that there are no additional regime changes during the simulation period. This means that it is often helpful to consider modelling an ecosystem by means of more than one ‘‘snapshot’’ Ecopath model, each describing a certain ecosystem state or characteristic period. Other shortfalls of EwE relate to the fixed link of Ecosim to Ecopath, which constrains the choices regarding growth of primary producers and the unexpected functional responses of predators to their prey (diet selection, competition among predators, the influence of regime shifts, disease- and parasite-transmission, parasitism and mutualism), reviewed in Cury et al. (2005). On the positive side, a major advantage that EwE models hold over MSVPA models, in particular, is that trophic interactions are reasonably well represented for all, including the lower trophic levels. By contrast, the latter are often lumped into a ‘‘catch-all’’ category such as ‘‘other food’’ in MSVPA models (Walters et al. 1997). MSVPA-type models are data-intensive, requiring extensive and time-consuming parameterization, associated with high uncertainties around MSVPA model simulations (Whipple et al. 2000). EwE, on the other hand, is less data-intensive and requires data that are more often available, and therefore EwE has been widely applied across well over a hundred diverse ecosystems (Whipple et al. 2000). As is necessary when using and interpreting the results and predictions of any model, the assumptions underlying EwE should be recognized, well explored and carefully considered. Plaga´nyi and Butterworth (2004) discuss the implications of the foraging arena hypothesis (which governs species interactions) in Ecosim. Aydin (2004) explored in depth the implications of the ‘‘fixed growth efficiency’’ function in Ecosim, which essentially means that EwE does not account adequately for changes in the population energetics as population size structure changes (e.g. due to heavy fishing or release of fishing pressure). The latter leads Ecosim to overestimate the biomass of top predators that could be supported in a pristine ecosystem, and to underestimate the prey biomass

280

M. Coll et al.

that may be supported if predation pressure is released by removal of top predators through fishing (Aydin 2004). Plaga´nyi and Butterworth (2004) foresee prudent application of EwE contributing to the inclusion of multispecies considerations in operational management procedures that are currently single-species-based. The authors stress that variability about predicted ecosystem responses should be carefully taken into account. EwE provides a useful and accessible format for policy exploration, but there are concerns related to the policy optimization routines. For example, the relative weightings allocated to the net economic value, employment value, mandated rebuilding and ecosystem structure objectives, required when searching for an ‘‘optimal’’ fishing strategy or when testing possible fishing policies, are open to much debate and subjectivity. A limitation of the policy optimization routine that was previously noted (and possibly still something of a concern despite the improvements in ecological objectives into mandated rebuilding and ecosystem structure objectives), was the lack of clarity in how to optimize ecosystem structure. Indeed, what is meant by optimization in an ecosystem context? There is certainly some subjectivity in the selection of groups to be given higher ecological weightings, or even to be ‘‘monitored’’ when testing for reasonable optimal strategies (Shannon 2002), and thus it remains to emphasize the importance of carefully considering and ultimately agreeing upon sensible fishing policies prior to exploring what their potential effects might be.

8.4.5 Future Directions for EwE A multitude of studies have demonstrated that EwE is a useful diagnostic tool. In addition, it has been shown to be of use in identifying or monitoring ecosystem changes over time (see Sections 8.3.1 and 8.4.3), an important building block if we are to respond appropriately through current or future fisheries management approaches. Thus, the diagnostic capabilities of EwE are considered to be a strength which could be further developed and used in the future. The future of EwE may well lie in the benefits gained from comparative analyses of EwE models. Several comparative studies have been discussed in Section 8.3.1, and have been shown to enhance our understanding of the way in which ecosystems function (generically and individually), thus affording us a new appreciation of the promise or failure of current and future fisheries management strategies. This approach is further enhanced by developing multiple models of the same system and comparing results. Fulton and Smith (2004) compared three models of Port Phillip Bay, Australia, an Ecosim model, and two bio-geochemical models and compared them under a range of fishing strategies. They found that some of their results were robust under different model formulations, whereas others were not, but

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

281

that model comparisons led to greater understanding of the ecosystem. Heymans and Baird (2000) compared results of an Ecopath model applied to the Northern Benguela ecosystem with results from a NETWRK model application. This work highlighted that even though different input methodologies for the two models resulted in differences in their outputs, in most cases the interpretation of results would lead to the same qualitative conclusion regarding network analysis of the ecosystem. Similarly, Savenkoff et al. (2001) compared an Inverse model with an Ecopath model (Bundy et al. 2000) of the Newfoundland-Labrador system. In this case, the combination of the optimisation algorithms of the Inverse method, together with the Ecopath software holds scope for a flexible approach to explore a statistical framework for Ecopath estimates. In 1981, trophic interactions were detailed in a study of the North Sea (the Year of the Stomach) and an Ecopath model was developed using this data. The main results were compared with those from a Multispecies Virtual Population Analysis (MSVPA) for the area (Christensen 1995b), indicating that consumption rates of prey by various gadoid species used within MSVPA were unrealistically low but that other parameters were very reasonable. The North Sea Ecopath model is currently being expanded and fitted to extensive time series data sets (Mackinson and Daskalov 2007). Shin et al. (2004) compared a selection of altered fishing simulations for the Southern Benguela using an OSMOSE model (a spatialized, size-based individual based model) to the same simulations using an EwE model (Shannon 2001), and found that the two were generally consistent in their results despite fundamentally different assumptions and theories underlying the models. Their conclusion was that cross-validation of model outputs such as these are useful means to evaluate robustness of model outputs for fisheries management purposes. The question remains as to whether EwE can, or should, be used in a predictive sense. The general consensus is that EwE is an exploratory tool for examining fishing strategies and their effects in the ecosystem context, and that it should not be seen as an all-empowering predictive glass ball looking into the future. However, several careful studies have demonstrated that when used in conjunction with other approaches, models and systematic tuning of EwE to empirical data, EwE can make valuable contributions to our understanding of ecosystems, and potentially towards management (Aydin et al. 2005, Martell et al. 2005, Bundy and Fanning 2005, Fulton and Smith 2004). In order to manage marine resources for long-term sustainable exploitation, it is vital that we aim to maintain the complexity and biodiversity of marine ecosystems. However, fisheries management is still mainly approached from the level of single-species management. It remains unclear as to how to begin to incorporate ecosystem considerations into fisheries management so that concerns regarding species interactions and spatial dynamics, which are thus far rarely addressed by current fisheries management strategies, can be explicitly taken into account. An ecosystem approach to fisheries will not be achievable without a solid scientific foundation, facilitating the assessment of the

282

M. Coll et al.

ecosystem effects of fishing, and also the effectiveness of the various management strategies that may be adopted in response to the ecosystem concerns identified and raised (Shannon et al. 2006). The world wide recognition of, and move towards the ecosystem approach to fisheries (e.g. Sinclair and Valdimarsson 2003, Sinclair et al. 2002, Pikitch et al. 2004, McCloud et al. 2005), has placed the spotlight on the scientific tools (in particular models) available to facilitate the process. Ecosystem models, by virtue of the large number of interactions being considered, have large inherent uncertainties and thus they are seen to be poor in terms of predictive power and far from robust (Shannon et al. 2004b). However, EwE now has the capacity for fitting ecosystem models to available time series data (Christensen and Walters 2004a, Walters and Martell 2004), and progress is being made towards incorporating uncertainty in model results (Bundy 2005, Gaichas 2006), building confidence in our ecosystem models as tools for use in the management arena. We believe that EwE is a promising approach, both in its usefulness for exploration of species interactions (EwE) and spatial dynamics (Ecospace) and their implications for fisheries in the ecosystem context, and in providing a means of testing alternative fishing strategies in an ecosystem modelling context. Of special relevance from spatial modelling with EwE are the predicted distribution maps of species or groups of species, Fig. 8.7, (Walters et al. 1999). Apart from validating model results, these maps can be used to gain knowledge from poorly known species within the ecosystem and how these species can change their distribution due to internal and external ecosystem features, as well as in response to the establishment of Marine Protected Areas (MPA). A future direction for development might be the use of spatial indicators such as species and species-fisheries overlap indices (Fre´on et al. 2005; Drapeau et al. 2004) that have been applied in GIS applications to fisheries data series. In this respect, EwE is seen as an important component of the ‘‘model toolbox’’ (Starfield et al. 1988) that will form the basis for construction of an ‘‘ecoscope’’ (Ulanowicz 1993, Cury 2004) to assemble ecosystem knowledge into a useable form for the provision of ecosystem management advice (Shannon et al. 2007). Further, it would be beneficial to expand the scope of EwE to couple EwE models to other models, or at least to provide some links between these models, which would, for example, facilitate a better description of lower TLs and impacts of climate change dynamics towards an End-to-End description of ecosystems. A few initiatives have recently been taken in this direction (Aydin et al. 2005, Libralato et al. 2005). Currently, a new generation of EwE software is being developed with the major focus on stakeholder participation and buy-in through communication and visualization of potential fisheries management options. This new approach aims to develop ecological models to be used to perform sophisticated simulations to predict and evaluate scenarios for sustainable management of fisheries and ecosystems and to directly influence the ecosystem management process (V. Christensen, pers. comm., www.lenfestoceanfutures.org).

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

283

References Ainsworth C, Heymans JJ, Pitcher T, Vasconcellos M (2002) Ecosystem models of Northern British Columbia for the time periods 2000, 1950, 1900 and 1750. Fisheries Centre Research Reports, 10(4): 41 pp. Allen R (1971) Relation between production and biomass. Journal of Fisheries Research Board of Canada, 28: 1573–1581. Arias-Gonzalez JE, Delesalle B, Salvat B, Galzin R (1997) Trophic functioning of the Tiahura reef sector, Moorea Island, French Polynesia. Coral Reefs, 16: 231–246. Arreguı´ n-Sa´nchez F (2000). Octopus-red grouper interaction in the exploited ecosystem of the northern continental shelf of Yucatan, Mexico. Ecological Modelling, 129: 119–129. Arreguı´ n-Sa´nchez F, Manickchand-Heileman S (1998) The trophic role of lutjanid fish and impacts of their fisheries in two ecosystems in the Gulf of Mexico. Journal of Fish Biology, 53(Suppl. A): 143–153. Arreguı´ n-Sa´nchez F, Arcos E, Cha´vez EA (2002) Flow of biomass and structure in an exploited benthic ecosystem in the gulf of California, Mexico. Ecological Modelling, 156: 167–183. Arreguı´ n-Sa´nchez F, Herna´ndez-Herrera A, Ramı´ rez-Rodrı´ guez M, Pe´rez-Espan˜a H (2004) Optimal management scenarios for the artisanal fisheries in the ecosystem of La Paz Bay, Baja California Sur, Mexico. Ecological Modelling 172: 373–382. Aydin KY (2004) Age structure or functional response? Reconciling the energetics of surplus production between single-species models and Ecosim. In. Shannon LJ, Cochrane KL, Pillar SC (Eds.), An ecosystem approach to fisheries in the southern Benguela. African Journal of Marine Science, 26: 289–301. Aydin KY, Friday N (2001) The early development of Ecosim as a predictive multi-species fisheries management tool. Document presented to the IWC Scientific Committee, July 2001. SC/53/E3: 8 pp. Aydin KY, McFarlane GA, King JR, Megrey BA, Myers KW (2005) Linking oceanic food webs to coastal production and growth rates of Pacific salmon (Oncorhynchus spp.) using models on three scales. Deep Sea Research II, 52: 757–780. Bianchi G, Gislason H, Hill L, Koranteg K, Manickshand-Heileman S, Paya I, Sainsbury K, Sa´nchez F, Jin X, Zwanenburg K (2000) Impact of fishing on demersal fish assemblages. ICES Journal of Marine Science, 57: 558–571. Booth S, Zeller D (2005) Mercury, food webs and marine mammals: implications of diet and climate change for human health. Environmental Health Perspectives, 113(5): 521–526. Bradford-Grieve JM et al. (2003) Pilot trophic model for subantarctic water over the Southern Plateau, New Zealand: a low biomass, high transfer efficiency system. Journal of Experimental Marine Biology and Ecology, 289: 223–262. Brando VE, Ceccarelli R, Libralato S, Ravagnan G (2004) Assessment of environmental management effects in a shallow water basin using mass-balance models. Ecological Modelling, 172(2–4): 213–232. Bundy A (1997) Assessment and management of multispecies, multigear fisheries: a case study from San Miguel Bay, the Philippines. Ph.D. Thesis. University of British Columbia. Bundy A (2001) Fishing on ecosystems: the interplay of fishing and predation in NewfoundlandLabrador. Canadian Journal of Fisheries and Aquatic Science, 58: 1153–1167. Bundy A (2004a) The ecological effects of fishing and implications for coastal management in San Miguel Bay, the Philippines. Coastal Management 32: 25–38. Bundy A (2004b) Mass balance models of the eastern Scotian Shelf before and after the cod collapse and other ecosystem changes. Canadian Technical Report of Fisheries and Aquatic Sciences, 2520. 205 pp. Bundy A (2005) Structure and function of the eastern Scotian shelf Ecosystem before and after the groundfish collapse in the early 1990s. Canadian Journal of Fisheries and Aquatic Science, 62(7): 1453–1473.

284

M. Coll et al.

Bundy A, Pauly D (2001) Selective harvesting by small-scale fisheries: ecosystem analysis of San Miguel Bay, Philippines. Fisheries Research 53: 263–281. Bundy A, Fanning P (2005) Can Atlantic cod recover? Exploring trophic explanations for the non-recovery of cod on the eastern Scotian Shelf, Canada. CJFAS 62(7): 1474–1489. Bundy A, Lilly G, Shelton P (2000) A mass balance model of the Newfoundland-Labrador shelf. Canadian Technical Report of Fisheries and Aquatic Sciences, 2310, 117 pp + App. Christensen V (1996) Managing fisheries involving predator and prey species. Reviews in Fish Biology and Fisheries 6: 417–442. Christensen V (1995a) A model of trophic interactions in the North Sea in 1981, The year of the stomach. Dana, 11(1): 1–28. Christensen, V (1995b) Ecosystem maturity – towards quantification. Ecological Modelling, 77: 3–32. Christensen V (1998) Fishery-induced changes in a marine ecosystem: insight from models of the Gulf of Thailand. Journal of Fish Biology, 53(Supplement A): 128–142. Christensen V (2000) Indicators for marine ecosystem affected by fisheries. Marine and Freshwater Research, 51: 447–450. Christensen V, Pauly D (1992) Ecopath II. A software for balancing steady-state models and calculating network characteristics. Ecological Modelling, 61, 169–185. Christensen V, Pauly D (Eds.) (1993) Trophic models of aquatic ecosystems. ICLARM, Manila, Philipinas, 26: 390 p. Christensen V, Pauly D (1995) Fish production, catches and the carrying capacity of the world oceans. Naga 18(3): 34–40. Christensen V, and Pauly D (1996) Ecological modelling for all. Naga 19(2): 25–26. Christensen V, Pauly D (1998) Changes in models of aquatic ecosystems approaching carrying capacity. Ecological Applications, 8(Suppl. 1): 104–109. Christensen V, Walters CJ (2000) Ecopath with Ecosim: methods, capabilities and limitations. In Pauly D, Pitcher TJ (Eds.), Methods for assessing the impact of fisheries on marine ecosystems of the North Atlantic. Fisheries Centre Research Reports 8: 79–105. Christensen V, Walters C (2004a) Ecopath with Ecosim: methods, capabilities and limitations. Ecological Modelling, 172(2–4): 109–139. Christensen V, Walters CJ (2004b) Trade-offs in ecosystem-scale optimization of fisheries management policies. Bulletin of Marine Science, 74(3): 549–562. Christensen V, Walters CJ, Pauly D (2000) Ecopath with Ecosim: users guide and help files. Fisheries Centre, University of British Columbia, Vancouver. Christensen V, Walters CJ, Pauly D (2005) Ecopath with Ecosim: a User’s guide. Fisheries Centre of University of British Columbia, Vancouver, Canada. 154 pp. Ciannelli L, Robson BW, Francis RC, Aydin KY, Brodeur RD (2005). Boundaries of open marine ecosystems: an application to the Pribilof Archipelago, Southeast Bering Sea. Ecological Applications, 14(3): 942–953. Coll M, Palomera I, Tudela S, Sarda` F (2006a) Trophic flows, ecosystem structure and fishing impacts in the South Catalan Sea, Northwestern Mediterranean. Journal of Marine Systems, 59: 63–96. Coll M, Shannon LJ, Moloney CL, Palomera I, Tudela S (2006b) Comparing trophic flows and fishing impacts of a NW Mediterranean ecosystem with coastal upwellings by means of standardized ecological models and indicators. Ecological Modelling, 198: 53–70. Coll M, Santojanni A, Arneri E, Palomera I, Tudela S (2007) An ecosystem model of the Northern and Central Adriatic Sea: analysis of ecosystem structure and fishing impacts. Journal of Marine Systems, 67: 119–154. Coll M, Bahamon N, Sarda` F, Palomera I, Tudela S, Suuronen P. (2008a) Ecosystem effects of improved trawl selectivity in the South Catalan Sea (NW Mediterranean). Marine Ecology Progress Series, 355: 131–147. Coll M, Palomera I, Tudela S, Dowd M (2008b) Food-web dynamics in the South Catalan Sea ecosystem (NW Mediterranean) for 1978–2003. Ecological Modelling, 217(1–2): 95–116.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

285

Cox SP, Essington TE, Kitchell JF, Martell SJD, Walters CJ, Boggs C, Kaplan I (2002) Reconstructing ecosystem dynamics in the central Pacific Ocean, 1952–1998. 2. A preliminary assessment of the trophic impacts of fishing and effects on tuna dynamics. Canadian Journal of Fisheries and Aquatic Science, 59: 1736–1747. Crawford RJM, Cruikshank RA, Shelton PA, Kruger I (1985) Partitioning of a goby resource amongst four avian predators and evidence for altered trophic flow in the pelagic community of an intense, perennial upwelling system. South African Journal of Marine Science 3: 215–228. Criales-Hernandez MI, Duarte LO, Garcı´ a CB, Manjarre´s L (2006) Ecosystem impact of the introduction of bycatch reduction devices in a tropical shrimp trawl fishery: insights through simulation. Fisheries Research, 77: 333–342. Cury P (2004) Tuning the ecoscope for the ecosystem approach to fisheries. In: Perspectives on ecosystem-based approaches to the management of marine resources. Marine Ecology Progress Series, 274: 272–275. Cury P, Shannon LJ, Roux J–P, Daskalov G, Jarre A, Pauly D, Moloney CL (2005) Trophodynamic indicators for an ecosystem approach to fisheries. ICES Journal of Marine Science, 62: 430–442. Dalsgaard J, Wallace SS, Salas S, Preikshot D (1998) Mass-balance model reconstruction of the Strait of Georgia: the present, one hundred, and five hundred years ago. Back to the Future: Reconstructing the Strait of Georgia Ecosystem. Fisheries Centre Research Report, 6(5): 72–91. Daskalov GM (2002) Overfishing drives a trophic cascade in the Black Sea. Marine Ecology Progress Series, 225: 53–63. Drapeau L, Pecquerie L, Fre´on P, Shannon LJ (2004) Quantification and representation of potential spatial interactions in the Southern Benguela ecosystem. In An Ecosystem Approach to Fisheries in the Southern Benguela. Shannon, L.J., K.L. Cochrane, and S.C. Pillar (Eds). African Journal of Marine Science 26: 141–159. FAO (2005) Review of the State of World Marine Fishery Resources. FAO Fisheries Technical Paper, 457, Rome, 235 pp. Fearon JJ, Boyd AJ, Schu¨lein FH (1992) Views on the biomass and distribution of Chrysaora hysoscella (Linne´, 1766) and Aequorea aequorea (Kosrkal, 1775) off Namibia, 1982–1988. Scientia Marina 56: 75–85. Fre´on P, Drapeau L, David JHM, Ferna´ndez Moreno A, Leslie RW, Oosthuizen H, Shannon LJ, Van der Lingen C (2005) Spatialized ecosystem indicators in the southern Benguela. ICES Journal of Marine Science, 62(3): 459–468. Fulton EA, Smith ADM (2004) Lessons learnt from a comparison of three ecosystem models for Port Phillip Bay, Australia. In. Shannon LJ, Cochrane KL, Pillar SC (Eds.), An ecosystem approach to fisheries in the southern Benguela. African Journal of Marine Science 26: 219–243. Gaichas SK (2006) Development and application of ecosystem models to support fishery sustainability: a case study for the Gulf of Alaska. PhD dissertation. University of Washington. 371 pp. Gasalla MA, Rossi-Wongtschowski CLDB (2004) Contribution of ecosystem analysis to investigating the effects of changes in fishing strategies in the South Brazil Bight coastal ecosystem. Ecological Modelling, 172: 283–306. Gucu AC (2002) Can overfishing be responsible for the successful establishment of Mnemiopsis leidyi in the Black Sea? Estuarine, Coastal and Shelf Science, 54: 439–451. Halfon E, Schito N (1993) Lake Ontario food web, an energetic mass balance. ICLARM conference proceedings. Manila 1993. Heymans JJ (2004) The effects of Internal and external control on the Northern Benguela ecosystem. In. Sumaila UR, Skogen SI, Boyer D (Eds.), Namibia’s Fisheries. Ecological, economic and social aspects. Eburon Academic Publishers. 29–52.

286

M. Coll et al.

Heymans JJ, Baird D (2000) Network analysis of the Northern Benguela ecosystem by means of NETWRK and Ecopath. Ecological Modelling, 131(2/3): 97–119. Heymans JJ, Shannon LJ, Jarre-Teichmann A (2004) Changes in the northern Benguela ecosystem over three decades: 1970s, 1980s and 1990s. Ecological Modelling, 172, 175–195. Hilborn R, Walters CJ (1992) Quantitative Fisheries Stock Assessment. Choice, Dynamics and Uncertainty. Kluwer Academic Publishers. 570 pp. Innes S, Lavigne DM, Earle WM, kovacs KM (1987) Feeding rates of seals and whales. Journal of Animal Ecology, 56: 115–130. Jarre-Teichmann A, Shannon LJ, Moloney CL, Wickens PA (1998) Comparing trophic flows in the southern Benguela to those in other upwelling ecosystems. In. Pillar SC, Moloney CL, Payne AIL, Shillington FA (Eds.), Benguela dynamics: impacts of variability on shelf-sea environments and their living resources. South African Journal of Marine Science 19: 391–414. Jennings S, Greenstreet SPR, Hill L, Piet GJ, Pinnegar JK, Warr KL (2002) Long-term trends in the trophic sturcture of the North Sea fish community: evidence from stable-isotope analysis, size-spectra and community metrics. Marine Biology, 141: 1085–1097. Kavanagh P, Newlands N, Christensen V, Pauly D (2004) Automated parameter optimization for Ecopath ecosystem models. Ecological Modelling, 172(2–4): 141–150. Kitchell JF, Kaplan IC, Cox SP, Martell SJD, Essington TE, Boggs CH, Walters CJ (2002) Ecological and economic components of alternative fishing methods to reduce by-catch of marlin in a tropical pelagic ecosystem. Bulletin of Marine Science, 74(3): 607–618. Lam VWY, Pauly D (2005) Mapping the global biomass of mesopelagic fishes. Sea Around Us Project Newsletter. 30: 4. Lalli CM, Parsons TR (1993) Biological oceanography: an introduction. Pergamon Press, Oxford. p. 296. Leontief WW (1951) The structure of the US economy. 2nd ed. Oxford University Press, New York. Libralato S, Cossarini G, Solidoro C (2005) Ecosystem approach with trophic web models: methodological evidences from the Venice Lagoon application. Proceedings of the 5th European Conference on Ecological Modelling, Pushchino, Russia. Libralato S, Christensen V, Pauly D (2006) A method for identifying keystone species in food web models. Ecological Modelling, 195(3–4): 153–171. Libralato S, Coll M, Tudela S, Palomera I, Pranovi F. (2008) A new index to quantify the ecosystem impacts of fisheries as the removal of secondary production. Marine Ecology Progress Series, 355: 107–129. Libralato S, Pastres R, Pranovi R, Raicevich S, Granzotto A, Giovanardi O, Torricelli P (2002) Comparison between the energy flow networks of two habitat in the Venice lagoon. P.S.Z.N. Marine Ecology, 23: 228–236. Lin H-J, Shao K-T, Hwang J-S, Lo W-T, Cheng I-J, Lee L-H (2004) A trophic model for Kuosheng Bay in Northern Taiwan. Journal of Marine Science Technology, 12: 424–432. Lindeman RL (1942) The trophic-dynamic aspect of ecology. Ecology, 23: 399–418. Lluch-Belda D, Crawford RJM, Kawasaki T, MacCall AD, Parrish RH, Schwartzlose RA, Smith PE (1989) World-wide fluctuations of sardine and anchovy stocks: the regime problem. South African Journal of Marine Science, 8: 195–205. Lluch-Belda D, Schwartzlose RA, Serra R, Parrish R, Kawasaki T, Hedgecock D, Crawford RJM (1992a) Sardine and anchovy regime fluctuations of abundance in four regions of the world oceans: a workshop report. Fisheries Oceanography, 1(4): 339–347. Lluch-Belda D, Hernandez-Vazquez S, Lluch-Cota DB, Salinas-Zavala CA, Schwartzlose RA (1992b) The recovery of the California sardine as related to global change. CalCOFI Reports, 33: 50–59. Mackinson S, Daskalov G (2007) An ecosystem model of the North Sea to support and ecosystem approach to fisheries management: description and parameterisation. Cefas Science Series Technical Report, 142: 195.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

287

Mackinson S, Vasconcellos M, Pitcher T, Walters CJ (1997) Ecosystem Impacts of harvesting small pelagic fish in upwelling systems: using a dynamic mass-balance model. Proceedings Forage Fishes in Marine Ecosystems. Alaska Sea Grant College Program, AK-SG-97-01. 731–748. Mackinson S, Blanchard JL, Pinnegar JK, Scott R (2003) Consequences of alternative functional response formulations in models exploring whale-fishery interactions. Marine Mammal Science, 19(4): 661–681. ´ M, Jiang H, Mackinson S, Daskalov G, Heymans JJ, Neira S, Arancibia H, Zetina-Rejon Cheng HQ, Coll M, Arreguin-Sanchez F, Keeble K, Shannon L Which forcing factors fit? Using ecosystem models to investigate the relative influence of fishing and changes in primary productivity on the dynamics of marine ecosystems. Ecological Modelling, in press. Magnu´sson KG (1995) An overview of the multispecies VPA – theory and applications. Reviews in Fish Biology and Fisheries, 5: 195–212. Magnusson KG (1999) Biological interactions in fish stocks: models and reality. Rit. Fiskideildar 16: 295–305. Margalef R (1968) Perspectives in Theoretical Ecology. The University of Chicago Press. Chicago, London. 111 pp. Martell S, Essington TE, Lessard B, Kitchell JF, Walters CJ, Boggs CH (2005) Interactions of productivity, predation risk and fishing effort in the efficacy of marine protected areas for the central Pacific. Canadian Journal of Fisheries and Aquatic Science, 62: 1320–1336. McAllister M.K., Pikitch E.K., Punt A.E., Hilborn R. (1994) A Bayesian approach to stock assessment and harvest decisions using the sampling importance resampling algorithm. Canadian Journal of Fisheries and Aquatic Sciences, 51(12): 2673–2687. McCloud KL et al. (2005) Scientific consensus statement on marine ecosystem-based management. Communication Partnership for Science and the Sea (http://compassonline.org/? q=EBM). Moloney C, Jarre A, Arancibia H, Bozec Y-M, Neira S, Roux J-P, Shannon LJ (2005) Comparing the Benguela and Humboldt marine upwelling ecosystems with indicators derived from inter-calibrated models. ICES Journal of Marine Science, 62(3): 493–502. Morrissette L (2007) Complexity, cost and quality of ecosystem models and their impact on resilience: a comparative analysis, with emphasis on marine mammals and the Gulf of St. Lawrence. PhD thesis. University of British Columbia. 278 pp. Mu¨ller F. (1997) State-of-the-art in ecosystem theory. Ecological Modelling, 100: 135–161. Murawski S.A. (1991) Can we manage our multispecies fisheries? Fisheries, 16(5): 5–13. Murawski S.A. (2000) Definitions of overfishing from an ecosystem perspective. ICES Journal of Marine Science, 57: 649–658. Neira S, Arancibia H (2004) Trophic interactions and community structure in the upwelling system off Central Chile (33–398S). Journal of Experimental Marine Biology and Ecology, 312(2): 349–366. Neira S, Arancibia H, Cubillos L (2004) Comparative analysis of trophic structure of commercial fishery species off central Chile in 1992 and 1998. Ecological Modelling, 172(1–4): 233–248. Neira S, Moloney C, Cury P, Arancibia H. (in prep.) Analyzing changes in the southern Humboldt ecosystem for the period 1970–2004 by means of trophic/food web modeling (in preparation to be submitted to Fish and Fisheries). Nilsson SG, Nilsson IN (1976) Number, food and consumption, and fish predation by birds in Lake Mockeln, Southern Sweden. Ornis Scandinavian, 7: 61–70. Odum EP (1969) The strategy of ecosystem development. Science, 104: 262–270. Odum WE, Heald EJ (1975) The detritus-based food web for an estuarine mangrove community. In. Cronin LE (Ed.), Estuarine Research, Vol. 1. Academic Press, New York. Okey T (2004) Shifted community states in four marine ecosystems: some potential mechanisms. PhD dissertation. University of British Columbia. 185 pp.

288

M. Coll et al.

Okey T, Pauly D (1999) A trophic mass-balance model of Alaska’s Prince William Sound ecosystem, for the post-spill period 1994–1996, 2nd edn. Fisheries Centre Research Reports, 7(4): 146 pp. Okey T., Wright BA (2004) Towards ecosystem-based extraction policies for Prince William Sound, Alaska: integrating conflicting objectives and rebuilding pinnipeds. Bulletin of Marine Science, 74(3): 727–747. Okey TA, Banks S, Born AF, Bustamante RH, Calvopin˜a M, Edgar GJ, Espinoza E, Farin˜a JM, Garske LE, Reck GK (2004) A trophic model of a Gala´pagos subtidal rocky reef for evaluating fisheries and conservation strategies Ecological Modelling, 172: 383–401. Ortiz M, Wolff M (2002a) Dynamical simulation of mass-balance trophic models for benthic communities of north-central Chile: assessment of resilience time under alternative management scenarios. Ecological Modelling, 148: 277–291. Ortiz M, Wolff M (2002b) Spatailly explicit trophic modelling of a harvested benthic ecosystem in Tongoy Bay (central northern Chile). Aquatic Conservation: Marine and Freshwater Ecosystems, 12: 601–618. Pauly D (1980) On the interrelationships between natural mortality, growth parameters, and mean environmental temperature in 175 fish stocks. Journal du Conseil, Conseil Inte´rnational pour l’Exploration de la Mer, 39: 175–192. Pauly D, Christensen V (1995) Primary production required to sustain global fisheries. Nature, 374: 255–257. Pauly D, Christensen V (2004) Ecosystem models. In. Hart PJB, Reynolds JD (Eds.), Handbook of Fish Biology and Fisheries. Blacwell. 211–227. Pauly D, Christensen V, Sambilay V (1990) Some features of fish food consumption estimates used by ecosystem modellers. ICES Council Meeting 1990/G:17, 8p. Pauly D, Christensen V, Walters C (2000) Ecopath, Ecosim and Ecospace as tools for evaluating ecosystem impact of fisheries. ICES Journal of Marine Science, 57: 697–706. Pauly D, Christensen V, Dalsgaard J, Froese R, Torres FJ (1998) Fishing down marine food webs. Science, 279, 860–863. Pikitch EK, Santora C, Badcock E., Bakun A, Bonfil R, Conover DO, Dayton P, Doukakis P, Fluharty D, Heneman B, Houde ED, Link J, Livingston PA, Mangel M, McAllister MK, Pope J, Sainsborry KJ (2004) Ecosystem-based fishery management. Science, 305: 346. Pimm SL, Lawton JH, Cohen JE (1991) Food web patterns and their consequences. Nature, 275: 542–544. Pinnegar JK (2000) Planktivorous fishes: links between the Mediterranean littoral and pelagic. PhD Thesis, University of Newcastle upon Tyne, UK. 213 pp. Pinnegar JK, Jennings S, O’Brien CM, Polunin NVC (2002) Long-term changes in the trophic level of the Celtic Sea fish community and fish market price distribution. Journal of Applied Ecology, 39: 377–390. Pitcher TJ (2001) Fisheries manager to rebuild ecosystems? Reconstructing the past to salvage the future. Ecological Applications, 11(2): 601–617. Pitcher TJ (2005) Back-to-the-future: a fresh policy initiative for fisheries and a restoration ecology for ocean ecosystems. Philosophical Transactions of the Royal Society B: Biological Sciences 360(1453): 107–121. Pitcher T, Cochrane K (2002) The use of ecosystem models to investigate multispecies management strategies for capture fisheries. Fisheries Centre Research Reports, 10(2): 156 pp. Pitcher TJ, Heymans JJ, Vasconcellos M (2002a) Ecosystem models of Newfoundland for the time periods 1995, 1985, 1900 and 1450. Fisheries Centre Research Reports, 10(5): 74 pp. Pitcher TJ, Buchary E, Hutton T (2002b) Forecasting the benefits of no-take human made reefs using spatial ecosystem simulation. ICES Journal of Marine Science, 59: S17–S26.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

289

Plaga´nyi E´E, Butterworth DS (2004) A critical look at the potential of Ecopath with Ecosim to assist in practical fisheries management. African Journal of Marine Science, 26: 261–288. Polovina JJ (1984) Model of a coral reef ecosystem. The ECOPATH model and its application to French Frigate Shoals. Coral Reefs, 3(1): 1–11. Power ME, Tilman D, Ester JA, Menge BA, Bond WA, Mills LS, Daily G, Castilla JC, Lubchenco J, Paine RT (1996) Challenge in the question for Keystones. Bioscience, 46(8): 609–620. Pranovi F, Libralato S, Raicevich S, Granzotto A, Pastres R, Giovanardi O (2003) Mechanical clam dredging in Venice lagoon: ecosystem effects evaluated with trophic massbalance model. Marine Biology, 143: 393–403. Rochet M-J, Trenkel VM (2003) Which community indicators can measure the impact of fishing? A review and proposals. Canadian Journal of Fisheries and Aquatic Science, 60: 86–98. Roux J-P, Shannon LJ (2004) Ecosystem approach to fisheries management in the northern Benguela: the Namibian experience. In. Shannon L, Cochrane KL, Pillar SC (Eds.), Ecosystem Approaches to Fisheries in the Southern Benguela. African Journal of Marine Science, 26: 79–94. Sa´nchez F, Olaso I (2004) Effects of fisheries on the Cantabrian Sea shelf ecosystem. Ecological Modelling, 172(2–4): 151–174. Savenkoff C, Ve´zina AF, Bundy A (2001) Inverse analysis of the structure and dynamics of the whole ecosystem in the Newfoundland-Labrador shelf. Canadian Technical Report of Fisheries and Aquatic Sciences, 2354: viii+56p. Scheffer M, Carpenter S, Foley JA, Folke C, Walker B (2001) Catastrophic shifts in ecosystems. Nature 413, 591–596. Schwartzlose RA, Alheit J, Bakun A, Baumgartner TR, Cloete R, Crawford RJM, Fletcher WJ, Green-Ruiz Y, Hagen E, Kawasaki T, Lluch-Belda D, Lluch-Cota SE, MacCall AD, Matsuura Y, Nevarez-Martinez MO, Parrish RH, Roy C, Serra R, Shust KV, Ward MN, Zuzunaga JZ (1999) Worldwide large-scale fluctuations of sardine and anchovy populations. South African Journal of Marine Science, 21: 289–347. Shannon LJ (2001) Trophic models of the Benguela upwelling system: towards an ecosystem approach to fisheries management. PhD Thesis, University of Cape Town. 319 pp. + appendices pages i–xxxv. Shannon LJ (2002) The use of ecosystem models to investigate multispecies management strategies for capture fisheries: Report on southern Benguela simulations. In. Pitcher T, Cochrane K (Eds.), The use of ecosystem models to investigate multispecies management strategies for capture fisheries. Fisheries Centre Research Reports 10(2): 118–126. Shannon LJ. Tropic structure and functioning of the southern Benguela ecosystem in the period 2000–2004. African Journal of Marine Science, in prep. Shannon LJ, Jarre-Teichmann A (1999a) A model of the trophic flows in the northern Benguela upwelling system during the 1980s. South African Journal of Marine Science, 21: 349–366. Shannon LJ, Jarre-Teichmann A (1999b) Comparing models of trophic flows in the northern and southern Benguela upwelling systems during the 1980s. p. 527–541 in Ecosystem approaches for fisheries management. University of Alaska Sea Grant, AK-SG-99-01, Fairbanks. 756 pp. Shannon LJ, Cury P (2003) Indicators quantifying small pelagic fish interactions: application using a trophic model of the southern Benguela ecosystem. Ecological Indicators, 3: 305–321. Shannon LJ, Cury P, Jarre A (2000) Modelling effects of fishing in the Southern Benguela ecosystem. ICES Journal of Marine Science, 57: 720–722. Shannon LJ, Field JC, Moloney C (2004a) Simulating anchovy-sardine regime shifts in the southern Benguela ecosystem. Ecological Modelling 172(2–4): 269–282.

290

M. Coll et al.

Shannon LJ, Christensen V, Walters C (2004b) Modelling stock dynamics in the Southern Benguella ecosystem for the period 1978–2002. African Journal of Marine Science, 26: 179–196. Shannon L J, Moloney C, Jarre-Teichmann A, Field JG (2003) Trophic flows in the southern Benguela during the 1980s and 1990s. Journal of Marine Systems, 39: 83–116. Shannon LJ, Coll M, Neira S, Cury PM, Roux J-P. The role of small pelagic fish in the ecosystem. In Checkley DM, Roy C, Alheit J, Oozeki Y (Eds.), Climate Change and Small Pelagic Fish, in press. Shannon LJ, Moloney CL, Cury P, Van der Lingen C, Crawford RJM, Cochrane KL (2007) Ecosystem modeling approaches for South African fisheries management. American Fisheries Society Symposium 2006. Fourth World Fisheries Congress May 2004, Vancouver, Canada. pp. 587–607, in press. Shannon LJ, Cury PM, Nel D, van der Lingen CD, Leslie RW, Brouwer SL, Cockcroft AC, Hutchings L (2006) How can science contribute to an ecosystem approach to pelagic, demersal and rock lobster fisheries in South Africa? African Journal of Marine Science 28(1): 115–157. Shin Y-J, Shannon LJ, Cury P (2004) Simulations of fishing effects on the Southern Benguela fish community using an individual-based model. Learning from a comparison with Ecosim. In. Shannon LJ, Cochrane KL, Pillar SC (Eds.), An Ecosystem Approach to Fisheries in the Southern Benguela. African Journal of Marine Science, 26: 95–114. Sinclair M, Valdimarsson G (Eds.), (2003) Responsible Fisheries in the Marine Ecosystem, Wallingford: CAB International. 426p. Sinclair M, Arnason R, Csirke J, Karnicki Z, Sigurjohnsson J, Rune Skjoldal H, Valdimarsson G (2002) Responsible fisheries in the marine ecosystem. Fisheries Research 58: 255–265. Salomon AK, Waller NP, McIlhagga C, Yung RL, Walters C (2002) Modeling the trophic effects of marine protected area zoning policies: a case study. Aquatic Ecology, 36(1): 85–95. Starfield AM, Shelton PA, Field JG, Crawford RJM, Armstrong MJ (1988) Note on a modelling schema for renewable resource problems. South African Journal of Marine Science, 7: 299–303. Stergiou KI, Karpouzi V (2002) Feeding habits and trophic levels of Mediterranean fish. Reviews in Fish Biology and Fisheries, 11: 217–254. Stokes TK (1992) An overview of the North Sea multispecies modelling work in ICES. In. Payne AIL, Brink KH, Mann KH, Hilborn R (Eds.), Benguela Trophic Functioning. South African Journal of Marine Science, 12: 1051–1060. Ulanowicz RE (1986) Growth and development: ecosystem phenomenology. Springer Verlag, New York. 203 pp. Ulanowicz RE (1993) Inventing the ecoscope. In Christensen V, Pauly D (Eds.), Trophic models of aquatic ecosystems. ICLARM Conference Proceedings 26: 9–10. Ulanowicz R E, Puccia CJ (1990) Mixed trophic impacts in ecosystems. Coenoses, 5: 7–16. Van der Lingen CD (1999) The feeding ecology of, and carbon and nitrogen budgets for, sardine sardinops sagax in the southern benguela upwelling ecosystem. PhD. Thesis, University of Cape Town. 202 pp. Venter GE (1988) Occurrence of jellyfish on the west coast of Soyth West Africa/Namibia. In MacDonald IAW, Crawford RJM (Eds.), Long term data series relating to southern Africa’s renewable natural resources. Report of South Africa’s National Scientific Programmes, 157: 56–61. Walters CJ, Martell S (2004). Harvest management for aquatic ecosystems. Princeton University Press. 420 pp. Walters C, Christensen V, Pauly D (1997) Structuring dynamic models of exploited ecosystems from trophic mass-balance assessments. Reviews in Fish Biology and Fisheries, 7: 139–172. Walters C, Pauly D, Christensen V (1999) Ecospace: predictions of mesoscale spatial patterns in trophic relationships of exploited ecosystems, with emphasis on the impacts of marine protected areas. Ecosystems, 2: 539–554.

8

Ecosystem Modelling Using the Ecopath with Ecosim Approach

291

Walters C, Pauly D, Christensen V, Kitchell JF (2000) Representing density dependent consequences of life history strategies in aquatic ecosystems: Ecosim II. Ecosystems, 3: 70–83. Walters CJ, Christensen V, Martell S, Kitchell JF (2005) Single-species versus ecosystem harvest management: ecosystem structure erosion under myopic management. ICES Journal of Marine Science, 62: 558–568. Watermeyer K (2007) Reconstructing the Benguela ecosystem for a time before man’s intervention. Masters Thesis. Zoology Department, University of Cape Town, South Africa. 161p. Watermeyer K, Shannon LJ, Griffiths CL (2008) Changes in the trophic structure of the southern Benguela before and after the onset of industrial fishing. African Journal of Marine Science, 30(2): In press. Watters G M, Olson RJ, Francis RC, Fiedler PC, Polovina JJ, Reilly SB, Aydin KY, Boogs CH, Essington TE, Walters CJ, Kitchell JF (2003) Physical forcing and the dynamics of the pelagic ecosystem in the eastern tropical Pacific: simulations with Enso-scale and global-warming climate drivers. Canadian Journal of Fisheries and Aquatic Science, 60: 1161–1175. Whipple SJ, Link JS, Garrison LP, Fogarty MJ (2000) Models of predation and fishing mortality in aquatic ecosystems. Fish and Fisheries, 1: 22–40. Wolff M (1994) A trophic model for Tongoy Bay – a system exposed to suspended scallop culture (Northern Chile). Journal of Experimental Marine Biology and Ecology, 182: 149–168. Wulff F, Field JG, Mann KH (1989) Network analysis in marine ecology. Methods and applications. Coastal and Estuarine Studies. Vol. 32. Springer-Verlag, New York. 134 pp. Zwanenburg KC (2000) The effects of fishing on demersal fish communities of the Scotian Shelf. ICES Journal of Marine Science, 57: 503–508.

Chapter 9

Image Recognition Thomas T. Noji and Ferren MacIntyre

9.1 Introduction Eleven years after this chapter was originally written it is still fair to say that image recognition by computers has proven much more difficult than optimists once hoped. Tasks, which are far more mechanical and possessed of rigid well known rules such as architecture and engineering, are now more realistically described as ‘‘computer-aided’’ rather than ‘‘computer-performed’’ operations. The reason is simple: silicon- and carbon-based intelligence behave very differently and are good at very different things. Neither is a replacement for the other, but learning to work together and how best to divide the work load is proving a difficult challenge. The standard response to requests for automated pattern recognition for the last three decades has been: ‘‘Tell me how you identify your friends, and I’ll teach the computer to do it.’’ We seem to have been pursuing the wrong approach, for we cannot, in fact, describe how we recognize our friends, any more than we can describe how we move a finger. It just happens—because the brain is designed for such operations. Face-recognition may be an unfair example, since the human brain is so highly optimized for facial recognition that it is able to see faces where no faces are (Gould 1994), such as the one on the Makapansgat Pebble carried by an australopithecine 3 million years ago (Morris 1994). One cell in a monkey’s brain fires when it ‘‘sees’’ upright monkey hands1 (Ornstein and Thompson 1984). Another cell is triggered by the front T.T. Noji (*) U.S. Department of Commerce, National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Northeast Fisheries Science Center, Sandy Hook, NJ 07732, USA e-mail: [email protected] This revision was written by T. Noji. While updating and revising the original text, I have attempted to retain the free and expressive style, which was crafted by one the original authors, Ferren MacIntyre. 1 Curiously, the cell is in region TE, not shown in Fig. 9.1, but physically close to TF and TH, and so presumably very high in the visual hierarchy.

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_9, Ó Springer ScienceþBusiness Media B.V. 2009

293

294

T.T. Noji and F. MacIntyre

views of faces (monkey and human), and another by facial profiles (Desimone 1991). It is an oversimplification to say that these cells are responsible for recognition, for other cells that aren’t being monitored may also be firing. And these cells seem to work by weighting inputs from perhaps 10,000 other neurons, each similarly connected (e.g. Young 1992). ‘‘Neural networks’’, which attempt to emulate in silicon the connectivity and structure of biological neurons, may be a better approach to achieving automated image recognition than are pixel algorithms. ‘‘Training’’ a neural network to recognize images does not consist of writing stepwise algorithms but of showing the network pictures, telling it whether its identification was right or wrong, and letting it adjust the weights it gives to its inputs (e.g. Rahin 1992). This rather magically avoids the problems inherent in ‘‘recognizing friends’’. On the down side, it is impossible to tell what this ‘‘black box’’ is doing, what the final weights ‘‘mean’’, or what it is that the net is really learning. The classical example is a military task involving photos of woods with and without tanks. The net learned easily enough to separate photos into 2 categories with all the tanks in one pile. But subsequent investigation showed that the 2 sets of photos had been taken during different weather conditions, and what the net had learned was to sort out shadows. It knew nothing of tanks. An extension of ‘‘simple’’ neural networks is multi-class pattern recognition, which is a system that accurately maps an input feature space to an output space of multiple pattern classes (Ou and Murphy 2007). Multi-class pattern recognition has a wide range of applications including handwritten digit recognition (Le Cun et al. 1989), object classification (Aha and Bankert 1997) and text categorization (Apte et al. 1994). Can neural networks be taught to recognize Calanus finmarchicus from other copepods? The answer appears to be, ‘‘Yes!’’ but with considerable constraints, as described in Section 3 Plankton Identification. If the progress in pattern recognition by silicon is still disappointing, keep in mind the disadvantages under which it is working. Figure 9.1 shows one view of our current understanding of how a functional pattern-recognition system is organized. The diagram contains many abstractions, and the purpose of most of the boxes is unknown. Edge detection, which is a fairly serious suite of algorithms for a desktop silicon computer, is handled by the boxes labeled Retina and V1. Not shown in Fig. 9.1 are several important—and indubitably complex—inputs from other systems. Visual circuitry, which distinguishes between a moving field representing motion of the environment, and the same field resulting from motion of the head and shutting off portions of the system during eye movements, does so by combining input from the proprioceptive system, at a high hierarchical level (Burr et al. 1994). Characteristic sounds predispose us to see certain patterns in certain directions: this requires substantial input from the auditory system. We have all had the experience of looking at something (often a reduced-dimension image) and not having any idea of what we were looking at for several seconds. This time lag represents the optical system calling upon other systems for help, most likely the file system, and appealing to an ever widening sphere of less often consulted processors for ideas. People in cultures, which do

9

Image Recognition

295

Fig. 9.1 Hierarchical view of working pattern-recognizing system: Connection diagram of the visual areas of the macaque brain. The names of the boxes are largely arbitrary and may be ignored. Each line represents 105–106 connecting neurons, Many connections are bidirectional and processing is concurrent: all boxes at a given level process the entire visual field simultaneously (After Crick’s (1994) redaction of Suzuki and Amaral’s modification of Felleman and Van Essen (1991), with scribal errors at each copying)

296

T.T. Noji and F. MacIntyre

not make graphical representations of humans, are usually unable to recognize people in photographs. None of these high-level linkages is indicated in Fig. 9.1. A little-appreciated feature of our vision is the remarkable concentration of receptors in the fovea, and their even more lopsided representation in the brain. Each foveal cell has about 4 times the cortical allocation that a peripheral cell has (Azzopardi and Cowey 1993) with the result that 0.01% of our retina gets to use 8% of the visual cortex. The object we are looking at in detail receives 800 times the attention that the same area of background gets (Solomon and Pelli 1994). Machine vision does not have this advantage, and must attach equal importance to each pixel. Figure 9.2 shows a silicon-based pattern recognizer. It will be appreciated that there is room for further development. We do not mean to imply that imitation of the behavior of carbon-based intelligence is the natural route for the development of silicon intelligence; rather that silicon must meet significant challenges to reach the level of complexity that pattern recognition seems to require. It is interesting to note that insects process visual information very much as do vertebrates despite the fact that insects have very much smaller brains. Back-to-back papers in Nature suggest that feature extraction in bees (Srinivasan et al. 1993) and dragonflies (O’Carroll 1993) might be comparable to, if less extensive than, mechanisms known to exist in mammals. This probably represents parallel evolution—that is, an experimental consensus on the best way to approach the problem—so that it might make very good sense to attempt to duplicate the insect optical processor in silicon. Workers at the University of Pennsylvania analyzed 128  128 pixel images on a workstation with a software neural net called NEXUS, containing ca. 1 million units and 100 million interconnections, in which each ‘‘unit’’ can itself be a complex circuit (Finkel and Sajda 1994) With this much power in a neural net, they were having some success with Gestalt perception, teaching the net to identify contours and assign them to surfaces with the help of concepts like good continuation, closure, similarity, proximity, and completion. Surprisingly, some of these concepts appear to be handled by the brain as early as V1 (Fig. 9.1).

Fig. 9.2 Desk-top silicon’s best current emulation of a pattern-recognition structure. Each line represents a 32-bit (or narrower) data bus running at <100 MHz and operation is doubly sequential: one algorithm at a time and one pixel at a time. Dotted boxes are inactive while the selected algorithm is running. Current cameras mimic only the simplest qualities of the retina. Current algorithms may mimic process as high as V2 in Fig. 9.1, but do not do everything that V1 does

9

Image Recognition

297

Despite the great difference in degree of development between carbon- and silicon-based systems, there are already certain areas in which silicon-based pattern recognition may be useful in fisheries research.

9.2 Image Preprocessing As can be seen from Figs. 9.3, 9.5, and 9.9, raw images are seldom optimized for pattern recognition, and machine enhancement can usually improve the end results. The possibilities are numerous. Most image-analysis systems offer a

Fig. 9.3 An original grey-scale image (A) of a common marine phytoplankter, automatically reduced to its binary outline (B) and measured for some primary image-analysis parameters (C) using a ZeusTM image analyzer. A taxonomic description (D) of the species is translated from the original German (after Drebes 1974, Marine Phytoplankton (Georg Thieme Verlag, Stuttgart). Arrows in (B) point to contour distortions arising from background material. The numbers in (C) are in pixels, but are also available from the machine in scaled form if desired

298

T.T. Noji and F. MacIntyre

variety of algorithms for each process such as contrast control, noise removal, smoothing, sharpening, edge detection, etc. There is no sharp line between the algorithms of an ‘‘artistic’’ program such as PhotoshopTM, and the ‘‘scientific’’ program of an image-analysis system. As law courts have recognized, photographs and videos are no longer trustworthy evidence, since they can be easily manipulated to show whatever is desired (Mitchell 1994). This imposes a certain constraint upon the user of an image-analysis system not to significantly alter the data but only to enhance the originally recorded features of the data. Examples of possible processes include some that affect the entire image, such as look-up tables (‘‘LUTs’’, usually pronounced ‘‘lutes’’) which replace one pixel-value with another. These are good for false color (mapping temperature to spectrum for instance), contrast enhancement, and ‘‘histogram flattening’’, which attempts to distribute information uniformly across the grey scale in pursuit of statistical merit. Neighborhood transformations, which replace the central pixel of a small square by a function of its neighbors, allow both integration (smoothing and noise removal) and differentiation (sharpening and edge detection). Processing time increases as the area of the neighborhood, but almost all transforms work better in a circular area than a square. The solution is to use the largest feasible square and fill its corners with zeros. It is usually possible to limit the operation of a neighborhood transform to a selected area of the screen. The first 12-bit board-level image-analysis systems bypassed the n2 time for neighborhood binary transforms by a clever trick: the hardware could load a 3  4 neighborhood into the memory location representing the central pixel. This could then be manipulated with a LUT in a single frame time (nominally 1/30 s—in practice, allowing for loading the appropriate LUTs, 1 s), allowing 3  3 neighborhood operations like shrinking, enlarging, skeletonization, etc. to be done stepwise while you watched. A 32-bit machine could do as much with a 5  5 neighborhood. It was this approach which made any sort of image analysis feasible on machines with 8-MHz clocks. Segmentation—the isolation of portions of the image—is often done on the basis of intensity (grey-scale thresholding), which requires that the features of interest lie in the same intensity band, and be separable from the background. Special methods are available for separating tangent objects in a formal manner. Once segmented, there are various approaches to extracting further information. One is to walk the perimeter of each defined object, calculating such things as area, perimeter, caliper diameters, and centroid ‘‘on the fly’’, pixel by pixel. Polygonal representation of the object can then be obtained by selecting a subset of the perimeter points. The convex hull (the polygon formed by a rubber band stretched around the object) requires a fast processor but is sometimes useful. Since pixel values are only numbers, it is possible to perform modulo-narithmetic on them, where n is the pixel depth. More useful is the ability to perform multi-image arithmetic, such as averaging the results of many video frames to reduce camera noise. The motion of a contrasting object can be followed, producing multiple time-lapse images in a single frame, for motion

9

Image Recognition

299

studies. Uneven illumination and mottling can sometimes be reduced by subtracting a blurred image from a sharp one. Background images can be stored for later subtraction from a series of data images. A final method of image preprocessing involves transformations of the entire image, of which the Fourier transform, described below, is the most generally useful. Useful books on the related but less ambitious subject of video microscopy are Inoue´ (1988) and Russ (1990).

9.3 Identifying Phytoplankton Classical image analysis works by discarding information until what remains can be dealt with by the tools available. We illustrate this process by comparing the information content of several approaches to identifying a phytoplankton cell. In Fig. 9.3 we show a grey-scale image which represents what the eye processes along with the outline of the binary image which is usually what a computer processes, as well as most of the parameters which can be automatically measured by machine, and a taxonomic description of the organism in question. Summarizing, the information content of the 4 descriptions can be ranked as shown in Table 9.1. Despite its numerical nature, such an exercise compares apples with oranges but is nonetheless instructive. The last line of Table 9.1 reveals how much information is thrown away by the abstractions, which are made by contemporary algorithms. What is more difficult is to estimate the amount of information used by the retina and cortex; this information is certain to be more than the total available from any of the components of Fig. 9.3. Even for the grey-scale image, a good deal of peripheral information is necessary to determine which aspects of the picture are relevant. The total relative information used by the retina and cortex (Felleman and Van Essen 1991; Finkel and Sajda 1994) might be many times larger than anything indicated in Table 9.1.

Estimated total bits Estimated useful bits n Comment on bit count n Shannon information = (n ln n) Relative information

Table 9.1 Relative information in Fig. 9.3 Greyscale (A) Contour (B) Image analysis (C)

Taxonomy (D)

524,288 130,000

34,864 34,864

171 1200

2160 15,000

75% of image is background 1.5106

Image can be reconstructed exactly 360103

Implied definitions

Implied definitions

8.5103

144103

176

42

1

17

300

T.T. Noji and F. MacIntyre

Little of this additional information is available to the image analyzer, as indicated in Table 9.1-C. The formal (Shannon) information is smaller and less reliable than the true amount for parts (C) and (D), because both of these descriptions assume additional information. In (C), one is expected to know what the various numbers signify; in (D), the words dinoflagellate, furca, armor, antapical, and horns imply awareness of special contextual definitions. The apparent mismatch between forward and right implies knowledge of the standard orientation of such specimens. We have multiplied the local information by 7 as an estimate of the true information conveyed by (C) and (D). Even so, the disparity between the information in the grey scale image and the numbers available from image analysis gives an honest measure of the challenging nature of the problem. An image-analysis (IA) system usually reports more parameters than we show in Table 9.1-C). Among these are measurements such as Length, Breadth, Equivalent Circular Diameter, Equivalent Circular Area, Convexity, Aspect Ratio, Smooth Perimeter, ‘‘Surface Area’’, and ‘‘Volume’’, but with the exception of a few more Feret (caliper) diameters at other angles, these measurements are derived by manipulating the few primary measurements listed in Table 9.1-C, and contain no new information. (There is also a small set of measures derived from local curvature, which are hard to name colloquially because they eye does not easily see what is being measured. They are also time-consuming to calculate, and at today’s pixel resolution, tend to be quite jagged and noisy.) Nevertheless, it is possible to obtain automatic separation of several phytoplankton species in mixed samples. But be warned: This is easier to do for some species mixes than others (Estep and MacIntyre 1989), because of the limited number of parameters available for comparison. The task is considerably less tractable, if the system is requested to separate natural communities containing many similarly shaped species and detritus, and the problem of variable shapes (from rotation of the subject or disposition of mobile body parts) remains challenging. Still, impressive progress has been made for the automatic identification of phytoplankton (Sieracki et al. 1998). In combination with other measurements, such as pigment analyses, image recognition techniques have been shown to be more effective (e.g. See et al. 2005) As this is written, Michael Sieracki2 is able to automatically identify up to about 70% of plankton taxa from the Gulf of Maine (personal communication). Still, researchers are faced with the need to manually identify phytoplankton, if 100% coverage is the goal. An ambitious international research project recently conducted gives an indication of the degree of success, one might expect, if the sample set is artificially constrained. In his report on the European ADIAC project (Automatic Diatom Identification and Classification), du Buf reports that results were excellent, and recognition rates were well above 90% (du Buf and Bayer 2002). 2

Michael Sieracki is the developer of the FlowCAM, an automatic plankton identification instrument.

9

Image Recognition

301

Dr. Kenneth Estep, co-developer of both Zeus3 and the de facto worldstandard4 taxonomic software Linnaeus II, sees these 2 projects as the basis of a more feasible hybrid approach for the foreseeable future.5 Linnaeus II is an interactive multimedia taxonomy system which incorporates text (species descriptions, synonyms, references, etc.), pictures (from electron micrographs to satellite images), sounds and sonograms, videos of behavior, distribution maps, etc. in a smoothly functional point-and-click interface. The goal is 2 million species described in a common format, so that one can obtain immediate answers to questions such as ‘‘what arthropods and grasses are common to the colder Argentine Pampas and the Russian Steppes?’’. Figure 9.4 shows the result of such a search, and—since the organism in question is relatively complex—suggests the difficulties of asking software to make a decision about an object of this nature. In the envisioned hybrid system—the next logical step in morphological taxonomy—the IA portion would produce images of the organism in question, and extract numerical parameters there from. An analysis system, something like Linnaeus II, running on the same front-end machine, would access a catalog of relevant images, plus taxonomic data keyed to the numerical IA parameters. It would scan the IA parameters automatically and search the data files for matches, presenting the viewer with a graded list of the most probable species. The final pattern matching of object and stored images would then be performed by the only system so far up to the task—the human brain—but a brain for which computers have done what they do best, namely, the time-consuming measurements and data searches.

9.4 Identifying Fish Stocks from Otolith Shape Differences A commonly available IA operation is transformation of an image from the spatial domain to the frequency domain via a (Fast) Fourier Transform, or (F)FT. Algorithm writers take great pride in writing an efficient FFT, because it lends itself well to clever coding, and the results not only look like black magic

3 The ZeusTM image analysis system was the first Macintosh-based user-friendly GUIoriented IA system. Its basic flaw was that it was based on the first available board-level IA system, which despite earnest appeals to the manufacturer, never appeared in a Macintosh version. Zeus was a hybrid system in which an IBM clone-on-a-board worked with a Macintosh front end. Both pieces worked well, but the combination was one computer more expensive than it needed to be, and the 2 uncoordinated CPUs spent an inordinate amount of time asking each other, ‘‘You done yet?’’ and responding ‘‘Go away. I’m busy!’’. 4 Arguably true at the time of publication of the first edition of this chapter. 5 At the time of writing of the first edition of this chapter, Dr. Estep was terminally ill and was not be able to pursue this project. He asked us to describe his vision in the hope that someone would recognize its merits and implement it.

302

T.T. Noji and F. MacIntyre

Fig. 9.4 Screen shots of Linnaeus II’s search module IdentifyItTM. Top: A match was requested to the test pattern at the lower right (the window needs to be scrolled to show all 22 items in the search pattern); the percentage of matching characteristics is shown in the leftmost column. Bottom: Clicking on a species name in the left window in the upper view will bring up the description and picture shown here. (Many other aspects of the system are not shown.) The illustrations are taken from an existing CD and are not truly relevant to the discussion of future possibilities. The point is that as the program stands, the search pattern could include IA parameters as easily as not. The more useful development discussed in the text would automatically search on IA parameters

9

Image Recognition

303

but are often useful. The best way to obtain a useful grasp of what is meant by ‘‘spatial’’ and ‘‘frequency’’ domains is to play with an IA system which performs FTs. Even if the arithmetic remains obscure, the process becomes familiar and one can begin to predict the effect of changes. However, short of experimenting with IA systems, we offer the analogy, below. Basically, the FT of an image is like a sonogram of a sound, in that it displays frequency and intensity. In fact, a sonogram is an FT, from the sound’s time domain to the acoustical frequency domain. The ‘‘frequency’’ of an image is not that of temporal repeats, but of spatial repeats: a picket fence has a strong signal at the spacing and orientation of the pickets. Some may know that any time-varying signal can be reconstructed by summing the proper assortment of pure sine waves with the correct amplitudes and phases. The FT is the device which deduces the correct amplitudes and phases. There are, however, severe limits on this process, of which the most troublesome is that the signal must be repetitive for the process to be valid. Such problems can be addressed mathematically (e.g. Rosky and Zahn 1972; Bloomfield 1976; Beauchamp 1987; Estep et al. 1994; Lombarte et al. 2006). Figure 9.5 illustrates the process. Figure 9.5C is a semi-logarithmic plot of the amplitudes of the Fourier coefficients of the contour of Fig. 9.5B. Figure 9.5D shows the points (some of which do not lie on the reconstructed contour) of Fig. 9.5B which were used in the Fourier analysis: either more or fewer points might have been chosen. The slanted line in 5B and 5D is machine constructed to run from the centroid of the figure, through the selected contour point closest to the centroid (or, if desired, toward the farthest point), and out to the mean radius of the contour. In Fig. 9.5C, the vertical lines span the distance between the sine (square dot) and cosine (other end) terms of the coefficient pairs. In this example, the system initially computed 16 coefficient pairs (determined either by operator request, or by the number of contour points selected). The system allows a choice of sine and cosine coefficients, or equivalent amplitude and phase coefficients. The latter set allows trivial rotation of complex objects, while the amplitudes form a rotation-, translation, and scale-independent descriptor of the contour. The projection on the left side of the otolith is not well reconstructed at this resolution. Nevertheless, such analyses are proving to be useful in shape analysis to identify fish stocks, which has been applied manually with moderate success (Begg and Brown 2000; Lombarte et al. 2006). The FT is an information-preserving process and if all coefficients are retained, will duplicate the original contour as exactly as digitization permits. The advantage of FTs is 2-fold. The first is conceptual: the FT converts a contour from an arbitrary visual object into a mathematical object, whose properties are well known and can be massaged by standard algorithms (e.g. Fredman and Goldsmith 1994). The other advantage is data compression, arising from the possibility of discarding the high-frequency (noise) terms and using only the significant remainder for further analysis.

304

T.T. Noji and F. MacIntyre

Fig. 9.5 Original grey-scale image of a fish otolith (A), which has been automatically reduced to its binary outline (B); FT analysis of the latter yields 16 complex coefficients (C), 7.5 of which have been reconstructed into a 2-D presentation in (D). Analysis was performed using a Zeus image analyzer

Further analysis treats the Fourier-coefficient data as a nearly singular rectangular matrix,6 as indicated graphically in Fig. 9.6. The only algorithm 6 The specimens are very similar, so their FTs will be similar. A matrix with two nearly identical lines is nearly singular and will cause many matrix-reduction methods to break, sometimes without telling you. The rank of a matrix is the maximum count of rows and columns; rounding errors and fuzzy data make rank determination a non-trivial exercise; the SVD ignores all problems and returns the best possible answer (although it may take a little longer than other methods which do not return the right answer). The SVD is most easily accessible via a button on the HP-48GX calculator.

9

Image Recognition

305

Fig. 9.6 A rank-deficient matrix rotated by the SVD. Three data points (the squares) lie inside an origin-centered ellipse (solid line) with major axis at angle V. Using the singular value decomposition, this ellipse can be rotated into US, which lies along the axis (dotted data), and then squeezed to lie inside the unit circle, with coordinates U. This is the basis of all Cartesian methods of ordination analysis. If relative and not absolute values of the data are of interest, the matrix may be centered (and normalized) before applying the SVD

which can deal with such difficult data reliably is the singular-value decomposition, or SVD (Golub and Van Loan 1983). The SVD converts the data matrix A into a set of 3 matrices such that A = USV’, where U contains eigenvectors, S eigenvalues, and V rotators, which can be converted into the less comprehensible ‘‘scores’’ and ‘‘loadings’’ of any of the usual varieties of ordination or factor analysis (examples include U, US1/2, or US for the ‘‘score’’ and V, VS1/2, or VS for the ‘‘loading’’, respectively). If fish breeding stocks cooperated by developing consistent differences in outline, however complex or non-obvious to the eye, such an ordination analysis of the components of an FT might allow detailed and automatic classification into different stocks. An aspect of matrix decomposition and ordination analysis which tends to get lost when computers do the arithmetic is that computers do not simultaneously do the science. Transforming a matrix adds nothing to the data, and science does not enter until the results of the transformation can be given meaningful names. Ordination analysis will return an impeccably correct ‘‘Factor 1’’ for any random matrix. Fortunately, we are not here interested in the ‘‘meaning’’ of the results of ordination analysis: for stock identification, it would suffice if we could identify a mean contour for a species, and a set of small, consistent modifications for each breeding stock. In any case, FTs are instructive, a great deal of fun to explore, and there will almost certainly be times when they are useful.

306

T.T. Noji and F. MacIntyre

Notably, nearly everything which has been said about otoliths can also be said about fish scales. The one major exception is that scales, being external, are subject to mechanical wear, which may confuse the issue. On the other hand, they are numerous, easily accessible, and sampling is not lethal, making it easy to construct an experimental time series of scale development which can be related to environmental conditions. Fin rays and sectioned teeth offer additional opportunities for pattern analysis.

9.5 Age Determination and Growth-Rate Measurements of Fish The growth of fish is regulated by physiology as well as food abundance and quality. Specifically, daily growth cycles are a function of diurnal patterns in feeding and metabolism, while annual cycles are dependent primarily upon seasonal oscillations in food supply. At least for some fish taxa, e.g. herring Clupea harengus and cod Gadus morhua, these cycles are evidenced in otolith microstructure, i.e. daily growth in larvae and annual growth in adults are recorded as layers of calcium and organic material, which appear as rings in thin-sectioned preparations. Distance between any two rings or increments is relative to the growth of the host during the time interval corresponding to production of the two increments. This fact is extremely convenient for fisheries biologists interested in the growth dynamics of fish. To determine age, the basic if oversimplified procedure is to sit down in front of a microscope, count the increments, and convert one-to-one to the appropriate calendar unit, i.e. days or years. Growth can be estimated by measuring the distances between increments. Consequently, growth rate is reflected by the ratio of inter-incremental distance to age. The problem lies chiefly in sample preparation and subjectivity during manual selection of rings. This can be alleviated through appropriate staining and etching techniques (Secor et al. 1991), which can enhance visibility of rings. Still, intercalibration exercises (Campana and Moksness 1991) recommend stringent protocol particularly with reference to researcher subjectivity, whereby the weakest points of otolith reading were reported to include otolith preparation, proper calibration of the optics and interpretation of the image— items largely determined by the operator. A properly calibrated video/imageanalysis system aids at least in reducing subjectivity and increasing accuracy and precision; effectiveness is correlated with image magnification (SEM if necessary) and quality. Ironically, in poorly prepared samples and/or using suboptimal optics and microscopes, identifying increments in otoliths—just like identifying faces—may rely as much on indescribable instantaneous flashes of ‘‘a-ha’’ resulting from years of experience, as on detecting well defined alternating black and white bands. Thus for the time being otolith reading remains a pragmatic partnership between high-tech image-analysis systems and experienced laboratory assistants.

9

Image Recognition

307

Otolith-ring analysis is another place where FTs can help. Two-dimensional (2-D) FTs can selectively emphasize contrast at the spatial frequency of the rings. The less processor-intensive 1-D FT can be used to count rings in an unbiased manner. Figure 9.8 converts a radial density scan (Fig. 9.8A) of the otolith of Fig. 9.7 into an FT whose coefficients are shown in Fig. 9.8B. The dotted box in the upper left is the operator’s way of selecting which coefficients are included. In this case, it is possible to omit the first few coefficients, which has the effect of removing long slow variation and waves across the transect. The ‘‘%Power’’ number indicates what percentage of available information is included inside the dotted box. The table below Fig. 9.8B is the initial part of an exportable spread-sheet output. Both Sine-Cosine and Amplitude-Phase values of the coefficients are shown. The zero-th coefficient is the only one which does not occur with both positive and negative frequencies, shown to the left and right of the ‘‘+n-’’ label in the heading. In Fig. 9.8C an inverse FT has reconstructed the density scan. A constant base line at grey-level 20 has been set; other choices are possible, such as a base line that is tangent to the deepest valleys. Peak heights and areas are measured from the baseline. The table below the diagram is again the start of an exportable spreadsheet. The machine will count either peaks or valleys, whichever is deemed appropriate. Pannella’s (1971) recognition of daily increments in otolith microstructure is thought to be analogous to the discovery of ancient Egyptian hieroglyphs in the sense that both show consistent and repetitive patterns reflecting processes (Secor et al. 1991). The hieroglyphs could not be read until the Rosetta Stone, which contained the same text in hieroglyphic, demotic, and Greek, was found and interpreted. Following Pannella’s breakthrough, modern-day fisheries

Fig. 9.7 Otolith from a c. 60-day-old herring larva showing daily growth rings. ¨ Photo by Oystein Paulsen, Institute of Marine Research, Marine Biological Station in Flodevigen, Norway

308

T.T. Noji and F. MacIntyre

Fig. 9.8 The same otolith shown in Fig. 9.7 was analyzed for number of daily rings and incremental distances. Analysis was conducted along a transect (thick white line in Fig. 9.7) from the center to outer edge of the otolith. Relative values for grey-level intensity (A) were converted by FFT (B) to mathematical derivations. From the inverse FFT reconstruction (C), valleys—corresponding to dark rings—were identified and labeled. The first three sets of numerical data constituting B and C are shown below the respective curves. Of particular interest to fisheries biologists are the number of valleys and inter-valley distances (Sep) reported in C. We note that by selecting 2 or more lines (thin white lines in Fig. 9.7) transecting less ‘‘noisy’’ parts of the otolith rings, the accuracy of otolith reading could be increased; measurement of a subset of common rings along each line would be necessary in order to intercalibrate incremental distances. All analyses were performed automatically using a Zeus image analyzer and figures A, B and C are screenshots of reported findings

9

Image Recognition

309

biologists are now challenged with constructing and deciphering their own Rosetta Stone.

9.6 Identifying Species from Echo Sounders The possibility of using image recognition on sonar records of fish schools is an approach which is being developed and has demonstrated some success (e.g. Denny and Simpson 1998; Horne and Jech 2005; Jech et al. 2005). The information available in such a display is an indication of the shape of a fish school, and intensity of return. School shape is to a certain extent species specific; intensity of return is both species and size dependent. Acoustic data from fish are proving to be useful for estimating the abundance as well as taxonomic identification of the target animals. It has now been recognized that a distinct advantage of acoustic recordings potentially lies in the primary form of the data, which are target-strength signals reflected back to the source by e.g. fish. Statistical correlation of such data with specific sources is akin to shape analyses on FTs of images. On the other hand, image analyses on graphic reconstructions of acoustic data are manageable, while for most people juggling target strengths over time and depth is too abstract a concept for true insight. Contemporary research in this field therefore considers both aspects during exercises in pattern recognition of acoustic data. Neural networks to support the interpretation of acoustic recordings are promising tools.

9.7 Needed Developments 9.7.1 Flat-Field Illumination Attempts at computer-aided microscopy have revealed two areas in which microscope development could help appreciably. The first is flat field illumination. The human eye is very good at noticing out-of-focus areas in a microscope field, so microscope builders have worked hard to design lenses which deliver a field in focus out to the very edges. This is a remarkable achievement, made possible only by computer-aided lens design. But the human eye is also very good at compensating for smoothly changing illumination, so there has been little pressure to develop substage optics which deliver uniform illumination in the object plane. Machine analysis of an image in which circular shading is completely ignored by a human observer may show that the edges of the field are actually 40 grey levels darker than the center. This is of no concern during visual analysis, but presents the computer with a severe problem.

310

T.T. Noji and F. MacIntyre

The principal information which the computer has to work with (even in a color system) is the absolute intensity of illumination. It is almost trivial to explain to silicon that you are interested in grey levels between 78 and 93, and have it highlight these pixels, count them, and tell you many things about their area and shape. It is much more difficult to ask it to do the same for areas which are 15 grey levels different from the background around their periphery and essentially impossible if the area in question is surrounded by objects of different intensity. The problem does not lie in finding 15-gray-level steps (over some agreed-upon distance), but in explaining the concepts of ‘‘object’’ and ‘‘background’’, which are properties of the whole image, to a device which deals only with individual pixels. Anyone thinking of purchasing a microscope for video microscopy should add illumination flatness to the list of preferred specifications. A corollary to the need for illumination flatness is that the substage illumination path must be adjusted properly. Experience shows that this is seldom the case. Instructions for obtaining Kohler illumination (the usual desideratum) come with most good microscopes. Nevertheless, in the course of 5 years of demonstrating image analysis7 with other people’s microscopes, we never once found a microscope in use which was properly adjusted. This inattention results directly from the eye’s indifference to shading. Consequently, the first step in a demonstration of what is possible with video microscopy is always to align the user’s substage optics. In a properly designed and adjusted microscope, there would be no need for clumsy data manipulation to correct for unevenly lit backgrounds.

9.7.2 Instantaneous, Calculation-Free FTs The brute-power algorithmic approach to FT works, after a fashion, but there is a more elegant alternative to obtaining frequency-domain images of microscopic samples. Between the objective and ocular of every microscope lies an optical plane in which the image itself is the Fourier transform of the object. Thus, perhaps with no more than a properly placed beam splitter and a sidemounted camera, it is possible to design a video microscope which would deliver both spatial- and frequency-domain information simultaneously and instantaneously, completely bypassing the need for mathematical manipulation if all that is required is frequency-domain information. Often, what one wants to do is modify the frequency-domain to improve the spatial-domain image. In this case, transformation back into the spatial domains is necessary, and it is possible to imagine a reflexive system in which all transforms are performed optically, with changes made to the image in one domain appearing instantaneously in the other. The ordinary tools—eraser, 7

As noted by Drs. MacIntyre and Estep.

9

Image Recognition

311

Fig. 9.9 Illumination shading (A) of marine detritus photographed with a standard CCDvideo camera and research-grade microscope. Grey-level intensity (B) was measured along the line indicated in A; analysis was conducted with the Zeus image analyzer. This is a particularly good separation of object and background: in less fortunate cases, background shading occurs across the object, making it impossible to separate object from background cleanly by grey-levels alone. (Note a tendency to 8-pixel-wide vertical bands. This is not caused by microscope illumination, but is a power-supply problem on early models of the IA board employed.)

Fig. 9.10 Sketch of an image analysis system helpful for pattern recognition in both spaceand frequency-domains. Transformation from one domain to the other is performed optically and in real time, rather than digitally after a delay. The ‘‘imager’’ might be related to the active LCD displays of contemporary notebook computers, with the same resolution as solid-state cameras. One function of the Control Bus is to switch the appropriate video memory to the Imager Control between frames

312

T.T. Noji and F. MacIntyre

pencil, brush, spray can, lasso, as well as more sophisticated processes—of a computer software drawing program could then be used on either display and the results simultaneously observed on the other. Moving a drawing tool onto one or the other monitor would readjust the flow of information properly so that the modified view was routed to the camera and its modified transform displayed as modifications were made in the original. This sort of immediate feedback will make FT microscopy an intuitive and useful tool for marine biologists. We note that both CCD cameras and active-matrix displays now include non-linear devices (transistors) at each pixel, offering a greatly improved dynamic range. In the near future it should be possible to build a ‘‘transform tube’’ like that shown centrally in Fig. 9.10, which will perform both direct and inverse FTs in a single video frame time. The feedback loop between transforms implicit in Fig. 9.10 might have curious properties such as rapid degradation from pixelization (binning of area) and digitization (binning of intensity). There might be a need for additional circuitry next to the camera and imager controls, where we have thoughtfully provided empty boxes.

9.8 Problems and Promises 9.8.1 Penrose The British mathematician Roger Penrose has a fine geometrical intuition. It was he who discovered the tribar of Fig. 9.11 and the Penrose Staircase that Escher used as the basis of Waterfall and Ascending and Descending, as well as a way to tile a plane with local, but not long-range, 5-fold symmetry. He explores his philosophical suspicion of ‘‘strong AI’’—the idea that one can mimic the processes of the human brain in algorithmic silicon, and the great white hope of the pattern-recognizing community—in The Emperor’s New Mind. His arguments range from the non-local paradoxes of quantum mechanics and the meaninglessness of simultaneity under special relativity, through a sketch of what the elusive ‘‘Correct Quantum Gravity’’ theory might be like, before drawing all of this into his conclusion.

Fig. 9.11 The Penrose tribar. What would a pattern-recognizing image-analysis system make of this?

9

Image Recognition

313

Disappointing to those who look to computers to do difficult things, Penrose concludes that judgment—the basis of our ability to recognize patterns—is no more algorithmic than is our ability to discern beauty. This may not mean that silicon is forever barred from recognizing the faces of its friends, but does suggest that it will not do so with designs based on the current Von Neumann (algorithmic) approach.

9.8.2 Hope for the Future In the third edition of his critique of artificial reason, (Dreyfus 1993) observes that in 20 years the book has changed from a controversial position, described by a critic as ‘‘short-sighted and unsympathetic’’, to a requiem for one of the great dreams of the last half of the 20th century. Fifty years of effort have lead to a degenerating research program, one that started off with a bang with Newell and Simon and their chess-playing machines in the late 1950s, and fizzling out in the 1990s with the unheralded demise of the grandiose Japanese Fifth-Generation Project. Dreyfus distinguishes 4 classes of intelligent activities (Dreyfus 1993, p. 292.) We abstract some points relevant to image recognition in the form of Table 9.2. By this classification, the minimally useful level of pattern recognition for fisheries appears to be the ability to deal with Complex-formal activities, recognizing complex patterns in the presence of noise. The state of the art is algorithm-based Simple-Formal analysis. True utility comes only at the Nonformal level, for which there are neither programs nor ideas about how to invent them. The state of the art may reach Complex-formal in the foreseeable future. Indeed, some time ago Feigenbaum and Feldman (1963) waxed enthusiastic about this possibility: In terms of the continuum of intelligence, the computer programs we have been able to construct are still at the low end. What is important is that we Table 9.2 Dreyfus’s classification of pattern-recognition activities I Associative II Simple- III Complex-formal IV Nonformal formal Learning mode

Learned by rote

Learned by rule

Learned by rule and practice

Example

Maze problems, trial and error Decision tree

Reading typed page

Recognition of complex patterns in presence of noise

Program Algorithm Search-pruning type heuristics Italics: Where we are today. Boldface: Where we want to be.

Learned by perspicuous examples Recognition of varied and distorted pattern None

314

T.T. Noji and F. MacIntyre

continue to strike out in the direction of the milestone that represents the capabilities of human intelligence. Is there any reason to suppose that we shall never get there? There is none whatever. Not a single piece of evidence, no logical argument, no proof or theorem has ever been advanced which demonstrates an insurmountable hurdle along the continuum (Feigenbaum and Feldman 1963).

9.9 Summary We have taken a rather negative view of the accomplishments of image recognition in fisheries, partly to offset the still prevalent impression of computer omnipotence. The human brain retains its superiority in this area. The most promising approaches seem to be those which attempt to apportion tasks properly between the machine (impossible to bore with rote processes) and the brain (clever at seeing gestalts but not very quantitative). We outline a hybrid approach. Capabilities needed for successful image recognition improve continually, and there is always room for new ideas. We passed rapidly over the related field of image analysis, indicating only in the broadest terms the sort of preprocessing that can be applied to enhance features of images prior to analysis for pattern recognition, and the features which can be measured. These are well covered in books on video microscopy. Of specific interest to fisheries are identification of fisheries-relevant taxa (e.g. target fisheries species, prey and predators); extraction of information (e.g. from otoliths) for stock identification, aging, and growth history; and the application of image recognition to sonar data from fish schools to estimate stock size and to identify taxonomic composition and behavior. Acknowledgments This chapter is dedicated to Dr. Kenneth Estep, marine protistan ecologist, researcher and identifier of the harmful algae Chrysochromulina spp., pioneer of computerized taxonomic identifications, co-founder of the Zeus image-analysis system, inspiration for the first edition of this chapter, and good friend to the authors. Dr. Estep passed away of AIDS in 1995 at the age of 42.

References Aha D, Bankert R (1997) Cloud classification using error-correcting output codes. Artificial Intelligence and Applied Natural Resources in Agricultural and Environmental Sciences 11(1):13–28 Apte C, Damera F, Weiss, SM (1994) Automated learning of decision rules for text categorization. Information Systems 12(3):233–251 Azzopardi P, Cowey A (1993) Preferential representation of the fovea in the primary visual cortex. Nature 361:719–721

9

Image Recognition

315

Beauchamp RG (1987) Transforms for Engineers. Oxford University Press, Oxford. Begg GA, Brown RW (2000) Stock identification of haddock Melanogrammus aeglefinus on Georges Bank based on otolith shape analysis. Transactions of the American Fisheries Society 129(4):935–945 Bloomfield P (1976) Fourier Analysis of Time Series. John Wiley, New York Burr DC, Morrone MC, Ross J (1994) Selective suppression of the magnocellular visual pathway during saccdic eye movements. Nature 371:511–513 Campana SE, Moksness E (1991) Accuracy and precision of age and hatch date estimates from otolith microstructure examination. ICES Journal of Marine Science 48:303–316 Denny G, Simpson P (1998) A broadband acoustic fish identification system. The Journal of the Acoustical Society of America 103(5):3069 Desimone R (1991) Face-selective cells in the temporal cortex of monkeys. Journal of Cognitive Neuroscience 3:1–24 Dreyfus HL (1993) What Computers Still Can’t Do. MIT Press, Cambridge, 354 pp du Buf H, Bayer MM (eds.) (2002) Automatic Diatom Identification. World Scientific, Series in Machine Perception and Artificial Intelligence, Vol. 51 Estep KW, MacIntyre F (1989) Counting, sizing, and identification of algae using image analysis. Sarsia 74:261–268 Estep KW, Nedreaas KH, MacIntyre F (1994) Computer image enhancement and presentation of otoliths. In Secor DH, Dean JM, Campana SE (eds.) Recent Developments in Fish Otolith Research. U. South Carolina Press, Columbia, SC Feigenbaum EA, Feldman J (eds.) (1963) Computers and Thought. McGraw-Hill, New York, p. 8 Felleman DJ, Van Essen DC (1991) Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex 1:1–47 Finkel LH, Sajda P (1994) Constructing visual perception. American Scientist 82:224–237 Fredman ML, Goldsmith DL (1994) Three stacks. Journal of Algorithms 17:45–70 Golub GH, Van Loan CF (1983) Matrix Computations (The Johns Hopkins Press, Baltimore MD) 475 pp Gould SJ (1994) Faces are special. The Sciences 34:36–37 Horne JK, Jech JM (2005) Models, measures, and visualizations of fish backscatter. In H. Medwin (ed.) Sounds in the Seas: From Ocean Acoustics to Acoustical Oceanography. Academic, New York, pp. 374–397 Inoue´ S (1988) Video Microscopy. Plenum, New York Jech JM, Foote KG, Chu D (2005) Comparing two 38-kHz scientific echo sounders. ICES Journal of Marine Science 62:1168–1179 Le Cun Y, Boser B, Denker J, Hendersen D, Howard R, Hubbard W, Jackel L (1989) Back propagation applied to handwritten zip code recognition. Neural Computation 1(4):541–551 Lombarte A, Chic O`, Parisi-Baradad V, Olivella R, Piera J, Garcı´ a-Ladona E (2006) A webbased environment from shape analysis of fish otoliths. The AFORO database. Scientia Marina 70:147–152 Mitchell WJ (1994) When is seeing believing? Scientific American 270:68–73 Morris D (1994) The Human Animal. BBC Books, London O’Carroll D (1993) Feature-detecting neurons in dragonflies. Nature 362:541–543 Ornstein R, Thompson RF (1984) The Amazing Brain. Houghton Miflin, Boston Ou G, Murphy YL (2007) Multi-class pattern classification using neural networks. Pattern Recognition 40:4–18 Pannella G (1971) Fish otoliths: daily growth layers and periodic patterns. Science (Wash., D.C.) 173:1124–1127 Rahin MG (1992) A self-learning neural-tree network for phone recognition. In Mammone RJ (ed.) Artificial Neural Networks for Speech and Vision. Chapman and Hall, London, pp. 227–239

316

T.T. Noji and F. MacIntyre

Rosky A, Zahn B (1972) Fourier descriptions for plane closed curves. IEEE Transactions on Computer, C-21 3 Mar 72:269–281 Russ JC (1990) Computer-Assisted Microscopy: The Measurement and Analysis of Images. Plenum, New York Secor DH, Dean JM, Laban EH (1991) Manual for Otolith Removal and Preparation for Microstructural Examination. The Belle W. Baruch Institute for Marine Biology and Coastal Research 1991–2001, 85 pp See JH, Campbell L, Richardson TL, Pinckney JL, Shen R, Guinasso NL (2005) Combining new technologies for determination of phytoplankton community structure in the northern Gulf of Mexico. Journal of Phycology 41:305–310 Sieracki CK, Sieracki ME, Yentsch CS (1998) An imaging in-flow system for automated analysis of marine microplankton. Marine Ecology Progress Series 168:285–296 Solomon JA, Pelli DG (1994) The visual filter mediating letter identification. Nature 369:395–397 Srinivasan MV, Zhang SW, Rolfe B (1993) Is pattern vision in insects mediated by ‘‘cortical’’ processes? Nature 362:539–540 Young ME (1992) Objective analysis of the topological organization of the primate cortical visual system. Nature 358:152–154

Chapter 10

Visualization in Fisheries Oceanography: New Approaches for the Rapid Exploration of Coastal Ecosystems Albert J. Hermann and Christopher W. Moore

10.1 Introduction Powerful new measurement technologies and numerical models are rapidly expanding our three-dimensional knowledge of the ocean and its biota. The effective visualization of this expanding database is not trivial, and its use in fisheries oceanography presents special challenges. Oceanic circulation and the life experience of marine fish are irrefutably three dimensional; despite this fact, most ocean data and model output has traditionally been viewed using twodimensional maps. Furthermore, a thorough analysis of any fish population ought to include both: (1) the local world experienced by representative fish (a pseudo-Lagrangian record of events, as in spatially explicit individual-based models); (2) the changing regional habitat of the population (an Eulerian description of the regional environment, as in spatially explicit lower trophic level models). In this chapter we survey visualization approaches, from the simple to the complex, which can facilitate such multifaceted analyses by researchers and managers. In particular we emphasize that modern approaches to scientific visualization include immersive techniques: by taking advantage of human binocular vision, these allow the user to experience and interact with measured and modeled features as ‘‘virtually real’’ objects in our three dimensional world. The experience is not unlike holding a part of the ocean basin in one’s hand, turning and examining it from different angles. In fisheries oceanography, such modern approaches can rapidly and effectively reveal: (1) the structure of spatially patchy prey fields, such as phytoplankton; (2) threedimensional flows near topography; (3) spatial tracks of individuals and the prey fields they experience along those paths. Recent developments allow immersive visualization through web servers, giving scientists the ability to collaboratively ‘‘fly through’’ three-dimensional data stored half a world away. We explore what additional insight is gained through immersive visualization, and describe how scientists of very modest means can easily A.J. Hermann (*) Joint Institute for the Study of the Atmosphere and Ocean, University of Washington, Seattle, WA 98115, USA

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_10, Ó Springer ScienceþBusiness Media B.V. 2009

317

318

A.J. Hermann and C.W. Moore

avail themselves of low-cost immersive technology. Further examples and resources are available at http://www.pmel.noaa.gov/people/hermann/vrml/ stereo.html.

10.1.1 3D Models vs. 2D Graphics All fisheries oceanographers have a compelling interest in developing clear ways of visualizing their data. Modern observational networks generate huge volumes of oceanographic data; numerical models generate even larger volumes of output (whether this output should strictly be called ‘‘data’’ is a semantic argument, but for our purposes we will refer to it as such). How to visualize these enormous data files? Two-dimensional plots along horizontal or vertical planes have served oceanographers nobly in the past three centuries (from Ben Franklin’s charts of the Gulf Stream to the present) but fail to capture many of the subtleties present in even twodimensional data sets. Consider the bathymetry of the ocean, which contains both large and small-scale features. Typically the large-scale features have a greater amplitude than the small-scale features. A simple contour plot of the ocean floor, with broad contour levels chosen based on the largest amplitude signals, will hence reveal the large scales but obscure the small (or alias them into the larger scales; see Fig. 10.1). By comparison, if we render the bathymetry as a surface in three-dimensional space, we reveal both large and small scales at once. In effect, a rendered surface is equivalent to having an infinite number of contour levels. In modern times it is easy to render such a three-dimensional surface and examine it on a computer monitor, or to print the result on flat page. We have seen many such illustrations in popular scientific journals. However, consider how primitive even this technique really is. Billions of years of evolution yielded our ability to perceive the depth of an object (a talent shared with many other species); we can easily tell when an object is further away from us, rather than simply smaller. We possess this ability because we have two eyes and a brain to combine the information from each (stereo vision). Viewing a three-dimensional surface printed on flat page is like viewing the world with one eye closed – the depth perception which stereo vision affords us is lost. Viewing a three-dimensional object with two eyes from far away – think of a mountain range viewed from a highaltitude jet airplane – is really not much different from the single-eye view. A much more powerful way to examine a surface or other threedimensional object is to view it ‘‘up close’’ with two eyes, from any angle, as if you were holding the object in your hands. With two eyes you clearly see which parts of the object lie in front of or behind other parts, and clearly distinguish up from down, left from right, and forward from

10

Visualization in Fisheries Oceanography

319

a

b

c

Fig. 10.1 Contour plots compared with surface rendering of gridded bathymetric data. These data contain significant sampling artifacts which include spurious fine-scale minima and maxima. A standard contour plot with only a few contour levels (a) misses these fine-scale artifacts. A contour plot with a very large number of contour levels (b) reveals many local features, but obscures their amplitude or sign. A rendered surface using all data points (c) simultaneously reveals both the large-scale structure and the small-scale irregularities. With proper software, this surface can be easily rotated and magnified; here, a VRML viewer was used for this purpose

backward. You cease to confuse small features as being further away simply because they are smaller. Regional planners have for many years utilized widely separated photographs of topography to achieve this ‘‘godlike’’ (albeit static) view of the land. Obviously it is not practical to create a plastic model of every three-dimensional surface one might want to view, and hold it physically. Instead, modern computing hardware and software make it easy to generate images on your computer monitor which are virtually identical in appearance – that is, ‘‘virtual reality’’. It is obviously not ‘‘reality’’ in the sense of what you encounter in your daily life, but rather what you would encounter in your field of view if you were to create a solid model of the surfaces, vectors, and volumes representing various aspects of the data set you wish to analyze.

320

A.J. Hermann and C.W. Moore

The quest for virtual reality is hardly new; 3-D photographs and movies have been with us for many years, beginning in the 19th century with stereo postcards. Their presence in popular culture has waxed and waned over time, including a few less-than-memorable films in the 1950s. The present popularity of IMAX 3D theaters suggests that 3D films have become a significant force in scientific education. What is new in the past decade is the tremendous power and low-cost of modern computing platforms, which allow interactive viewing of 3D worlds. Graphically speaking, what was inconceivable a decade ago has now become routine. Anyone with a low-cost personal computer can generate complex surfaces representing some aspect of a large (megabytes to gigabytes) data set – indeed, many do so on a daily basis as they play computer games. For a modest additional investment one can acquire the ability to view those surfaces in immersive stereo. In effect, one has the ability to fly through and rotate the virtual world with both eyes wide open. It is hard to describe the power of this illusion in words; picture yourself as a giant swimming through the ocean, able to view whatever feature you might choose, up close, far away, and from any angle. Our goal in this realm might be termed ‘‘scientific Virtual Reality for the masses’’. We are seeking ways to allow immersive visualization of numerical model output (or data) through inexpensive methods, powerful enough to handle large worlds and networked in a way which allows access to large datasets and collaboration with remote colleagues. We and our colleagues have not yet achieved all of these goals, but have made significant progress. Here, we begin with examples of fisheries phenomena whose illustration can benefit from the use of 3D graphics. We follow this with a brief summary of basic graphic concepts, and then a description of low-cost hardware and software which can be used to generate, explore, and interact with such immersive worlds.

10.1.2 Examples from Fisheries Oceanography Real fish live in a three-dimensional world, with complex fluid velocities and spatial paths. Here we present several examples of 3D features and phenomena relevant to fisheries oceanography, and emphasize how 3D visualization can communicate fundamentally new information, relative to simple 2D visualization techniques. Features of interest include the physical environment (e.g. seafloor bathymetry, currents, temperature, nutrients, and passive particle tracks), prey fields (e.g. zooplankton) and fish life histories (e.g. position, life stage, and size). Three dimensional visualization, and immersive visualization in particular, allow the fisheries scientist to quickly examine velocity fields and the animated spatial paths of measured or modeled fish in their native environment. Immersion facilitates the exploration of such data sets by quickly revealing spatial relationships (left from right, up from down, and closer vs. further away).

10

Visualization in Fisheries Oceanography

321

10.1.2.1 Bathymetry Ocean bathymetry contains features on many spatial scales which affect circulation and fish life history. Simple contour plots tend to reveal only the largescale features and hide the smaller ones. Stated more technically, a contour plot may show the global minimum and maximum of a particular area, but hide many significant local minima and maxima. In Fig. 10.1 we compare such simple contour maps with a rendered surface. Note how the rendered surface more clearly revealed many small scale errors (in this particular case, the errors were due to biases of a regridding scheme which had been used to interpolate from irregularly spaced bathymetry data to a regular latitude-longitude grid). If this bathymetry were used ‘‘as is’’ in a numerical model, the errors could have a substantial effect on the results. Here, rendering the data as a surface flagged these errors beforehand.

10.1.2.2 Fish Life History Fish eggs and larvae are essentially planktonic organisms, advected passively by the ocean currents. This advection is supplemented by vertical motions due to buoyancy and directed swimming. In both the real ocean and in numerical models, this can yield complicated three-dimensional trajectories (a.k.a. ‘‘individual life histories’’) through time. We have generated many such life histories in our modeling work (Hermann et al. 1996, 2001, Hinckley et al. 1996, 2001). This 3D information is only partially revealed when only two dimensions are displayed. Colored polygons can be used to represent the third dimension of depth on a 2D contour plot of bathymetry, but fail to give a visceral sense of a complex looping path through three dimensions and simultaneously communicate biological attributes of the fish. It has been our experience that the animation of a full 3D world is a far more instructive way to communicate such complex life histories (Fig. 10.2).

10.1.2.3 Circulation and Hydrography Both real and modeled oceans exhibit motions across a broad swath of spatial scales – indeed the real ocean exhibits scales of motion from thousands of kilometers down to a few millimeters. As a result both velocities and physical scalars such as temperature and salinity exhibit patchiness at many scales. Examples of 3D circulation phenomena familiar to fisheries scientists include coastal upwelling (which provides nutrients to the food chain), tidal motions (which provide nutrients through mixing, and interact with behavior to affect fish migration), and mixed layer turbulence (which affects feeding). We can represent some of these phenomena more effectively through the use of 3D isosurfaces and 3D vectors (see Sections 10.2.1.1, 10.2.1.2 and 10.2.1.3).

322

A.J. Hermann and C.W. Moore

Fig. 10.2 Example of 3D rendering to illustrate modeled walleye pollock (Theragra chalcogramma) life histories near Kodiak Island (in upper right of figure). Grey surface is ocean bathymetry used by the hydrodynamic model; maximum depth shown here (lower right) is 500 m. Blue line indicates coastline. Shape and color of fish indicate size and life stage. This frame, derived from the animation of a 6-month biophysical model hindcast, represents the population in mid-June, 1978. Animation of this world illustrates looping trajectories which would be obscured by simple 2-D graphics. Immersive viewing of this scene provides additional depth cues

10.1.2.4 Prey Fields Many fish consume planktonic species as a major prey item. Plankton fields are subject to the wide range of motions in the ocean; they are further affected by reproduction, predation and death. As a result, plankton fields can in fact be even patchier in space than physical scalars (e.g. Powell et al. 2006); like clouds in the atmosphere, they can be richly patchy in 3 dimensions. Contour plots fail to convey this structure as well as 3D graphics. Indeed, a 2D contour plot of atmospheric moisture reveals far less information (or at least very different information) about the spatial structure of clouds than a simple oblique view from the ground.

10.2 Computer Rendering of 3D Biophysical Worlds Here we describe basic objects rendered in 3D graphics, and the hardware needed to render them. This discussion is not specific to immersive visualization, but the advantages of immersion are noted where relevant.

10

Visualization in Fisheries Oceanography

323

10.2.1 3D Graphics Concepts 10.2.1.1 Isosurfaces An isosurface is the 3D equivalent of a 2D contour. A contour plot exhibits lines connecting points along a 2D surface which all have the same numerical value of some property. The most common example is a topographic map which shows lines of constant elevation of the land surface. Oceanographers frequently use contour maps of sea surface height or sea surface temperature. An isosurface is a surface connecting points which all have the same value of some property in three dimensions. For example, an equatorial oceanographer might wish to examine the surface along which the ocean has a temperature of 108C, to diagnose El Nino dynamics. A fisheries oceanographer could use similar information to examine the habitat of some species of fish which prefers a particular range of temperatures or prey density; the isosurface illustrates the boundaries of that habitat. 10.2.1.2 Volume Rendering In some circumstances the most effective way to represent a patchy 3D field is to render it opaque where values are highest. In nature, we see clouds where the density of water vapor is high; on a computer we may render virtual ‘‘fog’’ (and color it) where a scalar property exceeds a particular value. This is a very natural way to experience a patchy distribution of chlorophyll, for example. In Fig. 10.3 we demonstrate our use of these approaches to explore modeling results for the Gulf of Alaska (Hermann et al. 2008, Hinckley et al. 2008). 10.2.1.3 Vectors Oceanographers frequently represent velocity as a vector – graphically, as an arrow whose length and orientation illustrate the speed and direction of the currents at a particular location. Such arrows are usually drawn to represent horizontal velocity: eastward velocity, u, and northward velocity, v. 3D vectors (u, v, and vertical velocity, w) can be rendered as well. Such 3D vectors can be confusing on a 2D page, as it is difficult to discern whether a vector is actually short, or merely pointing away from the viewer. In such cases, the ability to rotate the 3D image on a computer, and/or to view it immersively, provides valuable depth cues which 2D plots lack. 10.2.1.4 Lighting Shadows cast by 3D features help reveal their spatial structure. Typically a scene is rendered using a single lighting source, which emanates from the coordinates of the viewer. Multiple lighting sources from the sides can add additional shadows and cues, just as in the real world.

324

A.J. Hermann and C.W. Moore

b a

c

Fig. 10.3 Modeled chlorophyll in the Gulf of Alaska, rendered using 2D versus 3D graphics. (a) Shaded contour plot of surface chlorophyll. (b) Oblique 3-D view (looking downward and to the southeast) of biological properties near Portlock Bank, southeast of Kodiak Island. Chlorophyll isosurface (green) is shown superimposed on bathymetry (grey) and a vertical slice of nutrients (blue/orange palette denotes low/high nutrient concentrations). (c) Oblique 3D view of chlorophyll rendered as a virtual fog. Note how 3D views simultaneously convey horizontal and vertical extent of patches. Note also how the isosurface view (b) is confusing without immersion or animation to provide spatial cues

10.2.1.5 Navigation Quick navigation – that is, easily changing one’s point of view – is essential for rapid exploration of complex 3D worlds. Indeed, this ‘‘ease-of-use’’ issue frequently makes all the difference between popular and unpopular software. In the natural world, we (and many predators) alter our point of view to get a

10

Visualization in Fisheries Oceanography

325

better sense of the spatial layout of our environment. Consider how a hunting cat bobs and weaves its head from side to side to judge the distance to its intended prey. In a virtual world the viewer may ‘‘fly’’ or ‘‘walk’’ through a rendered world, or manipulate an object in that world. These actions are accomplished by keyboard, mouse, or other input device. 10.2.1.6 Animation Animation of a 3D world has a double benefit. First, and most obvious, it reveals the path of an object (e.g. an individual modeled fish) through time. Second, it provides additional clues as to the location of objects, even if they are not appreciably mobile. In a sense, it is equivalent to changing one’s point of view for examination of a stationary object. 10.2.1.7 Collaborative Viewing Sophisticated yet inexpensive technologies now exist for collaborative viewing of 3D worlds. Such approaches are a natural extension of shared desktop software, and in fact share much in common with interactive online gaming. Collaborative viewing allows researchers at multiple institutions to examine and discuss the same 3D world. Viewers may be represented as ‘‘avatars’’, similar to virtual characters in a multiplayer game.

10.2.2 3D Computer Hardware Issues 10.2.2.1 Central Processing, Memory, and Graphics Processors As of this writing, anyone with a modern commodity PC can render and explore complex 3D worlds. Obviously, the more powerful the computer, the faster any rendering or navigation through a 3D world can take place. Less obviously, the bottleneck in such rendering may exist in the raw CPU, the memory to which it has fast access, or the power of the graphics processor. The latter typically exists as a separate element of the computer, and is responsible for the persective rendering of 3D objects (or more precisely, the individual polygons which make up a 3D digital object), given their location, color and texture. Driven by the relentless market for better and faster computer games, inexpensive yet powerful graphics cards have become widely available to consumers. Fisheries scientists obviously benefit from these developments. 10.2.2.2 Data Access A large amount of memory is required to render large artificial worlds in fine spatial detail. Some graphics programs (and many computer games) deal with this problem by storing on disk both coarse and fine-resolution versions of the

326

A.J. Hermann and C.W. Moore

world. The coarse version is loaded into memory and rendered when the observer is looking at the large-scale structure of the world (i.e. is ‘‘far away’’ from the world); fine-scale information is loaded into memory as necessary when the observer zooms in on a smaller region of the world. Data may be stored locally or on a remote server. Several popular geography and astronomy viewers utilize such interactive ‘‘data on demand’’ transfers. Extensive data sets are stored on a remote server; only those portions being viewed are downloaded to the local machine. In some cases the software anticipates and pre-fetches what data will be needed next, based on the trajectory of the viewer. 10.2.2.3 Parallel/Distributed Rendering Careful exploration of a virtual world requires both a broad spatial view and the ability to zoom in on fine detail. Even with efficient access to remote data, the rendering of complex worlds can be very taxing on a single CPU or graphics processor. One approach to handling complex worlds is to farm out separate pieces of the rendering to separate processors. This can be effected on multiple processors of a single computer, or even on widely separated processors connected through a network. These are stitched together by a central processor, or a central graphics processing unit. 10.2.2.4 Navigation Hardware In addition to keyboard and mouse control, navigation through virtual worlds can take place with simple game controllers, e.g. joysticks (Fig. 10.4). In some circumstances this provides the most natural way to ‘‘fly’’ through 3D worlds. In other cases, features are best explored through rotating the world with a standard mouse. Many specialized controllers have been built to facilitate navigation through three dimensions. Recently, simple handheld devices have become popular for electronic versions of games such as baseball and tennis. These measure the position, orientation, and acceleration of the user’s arm to manipulate virtual bats and rackets; we can easily imagine potential uses for scientific software as well.

10.3 Stereo-Immersive Approaches to Visualization Here we focus specifically on the hardware and software requirements for immersive visualization, which take advantage of the natural human capacity for binocular (‘‘stereo’’) vision. The hardware and software needed to render and immersively display 3D worlds, formerly available only at sophisticated visualization centers, is now accessible to scientists of modest means. Ever more sophisticated (and high-cost) technologies will doubtless emerge; here we focus on lower cost gear readily available today through local commodity computer

10

Visualization in Fisheries Oceanography

327

Fig. 10.4 Example of immersive joystick navigation through a virtual world on a laptop computer

dealers. Broadly speaking, the minimal requirements for a single user are a reasonably powerful CPU and a graphics card with a stereo driver. Immersive viewing by a group can be achieved with the addition of two projectors and an appropriate screen. The basic approach with all immersive techniques is to render two snapshots of the scene you wish to view, one for your left eye and one for your right. The differences between the two snapshots depend on how far away the viewed object is, and how widely spaced your ‘‘virtual’’ eyes are, relative to the size of the object. Once you have the two snapshots, you need to deliver each of them to the proper eye. The old technique for doing this, very popular around the beginning of the 20th century, is to produce an oversized postcard with the left eye’s view on the left half and the right eye’s view on the right half. A simple mechanical device then focuses each snapshot to the proper eye: the stereograph. With some training (but typically with some eye strain) you can focus the proper image from a stereo pair on the proper eye without any mechanical assistance. Sometimes pairs are rendered with the left eye’s view on the right and vice versa; the viewer then crosses her eyes to focus the proper image on the proper eye. Again, this involves some eye strain.

10.3.1 Stereo Hardware On present-day computers, the task of generating a 3D scene of graphical objects is handled by the CPU, central memory and graphics processor working together; the task of rendering two different views of that scene (and possibly sending them to different monitors or projectors) is typically handled by the graphics processor itself. Here we describe three different common approaches to stereo visualization on computers: anaglyphs, active shutterglasses, and passive polarization.

328

A.J. Hermann and C.W. Moore

10.3.1.1 Red-Blue Anaglyphs One simple technique for achieving immersive 3D is to print both the left and right eye’s view of a greyscale scene on top of one another, but in different colors (see Fig. 10.5). Such pictures are called anaglyphs. Inexpensive glasses with colored filters then screen out the left eye’s picture from the right eye and vice versa. This was the basic approach for 3D movies and comics of the 1950s. Fish (spheres) over bathymetry

Stereo version

Top view of fish locations

Velocity vectors in a submarine canyon

Fig. 10.5 Comparison of stereo and non-stereo images of particles and vectors in 3D worlds. For stereo viewing use red/blue anaglyph glasses with the red lens over your right eye and the blue lens over your left eye

10

Visualization in Fisheries Oceanography

329

Anaglyphs are in fact easy to display on computer monitors, and the results are suprisingly realistic for greyscale worlds. A limited degree of color is possible in such worlds as well. Shown in Fig. 10.5 are some examples of virtual worlds and greyscale stereo anaglyphs we have created from biophysical model output. The top figures illustrate positions of fish (represented as spheres) from a spatially explicit 3D individual-based model of walleye pollock (Theragra chalcogramma) in Shelikof Strait, Alaska (Hermann et al. 1996, Hinckley et al. 1996). Kodiak Island is the prominent feature in the right half of the top figure; our viewpoint looks to the northeast. The bottom figures are from a circulation model of Pribolof Canyon in the Bering Sea (Hermann et al. 2002). A cheap pair of red/blue glasses will cause the spheres and vectors to pop out of the page at you, revealing the true 3D location of the modeled fish, the steep slopes of the bathymetry, and the vertical motions near the submarine canyon. 10.3.1.2 Active Shutterglasses In the active shutterglass approach, still images and animations appear in full color on a computer monitor or projection screen. The approach is based on the fact that the computer, like a television set, refreshes the image on the screen many times per second. The computer alternately displays the left eye’s view, then the right eye’s view, while electrically controlled lenses (the application of a voltage renders them opaque) alternately allow the left eye to view the display, then the right eye, in sync with the alternating images. The synchronization signal is sent to the glasses by wire or through an infrared emitter. If the images are alternated quickly enough (say, 120 times per second), the viewer sees a steady image in stereo. There are three common techniques used with shutter glasses. (1) Frame Flipping: all available pixels on the monitor (or projector) are used to render each eye’s image in sequence. This approach was used in many high-end visualization environments through the 1990s, e.g. the VR-CAVE and the IMMERSADESK. (2) Interleaving: the monitor alternately displays the left and right eye’s view using the even-numbered columns (or rows) of pixels for the left-eye view, and the odd numbered columns (or rows) of pixels for the right-eye view. This requires less graphics memory than the frame flipping approach, but results in a lower resolution image for a given monitor. (3) Synch Doubling: the computer generates the left eye view on the top half of the screen, and the right eye view on the bottom half of the screen. A synch doubler then spreads out the top half image to fill the entire screen, and does the same with the bottom half image – they are alternately displayed at half the original refresh rate of the monitor. This requires no special graphics hardware within the computer itself, but as with interleaving results in a lower resolution image than frame flipping. A variation of the interleaving approach continuously displays left/right views on the odd/even columns, and deploys special lenses on the monitor itself

330

A.J. Hermann and C.W. Moore

to direct odd/even columns to the left/right eye of the viewer. This auto-stereo technique requires no special glasses, but typically requires the viewer to sit at a fixed location in front of the screen.

10.3.1.3 Passive Polarization: The Geowall A third and very appealing technique entails projecting the different images through two projectors fitted with polarizing filters, effecting different polarization for the left versus right eye’s view (Leigh et al. 2001). The two images are aligned and projected onto a polarization-preserving screen (see Fig. 10.6). Simple polarized lenses (much like sunglasses) are then used for stereo viewing, delivering the proper image to each eye. Unlike most anaglyphs, this allows full color images to be displayed. The low cost of both the projectors and the screen has made this approach very affordable; indeed, it has now become an attractive alternative to the shutterglass systems described in Section 10.3.1.2. DLP (Digital Light Processing) projectors are used because they naturally emit unpolarized light, which is subsequently polarized by the filters. This arrangement is usually termed a ‘‘Geowall’’, and many hundreds are now in use worldwide (for a partial list, see the Geowall consortium webpage: http://geowall.geo.lsa.umich.edu/). The system is compact and portable, can be installed in a single investigator’s office (Fig. 10.7), and has been very efficient for display of oceanographic and astrophysical data at

Fig. 10.6 Schematic of passive, dual-projector stereo display system (a ‘‘Geowall’’; see http:// geowall.geo.lsa.umich.edu/). Components include host computer (commodity PC or workstation), DLP projectors, polarizing filters, and polarization-preserving projection screen. Passive polarized glasses are worn by the viewer, to deliver the proper image to each eye. Thick arrows denote light path

10

Visualization in Fisheries Oceanography

331

Fig. 10.7 Setup of a Geowall system in a modest-sized office. Projectors with filters are placed at the back of the office, and driven with a linux PC outfitted with a graphics card supporting two displays. A polarization-preserving pull-down screen was installed in front of the window at the opposite end of the office. A model of the North Pacific (Curchitser et al. 2005) is displayed on the screen

conferences, workshops, and science fairs (Hermann and Moore 2004). The passive polarization approach has grown rapidly in popularity due to its simplicity and low cost, and is now the basis of many large-scale IMAX 3D theaters.

10.3.2 Stereo Software A large number of commercially available packages will render 3D worlds in stereo using the hardware described above. One of the earliest (and in our view still one of the best) open-source visualization packages for gridded data was originally developed by Bill Hibbard and colleagues at the University of Wisconsin: Vis5d (Hibbard and Santek 1990; http://vis5d.sourceforge.net/), which is featured in Figs. 10.3 and 10.7. When the C code for this package was originally developed, few computers could run the code speedily, and interactive viewing was a frustrating endeavor for all but the smallest of grids.

332

A.J. Hermann and C.W. Moore

However, times have changed (i.e. the hardware has caught up with the software), and modern PCs running Linux can execute this program handily with smooth navigation for grids with a million or more gridpoints. A related Javabased toolbox, VisAD (Hibbard 2002; http://www.ssec.wisc.edu/billh/ visad.html), has been used to develop software such as the Integrated Data Viewer (http://www.unidata.ucar.edu/software/idv/) with similar 3D capabilities. Such Java-based tools have the ability to communicate with large datasets through the Internet, which can be of great advantage when data is not stored locally. Our own experience has been that Java-based tools are slower than the old C code-based visualization of locally stored data – provided it all fits in local memory!

10.3.2.1 3D Through the Web: Virtual Reality Modeling Language The Virtual Reality Modeling Language (VRML) is a scene description language that describes three-dimensional environments in a simple ascii format, and which allows users to access, navigate, explore and interact with environmental data in three dimensions on the Web. An international open standard was accepted for VRML in 1997 (ISO/IEC 1997). This language has been very useful for the visualization of 3D oceanic data (Moore et al. 2000a, 2000b). Many of the virtual worlds shown in this chapter were constructed using this standard. VRML is scalable across platforms ranging from PCs to high-end workstations, and can be viewed either with a Web browser plug-in or with stand-alone software. A VRML world typically consists of polygonal surfaces that mimic the real environment. Objects rendered for oceanographic/atmospheric analyses include contoured and shaded slices, vector fields, isosurfaces, and spheres representing drifters or organisms. These objects can be touched, rotated, or animated using controls that the browser provides. VRML objects can be primitive (cubes, spheres, etc.), or user-defined (elevation grids, polygonal surfaces, lines), and can be given traits such as color, texture, sound, and video. Animations can be created by swapping surfaces of arbitrary shape using cached memory or morphing a surface by changing its defining coordinates over time. The user can define touch sensitive objects and assign actions (typically through simple Javascript routines) that allow the user to interact with the world. VRML can also interact with Java to create a myriad of 3D and 2D user interfaces. Free VRML plug-ins are currently available, and some PCs already have this capability preinstalled. A VRML file represents three-dimensional objects as a series of coordinates, independent of perspective; hence there are not stereo VRML files per se, but rather stereo viewers of VRML files. As noted earlier, the graphics card (with an appropriate driver) typically renders the two views of a particular scene. VRML worlds may be generated from popular data analysis software, such as Matlab. We have developed a web interface to serve up

10

Visualization in Fisheries Oceanography

333

VTK

Model

netCDF

LAS

Vis5D

VRML

Ferret

GIF

Server Client

Fig. 10.8 Implementation of Live Access Server for remotely stored data, delivering immersive graphic on demand (red/blue anaglyph) to viewer. Model generates data in netCDF file format, stored on remote server. Live Access Server (LAS) extracts graphical objects (here, isosurface of salinity and velocity vectors), using graphical software (VTK, Vis5D, Ferret) running on the server. Graphical objects are transferred as a VRML file, for immersive display on the local client

VRML worlds on demand for some of our data sets, through a Live Access Server (Hermann et al. 2003; Fig. 10.8).

10.3.3 Stereo APIs There are presently several software standards (Application Program Interfaces, APIs) for graphical manipulations at the level of computer hardware. Open Graphics Language (OpenGL) has emerged as a standard of choice for scientific applications, and many ‘‘professional’’ graphics cards now support it. Most modern graphic cards feeding the huge gamer’s market support an API known as DirectX. As noted earlier, such cards perform mathematical manipulations (matrix multiplications, ray tracing) needed for rotating and rendering objects on your display; this is more efficient than having to calculate everything on your CPU. It is worth noting that there are presently two different ‘‘flavors’’ of stereo, one based on DirectX (full-screen-stereo) and one based on OpenGL (stereo-in-a-window). Most consumer graphics cards (and stereo-enabled games) support at least the former; many currently available graphics cards are capable of rendering in either of these modes.

334

A.J. Hermann and C.W. Moore

10.4 The Future of Modeling and Visualization Our ability to model the ocean and its fisheries at relevant scales is partly dependent on the raw speed of computer hardware. According to Moore’s Law, computer processing speed increases 8-fold every 5 years. Our ability to resolve horizontal scales in the global ocean will double in that time frame, as it requires four times as many grid points and twice as many time steps for numerical stability. Our ability to store data inexpensively appears to be increasing even faster than Moore’s Law; at the time of this writing, the cost of storage is approximately $0.15/GB. Clearly, both model output and measured data will grow exponentially in the coming decades, and effective exploration of these results will pose a great challenge. Consider the real ocean: we cannot measure everything at once, and limit our attention to what is easily observed (e.g. surface properties) or what contains the largest information about the system as a whole (e.g. subsurface current measurements which are known to be well-correlated with larger circulation patterns). In the modeled ocean, we are not limited by technology to any particular depth, and the sheer immensity of what could be explored is daunting. Immersive visualization is a tool to facilitate more rapid exploration of real and modeled worlds, to help identify the new and unusual phenomena contained therein. It does not replace traditional graphics, but rather adds a powerful exploratory tool for research, collaboration, and communication of results to the public. While the raw processing speed of a single CPU has multiplied, the speedier execution of numerical models in this decade is in fact largely due to new massively parallel architectures, and more efficient parallelization of code. New visualization hardware and software are taking advantage of this parallelism as well. The emergence of dual- and quad-processor PCs, and ever more powerful graphics processing units, will further accelerate the graphical capabilities of a single desktop machine in the laboratory or office. Increasingly this obviates the need for remote (and potentially expensive) dedicated graphic centers. Indeed, it may be that immersive visualization will become a routine and widely available tool for many scientific endeavors. As computer hardware grows more powerful and less expensive, larger datasets will be rendered in finer detail. New (e.g. holographic) techniques will likely be developed, which will eliminate the need for special glasses of any type. If the past 10 years are any guide, the driving economic force behind these new technologies will be the computer gaming market; in particular, the market for massive multiplayer online games. As with the powerful graphics hardware already available, it behooves the scientific community to take advantage of these new technologies as they emerge. Acknowledgments We wish to acknowledge Nancy Soreide at NOAA/Pacific Marine Environmental Laboratory for her enthusiastic encouragement of this work over the years, as well as Glen Wheless, Nancy Lascara, and friends and colleagues within the GeoWall community for their continued enthusiasm and support. A special nod of thanks is also due the computer

10

Visualization in Fisheries Oceanography

335

gaming community, which has been the economic engine behind many developments in computer graphics. Much of the work described here was supported by the National Oceanic and Atmospheric Administration, through their High Performance Computing and Communications (HPCC) program. Additional support from the NSF-GLOBEC program (OCE0624490) is also gratefully acknowledged.

References Curchitser EN, Haidvogel DB, Hermann AJ, Dobbins EL, Powell TM (2005) Multi-scale modeling of the North Pacific Ocean I: Assessment and analysis of simulated basin-scale variability (1996–2003). J. Geophys. Res. 110 (C11021) doi:101029/2005JC002902. Hermann AJ, Hinckley S, Megrey BA, Stabeno PJ (1996) Interannual variability of the early life history of walleye pollock near Shelikof Strait, as inferred from a spatially explicit, individual-based model. Fish. Oceanogr. 5 (Suppl. 1): 39–57. Hermann AJ, Hinckley S, Megrey BA, Napp JM (2001) Applied and theoretical considerations for constructing spatially explicit Individual-Based Models of marine larval fish that include multiple trophic levels. ICES J. Mar. Sci. 58: 1030–1041. Hermann AJ, Stabeno PJ, Haidvogel DB, Musgrave DL (2002) A regional tidal/subtidal circulation model of the southeastern Bering Sea: Development, sensitivity analyses and hindcasting. Deep-Sea Res. II (Topical Studies in Oceanography) 49: 5495–5967. Hermann AJ, Moore CW, Dobbins EL (2003) Serving 3-D rendered graphics of ocean model output using LAS and VTK. In 19th International Conference on Interactive Information Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology, 2003 AMS Annual Meeting, Session 5.3, Long Beach, CA, 9–13 February 2003. Hermann AJ, Moore CW (2004) Commodity passive stereo graphics for collaborative display of ocean model output. In Proceedings of the 20th International Conference on Interactive Information and Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology, 2004 AMS Annual Meeting, Seattle, WA, 12–15 January 2004, paper 8.13. Hermann AJ, Hinckley S, Dobbins EL, Haidvogel DB, Mordy C (2008) Quantifying crossshelf and vertical nutrient flux in the Gulf of Alaska with a spatially nested, coupled biophysical model. Deep-Sea Research Part II, in press. Hibbard W, Santek D (1990) The Vis5D system for easy interactive visualization. In Proc. Visualization ’90, IEEE CS Press, Los Alamitos, Calif., pp. 28–35. Hibbard W (2002) Building 3-D user interface components using a visualization library. Comput. Graph. 36(1):4–7. Hinckley S, Hermann AJ, Megrey BA (1996) Development of a spatially explicit, individualbased model of marine fish early life history. Mar. Ecol. Prog. Ser. 139: 47–68. Hinckley S, Hermann AJ, Meir KL, Megrey BA (2001) The importance of spawning location and timing to successful transport to nursery areas: a simulation modeling study of Gulf of Alaska walleye pollock. ICES J. Mar. Sci. 58:1042–1052. Hinckley S, Coyle KO, Gibson G, Hermann AJ, Dobbins EL (2008) A biophysical NPZ model with iron for the Gulf of Alaska: Reproducing the differences between an oceanic HNLC ecosystem and a classical northern temperate shelf ecosystem. Deep-Sea Research Part II, in press. ISO/IEC 14772-1:1997 (1997) Information technology – Computer graphics and image processing – The Virtual Reality Modeling Language – Part 1: Functional specification and UTF-8 encoding Leigh J, Dawe G, Talandis J, He E, Venkataraman S, Ge J, Sandin D, DeFanti TA (2001) AGAVE: Access Grid Augmented Virtual Environment, Proc. AccessGrid Retreat, Argonne, IL, January 16, 2001.

336

A.J. Hermann and C.W. Moore

Moore CW, McClurg DC, Soreide NN, Hermann AJ, Lascara CM, Wheless GM (2000a) Exploring 3-dimensional oceanographic data sets on the web using Virtual Reality Modeling Language. In Proceedings of Oceans ’99 MTS/IEEE Conference, Seattle, WA, 13–16 September. Moore CW, Soreide NN, Hermann A, Lascara C, Wheless G (2000b) VRML techniques and tours: 3D experiences of oceans and atmospheres. In Proceedings of the 16th International Conference on IIPS for Meteorology, Oceanography, and Hydrology, AMS, Long Beach, CA, 9–14 January 2000, 436–438. Powell TM, Lewis CVW, Curchitser E, Haidvogel D, Hermann A, Dobbins E (2006) Results from a three-dimensional, nested biological-physical model of the California Current System: Comparisons with Statistics from Satellite Imagery. J. Geophys. Res. 111 (C07018), doi:10.1029/2004JC002506.

Chapter 11

Computers in Fisheries Population Dynamics Mark N. Maunder, Jon T. Schnute and James N. Ianelli

11.1 Introduction Theories of fish population dynamics have always involved computation (Quinn 2003). For example, the classical analysis of Beverton and Holt (1957) uses straightforward bookkeeping to represent the biological processes of mortality, growth, and fishing. The tools available for such calculations can, however, have a dramatic effect on the analyses attempted. In the first edition of this book, Hilborn (1996) cited a remark by Beverton and Holt (1957, p. 309) that ‘‘an experienced computer’’ could complete a certain yield calculation ‘‘in about three hours.’’ In that context, a ‘‘computer’’ referred to a person ‘‘using a hand calculating machine.’’ Such a person would be unlikely to try several hundred similar calculations to test variations in the underlying parameter values. With electronic computers available now, a sensitivity analysis becomes almost trivial, and a modern analyst might devise something even more elaborate to assess uncertainty in the results. Computers have greatly influenced current theories of fish population dynamics. What can be done guides the development of theory, and ultimately governs what actually is done. A student entering the field today faces a daunting multiplicity of software environments and methodologies, many of them developed since the first edition of this book was published in 1996. In this chapter, we provide a guide to the current landscape of possibilities. Although the terminology has changed since 1957, so that a calculating machine has become a computer and a computer has become an analyst, the fundamental scenario remains the same. A person uses computational equipment as an aid to thinking about how fish populations work. Given that modern computers make it possible to conduct broad explorations of the possibilities, do these explorations really enhance our understanding? Can they sometimes be misleading? We return to such basic questions in our final discussion.

M.N. Maunder (*) Inter-American Tropical Tuna Commission, 8604 La Jolla Shores Drive, La Jolla, CA, 92037-1508, USA

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_11, Ó Springer ScienceþBusiness Media B.V. 2009

337

338

M.N. Maunder et al.

Essentially, a modern computer upgrades a mechanical calculator by providing the ability to run programs and implement algorithms. This shifts the focus from restrictive analytical results and bookkeeping to cleverly designed algorithms that use intensive computation to achieve a desired result. Efron and Tibshirani (1993) express this point of view in the preface to their textbook on bootstrap methods: Statistics is a subject of amazingly many uses and surprisingly few effective practitioners. The traditional road to statistical knowledge is blocked, for most, by a formidable wall of mathematics. Our approach here avoids that wall. The bootstrap is a computer-based method of statistical inference that can answer many real statistical questions without formulas. Our goal in this book is to arm scientists and engineers, as well as statisticians, with computational techniques that they can use to analyze and understand complicated data sets.

As computing speed increases, partly through distributed computing across multiple Central Processing Units (CPUs), the potential grows for more elaborate analyses and increasingly clever algorithms. Reduced computation time facilitates the investigation, testing, and comparison of old and new methods. Theories of fish population dynamics and their applications have changed significantly in response to these developments, particularly in the field of Bayesian statistics (Punt and Hilborn 1997). Like bootstrap methods, which belong properly to the frequentist school of statistics, Bayesian methods can be implemented with great generality using computer intensive calculations. Consequently, as expressed by Clifford (1993, p. 53), ‘‘we can compare our data with the model that we actually want to use rather than a model which has some mathematically convenient form.’’ In part due to its adaptability, the Bayesian approach has found a wide audience. The comprehensive series of reference volumes on classical (largely frequentist) statistics by Kendall and various co-authors has now been extended to include a new volume on Bayesian statistics (O’Hagan and Forster 1994; 2004). A recent compilation of Monte Carlo methods for sampling Bayesian posterior distributions (Robert and Casella 2004) extends to more than 600 pages. Computers have also influenced management of fish populations directly by providing computationally intensive methods for assessing management policy (de la Mare 1986; Butterworth et al. 1997; De Oliveira et al. 1998; Butterworth and Punt 1999; Smith et al. 1999; Schnute and Haigh 2006). Formally, a stock assessment often boils down to an algorithm for determining a policy decision, such as a catch quota, based on the available data. Perhaps hundreds or thousands of historical data values (e.g., abundance indices, catches, proportions by age and/or size) get reduced to a single policy variable. If the data reduction algorithm (a management strategy) can be completely formalized, then it can be tested using simulation models (called operating models) with high levels of presumed biological complexity. The modern technique of management strategy evaluation (MSE) tries to identify good control algorithms robust

11

Computers in Fisheries Population Dynamics

339

to various biological scenarios. Although this idea has roots in the classical literature on dynamic system control (e.g., Luenberger 1979), it takes quite specialized forms in the context of fishery management. Schnute and Haigh (2006) provide technical details for a simple case. In the sections that follow, we start by examining various software environments available for implementing fisheries models. All of these have potential application in many fields. We also discuss several packages explicitly tailored to deal with fishery data analysis. Later in the paper, we use this material as background for a review of recent advances in model design and application. Our final discussion includes a critique of current software, theories, and applications, along with a few speculations and suggestions for future progress.

11.2 Modeling Environments A large number of software environments have been applied to the quantitative analysis of fisheries population dynamics. These include traditional programming languages (e.g., Fortran, BASIC, and C), common spreadsheet programs (e.g., Excel), and fourth-generation languages that address specific problems for quantitative analysis of fisheries population dynamics (e.g., AD Model Builder, discussed below). Important characteristics of these fourth generation languages relate to finding parameter estimates by fitting a model to data. An appropriate choice of software for investigating fish population dynamics depends on the purpose. For example, some software may be highly efficient for instructional or educational purposes, but inadequate for providing reasonable scientific advice for fisheries managers. Different modeling environments vary in their computational efficiency, ease of use, and quality of available documentation. Most analysts adopt the axiom: ‘‘The best software in the world is the package you know how to use’’. The modern stock assessment practitioner needs to balance time between learning new software systems and keeping abreast of growing scientific fishery issues. This dilemma underscores the importance of proficiency trade-offs among the wide variety of choices. Although we discuss several of the more popular environments, we certainly don’t provide a comprehensive survey of all available software.

11.2.1 Programming Languages Standard programming languages have traditionally played an important role in quantitative analyses of fisheries population dynamics. For example, the original application of the Pella-Tomlinson surplus production model (Pella

340

M.N. Maunder et al.

and Tomlinson 1969) used Fortran to fit the model to available data. However, the use of such languages traditionally required programming all components of the analysis. Currently, repositories of reliable libraries to carry out most essential modeling components (e.g., non-linear optimization, random number generation, matrix algebra) are readily available. For example, the NETLIB and STATLIB websites (http://www.netlib.org/, http://lib.stat.cmu.edu/) provide access to a wide number of useful libraries for many computer languages. The general stock assessment model CASAL (see below) is written in C++ and uses the freely-available ADOL-C automatic differentiation library (Walther et al. 2005). Programming in a third-generation language like Fortran, C, or C++ requires considerable time and familiarity with the code, with corresponding low ease of use and utility for instruction. Applications often require a long development time, but savings on computer memory and CPU cycle efficiency can be quite high, depending on the practitioner’s proficiency with the language. Fishery models written in these languages tend to be reasonably scalable. Thus, they can handle increased size or complexity, perhaps with modest additional resources (such as computer memory).

11.2.2 AD Model Builder Currently, AD Model Builder (ADMB – http://otter-rsch.com/admodel.htm) plays a major role as a software environment for estimating the parameters of complex, highly-parameterized fisheries stock assessment models. Maunder (2007) has compiled an extensive list of references that use this software. Designed explicitly for parameter estimation in such models, ADMB uses automatic differentiation (AD) to provide the function optimizer with exact derivatives. It takes advantage of C++ class structures to collect intermediate results and perform internal calculations that implement the AD algorithm, which is based on the chain rule (often learned and forgotten by calculus students). The ‘‘reverse’’ AD method used by ADMB performs much more quickly and accurately than a numerical derivative calculation based on finite differences. For example, if a computer requires time T to evaluate a function with n parameters, then it will need time nT to evaluate the numerical gradient, because one evaluation is needed for an offset in each of n coordinate directions. Reverse AD does the same job in a time of about 5T, regardless of the number n, without the round-off error involved in finite differences. This result seems almost miraculous until one understands the main idea of the proof (Griewank 2000). Like other modern computing algorithms, reverse AD uses extensive computer memory to achieve computational efficiency. ADMB also includes a highly-configurable Bayesian MCMC algorithm, likelihood profile calculations, simulation capabilities, and extensive matrixalgebra functionality. A recent extension of ADMB adds the capability of

11

Computers in Fisheries Population Dynamics

341

modeling random effects, using the Laplace approximation or importance sampling (Skaug 2002; Skaug and Fournier 2006). ADMB has allowed the use of nonlinear dynamic models with thousands of parameters (e.g., Maunder and Watters 2003a) or hundreds of thousands of data points (e.g., Maunder 2001a), problems of a scale that could not be solved previously. In somewhat outdated comparisons, ADMB has been shown to perform much more efficiently than other modeling tools (Schnute et al. 1998). While the learning curve for ADMB can be particularly steep, the documentation continues to improve and the examples are helpful. ADMB is based on C++ coding with additional structure such as keywords, functions, and operators. Knowledge of C++ is not a prerequisite for a successful ADMB user, because a simplified coding environment shields a user from advanced C++ concepts. However, some understanding of standard C/C++ syntax rules is helpful, and an understanding of the principles of C++ (e.g., operator overloading) can provide insight into aspects of ADMB. More information about AD Model Builder can be found at the newly founded ADMB Foundation http://admb-foundation.org/.

11.2.3 R, S, and S-PLUS Venables and Ripley (2000) describe S as ‘‘a high-level language for manipulating, analyzing, and displaying data. It forms the basis of two widely used data analysis software systems, the commercial S-PLUS and the Open Source R.’’ These closely related software packages provide access to a large number of algorithms useful for building statistical models. They offer beginners a less intimidating environment than third-generation programming languages, like Fortran and C/C++. The graphical user interfaces are advanced, and the documentation is easily accessible and usually clear. Perhaps most importantly, users quickly find the advantages of scripting (writing sets of commands that apply) as a means for accumulating skills and documenting progress on specific projects. We strongly encourage fishery data analysts to become at least somewhat familiar with the free package R (available at http://www.r-project.org/). It offers excellent facilities for graphical representation of data, and (thanks to its native support for many algorithms) a broader range of analyses are available than for spreadsheet programs. In our opinion, the scripting part of R (and S-PLUS) tends to be simpler to use than the scripting capabilities within Excel (based on Visual Basic). Many fishery applications use R extensively, either by itself or as a means for analyzing and displaying results from other programs. Stock assessment practitioners can’t operate effectively without tools for producing coherent summaries quickly. Availability of support, documentation, and extended features plays a key role in determining the usefulness of any program. R benefits particularly from books with detailed descriptions of programming techniques and applications (e.g., Venables and Ripley 1994; 1999; 2000; Dalgaard 2002; Maindonald and

342

M.N. Maunder et al.

Braun 2003; Murrell 2006; Wood 2006). Furthermore, R continues to grow through a formal system of user contributed libraries that can be downloaded from the Comprehensive R Archive Network (CRAN, http://cran.r-project.org/). Users who submit these packages can write code in native R, which runs interpretively, or use C or Fortran to improve the performance of computationally intensive algorithms. The latter code gets compiled automatically for a variety of operating systems when a package is submitted to CRAN. Rigorous standards, enforced by testing procedures, ensure a reasonable quality of documentation and software operation. Each book cited above has an associated electronic file with code and data for all worked examples. Our references give details for obtaining these files as CRAN libraries or compressed files at an explicit web site. At the time of writing this chapter (May 2007), CRAN provides over 1,000 packages that implement a wide variety of algorithms from diverse fields. The recently-initiated Fisheries Library in R (FLR, http://flr-project.org/, Kell et al. 2007) provides a collection of R tools that facilitate the construction of models representing fisheries and ecological systems. In particular, the package focuses on evaluating fisheries management strategies. It includes various models for assessing stocks and simulating population dynamics. The software comes as a suite of R packages and other toolkits that can be downloaded from the project web site, which also includes useful tutorials. Some components are written entirely in R, while others are written in C++ or Fortran to accommodate existing programs or to recode programs for greater efficiency. The R code takes advantage of the language’s class structure (technically called S4 classes) to facilitate integration among the model components. One of us (JTS) has contributed two libraries to CRAN. The package PBS Mapping (Schnute et al. 2004) supports various spatial algorithms and represents information on coastal maps. Although many assessment models ignore spatial effects, except possibly for subdividing populations into regional components, data from modern fisheries often include precise latitude and longitude coordinates of individual fishing events. Consequently, stock assessments can now take account of spatial phenomena, such as the concentration of fishing effort into smaller areas (Fig. 11.1). A PBS Mapping simulation (Fig. 11.2) illustrates a calculation of the area impacted by tows along the sea floor. Each tow corresponds to a simple rectangle, but the union of tows gives a complicated polygon with many vertices and holes (rather like the profile of dropped pick-up sticks). The union has an area less than the sum of its parts. To address problems like this, stock assessments now use complex algorithms from computational geometry, in addition to statistical theory. Another package PBS Modelling (Schnute et al. 2006) makes it easy to build models with graphical user interfaces (GUIs) that facilitate model exploration and data analysis. For example, the yield-per-recruit calculation (Beverton and Holt 1957; Ricker 1975) mentioned in the first paragraph of this chapter depends on natural mortality M, four growth parameters (W1, K, t0, b), and

Computers in Fisheries Population Dynamics

343

A

B

5700

UTM Northing (km) 5800 5900

6000

11

CPUE (kg/h) Area (km²) 0–100 100–500 500–1000 > 1000

200

CPUE (kg/h) Area (km²)

3,888 2,112 2,000 1,856

300

0–100 100–500 500–1000 > 1000

400 UTM Easting (km)

500

200

3,040 2,384 1,728 1,488

300

400 UTM Easting (km)

500

Fig. 11.1 Annual distribution of Pacific ocean perch (Sebastes alutus) catch rates (kg/h) by the groundfish trawl fishery in northern British Columbia, Canada. Axes show coordinates (km) in Universal Transverse Mercator (UTM) zone 9. Grid blocks have size 4 km  4 km. Red, blue, and black contours indicate depths of 200 m, 500 m, and 1,000 m, respectively. The figure, produced with the PBS Mapping package for R, represents 2 years: (left panel) 1997 and (right panel) 2005. A moderately-productive region near the center of the 1997 map seems no longer active in 2005

a reference age a (Schnute 2006, Table 11.2). We mentioned that an ‘‘experienced computer,’’ who needs three hours to do one calculation, would be hard pressed to test sensitivity of the results to variations in the input parameters. 100

100

80

80

60

60

40

40

20

20

0

0 0

20

40

60

80

100

0

20

40

60

80

100

Fig. 11.2 Results of a simulation from the R package PBS Mapping, showing 40 random tows in a square with side length 100. Each tow has width 1, and the rectangle encompasses 10,000 square units. Left: The individual rectangles, with 160 vertices, have areas that sum to 2,243 square units. Right: The union includes a complex polygon (red) and two isolated polygons (green, blue) that cover only 1,919 square units. The complex polygon (red) has 730 vertices and 123 holes. The green polygon has 9 vertices

344

M.N. Maunder et al.

Fig. 11.3 A yield-per-recruit (YPR) calculation similar to that proposed by Beverton and Holt (1957), based on equations summarized by Schnute (2006, Table 2). Biomass yield per recruit at reference age a = 1 y depends on a balance between natural mortality M = 0.2 y–1 and fish growth, as determined by four von Bertalanffy parameters: W1 = 1,209 g, K = 0.2 y–1, t0 = –1 y, b = 3. Left: A graphical user interface (GUI) in the R package PBS Modeling allows entry of the biological parameters (blue controls). Policy options correspond to the selected ranges of fishing mortality F and recruitment age tR (i.e., age of first capture). Points on the eumetric curve correspond to the maximum yield for a specified fishing mortality F, where two such values of F can be chosen in the GUI. Numbers highlighted in red show calculated values for the corresponding recruitment age tR and maximum YPR. Right: Clicking the green ‘‘Plot’’ button causes PBS Modelling to calculate YPR (g) at 500  500 = 250,000 points in policy space (F, tR) and to draw the interpolated contours shown here. The eumetric curve appears in blue, and red points with vertical dashed lines correspond to the two values of F selected in the GUI

PBS Modelling includes a GUI (Fig. 11.3) to adjust the input parameters and produce a corresponding contour plot of yield per recruit, which depends on management choices of fishing mortality F and recruitment age tR. A simple text file with about 20 lines produces the GUI itself, as described in the user’s guide (Schnute et al. 2006). Almost every stock assessment analysis can benefit from a similar facility to experiment with alternatives, and PBS Modelling makes it easy to construct a customized GUI for this purpose. The package includes other examples with interfaces to (1) run the simulation portrayed in Fig. 11.2, (2) conduct Bayes posterior sampling, (3) find maximum likelihood estimates, and (4) run simulations based on differential equations. Many of these examples depend on other contributed libraries. Consequently, they illustrate how extendible the R environment can be.

11

Computers in Fisheries Population Dynamics

345

11.2.4 Matlab, Scilab, and Others The commercial program Matlab and its free counterparts Scilab (http:// www.scilab.org/) and Octave (http://www.gnu.org/software/octave/) offer another platform for simulation and data analysis that is similar to R and Splus, but with roots in engineering more than statistics. This platform takes advantage of a rich literature of algorithms in linear algebra, differential equations, random number generation, control theory, optimization, and many other fields. Like R, Scilab has reasonable documentation and sample code available from both online sources and comprehensive books (e.g., Gomez 1999, Campbell et al. 2006). Libraries (called toolboxes) provide language extensions, where the Scicos toolbox makes it possible to design a simulation using a visual flow diagram. On a Windows pltatform, the open source Octave package requires the UNIX emulator Cygwin (http://www.cygwin.com/). Matlab is used extensively in a number of fisheries laboratories and remains popular among scientists with backgrounds in physical oceanography. For example, analysts at the Alaska Fisheries Science Center (AFSC) use Matlab for evaluating surveys based on echo-integration data (acoustics). Another product Gauss (http://www.aptech.com/index.html), sold as a ‘‘Mathematical and Statistical System,’’ has been used effectively for developing complex migration models (e.g., Heifetz and Fujioka 1991). It remains popular for its speed, ease of use, scripting facilities, and power for simulation-estimation experiments. Finally, symbolic interpreted software environments, such as MATHCAD (http://www.mathsoft.com/), have also proven effective in fisheries analyses (Thompson 1992; 1999). This software allows users to write equations in symbolic form so that analyses can be fully documented and easily read and understood without the need to learn scripting languages and seemingly arcane syntax rules.

11.2.5 BUGS, WinBUGS, and OpenBUGS The BUGS Project (Bayesian inference Using Gibbs Sampling; http://www.mrcbsu.cam.ac.uk/bugs/) provides software that makes Bayesian sampling techniques available to the average analyst. The software has appeared in two incarnations, known as WinBUGS (Spiegelhalter et al. 2004) and OpenBUGS, both accessible through the BUGS web site. Currently, R users automatically get the OpenBUGS package by installing the BRugs library (Thomas 2004). The PBS Modelling library includes examples that illustrate the use of OpenBUGS to generate Bayesian posterior samples. The simplest deals with a markrecovery experiment in which M fish are marked and released, S are sampled from the population, and the sample produces R recoveries. The software generates a posterior distribution of the unknown population size N. The OpenBUGS model contains two key lines:

346

M.N. Maunder et al.

p 5 M=N R  dbinðp; SÞ Stated verbally, these lines say: ‘‘p is the proportion of fish marked, and R is distributed binomially with probability p and sample size S.’’ Another more complex example, which deals with catch curve analysis (Schnute and Haigh 2007), ends with the key model line: y½1:g  ddirchðalpha½1:gÞ which says that: ‘‘the vector y of observed proportions follows the Dirichlet distribution with parameter vector , and both these vectors have length g.’’ These two examples illustrate a feature central to the BUGS language. It contains a predefined vocabulary of distributions, such as the binomial and Dirichlet. Consequently, a user doesn’t need to think about analytic formulae that define them. This makes coding easy, as long as the required distributions are supported. Current documentation lists 23 distributions in the language, including univariate and multivariate cases, both discrete and continuous. BUGS runs slowly with complex models, and the language was designed for models rather different than those commonly used for stock assessment. Technically, a BUGS model must correspond to a directed acyclic graph (DAG), in which random variables successively influence each other to produce the observed data. Fairly complex models can be developed with only a small amount of code that follows syntax akin to R. BUGS code has been developed for at least two relatively complex assessment models (Nielsen 2000; Lewy and Nielsen 2003).

11.2.6 Spreadsheets Spreadsheet programs, now familiar to most computer users, have contributed almost inevitably to fish population dynamics. Beverton and Holt (1957, pp. 309–311) presented their ‘‘three-hour’’ yield calculations in a series of worksheets, which ‘‘the authors have found by experience to be satisfactory for the computation of most of these equations.’’ Their orderly approach to bookkeeping almost prophetically mimics programs like Microsoft ‘‘Excel’’ or the free OpenOffice program ‘‘Calc’’ (http://www.openoffice.org/product/calc.html). People with little or no programming experience often find spreadsheet programs appealing, due to the intuitive layout of calculations. The immediate display of results can facilitate learning and debugging (Prager and Williams 2003). However, spreadsheets can be prone to programming errors, and their code is difficult to document, review, and maintain (Prager and Williams 2003). Haddon (2001) uses spreadsheets extensively in his book on quantitative methods in fisheries. He provides a useful tutorial on using Excel for fisheries

11

Computers in Fisheries Population Dynamics

347

applications in an appendix, and all examples in the book appear as Excel workbooks ready for download from his web site. Punt and Hilborn’s (1996; 2001) FAO manuals on biomass dyamic models and Bayesian stock assessment models also provide extensive examples in Excel. Spreadsheet programs typically have additional features useful for fishery data analysis, such as random number generation and (somewhat limited) matrix algebra. In particular, the nonlinear solver in Excel allows parameter estimation by function optimization. Although this algorithm sometimes works acceptably well, it can perform poorly on ill-conditioned problems with many parameters. Frontline Systems (http://www.solver.com/), the company that produces the Excel solver, sells a more advanced version that can be purchased separately. The Excel add-on ‘‘Pop Tools’’ (http://www.cse.csiro.au/poptools/) provides additional functionality designed explicitly for modeling population dynamics. In particular, it allows for iterative spreadsheet calculations associated with Monte Carlo simulations, randomization tests, and bootstrap statistics. Spreadsheet programs often have comprehensive macro languages that provide additional functionality and allow programs to be embedded in the bookkeeping calculations. For example, Excel uses a version of Visual Basic for its macro language, and Haddon (2001, p. 367) provides an example to conduct bootstraps for a catch-at-age assessment model. Applications can be designed in which the spreadsheet is used only for data input or display of results, and the macro language or another software product used to implement the population dynamics model. For example, the general stock assessment model Coleraine (discussed below) uses AD Model Builder to implement the population dynamics model, but uses Excel as the user interface and to present results. In summary, spreadsheets have become de rigueur for many aspects of fisheries analyses, and are particularly effective for instructional purposes. They are simple and straightforward to learn, and their use in everyday computing for a broad number of tasks enhances support and usability features. For most serious stock assessment applications, however, spreadsheets are plagued with difficulties. They lack facilities for estimating uncertainty, have slow execution times for complex models (especially for non-linear parameter estimation), and have poor scalability (i.e., difficulty expanding to larger problems).

11.2.7 Summary Each modeling environment has its own advantages and disadvantages. An analyst usually requires several of them for efficient work. For example, stock assessment scientists often use a combination of ADMB, R (or S-PLUS), and Excel. ADMB is used for the final full analysis, but error checking is done by duplicating the model in Excel or R (often without parameter estimation). Coding the model in two completely different ways reduces the chances for

348

M.N. Maunder et al.

error, because the same mistakes probably will not occur in both environments. Both R and Excel are commonly used to process the input data, collate model results, and produce graphical displays. R works well for complex data analysis and manipulation, and it can generate publication-quality figures. Excel sometimes provides a more convenient tool to accomplish the same things, especially if the analyses are relatively simple.

11.3 Customized Stock Assessment Software The move toward better management of fish stocks has led to the need for more stock assessments. However, due to limited human resources available to conduct these assessments, people have devised software packages to implement a broad range of stock assessment models. The package designer attempts to include as many types of data as possible and to deal with the diversity of assumptions that might sometimes be appropriate. A recent workshop (Maunder 2005) compared three of these software packages (CASAL, MULTIFAN-CL, Stock Synthesis 2) with a standard assessment method (A-SCALA) used at the Inter-American Tropical Tuna Commission (IATTC).

11.3.1 Stock Synthesis Stock Synthesis (Methot 1990) provided one of the earliest general models that incorporated numerous types of data. It used an ad-hoc function optimizer with numerical derivative approximations, and therefore performed relatively slowly. The developer has since completely revised the code, based on a model programmed with ADMB, to produce Stock Synthesis 2 (SS2), a component of the NOAA Fisheries Toolbox (NFT – http://nft.nefsc.noaa.gov/). SS2 inherits functionality from ADMB: efficient and stable parameter estimation, as well as the ability to do Bayesian analyses using Markov chain Monte Carlo (MCMC) methods. In addition, SS2 has facilities for bootstrapping and forward projections. It allows differential growth among groups of fish, so that length-specific selectivity can change the length-at-age distribution. Throughout its architecture, SS2 supports a general parameter structure. This allows most estimated parameters to have priors and to be conditioned in various ways by environmental variables, time periods, fish sex, or other factors. As a result, the software provides a valuable method of testing hypotheses related to factor differences. SS2 has been used primarily to assess U.S. Pacific groundfish stocks, but it has also being applied to other species (e.g., tuna and billfish). An extensive graphical user interface has been developed to help users interpret the results (Fig. 11.4). The third generation of Stock Synthesis is now available and has the capacity to include tagging data.

11

Computers in Fisheries Population Dynamics

Fig. 11.4 (continued)

349

350

M.N. Maunder et al.

11.3.2 Coleraine Coleraine (Hilborn et al. 2000) was the first general stock assessment package developed with the ADMB environment. One of us (MNM) did most of the ADMB coding. It also provided the first general Bayesian stock assessment model, including priors and forward projections for management strategy simulation and decision analysis. Coleraine is, however, much less general than SS2. Coleraine uses Microsoft Excel to provide the user interface and to display results. It has been applied to several stocks in Australia, Canada, Chile, Iceland, New Zealand, and the United States (e.g., Field and Ralston 2005). It has been used as a teaching tool in university stock assessment courses, and it was included in the Food and Agriculture Organization of the United Nations’ Bayesian stock assessment user’s manual (Punt and Hilborn 2001).

11.3.3 MULTIFAN-CL MULTIFAN-CL (MFCL – http://www.multifan-cl.org/) is an extension of an earlier MULTIFAN program used to analyze length-frequency data (Fournier et al. 1990). MFCL was developed to analyze tuna populations (Fournier et al. 1998) and therefore a major focus was the inclusion of spatial structure in the population dynamics and fitting to tagging data (Hampton and Fournier 2001). It was developed using AUTODIF, which is the automatic differentiation software underpinning ADMB. The use of random effect-type parameters to model fishing mortality and changes in catchability result in applications with thousands of parameters. These are among the most complex and highly parameterized stock assessment models currently used for management advice. A Java program is available to allow users to view results (Fig. 11.4; Fabrice Bouye, Secretariat of the Pacific Community).

11.3.4 CASAL Of the general packages consistently used for stock assessment, CASAL (Bull et al. 2005) probably offers the greatest generality. It combines features

Fig. 11.4 User interfaces for general stock assessment models. The SS2 graphical user interface (upper panel) developed by Alan Seaver, U.S. National Marine Fisheries Service, allows users to specify values to control the parameter estimation procedure including parameter bounds, information about priors, initial values, and phase of estimation. The MFCL Java viewer (lower panel) developed by Fabrice Bouye, Secretariat of the Pacific Community, allows users to view details of the model fit to data. The yellow bars are the observed length frequencies, and the red lines represent the component of the model predicted length frequencies attributed to each cohort

11

Computers in Fisheries Population Dynamics

351

available in SS2 and MFCL, with support for the spatial structures need to fit tagging data. The implementation of CASAL uses the ADOL-C library (Walther et al. 2005) to produce a working environment similar to AUTODIF. CASAL has been used for the assessment of fish and shellfish stocks in New Zealand. Some of these assessments have been conducted using both CASAL and Coleraine to obtain comparative results between the two platforms.

11.3.5 Others Numerous other general models are actively used for stock assessments, and we mention a few of them here. Some of them have a narrow focus, while others address broader issues. For example, ASPIC (Prager 1995) implements a nonequilibrium stock-production model focused explicitly on fish catch and relative abundance data. Toward the other extreme, Gadget (Begley and Howell 2004) provides a statistical toolbox for investigating species interactions and spatially structured population dynamics. Some of these models are used extensively, like the extended survivors analysis (XSA) employed by member nations of the International Commission for the Exploration of the Sea (Shepherd 1999). Other models play a role in generating alternative perspectives. For example, following the FAO framework by Hoggarth et al. (2006), MRAG Ltd. has developed five software packages as part of the Fisheries Management Science Programme for the UK Department for International Development. These include LFDA (Length Frequency Distribution Analysis), CEDA (Catch Effort Data Analysis), Yield, ParFish and EnhanceFish. Similarly, in conjunction with the European Union, FAO has also supported the development of FiSAT, a software package developed mainly for the analysis of length-frequency data (http://www.fao.org/fi/statist/fisoft/fisat/index.htm). The NOAA Fisheries Toolbox (http://nft.nefsc.noaa.gov/) includes several packages besides SS2, some of them similar to SS2 but with fewer options. One of these (AMAK) provides enhanced flexibility in specifying time-varying selectivity (e.g., as in Butterworth et al. 2003), a feature unavailable in other packages. The International Commission for the Conservation of Atlantic Tunas (ICCAT) uses several software packages for stock assessment, including a generalized virtual population analysis (VPA) that can model two intermixing populations and sex-structure http://www.iccat.int/AssessCatalog.htm. Some models primarily offer simulation tools. For example, TEMAS is an open-source generic model for management scenarios programmed in Visual Basic by Per Sparre at the Danish Institute of Fisheries Research. It uses the FAO BEAM models (http://www.fao.org/fi/statist/fisoft/BEAM4.asp) to represent multiple fisheries with harvest and economic components. ISIS (Mahe´vas and Pelletier 2004) provides a similar model, written in Java, with spatially explicit components.

352

M.N. Maunder et al.

11.3.6 Advantages A general package offers a streamlined development path for implementing stock assessments. It eliminates the need for developing customized code that must be debugged and tested. End users need not be highly skilled programmers or analysts who work at the forefront of method development. However, they still must have an adequate understanding of fisheries stock assessment methods to ensure that they apply the general model appropriately. If the model receives consistent use for numerous assessments, the interest groups will gain a better understanding and accept the results more readily. Presumably, general models have already been tested in multiple applications, with a much lower risk of programming errors than in custom models.

11.3.7 Disadvantages Unfortunately, the advantages of a general package can sometimes turn into disadvantages. Availability can promote misuse through a lack of understanding. Someone with limited experience and expertise might adopt general software and use the results naively to generate management recommendations. A user might invoke inappropriate assumptions for the stock in question. Similarly, unfamiliarity with the software could lead to incorrect settings for the control variables. Improvements in documentation, user-friendly graphical interfaces, and automated error and consistency checking may help resolve such problems. In some cases, a ‘‘general’’ package may still lack a feature unique to a particular assessment context. The user is then forced to shoe-horn the assessment into the model, perhaps with unwarranted assumptions. Even if everything works as planned, general stock assessment software packages typically have several unused features for each modeling application. This can result in added overhead and cause the model to perform more slowly than one written specifically for the problem at hand.

11.3.8 Summary Fishery scientists have invested substantial effort in developing software packages that can deal with most common stock assessment situations. If understood properly and applied correctly, these programs could potentially make it possible to spread limited resources across more stocks, including some that would otherwise not be assessed. For example, in 2005 SS2 was used to assess 15 U.S. Pacific groundfish stocks. The software facilitated not only the production of stock assessments, but also the subsequent review processes. Innovations have the potential to remove some disadvantages of applying a

11

Computers in Fisheries Population Dynamics

353

general model, such as online tutorials and smart systems designed to avoid pitfalls. A new generation of software might even help practitioners build and test their own assessment models, with a greater appreciation for the concepts involved (Schnute et al. 2007).

11.4 Recent Advances in Model Design and Application The increased power of computers and software has greatly extended the range of analytical methods available for stock assessment. In this section, we outline some of the important developments that have taken place in the last two decades.

11.4.1 Integrated Analysis Fisheries stock assessment models have grown to accommodate increasing quantities and types of data (Quinn 2003). Historically, an independent analysis applied to each data type, with a final comparison across the spectrum of results. In some cases, the output from one analysis became input for another analysis. Advances in parameter estimation methodology and increases in computing power have allowed the integration of multiple data sets or analyses into a single analysis (e.g., Fournier and Archibald 1982; Methot 1990; Hampton and Fournier 2001). The historical two-step approach has a number of serious potential problems, such as loss of information in the summaries, inconsistent assumptions, undefined error structure; and reduced diagnostic ability (Maunder 1998; 2001a). An integrated approach can overcome such problems or at least identify the need for their resolution. However, it also introduces new technical challenges, including high computational requirements, robust convergence methods, confounded parameters, model selection (with potential misspecification), weighting among data sets, and a suitable level of data abstraction. In cases where catch-at-age data are not directly available, recent models take direct advantage of catch-at-length data (e.g., Fournier et al. 1998), which are usually easier to obtain. This single step process eliminates the need for an intermediate step of converting length-frequency data into age-frequency data, with the usual difficulties that accompany multi-stage analyses. Other composite analyses include:

 adding stock-recruitment relationships to assessment models (Francis 1992; Smith and Punt 1998; Ianelli 2002),

 incorporating tagging models into assessment models (Maunder 1998; 2001b; Hampton and Fournier 2001),

 embedding the CPUE standardization process within assessment models (Maunder 2001a; Jiao and Chen 2004; Maunder and Langley 2004),

354

M.N. Maunder et al.

Biomass of fish 0.75+ years old (thousands of tons)

1500

Longline Floating object Small discards Biomass with fishing Biomass with no fishing

1000

500

0 75

77

79

81

83

85

87

89

91 Year

93

95

97

99

01

03

05

Fig. 11.5 Fishery impact plot for bigeye tuna in the eastern Pacific Ocean (from Maunder and Hoyle 2007). The solid black line represents the estimate of abundance over time. The dashed black line represents the abundance that would have existed if fishing did not occur. The difference between the two lines represents the impact of the fishery. The change in the dashed line over time represents the effect of the environment on the population. The shaded areas represent the proportions of the fishery impact that were due to the different fishing methods

 integrating environmental data into assessment models (Maunder and Watters 2003b), and

 using growth increment data in a length-based model (Bentley et al. 2001). The integrated approach gives a more comprehensive analysis of the data and usually leads to a better presentation of results. For example, the assessment of tunas in the eastern Pacific Ocean (Maunder and Watters 2003a) separates the causes of abundance variation among the environment, the fishery, and different fishing methods within the fishery (Fig. 11.5).

11.4.2 Process Error Analyses often ignore large sources of process error in fisheries population dynamics, which include annual variations in recruitment, natural mortality, fishery selectivity, and growth. For example, many age-structured stock assessment models include process error in recruitment only, perhaps modeled by a temporal deviate from a lognormal distribution (e.g., Ianelli and Zimmerman 1998). The standard deviation of this distribution is typically fixed at a

11

Computers in Fisheries Population Dynamics

355

predetermined level because appropriate methods to estimate this parameter require integration across the random effect, a computationally intensive process (Maunder and Deriso 2003). The package ADMB-RE (Skaug 2002; Skaug and Fournier 2006) includes both Laplace approximation and importance sampling to make such analyses possible. Alternatively, a full Bayesian MCMC approach automatically integrates across the random effect, but requires priors for all model parameters.

11.4.3 Estimating Uncertainty Uncertainty in estimated quantities of interest, including future projections, plays an important role in the quantitative analysis of fisheries population dynamics (Figs. 11.6 and 11.7). Common methods to estimate uncertainty (Restrepo et al. 2000; Patterson et al. 2001) include (1) a normal approximation obtained from the parameter covariance structure, (2) a bootstrap, (3) a confidence region from a likelihood profile, and (4) a full Bayesian analysis from a posterior sample. The normal approximation method is the least computationally intensive; it requires

600000

Tons

400000

200000

0 75 77 79 81 83 85 87 89 91 93 95 97 99 01 03 05 07 09 11

Year Fig. 11.6 Estimates of historic and future abundance of bigeye tuna in the eastern Pacific Ocean (from Maunder and Hoyle 2007). The large red circle indicates the current estimate of abundance (start of 2006). The model is projected forward in time using current effort levels. The shaded area represents the approximate 95% uncertainty intervals based on the delta method

356

M.N. Maunder et al. 1.00

Cumulative probability

0.75 2005 2008 2010

0.50

Constant catch of 1.3 million tons 0.25

0.00 0

1 Biomass relative to B40%

2

Fig. 11.7 Estimated cumulative probability that projected female spawning biomass levels of eastern Bering Sea walleye pollock will drop below B40% based on a fixed constant catch level of 1.3 million tons. The cumulative probability is estimated from the marginal distribution of the full joint posterior distribution based on a thinned MCMC chain. Corresponding expected values (means) are shown by the vertical lines terminated with closed circles (from Ianelli et al. 2005)

only the parameter estimates and an estimate of the Hessian matrix at the mode. However, it often performs poorly in realistic problems with departures from colinearity among parameters and/or a skewed likelihood surface. The remaining methods tend to require intensive computation. For example, bootstrap methods require the objective function to be optimized hundreds of times. The profile likelihood requires the optimization of the objective function tens of times for each quantity of interest. Bayesian posterior sampling methods require the function to be evaluated millions of times (Punt and Hilborn 1997). For some modern stock assessments that contain thousands of parameters, only the first method is practical, and an analyst must resort to the normal approximation (Maunder et al. 2006). All methods of assessing uncertainty depend on approximations and unknown sampling properties. The normal theory of likelihood approximation is actually violated in most fishery models, where the number of parameters increases with the number of observations. Bootstrap and Bayesian samples must be ‘‘large enough’’, but how large is that? Simulations can be used to test a method of measuring uncertainty (e.g., Maunder et al. 2006), although these typically require very substantial computing resources.

11

Computers in Fisheries Population Dynamics

357

11.4.4 Weighting of Data Sets Integrated analyses of multiple data sets and the rising demands for measures of uncertainty have focused attention on a fundamental issue. What relative weight should the model give to each data set analyzed? A strict adherence to likelihood theory gives weights associated with variance parameters; although the classical theory of errors-in-variables shows that in some cases these are not all estimable without further information (Schnute and Richards 1995). Furthermore, even this analysis depends on the distribution used in the model to represent proportions associated with frequency data. Schnute and Haigh (2007) point out that the multinomial represents a poor choice, and they examine the Dirichlet and logistic-normal as alternatives. Research continues on methods of assigning weights to model components, based on combinations of measurement and process error (Pennington and Volstad 1994; McAllister and Ianelli 1997; Pennington et al. 2002; Francis et al. 2003; Miller and Skalski 2006; Deriso et al. 2007).

11.4.5 Bayesian Analysis Many fishery scientists have now adopted a Bayesian statistical framework for their analyses (Punt and Hilborn 1997). A standard implementation requires some method of posterior sampling (e.g., SIR, MCMC, Gibbs). All of these are computationally intensive, sometimes even impractical for highly-parameterized stock assessment methods (e.g., Fournier et al. 1998). MCMC has become a preferred method, partly due to its generality, ease of use, and scalability to larger problems. Furthermore, a large community currently uses MCMC for a variety of problems (Gelman et al. 1995). ADMB includes a convenient implementation. WinBUGS, as the package name (Bayesian inference Using Gibbs Sampling) indicates, starts with the Gibbs method, although it may use other algorithms in appropriate circumstances. Serious ADMB users might quickly dismiss WinBUGS as a mere toy, due to its slow operation, but both of these packages have greatly increased the understanding and application of Bayesian techniques. The introduction of MCMC sampling has made Bayesian analysis applicable for situations in which more traditional methods are impractical. For example, classical random effects estimation in non-linear models requires both maximization and integration, a challenging and time consuming numeric process. In such cases, non-linear hierarchical Bayesian analysis often becomes a method of convenience, without much regard to the role of Bayesian theory in the calculation. Because integrated analysis strives to account for all available data, creating data-based priors becomes problematic (Maunder 2003). If all data from the population of interest have been integrated into the analysis, prior information

358

M.N. Maunder et al.

can come only from other similar circumstances or expert judgments (MinteVera et al. 2004). Bayesian analysis always requires priors for all parameters, and well designed software packages like WinBUGS force the user to specify them. In a problem like simple linear regression, it can be a bit startling to realize that you really do need to give priors for the intercept a and the slope b. For cases with no legitimate sources of prior information, a broad uniform or other diffuse distribution can serve as an ‘‘uninformative’’ prior. Then the analyst can at least check each parameter to see whether or not the posterior differs notably from the prior. If not, it seems safe to conclude that the model and data offer little information about that parameter. For example, Schnute and Haigh (2007) illustrate a simple example with a selectivity parameter that seems blithely unconcerned about the structure imposed by the model-data combination. However, Maunder (2003) describes fisheries stock assessment examples for which the choice of the vague prior and the model parameterization can substantially influence the results of derived quantities. We admit that this relaxed attitude may not always be appropriate. A great deal of research into ‘‘objective Bayes’’ statistics (Bayarri and Berger 2004) tries to determine which default priors have reasonable properties according to various criteria. However, these methods can be difficult to apply in the highly-parameterized nonlinear dynamic models used in fisheries, and little work has been done in this area of fisheries research (Millar 2002). One promising method uses confidence distributions generated from bootstraps and profile likelihoods as objective Bayes posteriors. These have good frequentist properties and are invariant to parameter transformations (Schweder and Hjort 2002; Maunder et al. 2006).

11.4.6 Meta-Analysis As we discussed earlier, the availability of general stock assessment software has increased the number of stock assessments that can actually be conducted. Recent concerns about species at risk have mandated new assessments for species that have low commercial importance and limited available data. These assessments require relevant information to be collected from any source possible, via shared information from other populations or similar species. The available information can be encapsulated as a prior distribution for Bayesian analysis (Hilborn and Lierman 1998; Dorn 2002; Minte-Vera et al. 2004). The prior can be developed by simultaneously analyzing many data sets from relevant populations and estimating the distribution of a key parameter, such as the steepness of the stock-recruitment relationship (Myers et al. 1999). Most meta-analyses use simple models or summarized data. For example, estimates of spawning stock size and recruitment from stock assessment models are fit using stock-recruitment models with a parameter of the stock-recruitment model treated as a random effect that generates a distribution for the

11

Computers in Fisheries Population Dynamics

359

steepness of the stock-recruitment relationship (e.g., Myers et al. 1999). However, this method generally does not account for the uncertainty in the estimates of recruitment or spawning stock size. A more comprehensive method would simultaneously take account of all sources error in the meta-analysis when generating a Bayesian model for the new stock. Not surprisingly, such elaborate extensions can become computationally intensive.

11.4.7 Spatial Structure New data collection methods have greatly increased the quantity and detail of available data. In particular, recent databases include volumes of information with spatial coordinates at a course or fine level. Models that incorporate spatial structure and fish movement have much greater complexity and computational requirements than classical models for a single region. Spatially-structured models that estimate parameters by including detailed tagging data have already become standard tools for some stock assessments (Maunder 1998; 2001b; Hampton and Fournier 2001). They also play a role in Alaska groundfish management, where fine-spatial scales are required to protect sensitive ecosystem features, such as endangered stocks of Steller sea lions. One such model links age-structured stock assessment estimates with spatial patterns of bycatch and fleet behavior to evaluate alternative management scenarios (Anon. 2004).

11.4.8 Model Selection and Averaging The increased focus on uncertainty in providing advice to managers has led to including model uncertainty in stock assessments. For example, a model might include either a Ricker or a Beverton-Holt recruitment curve, and the data may or may not be informative as to which choice is better (Spiegelhalter et al. 2002; Wilberg 2005). One method of evaluating model uncertainty involves a sensitivity analysis of the model structure. However, a composite Bayesian analysis can average over model choices to produced measures of uncertainty that account for (predefined) alternatives (Parma 2001). The technique of Management Strategy Evaluation (MSE) deals with model uncertainty by employing operating models configured to a variety of potential specifications.

11.4.9 Role of Environmental Data An old debate (Skud 1975) persists about the relative influence of fishing versus the environment on fish population dynamics. Many current analyses consider

360

M.N. Maunder et al.

environmental influences as components of dynamic models. For example, some applications include environmental covariates within the stock assessment itself (Ianelli et al. 1999; Maunder and Watters 2003b) or within the CPUE standardization process (Maunder and Punt 2004). Because environmental changes often work on small spatial and temporal scales, computationallyintensive spatially-structured models also play a role (Lehodey et al. 1997). In such spatially explicit models, it can be difficult to estimate all required model parameters or to perform hypothesis tests. The legendary ‘‘curse of dimensionality’’ strikes again!

11.4.10 Policy Design and Evaluation Increased computing power and available software has stimulated a dramatic increase in fishery model complexity. These developments have forced researchers and managers to realize that management advice contains much more uncertainty than previously recognized. Consequently, the focus has shifted away from methods designed to produce the best estimates of stock status toward developing management advice that is genuinely robust to uncertainty (de la Mare 1986; Butterworth et al. 1997; De Oliveira et al. 1998; Butterworth and Punt 1999; Smith et al. 1999; Schnute and Haigh 2006). Management Strategy Evaluation (MSE) defines a set of potential management strategies, which are then tested against realistic biological scenarios. Strategies that prove robust to uncertainty and provide desirable outcomes become acceptable candidates for actually managing the stock. The management strategy defines the data to be collected, the method used to analyze these data, and the management actions taken based on those results. Given criteria for evaluating the outcome, MSE acts like a job interview: can you really do the job or not? Does it really matter whether or not we include random effects or any of the other recent proposed innovations? Can we use some simple old procedures that have been around for decades? When does added complexity actually help? The answer from MSE is simple: let’s conduct the job interviews and find out. MSE is computationally intensive (Fig. 11.8). For each year of the simulation to test a management strategy, artificial data are created and an assessment is carried out. This cycle gets repeated iteratively through a series of years. Furthermore, the entire iterative process must be repeated many times for each potential biological scenario. When complete, these extensive calculations represent results from random measurement error in generating the data and process error in the population dynamics. For example, in a 10-year projection, a full stock assessment must be carried out 10 times, and this cycle must be repeated many times for each potential biological scenario. The full

11

Computers in Fisheries Population Dynamics

Management Procedure

361

Slave computer

Master computer Apply assessment model to develop states of nature

Repeat for 10 years

Generate Artificial data

Data collection scheme

Assessment scheme

Management rule

Performance criteria

Simulate population one year

Apply assessment

Repeat 500 times: If slave computer available, randomly draw one state of nature and send to slave computer

Apply management rule

Evaluate Performance And send results to master computer

Synthesize results and determine most robust management procedure

Fig. 11.8 Flow of information in a management strategy evaluation implemented using a distributed computing environment. The master computer controls the analysis and sends individual analyses to slave computers when they become available

simulation may require the stock assessment to be conducted thousands of times for each management strategy, and there may be several candidate management strategies. Obviously, we’ve suggested a huge computational project. In its initial phases, MSE tends to involve relatively simple assessment methods that don’t rely on statistical age-structured models. Some studies have shown that simple assessment methods perform better than more complex methods, even when complex methods are used to generate the data (Ludwig and Walters 1985; Hilborn 1979; Punt 1988, 1993). At the time of writing this chapter, however, simple methods have not been compared systematically with the more complex methods widely used for modern stock assessments. Any evaluation must begin with a presumed set of biological scenarios, including initial population states. This choice introduces just one of many dimensions to be explored in a comprehensive test. What possibilities should be considered, and with what probability? One method uses simulated results from a dynamic model comparable with a modern stock assessment model that has substantial complexity in the population dynamics and Bayesian priors on potential model parameters and structural assumptions (Parma 2002). The posterior distribution could then represent a spectrum of biological scenarios to be used in MSE tests.

362

M.N. Maunder et al.

11.5 Discussion 11.5.1 Intuition and Common Sense We have emphasized that computers influence fisheries population models by implementing complex algorithms. Increases in computing speed, improvements in software, and the design of new algorithms have made it possible to conduct sophisticated analyses of fishery data. While the results certainly have the potential to reveal new interpretations of the available information, they can also detract from intuitive understanding. The authors of this chapter have all attended stock assessment meetings dominated by heated debates about structural assumptions, weights among likelihood components, model choices, and many other technical issues (e.g., Maunder and Starr 2002). Managers and other stakeholders usually get left behind in such discussions, and a clear link between input data and policy options sometimes gets lost. Although the clearly-defined analyses in classical textbooks (e.g., Beverton and Holt 1957; Ricker 1975) may seem too restrictive by modern standards, they have the advantage of providing definite computational goals with clear biological interpretations. Schnute (2006) reviews Ricker’s contributions to quantitative fishery models, with concepts that evolved through three influential handbooks. In the preface to his second handbook, Ricker (1958, p. 14) offers timeless advice to every fishery scientist: . . . the practising biologist quickly discovers that the situations he has to tackle tend to be more complex than those in any Handbook, or else the conditions differ from any described to date and demand modifications of existing procedures. It can be taken as a general rule that experiments or observations which seem simple and straightforward will prove to have important complications when analyzed carefully – complications which stem from the complexity and variability of the living organism, and from the changes which take place in it, continuously, from birth to death.

Ricker addressed this problem by starting with the simplest interpretations and proceeding to more complex analyses. His approach used intuition to guide the process of deeper understanding. Following a tradition in Kendall’s books on the advanced theory of statistics, O’Hagan and Forster (2004) begin their volume on Bayesian statistics with a quote from a novel, in this case The Tin Men (Frayn 1965). The protagonist Goldwasser has a ‘‘sort of cerebral hypochondria.’’ He fears that his brain might have peaked at the age of 30 and that his mental performance might now be declining with age. To test this possibility, he borrowed sets of IQ tests from his colleagues, and timed his performances, plotting the results on graphs. When he produced a graph with a curve that went downward, he assured himself that it was merely a misleading technique; and when he produced one with a curve that went upwards, he told himself skeptically that it must be the result of experimental error.

11

Computers in Fisheries Population Dynamics

363

Like Goldwasser, fishery data analysts can usually find reasons to doubt their own conclusions. The data, usually inadequate, often have multiple interpretations (Schnute 1987; Richards 1991; Schnute and Hilborn 1993). No arguments about model choices, weighted likelihood components, or other sophisticated approaches can gloss over this problem. Even the most sophisticated software won’t fix it. For this reason, we consider it important not to let the analysis stray too far from the input data. A really useful model might only suggest techniques for graphing the raw data so that the conclusions (or ambiguities) become fairly obvious to a general audience. The fishery on Pacific ocean perch (Sebastes alutus) off the coast of British Columbia, Canada, illustrates some of the issues that we have in mind. Suppose that a typical catch-at-age model were used to set the quota. Following common practice, the data might be indexed by year and lumped across a broad geographic region, such as that shown in Fig. 11.1. However, for this particular fishery, an observer program gives spatial coordinates for all commercial tows. The two panels in Fig. 11.1 show a changing pattern in the fishery between 1997 and 2005, where a moderately-productive region near the center of the map in 1997 seems no longer active in 2005. Why? Have fish populations declined? Have catch restrictions on other species caused fishermen to avoid this region? Without further information, we simply can’t know the answer, although we could build models supporting these hypotheses and many others. A standard lumped model would contain none of the spatial information in the figure, so that such questions might not even be discussed in that context. Although more sophisticated spatial models could be applied (e.g., Hampton and Fournier 2001), even they would required further information to resolve the ambiguities inherent in the data. If the analysis also has strong intuitive meaning, it might still be possible to consider important external information when making a management decision.

11.5.2 Software and Computing Choices Two developments in networking have greatly increased the computer power available for some of the analyses that we have detailed above. Parallel processing now allows a single program to run across multiple computers. For example, in a spatially-structured stock assessment with tagging data, each tag cohort is modeled as a separate population. Within the parameter optimization process, each tag cohort could be analyzed simultaneously on a separate computer (processor) and the results collated as the objective function is calculated (D. Fournier pers. com). AD Model Builder has parallel processing capability, but this feature still has not received widespread use. The second development, distributed computing, is much more accessible to most researchers. It allows separate programs to run on multiple computers under the control of a single primary computer. For example, a distributed

364

M.N. Maunder et al.

computer system can run a stock assessment model on employees’ computers during idle periods, such as evenings or lunch breaks. Distributed systems are ideal for situations in which an analysis must be run many times with slight changes, such as bootstraps, MSE simulations, profile likelihood calculations, ecosystem simulations, and sensitivity analyses. We have found condor (http://www.cs.wisc.edu/condor/) very useful for simulation testing of new statistical methodologies. E.J. Dick (pers. comm.) has developed a distributed computing system at the Southwest Fisheries Science Center (SWFSC) in Santa Cruz. He used it to obtain jackknife CVs for delta-GLM abundance indices. Microsoft has recently released Windows Compute Cluster Server for High-Performance Computing, which may make distributed computing much more available to general computer users. Software usage depends greatly on its availability and ease of application. We have seen an explosion in the use of general purpose free software, such as R and OpenBUGS. Due to their scientific and academic nature, these products receive extremely high levels of development and support. Some software gets limited use due to narrow specialization (e.g., MULTIFAN CL) or commercial cost (e.g., S-PLUS, Gauss, MATLAB). Progress in fisheries population dynamics software often comes through the development of freely-available software tools and general models. Unfortunately, the success has been much more limited than we would hope. Following the R and OpenBUGS framework, the underlying tools need to be general and free. Current impediments to progress include a limited number of qualified quantitative fisheries scientists and a lack of broad-based support by many institutions. A viable open-source project must include a rigorous framework for testing new packages to ensure at least minimal standards of documentation and operation. Because R already has these features, its library system offers a potential route toward this goal. The Fisheries Library in R represents a significant step in the right direction (Kell et al. 2007; Schnute et al. 2007). As analytical methods converge between fisheries and other areas of ecology (e.g., Maunder 2004), software that crosses these disciplines will become more successful.

11.5.3 Future Prospects We see some fairly clear trends in the direction of quantitative analysis. New data sources will become available, and models will evolve in complexity to incorporate new ideas available from richer data sets. This process has already begun in regard to spatially-disaggregated data (Fig. 11.1). Methods will also be developed to share information from other stocks through the use of priors or more complex meta-analyses, perhaps in the form of multi-species models. Environmental data will be included into the analyses on a scale appropriate to the processes that are affected. New statistical methods (illustrated currently

11

Computers in Fisheries Population Dynamics

365

by the Laplace approximation, importance sampling, and efficient MCMC methods) will allow analysts to handle this added complexity and represent dynamic processes more realistically. Many current models lack the ability to deal adequately with process error and demographic stochasticity. We foresee a broader framework for representing all sources of uncertainty, including uncertainty about the model itself. Although no magic wand exists to resolve all ambiguities in ecological data, we anticipate improved methods that bring these ambiguities into better focus and allow scientists to speculate meaningfully about new data acquisition programs. With the increased power of computers, new methods can be tested extensively, using simulations. These improvements will particularly impact the science of MSE. Our discussion of population dynamics has focused particularly on singlespecies stock assessment models. Computers also find extensive use in many other areas of fisheries population dynamics. For example, increases in computing power have greatly improved our ability to apply models at the level of individuals (agent-based models, http://en.wikipedia.org/wiki/Agent_based_ model), single species, multiple species (Stefansson 1998; Jurado-Molina et al. 2005), and entire ecosystems (Chapters 8 and 12). Theoretical studies based on differential equations similarly require computers to generate numerical solutions, and both computing power and improved algorithms have advanced this method of concept development. Linkages work both ways. Our field has benefited greatly from scientific developments elsewhere, and conversely stock assessment methods have found their way into other areas of research, from wildlife sciences to commodity modeling. As this convergence increases, we anticipate a better exchange of ideas among fields to improve methods used in fisheries. The future of fisheries population dynamics looks bright. The developments discussed here indicate potential paths toward progress. Unfortunately, fishery modeling often occurs in isolation, with results that can’t readily be applied elsewhere. Perhaps science always develops in disparate ways, with many independent discoveries of the same results. Nevertheless, real progress requires accumulation, in which new discoveries build on previous results and generalities begin to emerge. For example (Schnute and Haigh 2006), after extensive tests of management strategies in diverse laboratories and universities, what general results do we now know? What are some really good strategies, and under what biological conditions can we expect them to perform well? Under what circumstances can we say convincingly that the data won’t suffice to make a strategy feasible? Could future students be taught standards for stock assessment, based on extensive, replicated simulation experiments? Could training follow the example of engineers, who learn construction standards from known properties of materials? To answer such questions, we believe that a comprehensive collaboration will be required that brings together the limited expertise spread thinly throughout government research organizations and universities. First and foremost, the project should follow the open-source concepts that have made R and other

366

M.N. Maunder et al.

projects so successful. This will allow a much greater involvement of scientists world-wide and a greater uptake by the scientific community at large. The project should take advantage of contemporary algorithms and computing systems, and it should provide the essential tools for developing fisheries population dynamics models. While encouraging new developments, it should also provide a general stock assessment model that is flexible adaptable, scalable, and well documented. The documentation should include books, written independently by multiple authors, similar to the literature currently available for R. Perhaps an expert system could be created for this model to help develop assessments, make linkages to a framework for MSE, and avoid improper usage. Operating on a computer cluster freely available to remote users, it would encourage replicable experiments that generate scientific results reported in the literature, perhaps in a special online journal. Then, at last, stock assessment might progress as a real science, rather than a disparate collection of adhoc techniques. Acknowledgments We thank Rowan Haigh for help producing Figs. 11.1, 11.2 and 11.3 and giving helpful comments on earlier versions of this chapter. Clara Ulrich-Rescan provided information on simulation models used in Europe. Adam Langley provided the MFCL viewer screen shot.

References Anon (2004) Alaska Groundfish Fisheries Final Programmatic Supplemental Environmental Impact Statement. http://www.fakr.noaa.gov/sustainablefisheries/seis/intro.htm. 7000 pp Bayarri MJ, Berger J (2004) The interplay between Bayesian and frequentist analysis. Statistical Science 19:58–80 Begley J, Howell D (2004) An overview of Gadget, the Globally applicable Area-Disaggregated General Ecosystem Toolbox ICES CM 2004/FF:13. See http://www.hafro.is/gadget/ Bentley N, Breen PA, Starr PJ, Kendrick T (2001) Assessment of the CRA3 and NSS stock of red rock lobster (Jasus edwardsii) for 2000. New Zealand Fisheries Assessment Report 2001/69 Beverton RJH, Holt SJ (1957) On the dynamics of exploited fish populations. U.K. Ministry of Agriculture, Fish & Fisheries Investigations (Series 2) 19 Bull B, Francis RICC, Dunn A, McKenzie A, Gilbert DJ, Smith MH (2005) CASAL (C++ algorithmic stock assessment laboratory): CASAL user manual v2.07-2005/08/21. NIWA Technical Report 127. See http://www.niwascience.co.nz/ncfa/tools/casal Butterworth DS, Cochrane KL, De Oliveira JAA (1997) Management procedures: a better way to management fisheries? The South African experience. In: Pikitch EL, Huppert DD, Sissenwine MP (eds) Global Trends: Fisheries Management. American Fisheries Society Symposium 20, Bethesda, pp 83–90 Butterworth DS, Punt AE (1999) Experiences in the evaluation and implementation of management procedures. ICES Journal of Marine Science 56:985–998 Butterworth DS, Ianelli JN, Hilborn R (2003) A statistical model for stock assessment of southern bluefin tuna with temporal changes in selectivity. African Journal of Marine Science 25:331–361 Campbell SL, Chancellier JP, Nikoukhah R (2006) Modeling and simulation in Scilab/Scicos. Springer, New York (Software available at http://www.scicos.org/book.html)

11

Computers in Fisheries Population Dynamics

367

Clifford P (1993) Discussion on the meeting on the Gibbs sampler and other Markov chain Monte Carlo methods. Journal of the Royal Statistical Society (Series B) 55:53–102 Dalgaard P (2002) Introductory statistics with R. Springer-Verlag, New York (Software in the R package ‘‘ISwR’’) de la Mare WK (1986) Simulation studies on management procedures. Reports of the International Whaling Commission 36:429–450 De Oliveira JAA, Butterworth DS, Johnston SJ (1998) Progress and problems in the application of management procedures to South Africa’s major fisheries. In: Funk F, Quinn II TJ, Heifetz J, Ianelli JN, Powers JE, Schweigert JJ, Sullivan PJ, Zhang CI (eds) Fishery Stock Assessment Models. Alaska Sea Grant College Program Report No. AK-SG-98-01, University of Alaska Fairbanks, pp 513–530 Deriso RB, Maunder MN, Skalski JR (2007) Variance estimation in integrated assessment models and its importance for hypothesis testing. Canadian Journal of Fisheries and Aquatic Science 64:187–197 Dorn MW (2002) Advice on West Coast rockfish harvest rates from Bayesian meta-analysis of stock–recruit relationships. North American Journal of Fisheries Management 22:280–300 Efron BE, Tibshirani RJ (1993) An Introduction to the Bootstrap. Chapman & Hall, New York Field JC, Ralston S (2005) Spatial variability in rockfish (Sebastes spp.) recruitment events in the California Current System. Canadian Journal of Fisheries and Aquatic Science 62:2199–2210 Fournier D, Archibald CP (1982) A general theory for analyzing catch at age data. Canadian Journal of Fisheries and Aquatic Science 39:1195–1207 Fournier DA, Sibert JR, Majkowski J, Hampton J (1990) MULTIFAN a likelihood-based method for estimating growth parameters and age-composition from multiple length frequency data sets illustrated using data for southern bluefin tuna (Thunnus maccoyii). Canadian Journal of Fisheries and Aquatic Science 47:301–317 Fournier DA, Hampton J, Sibert JR (1998) MULTIFAN-CL: a length-based, age-structured model for fisheries stock assessment, with application to South Pacific albacore, Thunnus alalunga. Canadian Journal of Fisheries and Aquatic Science 55:2105–2116 Francis RICC (1992) Use of risk analysis to assess fishery management strategies: a case study using orange roughy (Hoplostethus atlanticus) on the Chatham Rise, New Zealand. New Zealand Journal of Marine and Freshwater Research 49:922–930 Francis RICC, Hurst RJ, Renwick JA (2003) Quantifying annual variation in catchability for commercial and research fishing. Fishery Bulletin 101:293–304 Frayn M (1965) The Tin Men. Collins, London Gelman A, Carlin BP, Stern HS, Rubin DB (1995) Bayesian Data Analysis. Chapman and Hall, London Gomez C (ed.) (1999) Engineering and scientific computing with Scilab. Birkha¨user Boston (Book includes software on CD ROM.) Griewank A (2000) Evaluating derivatives: principles and techniques of algorithmic differentiation. Frontiers in Applied Mathematics 19. Society for Industrial and Applied Mathematics, Philadelphia, PA. Haddon M (2001) Modelling and quantitative methods in fisheries. Chapman & Hall/ CRC Press, Boca Raton (Software available at http://www.crcpress.com/e_products/downloads/download.asp?cat_no=C1771) Hampton J, Fournier DA (2001) A spatially disaggregated, length-based, age-structured population model of yellowfin tuna (Thunnus albacares) in the western and central Pacific Ocean. Marine and Freswater Research 52:937–963 Heifetz J, Fujioka JT (1991) Movement dynamics of tagged sablefish in the northeastern Pacific. Fisheries Research 11:355–374 Hilborn R (1979) Comparison of fisheries control systems that utilize catch and effort data. Canadian Journal of Fisheries and Aquatic Science 36:1477–1489

368

M.N. Maunder et al.

Hilborn R (1996) Computers in fisheries population dynamics. In: Megrey BE, Moksness E (eds) Computers in Fisheries Research. Chapman and Hall, London, pp 176–189 Hilborn R, Liermann M (1998) Standing on the shoulders of giants: learning from experience in fisheries. Reviews in Fish Biology and Fisheries 8:273–283 Hilborn R, Maunder MN, Parma A., Ernst B, Payne J, Starr PJ (2000) Documentation for a general age-structured Bayesian stock assessment model: code named Coleraine. Fisheries Research Institute, University of Washington, FRI/UW 00/01. Software available at http://fish.washington.edu/research/coleraine/ Hoggarth DD, Abeyasekera S, Arthur RI, Beddington JR, Burn RW, Halls AS, Kirkwood GP, McAllister M, Medley P, Mees CC, Parkes GB, Pilling GM, Wakeford RC, Welcomme RL (2006) Stock assessment for fishery management: a framework guide to the stock assessment tools of the Fisheries Management Science Programme. FAO Fisheries Technical Paper 487 (ftp://ftp.fao.org/docrep/fao/009/a0486e/a0486e00.pdf). Software available at http://www.fmsp.org.uk/Software.htm Ianelli JN (2002) Simulation analyses testing the robustness of productivity determinations from West Coast Pacific ocean perch stock assessment data. North American Journal of Fisheries Management 22:301–310 Ianelli JN, Zimmerman M (1998) Status of the Pacific Ocean Perch Resource in 1998. Pacific Fishery Management Council, Portland Ianelli JN, Fritz L, Walters G, Williamson N, Honkahleto T (1999) Bering Sea-Aleutian Islands Walleye Pollock Assessment for 2000. In: Stock Assessment and Fishery Evaluation Report for the Groundfish Resources of the Eastern Bering Sea and Aleutian Island Region, 2000. North Pacific Fishery Management Council, Anchorage Ianelli JN, Barbeaux S, Honkalehto T, Lauth B, Williamson N (2005) Assessment of Alaska pollock stock in the eastern Bering Sea. In: Stock Assessment and Fishery Evaluation Report for the Groundfish Resources of the Bering Sea/Aleutian Islands Regions. North Pacific Fishery Management Council, Anchorage, pp 31–124 Jiao Y, Chen Y (2004) An application of generalized linear models in production model and sequential population analysis. Fisheries Research 70:367–376 Jurado-Molina J, Livingston PA, Ianelli JN (2005) Incorporating predation interactions in a statistical catch-at-age model for a predator-prey system in the eastern Bering Sea Canadian Journal of Fisheries and Aquatic Sciences 62:1865–1873 Kell LT, Mosqueira I, Grosjean P, Fromentin J-M, Garcia D, Hillary R, Jardim E, et al. (2007). FLR: an open-source framework for the evaluation and development of management strategies. ICES Journal of Marine Science, 64: 640–646 Lehodey P, Bertignac M, Hampton J, Lewis A, Picaut J (1997) El Nin˜o Southern Oscillation and tuna in the western Pacific. Nature 389:715–718 Lewy P, Nielsen A (2003) Modeling stochastic fish stock dynamics using Markov Chain Monte Carlo. ICES Journal of Marine Science 60:743–752 Ludwig D, Walters CJ (1985) Are age-structured models appropriate for catch-effort data? Canadian Journal of Fisheries and Aquatic Science 42:1066–1072 Luenberger DG (1979) Introduction to dynamic systems: theory, models, and applications. John Wiley and Sons, New York Mahe´vas S, Pelletier D (2004) ISIS-Fish, a generic and spatially explicit simulation tool for evaluating the impact of management measures on fisheries dynamics. Ecological Modelling 171:65–84 Maindonald JM, Braun J (2003) Data analysis and graphics using R: an example-based approach. Cambridge University Press, Cambridge (Software in the R package ‘‘UsingR’’.) Maunder MN (1998) Integration of Tagging and Population Dynamics Models in Fisheries Stock Assessment. PhD dissertation, University of Washington Maunder MN (2001a) A general framework for integrating the standardization of catch-perunit-of-effort into stock assessment models. Canadian Journal of Fisheries and Aquatic Science 58:795–803

11

Computers in Fisheries Population Dynamics

369

Maunder MN (2001b) Integrated Tagging and Catch-at-Age ANalysis (ITCAAN). In: Kruse GH, Bez N, Booth A, Dorn MW, Hills S, Lipcius RN, Pelletier D, Roy C, Smith SJ, Witherell D (eds) Spatial Processes and Management of Fish Populations, Alaska Sea Grant College Program Report No. AK-SG-01-02, University of Alaska Fairbanks, pp 123–146 Maunder MN (2003) Paradigm shifts in fisheries stock assessment: from integrated analysis to Bayesian analysis and back again. Natural Resource Modeling 16:465–475 Maunder MN (2004) Population viability analysis, based on combining integrated, Bayesian, and hierarchical analyses. Acta Oecologica 26:85–94 Maunder MN (ed.) (2005) Inter-American Tropical Tuna Commission Workshop On Stock Assessment Methods, La Jolla, California, 7–11 November 2005. Available at http://iattc.org/PDFFiles2/Assessment-methods-WS-Nov05-ReportENG.pdf Maunder MN (2007) A reference list of papers that use AD Model Builder and its precursor AUTODIF. Available at http://admb-project.org/community/bibliography Maunder MN, Starr PJ (2002) Industry participation in stock assessment: The New Zealand SNA1 snapper (pagrus auratus) fishery. Marine Policy 26:481–492 Maunder MN, Deriso RB (2003) Estimation of recruitment in catch-at-age models. Canadian Journal of Fisheries and Aquatic Sciences 60:1204–1216 Maunder MN, Langley AD (2004) Integrating the standardization of catch-per-unit-of-effort into stock assessment models: testing a population dynamics model and using multiple data types. Fisheries Research 70:389–395 Maunder MN, Punt AE (2004) Standardizing catch and effort data: a review of recent approaches. Fisheries Research 70:141–159 Maunder MN, Watters GM (2003a) A-SCALA: an age-structured statistical catch-at-length analysis for assessing tuna stocks in the eastern Pacific Ocean. Inter-American Tropical Tuna Commission Bulletin 22:433–582 (in English and Spanish) Maunder MN, Watters GM (2003b) A general framework for integrating environmental time series into stock assessment models: model description, simulation testing, and example. Fishery Bulletin 101:89–99 Maunder MN, Harley SJ, Hampton J (2006) Including parameter uncertainty in forward projections of computationally intensive statistical population dynamic models. ICES Journal of Marine Science 63:969–979 Maunder MN, Hoyle SD (2007) Status of bigeye tuna in the eastern Pacific Ocean in 2005 and outlook for 2006. Inter-American Tropical Tuna Commission Stock Assessment Report 7:117–248 McAllister MK, Ianelli JN (1997) Bayesian stock assessment using catch-age data and the sampling/ importance resampling algorithm. Canadian Journal of Fisheries and Aquatic Sciences 54:284–300 Methot RD (1990) Synthesis model: an adaptable framework for analysis of diverse stock assessment data. International North Pacific Fishery Commission Bulletin 50:259–277 Millar RB (2002) Reference priors for Bayesian fisheries models. Canadian Journal of Fisheries and Aquatic Science 59:1492–1502 Miller TJ, Skalski JR (2006) Integrating design- and model-based inference to estimate length and age composition in North Pacific longline catches. Canadian Journal of Fisheries and Aquatic Sciences 63:1092–1114 Minte-Vera CV, Branch TA, Stewart IJ, Dorn MW (2004) Practical application of metaanalysis results: avoiding the double use of data. Canadian Journal of Fisheries and Aquatic Sciences 62:925–929 Murrell P (2006) R Graphics. Chapman & Hall/CRC Press, Boca Raton (Software in the R package ‘‘RGraphics’’) Myers RA, Bowen KG, Barrowman NJ (1999) Maximum reproductive rate of fish at low population sizes. Canadian Journal of Fisheries and Aquatic Sciences 56:2404–2419

370

M.N. Maunder et al.

Nielsen A (2000) Fish Stock Assessment using Markov Chain Monte Carlo. MSc Thesis, University of Copenhagen (http://www.dina.kvl.dk/anielsen/master/) O’Hagan A, Forster J (1994; 2004-2nd edition) Kendall’s advanced theory of statistics, Volume 2B: Bayesian inference. Arnold, London Parma AM (2001) Bayesian approaches to the analysis of uncertainty in the stock assessment of Pacific halibut. American Fisheries Society Symposium 24:111–132 Parma AM (2002) In search of robust harvest rules for Pacific halibut in the face of uncertain assessments and decadal changes in productivity. Bulletin of Marine Science 70:423–453 Patterson K, Cook R, Darby C, Gavaris S, Kell L, Lewy P, Mesnil B, Punt A, Restrepo V, Skagen DW, Stefansson G (2001) Estimating uncertainty in fish stock assessment and forecasting. Fish and Fisheries 2:125–157 Pella JJ, Tomlinson PK (1969) A generalized stock production model. Inter-American Tropical Tuna Commission Bulletin 13:421–458 Pennington M, Volstad JH (1994) Assessing the effect of intra-haul correlation and variabledensity on estimates of population characteristics for marine surveys. Biometrics 50:725–732 Pennington M, Burmeister LM, Hjellvik V (2002) Assessing the precision of frequency distributions estimated from trawl-survey samples. Fisheries Bulletin 100:74–80 Prager MH (1995) User’s manual for ASPIC: a stock-production model incorporating covariates, program version 3.6. NMFS, Southeast Fisheries Science Center Miami Lab Document MIA-92/93-55, Miami Prager MH, Willaims EH (2003) From the golden age to the new industrial age: fishery modeling in the early 21st century. Natural Resource Modeling 16:477–489 Punt AE (1988) Model Selection for the Dynamics of Southern African Hake Resources. MSc thesis, University of Cape Town Punt AE (1993) The comparitive performance of production-model and ad hoc tuned VPA based feedback-control management procedures for the stock of Cape hake off the west coast of South Africa. In: Smith SJ, Hunt JJ, Rivard D (eds) Risk Evaluation and Biological Reference Points for Fisheries Management. Canadian Special Publication in Fisheries and Aquatic Science 120, pp 283–299 Punt AE, Hilborn R (1996) Biomass Dynamic Models – User’s Manual. FAO Computerized Information Series (Fisheries) No. 10, Rome Punt AE, Hilborn R (1997) Fisheries stock assessment and decision analysis: the Bayesian approach. Reviews in Fish Biology and Fisheries 7:35–63 Punt AE, Hilborn R (2001) BAYES-SA – Bayesian Stock Assessment Methods in Fisheries – User’s Manual. FAO Computerized Information Series (Fisheries) No. 12, Rome (http://www.fao.org/docrep/005/Y1958E/Y1958E00.HTM) Quinn TJ II (2003) Ruminations on the development and future of population dynamics models in fisheries. Natural Resource Modeling 16:341–392 Restrepo NR, Patterson KR, Darby CD, Gavaris S, Kell LT, Lewy P, Mesnil B, Punt AE, Cook RM, O’Brien CM, Skagen DW, Stefa´nsson G (2000) Do different methods provide accurate probability statements in the short term? ICES CM 2000/V:08 Richards LJ (1991) Use of contradictory data sources in stock assessments. Fisheries Research 11:225–238 Ricker WE (1958) Handbook of computations for biological statistics of fish populations. Bulletin of the Fisheries Research Board of Canada No. 119 Ricker WE (1975) Computation and interpretation of biological statistics of fish populations. Bulletin of the Fisheries Research Board of Canada No. 191 Robert CP, Casella G (2004-2nd edition) Monte Carlo statistical methods. Springer, New York, NY Schnute J (1987) Data uncertainty, model ambiguity, and model identification: ‘‘Mirror, mirror on the wall, what model’s fairest of them all?’’. Natural Resource Modeling 2:159–212

11

Computers in Fisheries Population Dynamics

371

Schnute JT (2006) Curiosity, recruitment, and chaos: a tribute to Bill Ricker’s inquiring mind. Environmental Biology of Fishes 75:95–110 Schnute JT, Haigh R (2006) Reference points and management strategies: lessons from quantum mechanics. ICES Journal of Marine Science 63:4–11 Schnute JT, Haigh R (2007) Compositional analysis of catch curve data with an application to Sebastes maliger. ICES Journal of Marine Science 64:218–233 Schnute JT, Hilborn R (1993) Analysis of contradictory data sources in fisheries stock assessment. Canadian Journal of Fisheries and Aquatic Sciences 50:1916–1923 Schnute JT, Richards LJ (1995) The influence of error on population estimates from catch-age models. Canadian Journal of Fisheries and Aquatic Sciences 52:2063–2077 Schnute JT, Richards LJ, Olsen N (1998) Statistics, software, and fish stock assessment. In: Funk F, Quinn II TJ, Heifetz J, Ianelli JN, Powers JE, Schweigert JJ, Sullivan PJ, Zhang CI (eds) Fishery Stock Assessment Models. Alaska Sea Grant College Program Report No. AK-SG-98-01, University of Alaska Fairbanks, pp 171–184 Schnute JT, Boers NM, Haigh R (2004) PBS Mapping 2: User’s Guide. Canadian Technical Report of Fisheries and Aquatic Sciences 2549 (Software in the R package ‘‘PBSmapping’’, which includes this report as ‘‘PBSmapping-UG.pdf’’) Schnute JT, Couture-Beil A, Haigh R (2006) PBS Modelling 1: User’s Guide. Canadian Technical Report of Fisheries and Aquatic Sciences 2674 (Software in the R package ‘‘PBSmodelling’’, which includes this report as ‘‘PBSmodelling-UG.pdf’’) Schnute JT, Maunder MN, Ianelli JN (2007) Designing tools to evaluate fishery management strategies: can the scientific community deliver? ICES Journal of Marine Science, 64: 1077–1084 Schweder T, Hjort NL (2002) Confidence and likelihood. Scandinavian Journal of Statistics 29:309–332 Shepherd JG (1999) Extended survivors analysis: an improved method for the analysis of catch-at-age data and abundance indices. ICES Journal of Marine Science 56:584–591 Skaug H (2002) Automatic differentiation to facilitate maximum likelihood estimation in nonlinear random effects models Journal of Computational and Graphical Statistics 11:458–470 Skaug H, Fournier D (2006) Automatic approximation of the marginal likelihood in nonlinear hierarchical models. Computational Statistics and Data Analysis 51:699–709 Skud BE (1975) Revised estimates of halibut abundance and the Thompson-Burkenroad debate. International Pacific Halibut Commission Scientific Report 56:36 pp (Available at http://www.iphc.washington.edu/HALCOM/pubs/scirep/SciReport0056.pdf) Smith ADM, Punt AE (1998) Stock assessment of gemfish (Rexea solandri) in eastern Australia using maximum likelihood and Bayesian methods. In: Funk F, Quinn II TJ, Heifetz J, Ianelli JN, Powers JE, Schweigert JJ, Sullivan PJ, Zhang CI (eds) Fishery Stock Assessment Models. Alaska Sea Grant College Program Report No. AK-SG-98-01, University of Alaska Fairbanks, pp 245–286 Smith ADM, Sainsbury KJ, Stevens RA (1999) Implementing effective fisheries-management systems: management strategy evaluation and the Australian partnership approach. ICES Journal of Marine Science 56:967–979 Spiegelhalter DJ, Best NG, Carlin BP, van der Linde A (2002) Bayesian measures of model complexity and fit. Journal of the Royal Statistical Society (Series B) 64:583–639 Spiegelhalter D, Thomas A, Best N, Lunn D (2004) WinBUGS User Manual, version 2.0 (http://mathstat.helsinki.fi/openbugs/) Stefansson G (1998) Comparing different information sources in a multispecies context. In: Funk F, Quinn II TJ, Heifetz J, Ianelli JN, Powers JE, Schweigert JJ, Sullivan PJ, Zhang CI (eds) Fishery Stock Assessment Models. Alaska Sea Grant College Program Report No. AK-SG-98-01, University of Alaska Fairbanks, pp 741–758 Thomas N (2004) BRugs User Manual (the R interface to BUGS), version 1.0 (http:// mathstat.helsinki.fi/openbugs/, Software in the R package ‘‘BRugs’’)

372

M.N. Maunder et al.

Thompson GG (1992) A Bayesian approach to management advice when stock-recruitment parameters are uncertain. Fishery Bulletin 90:561–573 Thompson GG (1999) Optimizing harvest control rules in the presence of natural variability and parameter uncertainty. NOAA Tech. Memo. NMFS-F/SPO-40 Venables WN, Ripley BD (1994, 1999-3rd edition) Modern applied statistics with S-PLUS. Springer Verlag New York (Software in the R package ‘‘MASS’’) Venables WN, Ripley BD (2000) S Programming. Springer Verlag, New York (Software available at http://www.stats.ox.ac.uk/pub/MASS3/Sprog/) Walther A, Kowarz A, Griewank A (2005) ADOL-C: a package for the automatic differentiation of algorithms written in C/C++, version 1.10.0. Report and software available from: http://www.math.tu-dresden.de/%04adol-c/ Wilberg MJ (2005) Improving statistical catch-at-age stock assessments. PhD Dissertation, Michigan State University Wood S (2006) Generalized Additive Models: An Introduction with R. Texts in Statistical Science 67. Chapman & Hall/CRC Press, Boca Raton (Software and data available in the R packages ‘‘mgcv’’ and ‘‘gamair’’)

Chapter 12

Multispecies Modeling of Fish Populations Kenneth A. Rose and Shaye E. Sable

12.1 Introduction Computers played a critical role in the initial development of multispecies fisheries modeling in 1970s, and will likely play a major role with the recent resurgence of interest in multispecies modeling. The initial availability of digital computing in the 1970s opened the door for multispecies modeling. Prior to digital computing, much effort was spent on obtaining analytical (closed-form) solutions of fisheries models or on making major simplifying assumptions to enable approximate solutions on a calculator. Considering models that involved multiple species was out of the question from a numerical point of view. Fisheries scientists had long recognized that multispecies interactions (competition and predation) were important to population dynamics, but solving multispecies models was simply too difficult or the assumptions required for a solution too restrictive. Once digital computing became available, there was a surge of activity involving multispecies models because numerical simulation became a widely available option. Most of these initial models were presented as demonstrations or illustrations of how inter-specific interactions could affect population dynamics (e.g., May et al. 1979). After the initial surge of activity, multispecies modeling retreated because such modeling was generally considered to stretch and overstep the available empirical information, and because management focused on population approaches. The multispecies modeling since the 1970s was generally tilted towards theoretical analyses (e.g., Matsuda and Katsukawa 2002). Fisheries management was almost completely focused on population modeling. In the US, the guidelines for the implementing the Magnuson-Stevens Act resulted in the exclusive use of population models for assessing the status of harvested stocks and for setting quotas and evaluating other management actions (Rose and Cowan 2003).

K.A. Rose (*) Department of Oceanography and Coastal Sciences, Louisiana State University, Baton Rouge, LA 70803, USA

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_12, Ó Springer ScienceþBusiness Media B.V. 2009

373

374

K.A. Rose and S.E. Sable

Interest in multispecies modeling is presently undergoing resurgence (Latour et al. 2003; Link 2002). This resurgence is due to the desire to move towards ecosystem-based fisheries management (Alaska Sea Grant 1999; NMFS 1999), the convergence of the fisheries and oceanography disciplines, increases in computing power, new measurement methods, and new ideas about system dynamics. Ecosystem-based fisheries management simply and absolutely requires multispecies models in order to address how food web and environmental variables affect the dynamics of the population(s) of interest. Also, there is growing interest in interdisciplinary approaches that combine fisheries and oceanography (e.g., Kendall et al. 1996; Runge et al. 2004). While this interest does not necessarily require multispecies approaches, it does add to the general momentum of thinking beyond the population-level approach. The power and availability of desktop computing have accelerated greatly, with powerful desktop personal computers now a staple on most every scientist’s desk. There is some hope that new measurement techniques (e.g., tagging of individuals, stable isotopes) may enable us to get over the hurdle that stopped the 1970s surge of interest when multispecies modeling quickly exceeded the available empirical knowledge. Finally, modeling methods have also expanded (e.g., individual-based approach), and we have expanded how to view ecosystem dynamics. It seems that the recent ideas of complex systems theory (Ban-Yam 1997) may help us better understand multispecies responses to perturbations, and may begin to explain the unexpected responses of populations that ‘‘surprise’’ ecologists (Paine et al. 1998). This chapter explores the role of computers in multispecies modeling of fish populations. First, we briefly summarize multispecies interactions and how they have been viewed in fisheries. We then review the general types of multispecies models that are available. This review is an update of the previous review in Rose et al. (1996). Third, we present an example analysis that compares population and predator-prey versions of matrix projection models against the predictions of an individual based model (IBM). Matrix projection modeling has been widely used for decades for both theoretical analysis and management of single species populations, and projection matrix-like calculations are often in age-structured multispecies models. However, the performance of multispecies matrix models has rarely been evaluated under controlled conditions. We conclude with a report card revisit to the future directions we suggested 10 years ago in the first edition (Rose et al. 1996), and a brief discussion of new, ‘‘updated’’ future directions.

12.2 Multispecies Interactions Multispecies interactions generally involve predation or competition. While predation is generally considered to be an interspecific interaction (except cannibalism), competition can be either intraspecific or interspecific. Both

12

Multispecies Modeling of Fish Populations

375

predation and competition can lead to density-dependent responses in the population of interest (Rose et al. 2001). Indeed, this is often how the complexity of interspecific interactions is simplified for use in single species models. Whether one can sufficiently capture the complexity of interspecific interactions in a single-species model is questionable, and the controversy that surrounds the specification of density-dependent survival, growth, and reproduction in single species models (Rose et al. 2001) confirms how difficult it is to formulate single species models for populations in highly-coupled food webs. The importance of interspecific interactions among fish populations is well recognized. The focus on single-species population models, especially for management, was not due to fisheries scientists being unaware of the roles of predation and competition in affecting population dynamics. Fisheries scientists know that fish live in a food web and that competition and predation are important. Rather, the emphasis on the population-level approach was a pragmatic decision. Multispecies models simply were anticipated to be too difficult to defend when the subsequent scrutiny was going to be triggered by unpopular management decisions. In many situations, predation mortality on the species of interest in marine systems is known to vary in time and space, and its magnitude is often equivalent to, or exceeds, harvesting rates; yet, natural mortality is often treated as a constant in many population models (Bax 1998). Specifying the nature of the predator-prey relationship involves the functional response and this itself has been controversial (Abrams and Ginzburg 2000). We seem to know more about predation mortality in freshwater lakes because they are enclosed systems and easier to sample. The idea of the trophic cascade came from considering food web effects of fish removals and additions in lakes (Carpenter and Kitchell 1993). Recently, such cascading effects from the top predators to nutrients have been documented in a coastal ecosystem (Frank et al. 2005). Large-scale perturbations to marine ecosystems, such as the removal of certain fish species via harvesting, have lead to dramatic changes in the food web, including changes in other fish species (e.g., Fogarty and Murawski 1998). Cury et al. (2000) suggested that small pelagic fishes in upwelling ecosystems exert top-down control on their zooplankton prey and bottom-up control on their predators. Yet, a meta-analysis of laboratory and field studies questioned the degree of coupling from fish all the way down to phytoplankton (Micheli 1999). Link (2002) compared marine food webs with freshwater and terrestrial food webs and suggested that marine food webs are highly connected with many weak species interactions and may require some re-thinking of the classical food web paradigms. Multispecies modeling of fish has been limited not because of ignorance about the importance of predation and competition, but because of limited information. The scientific evidence for the importance of interspecific interactions is strong, obvious, and well known. The relatively few truly structured multispecies modeling efforts [excluding Ecopath with Ecosim (EwE) which we do not consider to be structured for fish] have been focused on well-studied

376

K.A. Rose and S.E. Sable

salmonids in streams and rivers (Clark and Rose 1997a; Strange et al. 1993) and food webs in selected locations where extensive long-term data exist (e.g., Barents Sea – Tjelmeland and Bogstad 1998).

12.3 Multispecies Models Any taxonomy of models is dangerous. Once distinct categories are defined, some models always seem to then become classified as hybrids because they are intermediate between two categories or have features that span multiple categories. For convenience and consistency with the previous review (Rose et al. 1996), we discuss multispecies fisheries models under the categories of: budget models, coupled single-species models, individual-based bioenergetics models, and holistic or ecosystem models. Budget models are typically representations of biomass or energy flows among the compartments of a food web, and can be static or time-varying. Coupled models generally use the single species models and couple them to form, usually simple, community models. The third category of individualbased, bioenergetics models is a new category that was not explicitly defined in our previous review but was the subject of the example analysis in that review and was elevated to a distinct category in the review by Latour et al. (2003). Holistic models include environmental and non-fish biological variables as explicit compartments (e.g., nutrients, zooplankton). We use the term holistic rather than ecosystem because the oceanographers consider nutrient-phytoplankton-zooplankton (NPZ) models (i.e., no fish) to be ecosystem models. The distinction between budget models that include many compartments and holistic models is arbitrary. Budget models tend to treat the functions that determine the flow rates among compartments quite simply, sometimes simply a fixed rate, and without much within-population structure. Holistic models add biological detail to the functions that determine the flow rates among compartments, often with feedbacks built in. The review by Whipple et al. (2000) included a category of size-spectrum models; we do not consider them here because they do not deal with species as state variables. The popular EwE set of models (see Chapter 8) immediately offers an illustration of the problems with model taxonomies. In our taxonomy, one could argue that Ecopath started out as a budget model, and with the development and continued refinement, has evolved into a holistic model (Walters et al. 1997). There have been several recent reviews of multispecies modeling of fish (Latour et al. 2003; Whipple et al. 2000; Hollowed et al. 2000; Bax 1998). We present a few examples in each of our four categories to illustrate the various approaches. We repeat many of the same examples as presented before in Rose et al. (1996), and, where appropriate, we update these with some more recent examples.

12

Multispecies Modeling of Fish Populations

377

12.3.1 Budget Models Budget models have been used to model a variety of fish communities. Georges Bank was represented with about 10 compartments ranging from bacteria to several fish compartments (Cohen et al. 1982; Fogarty et al. 1987). These models estimated that fish consume 60–90% of their own production. Walsh (1981) compared a 13-compartment carbon flow budget of the Peruvian coastal ecosystem before and after overfishing of anchovy. He concluded that the decline in anchovy grazing pressure as a result of overfishing apparently led to increased plankton, sardine, and hake standing stocks, and ultimately caused a decline in water concentrations of oxygen and nitrate. Jarre et al. (1991) also presented a budget model of the Peruvian upwelling system. They included eight fish compartments and compared budgets and system-level indices among three periods with different anchovy biomass. They discussed the changing role of anchovy in the budget and how its role as prey for top fish predators was replaced by other fish species. Pace et al. (1984) constructed a 17-compartment (two fish compartments) energy flux budget of a generic continental shelf food web. They then explored the time-dependent version of the budget by constructing a differential equation for each compartment and simulated how fish production would increase much more under gradually increased nitrogen inputs versus pulsed nitrogen inputs. In a series of three papers (Polovina 1984; Atkinson and Grigg 1984; Grigg et al. 1984), a 12-compartment (four fish compartments) budget of biomass flows of coral reef ecosystem was constructed and analyzed. They concluded that the coral reef ecosystems were top–down regulated by predation rather than nutrient limited. These papers were also important in that they were the predecessor to the Ecopath model, which has subsequently been applied to many ecosystems (see Chapter 8).

12.3.2 Coupled Single-Species Models State variable and structured single-species models have been coupled to represent the dynamics of fish communities. A frequently cited paper is that of May et al. (1979) who used a series of Lotka-Volterra-like predator-prey models to examine the effect of harvesting in various combinations of food webs involving krill, cephalopods, baleen whales, and sperm whales. They concluded that the concept of maximum sustainable yield (MSY) was not useful for species other than the top predators, that different populations operated on different inherent time scales, and that already exploited populations were less resilient than virgin populations. The ideas present in this paper are still highly relevant today as we struggle to formulate models for ecosystem-based fisheries management.

378

K.A. Rose and S.E. Sable

Collie and Spencer (1994) were motivated by fluctuations in Pacific hake and Pacific herring and extended a predator-prey model proposed by Steele and Henderson (1981). The approach is similar to the models of May et al. (1979); coupled differential equations with logistic growth of prey and functional response for predation. Collie and Spencer included a stochastic variable that had first-order autocorrelation and affected the mortality rate of the predator. Model simulations showed multiple equilibria and stochasticity caused shifts between high and low abundances. Spencer and Collie (1995) modified the model for a case of the predator eating other prey than the modeled prey, and examined the spiny dogfish-haddock interaction on Georges Bank. Simulations with stochasticity also caused shifts between alternative equilibria, and showed that predator biomass could increase even when modeled prey biomass was low. The effects of harvesting on prey and predator populations and yields were compared. Allen and McGlade (1986) combined a multispecies fish model based on Lotka-Volterra representation with a dynamic model of fisheries, and embedded both in a spatially-explicit grid of cells. Their emphasis was on nonequilibrium analysis and the importance of not only modeling the fish but also the behavior of the fishers. Using coupled logistic growth models, Pope (1976) concluded that the total yield from an interactive two-species system would be less than the sum of the MSYs of each species. This was questioned by Brander and Mohn (1991) who showed that the result depended on how predation mortality was represented in the single-species calculations. Yodzis (1994) eloquently captured the fundamental drawback to the coupled single species models approach by showing that the choice of the functional form of the interaction terms can greatly influence model predictions. Subsequently, Yodzis (1998) developed a general modeling approach that used coupled differential equations, specific to prey and predators. He applied this model to the issue of the effects of culling fur seals on the fisheries in the Benguela ecosystem by representing the dynamics of 29 functional groups or species. His analysis was somewhat theoretical and used some simplifying assumptions (e.g., constant harvesting rates) to illustrate how the community responded to press-type perturbations and how the modeled community could be reduced in complexity. Koen-Alonso and Yodzis (2005) then applied the same modeling approach to a four species system of the Patagonia region (Argentine continental shelf) of the South Atlantic. Squid and anchovy were represented as prey (termed basal equations) and hake and sea lions as consumers. Sea lions also consumed hake. The prey differential equations were logistic growth with additional losses due to predation and other sources of mortality. The consumer equations used a functional response to determine the rate of prey consumption with additional losses due to predation upon them and density-dependent mortality. They compared alternative formulations for the function response, including the formulation used by EwE. Fits to observed data implied that all formulations could be fit to data. The addition of parameter uncertainty under conditions of increasing harvest rates showed that the

12

Multispecies Modeling of Fish Populations

379

different formulations resulted in different predictions of the effects of harvesting on the community. The long-term population oscillations of pelagic species that show replacement over time were modeled by Matsuda and Katsukawa (2002) using three coupled difference equations. Their motivation was the well-documented cycles among sardine, anchovy, and chub mackerel. They showed that a fishery that switches to the dominant species could avoid overexploitation and stabilize the catch. Structure in terms of age or stage classes can be added to each of the coupled population models. Each population is then represented by multiple compartments. Murawski (1984) proposed a multispecies, mixed fishery model using coupled population models with age-structure. The model was applied to the four-species, six-fishery community on Georges Bank (Murawski 1984) and to the 15-species, six-fishery demersal community of the Gulf of Maine (Murawski et al. 1991). They concluded that discards were important and affected model predictions even if some of the fisheries were curtailed. Wilson et al. (1991a,b) used an age-structured, five-species model to invoke the idea of complexity theory (chaotic behavior) in population dynamics of exploited species. Numbers at age were followed annually with each new yearclass predicted from species-specific spawner-recruit relationships; mean weight at age was assigned from Von Bertalanffy curves. The species were linked via a community predation term that acted to adjust the survival of the recruits. Total biomass over all ages and species was computed each year. Whenever the total biomass exceeded the specified carrying capacity biomass, biomass of recruits from species were reduced sequentially up to 75% of the bloom species, then up to 50% of first cod, then Pollack, and then haddock, and finally up to 25% of redfish. This sequential reduction in species-specific recruitment was repeated until total biomass no longer exceeded the carrying capacity or until all recruitment was lowered to zero. Simulations showed that, under some conditions, total biomass was relatively stable while the annual biomasses of individual species varied chaotically. Hilborn and Gunderson (1996) objected to the analysis, especially on the subsequent interpretation by Wilson et al. (1994) that their chaotic population dynamics results implied that single-species management focused on controlling harvest or effort was misguided. Hilborn and Gunderson (1996) discussed what they believed to be several unrealistic assumptions in the Wilson et al. model. Thomson et al. (2000) modeled the predator-prey interactions of Antarctic fur seals and krill, and how harvesting of krill would effect the seals. They divided the fur seal population into three classes (pups, subadults, and adults) and linked their survival rates to the available biomass of krill. Krill were represented with a simple age-class model with a ‘‘spawner-recruit’’ relationship used to initiate each year-class. Based on equilibrium analysis, they concluded that harvesting of krill at levels within those recently recommended would reduce the seal population by 50% from the level obtainable under no krill harvesting.

380

K.A. Rose and S.E. Sable

Strange et al. (1993) coupled age-structured matrix projection models for a three-species system of brown trout, rainbow trout, and Tahoe sucker. The model predicted annual abundances by age for all three species. Rainbow trout and brown trout competed for food; adult brown trout consumed sucker. Month-specific stream discharge affected early life stage survival of all three species. Mean lengths at age were followed, which then determined maturity; growth effects on fecundity of rainbow trout due to competition with brown trout were indirectly modeled as dependent on the ratio of brown trout to rainbow trout abundances. Brown trout consumed young-of-the-year (YOY), age-1, and age-2 suckers, and piscivory resulted in increased brown trout fecundity. Strange et al. compared averaged and altered high flow stream discharge conditions in two key months for two sets of initial conditions. Increased frequency of higher discharge in the different months affected the relative abundances of he species, and alternative initial conditions affected the rate of decline of Tahoe sucker. Jackson (1996) adapted the model of Jones et al. (1993) to simulate PCBs in the Lake Ontario food web. The biological model represented 6 fish species with age-structured models that were coupled via predator-prey interactions. Prey fish species were renewed each year via a spawner-recruit relationship, and their mortality from predator fish species in the model was represented using a functional response relationship. Predation mortality between pairs of predator and prey species depended on a function that was derived from prey and predator swimming speeds, prey to predator length ratio, and habitat overlap. Annual recruitment of the predator fish species was treated as a constant. The results of the original model (Jones et al. 1993) showed that alewife biomass was sustainable under current predator demand levels and that the prey fish community was sensitive to overwinter survival of the prey species alewife. Jackson (1996) further explored the role of overwinter mortality and how stocking of different predator species would affect future PCB concentrations in fish tissue. Bogstad and colleagues (Bogstad et al. 1997; Tjelmeland and Bogstad 1998) developed an age-structured model of capelin, herring, cod, harp seal, and minke whale interactions. Growth in length and in weight depended on water temperature, fish size, and feeding level. Predator consumption of specific prey species depended on fish lengths; plankton as food was forced. The Barents Sea was divided into 7 areas and migration of biomass was determined via transition probabilities, which varied by month and age-class but were constant over time. Fish species were started each year with spawner-recruit relationships; marine mammals used either simple fertility and maturity calculations or recruitment was forced. The model is used to estimate the spawning stock size of capelin. Exploratory simulations suggested that increased whale stock size would have greatest effect on herring, whereas an increased harp seal population would have relatively larger effects on capelin and cod. Ault et al. (1999) developed a predator-prey model of spotted seatrout and pink shrimp that was imbedded in 2-dimensional hydrodynamic model.

12

Multispecies Modeling of Fish Populations

381

Circulation was used to move around concentrations of presettlement shrimp and eggs and yolk-sac larvae of seatrout; active behavior was modeled for settled shrimp and juvenile and adult seatrout. Shrimp grew based on the cumulative temperature they were exposed to; seatrout grew based on bioenergetics with consumption based on ingested shrimp. One year simulations were reported that illustrated model behavior and its usefulness for examining how environmental and biological factors affect predator-prey dynamics. June-spawned seatrout were transported more into the bay, settled over a wider range of habitats, and grew faster than an August-spawned cohort. They also showed that spatial variation in habitat quality can affect growth rates of fish, even causing difference among individuals from the same spawning cohort.

12.3.3 Individual-Based, Bioenergetics Models Individual-based models (IBMs) was not a category in Rose et al. (1996) because there were only a few examples at that time. Latour et al. (2003) gave them full category status in their review, although then added they knew of only one (presumably marine) example. There are several individual-based models that examined the population dynamics of multiple species in freshwater systems. Rose and his collaborators developed individual-based models of brook and rainbow trout competition in streams (Clark and Rose 1997a,b,c), yellow perch and walleye predator-prey interactions in Oneida Lake (Rose et al. 1999; Rutherford et al. 1999), and a six-species model for Lake Mendota (McDermot and Rose 1999). These models all shared the same basic approach of bioenergetics determining growth of individuals, and reproduction and mortality determining the numbers of individuals. Clark and Rose’s model also dealt with a fairly detailed spatial model of the stream consisting of a series of linked cells defined by their dimensions as pool, run, or riffle cells. In all three examples, the role of interspecific competition or predation greatly influenced the results. Shin and Cury (2001) recently proposed a more general simulator of fish communities (OSMOSE) that used an individual-based approach. Predator species consumed prey species if they co-occurred together in the same spatial cell and the prey were vulnerable based on the predator-prey size ratio. Growth in length was based on the prey consumed and Von Bertalanffy equations, rather than on true bioenergetics. In addition to predation, there were terms for starvation and harvesting mortality. Reproduction closed the life cycle by initiating new individuals based on the surviving spawners; larval and juvenile survival determined recruitment. They presented some example simulations with seven fish species and investigated the effects on community composition and biomass of harvesting, the intensity of species interactions (by adjusting the cell sizes in the grid), and redundancy in species. Shin and Cury (2004) applied

382

K.A. Rose and S.E. Sable

the model to a 100-species theoretical fish community and examined how the size spectrum approach can help in assessing fishing effects at the community level. A general individual-based simulator for fish communities in lakes was proposed by van Nes et al. (2002). As with the OSMOSE model, the user decides on the number and linkages among the fish species. Growth of individuals was via bioenergetics with consumption of prey (including other fish species in the model) dependent on a functional response relationship. Mortality arises from starvation, fishing, and bird predation. Prey not represented as individuals can be read in as driving variables or modeled using logistic population growth formulation for biomass. They demonstrated the model by examining how alternative fishery scenarios affected the community-moderated interactions between pikeperch and bream in a shallow Netherlands lake.

12.3.4 Holistic Models Many models could be included here, including some already discussed under our other (especially budget) categories. Patten et al. (1975) and Patten (1975) developed a holistic model that included six fish compartment among the 33 total compartments. The model was linear, donor-controlled with time-varying driving variables. They examined the predicted changes in compartment biomasses under constant lake levels, thermal pollution causing a 3oC increase, eutrophication causing increased nitrogen and phosphorus inputs, and the introduction of a new piscivore. Ploskey and Jenkins (1982) simulated the biomasses of six fish types defined by their source of food (e.g., plants, benthos, detritus), treating YOY separately from older fish. They applied the model to the DeGray Reservoir in Arkansas and concluded that reservoir fishes were efficient grazers that can overrun their food supply. In the marine environment, Laevastu and his colleagues have formulated and applied a family of biomass-based models to the Bering Sea, western Gulf of Alaska, Balsfjord (Norway), and Georges Bank (Bax 1985; Bax and Eliassen 1990; Laevastu and Larkins 1981). These models were biomass-based and, in some applications, spatially-explicit. Ten to more than twenty groups were often represented, including multiple fish species, zooplankton, and phytoplankton. Difference equations were numerically solved under equilibrium assumptions to obtain annual values of biomasses from specified growth rates, mortality rates, and consumption rates by other groups in the model. Biomasses were also simulated through time by using a monthly time step and with perturbations from equilibrium attributed to changes in the environmental conditions or species biomasses (e.g., sudden rise of the top predator biomass). Analyses focused on how energy was partitioned in the food web, and how changes in fishing would affect community structure. Interestingly, Bax and Laevastu (1990) state that Polovina (1984) adapted the approach and developed

12

Multispecies Modeling of Fish Populations

383

a solution method using standard software; recall Polovina (1984) is considered the originator of ECOPATH. These analyses focused on the whole ecosystem in a spatially-explicit grid and, in many ways, were ahead of their time. What is now considered a classic paper, Andersen and Ursin (1977) developed a simulation model that included nutrients, phytoplankton, zooplankton, detritus, and age-structured Beverton-Holt-like models of the average weight and numbers of fish in age classes. The full model was comprised of 308 coupled differential equations. They applied a 21-variable version (11 fish species) to the North Sea and estimated the state of the virgin stocks (i.e., no fishing), noting that cod replaced man as the dominant predator, and that maximum possible yield, under suspiciously high effort in their opinion, was about twice the actual 1970 yield. Anderson and Ursin’s analysis not only is an excellent example of the holistic approach (done 30 years ago!), but also laid the groundwork for multispecies virtual population analysis (MSVPA). MSVPA can be viewed as a simplification of the Andersen and Ursin model (Sparre 1991), and has been the analytic tool of focus of the Multispecies Assessment Working Group of ICES (Anon 1990; Pope 1989) and its subsequent working groups for the North Sea and Baltic Sea. MSVPA is continuing to be applied to intensively studied areas, such as Georges Bank (Tsou and Collie 2001), and has been expanded to include technical interactions and spatial aspects (Vinther et al. 2001). Interestingly, there has also been, what we would call, a coupled, single-species approach (population as biomass) derived from the age-structured Andersen and Ursin model (Horbowy 1996).

12.4 Comparison of Individual-Based and Matrix Models 12.4.1 Background In this section we use an IBM to evaluate the performance of population and predator-prey matrix projection models. There is no consensus on which modeling approach to use for multispecies analysis of fish population dynamics, and there likely never will be. This is appropriate, as the decision of which modeling approach (or combination of approaches) to use depends on the: (a) question or hypothesis of interest, (b) availability of data, and (c) use that will be made of model predictions and forecasts. The popularity of the EwE model is undeniable and admirable, although we would suggest that no single model or modeling approach is universally applicable. The availability of user-friendly software for the EwE has made it accessible to many. However, in our opinion, the EwE model sill lacks the structure (age or stage) within the fish populations and options for representing within-year resolution (seasonality) that is necessary for some fish population dynamics questions. We therefore decided to compare

384

K.A. Rose and S.E. Sable

population and predator-prey matrix projection models against an IBM to determine if explicit representation of the predator’s dynamics improved matrix model predictions. Both individual-based models (IBMs) and matrix projection models are commonly used to simulate fish population dynamics. Matrix models have been widely used for decades for fish and other taxa (Caswell 2001), and form the basis for much of fisheries stock assessment and management (Quinn and Deriso 1999). IBMs have been gaining popularity in ecology due to increasing computing power and their potential for better understanding the complex dynamics exhibited by populations and communities (DeAngelis and Mooij 2005). Each approach offers advantages and disadvantages (DeAngelis and Rose 1992; Tyler and Rose 1994; Caswell 2001). Developing multispecies IBMs puts even more demands on data than the already data-hungry singlespecies versions. Matrix models make use of readily available age or stage specific demographic data on survival, growth, and reproduction. Little attention has be played to how single and multispecies versions perform relative to some assumed truth.

12.4.2 General Approach We based our evaluation of how well matrix projection models can be expanded to two species on a previously developed IBM of yellow perch and walleye in Oneida Lake. We treated the output of the IBM as data (truth) and estimated the parameters of a single-species population version (yellow perch only) and a predator-prey version (yellow perch and walleye) from the output of a 200-year baseline simulation of the IBM. We then applied the same two stresses to the IBM and the two matrix models and compared the predicted responses of yellow perch among the three models. The idea is whether explicitly including the dynamics of the predator results in similar predictions of prey responses as a purely prey population dynamics model. More details about the methods and additional analyses and explanations can be found in Sable and Rose (2008, in press). We summarize a subset of the results of their analysis in this chapter. Oneida Lake is an excellent system for testing models because the population dynamics and trophic interactions of walleye and yellow perch in Oneida Lake have been studied for the past 50 years. These data were used to configure and evaluate the IBM used here (Rose et al. 1999). Thus, we have some confidence that the output from the IBM, which we treat as ‘‘truth’’ in our matrix model analysis, provides a reasonably realistic portrayal of yellow perch and walleye population dynamics, and perhaps is also generally representative of simple fish communities with size-structured predator-prey dynamics (Ebenman and Persson 1988).

12

Multispecies Modeling of Fish Populations

385

12.4.3 Individual-Based Model The Oneida Lake IBM simulated the predator-prey dynamics of yellow perch and walleye over 200 years using a daily time step. All dynamics took place in a single, well-mixed spatial box. The model began each year (April 10) with the spawning of individual yellow perch and walleye. Each mature individual generated a cohort of eggs based on their body weight and a birth date dependent on water temperature. Each cohort was followed daily with the numbers of individuals in the cohort decreased by mortality rate and development based on water temperature. Eggs hatched into yolk-sac larvae, and then entered the larval stage when they initiated exogenous feeding, at which time they were followed as individuals in the model. The species, length, weight, sex, maturity status, and other traits were tracked for each model individual over their life time (age-10 for yellow perch and age-12 for walleye). Length and weight of each model individual were updated daily based on growth predicted by a bioenergetics equation. Yellow perch were assumed to eat zooplankton and benthos throughout their life, while walleye ate zooplankton, benthos, forage fish, and other YOY and yearling model individuals (Table 12.1). The availability of yellow perch and walleye YOY and yearlings to walleye predation depended on their relative lengths. Each day, available prey was combined into a consumption term that was constrained to be below some maximum value based on the individual predator’s weight and water temperature. The prey were combined using a multispecies functional response in which the amount of each prey type eaten depended on specified vulnerability factors and feeding efficiency parameters. Respiration, egestion, and reproduction were then subtracted from consumption, and the resulting change in weight was added to the individual’s weight to obtain a new weight. A new length was determined from the new weight as long as the weight gain was positive. Mortality rates of larvae of yellow perch and walleye were fixed. Yellow perch and walleye juveniles and yearlings died from being eaten by modeled Table 12.1 Predator-prey interactions in the Oneida Lake IBM of yellow perch and walleye Predator Yellow perch Walleye Larvae YOY and Adults Larvae YOY Yearlings and Prey yearlings adults Zooplankton X X X X X Benthos X X X X X Forage Fish X X YOY YP X X X YOY WA X X Yearling YP X Yearling WA X Note: YOY refers to YOY juveniles (20 mm and longer). YP=yellow perch, WA=Walleye.

386

K.A. Rose and S.E. Sable

walleye. Natural and fishing mortality rates of age-2 and older yellow perch were constants, while adult walleye fishing mortality rate depended on forage fish biomass (including young yellow perch and walleye). In the matrix models, we represented age-4 through age-12 walleye mortality rates as dependent on yearling yellow perch abundance and adult walleye biomass. IBM predictions of abundances, survival rates, and mean lengths by life stage and on specific days when field data were available were consistently close to averaged values from long-term monitoring data (Rose et al. 1999). Oneida Lake has undergone several, roughly decade-long, periods that correspond to high forage fish abundances, high Mayfly abundances (prey for perch and walleye), zebra mussel invasion, high cormorant abundances (who eat perch and walleye), and baseline conditions when all of these were at low abundances. We used the baseline period because this is when yellow perch and walleye are likely most coupled as predator-prey, as alternative prey and predators were not at abundant levels.

12.4.4 Matrix Projection Models We used stage-within-age matrix models for the population and predator-prey versions. YOY stages (egg, yolk-sac larvae, feeding larvae, YOY juveniles) were simulated using a daily time step; age-1 and older age-classes were simulated using an annual time step. In the predator-prey version, yellow perch and walleye each had their own matrix, with certain elements dependent on combinations of their own biomass or abundance and the biomass or abundance of the other species. In the population version, these elements could only depend on the biomass or abundance of yellow perch (i.e., density-dependence only). In both versions, density-dependence and interspecific dependencies were updated on an annual time step (i.e., matrix elements were adjusted once per year at the beginning of each year). Yellow perch were followed until age-10 and walleye until they reached age-12. The year began on April 10 (calendar day 100). We used a non-standard modification to the usual matrix projection models to accommodate density-dependent adult growth and its effects on maturity and fecundity. In the classical matrix projection models, fraction mature and fecundity are used to determine the first row in the matrix (Caswell 2001), and are assumed to be fixed with age. Thus, age-4 individuals implicitly have the same length and weight over years, which determine maturity and fecundity. In the IBM, maturity was specified as a function of length and fecundity was specified as function of weight. Both weight and length varied with age in the baseline IBM simulation and, at times, weight varied with length because individuals were allowed to lose weight but not length. We therefore modified the classical matrix model by simulating weight and length at age via agespecific functions that related annual growth increments of yellow perch to yellow perch abundances (density-dependent competition) and walleye growth

12

Multispecies Modeling of Fish Populations

387

increments to yellow perch (predator-prey) and walleye (competition) abundances. Each year in the matrix model simulation, length and weight were updated and used to determine new maturity and fecundity values by age such that the elements in the top row of the matrices varied each year. We estimated the elements of both versions (3 matrices), including density and interspecific dependencies, from the daily and annual output of the baseline simulation of the IBM. Daily values were averaged to obtain single values for each year of the simulation. We systematically explored the IBM output for relationships between stage-specific or age-specific vital rates that determine the matrix elements and various combinations of annual abundances and biomasses of yellow perch and walleye. For the predator-prey version, we then used stepwise regression with all of the important explanatory variables to see if multivariable models should be used. Only a subset of the possible combinations was deemed important enough to include in the matrix models (Table 12.2). An example is the juvenile daily mortality rate of yellow perch, which was dependent on total yellow perch eggs produced (Y = 0.0138 + 5.35  1010*X1) in the population version and dependent on yellow perch eggs and adult walleye biomass at the beginning of the year (Y = 0.0091 + 4.4  1010*X1 – 3.61  108*X2) in the predator-prey version. Each year of the simulation, the values of explanatory variables were used to determine the values of the response variables (Table 12.2), which were then used to determine the values for the upcoming year of the appropriate elements Table 12.2 Combinations of response and explanatory variables examined to specify densitydependent and interspecific relationships for the population and predator-prey versions of the matrix projection models. In all cases, IBM output was averaged to obtain one value for each year of the simulation for all response and explanatory variables Response variable Population model Predator–prey model Yellow Perch Larval mortality Juvenile mortality Yearling mortality Growth by age

YP egg production YP egg production YP yearlings YP adult abundance

YP egg production YP egg production and adult WA biomass YP yearlings and adult WA Biomass YP yearlings and YP adult Biomass

Walleye Juvenile mortality NA WA egg production and YP Yearlings Yearling mortality NA Adult WA biomass Growth (ages 1, 3) NA YP yearlings Growth (ages 2, 5–8, 10) NA Adult WA biomass and YP Yearlings Growth (ages 4, 9) NA Adult WA biomass Mortality (ages 4–10) NA YP yearlings and adult WA biomass Note: Yearling refers to the numbers of yearlings at the beginning of the year (April 10), which is the same as the number of surviving YOY from the previous year. Adult abundances and biomass refer to age-2 and older at the beginning of the year. Growth is the annual age-specific increment in length and in weight. YP=yellow perch; WA=walleye; NA=not applicable.

388

K.A. Rose and S.E. Sable

of the matrices. Larval and juvenile mortality rates and stage durations were used to define the diagonal and subdiagonal elements of the YOY portion of the population and predator-prey versions. Stochasticity was added by generating deviates from probability distributions that were used to multiply the value of the response variable determined from density-dependent and inter-specific relationships. The residuals from the fitted regression models to the IBM output were used to specify the probability distributions. A stochastic multiplier was applied to each response variable at the beginning of each year. Because the matrix models were nonlinear and stochastic, they were solved numerically as difference equations.

12.4.5 Design of Simulations We first compared the predictions of population and predator-prey matrix models to the IBM output under baseline conditions. Agreement would confirm that our estimation procedure we used with the matrix models, which was based on the baseline output of the IBM, was reasonable. The matrix models were run for 200 years and compared to the baseline run of the IBM. Once we were convinced that the matrix models were reasonably well estimated based on their agreement with the baseline simulation of the IBM, we applied the same stressors to all three models (IBM, population matrix, predator-prey matrix). The stressors were a 50% decrease in the annual egg production of yellow perch and a 10% decrease in annual survival of yellow perch age-2 and older classes, both applied every year of the stressed simulations. Each of the stressed simulations resulted in annual values of yellow perch, which we averaged over the last 185 years of each 200-year simulation. To compare the predicted response of the three models to the stressed conditions, we computed the percent change between the averaged key output variables of yellow perch for each stressed simulation (YS) and the averaged baseline value (YB): 100*(YS – YB)/YB.

12.4.6 Results and Implications Under baseline conditions, the population matrix model generated reasonably similar dynamics of annual numbers of yellow perch spawners as the IBM and the predator-prey matrix model generated similar dynamics of yellow perch and walleye spawners (Fig. 12.1). This is an expected result, as the IBM output was used to formulate the matrix models. However, it is reassuring that it appears we captured the major dynamics of yellow perch in all three models and of walleye in the IBM and predator-prey version. These were almost ideal conditions for success because we had 185 years of output from the IBM, and even so,

12

Multispecies Modeling of Fish Populations 280

(a)

389

Walleye

40

240

30

200

20

160

10

Yellow Perch

Spawner Abundance

120 280

0

(b)

240 200 160 120 280

(c)

40

240

30

200

20

160

10

120

0 30

60

90

120 Year

150

180

Fig. 12.1 Annual spawner abundance (number in modeled box) of (a) yellow perch and walleye for the IBM, (b) yellow perch for the population matrix model, and (c) yellow perch and walleye for the predator-prey matrix model under baseline conditions

we still had some difficulty in specifying density-dependent and interspecific relationships. When asked to simulate the stressed conditions, both the population and predator-prey matrix models showed similar and generally good agreement to the IBM (Fig. 12.2). Predicted responses of yellow perch abundances (spawners and age-2 recruits) to reduced egg and adult survival were in the same direction as the IBM but were underestimated by both matrix models. Predicted responses in YOY survival (net effect of larval and juvenile mortality rate and stage duration responses) to both stressors were similar for all three models (i.e., about 60% increase). Small reductions in yearling survival were predicted by the IBM, while the matrix models both predicted small increases. Changes in the fraction of age-5 that were mature were all positive, with the IBM predicting larger responses.

390

K.A. Rose and S.E. Sable

Fig. 12.2 Predicted percent response relative to baseline of model output variables for the IBM, population matrix model, and predator-prey matrix model for the (a) reduced yellow perch egg survival and (b) reduced yellow perch adult survival stress simulations

Our analysis showed that the idea of coupling matrix models to simulate multiple fish species is a promising approach, but also that sometimes a population model can perform as well as a predator-prey model. This was surprising, considering that yellow perch and walleye represent a highly coupled predatorprey system in the virtual world of the IBM. We had hoped the predator-prey model would outperform the population model. Density-dependent relationships have always been difficult to estimate with confidence (Barnthouse et al. 1984); yet, density-dependence is critical for modeling realistic responses (Rose et al. 2001). We struggled somewhat even with 185 years of ‘‘data’’ generated from another model that does not include the many sources of variation that would further cloud detection of density-dependent relationships. Having to additionally specify inter-specific relationships is a further challenge. One interpretation of our results is that with 185 years of data, one can capture the important dynamics of the prey with density-dependent relationships as well as by explicitly considering the predator dynamics. We suspect that similar analyses using reduced data sets might show greater differences between the population and predator-prey models. Perhaps we are again faced with situation

12

Multispecies Modeling of Fish Populations

391

from decades ago – we have the modeling technology to simulate multispecies situations but we are limited by the data and our knowledge.

12.5 Future Directions This chapter briefly reviewed examples of multispecies modeling of fish populations, and offered a glimpse into an analysis demonstrating the potential for multispecies modeling using a variation of the classic matrix projection modeling approach. After the initial surge of interest in multispecies modeling the 1970s with the advent of digital computing, there was a lull due to limitations imposed by empirical information. There is now resurgence of interest in multispecies modeling because of the pressures to use ecosystem-based fisheries management rather than single-species approaches. While computing power and measurement techniques both advanced greatly in the past 10 years, the limiting factor in developing multispecies models and applying them to management situations is still the lack of sufficient information. As before, there are a few locations that have been subjected to extensive, specialized study (e.g., Georges Bank, North Sea), and these were generally where multispecies approaches were first used in the 1970s and they continue to be where multispecies approaches are used now. The development of demonstration type multispecies models has continued (e.g., Matsuda and Katsukawa 2002) and the call for multispecies approaches has gotten louder (e.g., Latour et al. 2003), but the actual use of multispecies models for management advice has not spread much beyond these initial strongholds. One notable exception is the widespread application of EwE to many locations (see Chapter 8), although that is a topic for an entirely different paper. If one requires multispecies modeling to lead to actual stock-specific or site-specific management advice, then we stand by our statement that, despite the hyperbole, we have not progressed very much. We decided to go back to our chapter in the first edition and see how we did on our ‘‘future directions’’ statements from about 10 years ago. 1. We correctly forecasted the renewed interest in multispecies modeling, but we did not foresee the strong impetus for resurgence coming from the calls for ecosystem-based fisheries management. We thought species diversity issues and climate change concerns would be the drivers behind a resurgence in multispecies modeling. 2. We also suggested that IBMs were going to be widely used for multispecies modeling of fish populations, which we would have to say 10 years later that we were only partially correct. There are more examples of multispecies IBMs than 10 years ago, but not enough to deserve the phrase ‘‘widely used.’’ 3. We said we expected little advances in software development, despite the interest in object-oriented programming. We would say we were correct. With a few exceptions that have not spread widely (e.g., Shin and Cury 2001), most fisheries models and IBMs beyond simple demonstration models

392

K.A. Rose and S.E. Sable

still mostly require cumstomized coding, especially when applied in the management arena where site-specific details are important. 4. We suggested that parallel computing and hardware advances would impact modeling, especially IBMs. Desktop computing power has advanced beyond our imagination. While there are many more parallel machines today than 10 years ago, they still lie within the domain of the computer experts. We would say that we were mostly wrong, as hopes for significant software developments to enable more widespread use of parallel computing have been sluggish. While forecasting computing-related advances is always a tricky prospect, we would have to say our predictions of future directions from the first edition were moderately on target. What do we expect in the next 10 years? When viewing our new predictions of future directions, the reader might want to factor in how we fared last time. We expect the following developments: 1. Ever increasing pressure to use multipseices approaches for modeling of fish and fisheries (easy prediction). 2. Continued increases in computing power (no brainer), but we are no longer hopeful that advances in software will make the development of complicated multispecies models easy nor will high-end computing become available to non-experts. Customized coding will be needed, especially for capturing the site-specific details inherent in managment applications. 3. The availability of data from new measurement methods will enable further development of multispecies models. Unfortuneately, this increase in information will likely augment the information at already well-studied locations, and not elevate sites with limited long-term baseline (historical) information to the level to permit extensive multispecies modeling. 4. We predict that the accelerating advances in measurement methods will continue, and that in 10 years from now we will be ready for multispecies efforts that influence management decisions on a surgical basis (certain key life stages for certain species in specific locations). Likely, multispecies assessments will be in support of traditional single-species assessments (Hollowed et al. 2000). We do not want to set the bar too low. The time is now for theoretical applications that are grounded in reality by being based on real species and life histories and actual environmental conditions. The computing power is available, although the software is still limited with a few exceptions to custom coding in sequential languages. We demonstrated the use of individual-based modeling for simple multispecies situations in the first edition, and here we showed that structured matrix models in the same predator-prey situation can be extended to multispecies situations. Let’s push these modeling approaches, and others, by applying them to real conditions, but without the pressures and scrutiny that comes with the model

12

Multispecies Modeling of Fish Populations

393

results being used as part of difficult and controversial management decisions. We need some breathing room in order to fully explore what types of models and approaches do not work well in which situations (‘‘safe places for play’’, Doll 1993). Given the adversarial tone in fisheries management these days, we have difficulty seeing how multispecies approaches, other than in those few, well-studied locations with ongoing mutlispecies programs, can be defended. We strongly urge the continued exploration of multispecies models in the arena between purely theoretical and truly site-specific. Computing power is not limiting, and while software does limit the participants, the major area of concentration needs to be on developing models under conditions with sufficient room for exploration yet grounded in reality in order to pinpoint exactly what information is lacking and critical in which situations.

References Abrams PA, Ginzburg LR (2000) The nature of predation: prey dependent, ratio dependent or neither? Trends in Ecology and Evolution 15:337–341 Alaska Sea Grant (1999) Ecosystem approaches for fisheries management. University of Alaska Sea Grant Program, Report No. 99-01, Fairbanks Allen PM, McGlade JM (1986) Dynamics of discovery and exploitation: the case of the Scotian Shelf groundfish fisheries. Canadian Journal of Fisheries and Aquatic Sciences 43:1187–1200 Andersen KP, Ursin E (1977) A multispecies extension to the Beverton and Holt theory of fishing, with accounts of phosphorus circulation and primary production. Meddeleser fra Danmarks Fiskeri-og Havundersogelser 7:319–435 Anon (1990) Report of the Multispecies Assessment Working Group. ICES C.M. 1991/ Assess:7 Atkinson MJ, Grigg RW (1984) Model of a coral reef ecosystem, II. Gross and net benthic primary production at French Frigate Shoals. Coral Reefs 3:13–22 Ault JS, Luo J, Smith SG, Serafy JE, Wang JD, Humston R, Diaz GA (1999) A spatial dynamic multistock production model. Canadian Journal of Fisheries and Aquatic Sciences 56(Suppl.1):4–25 Ban-Yam Y (1997) Dynamics of complex systems. Addison-Wesley, Reading, Massachusetts Barnthouse LW, Boreman J, Christensen SW, Goodyear CP, Van Winkle W, Vaughan DS (1984) Population biology in the courtroom: the Hudson River controversy. Bioscience 34:14–19 Bax NJ (1985) Application of multi- and univariate techniques of sensitivity analysis to SKEBUB, a biomass-based fisheries ecosystem model, parameterized to Georges Bank. Ecological Modelling 29:353–382 Bax NJ (1998) The significance and prediction of predation in marine fisheries. ICES Journal of Marine Science 55:997–1030 Bax N, Eliassen JE (1990) Multispecies analysis in Balsfjord, northern Norway: solution and sensitivity analysis of a simple ecosystem model. Journal du Conseil International pour l’Exploration de la Mer 47:175–204 Bax NJ, Laevastu T (1990) Biomass potential of large marine ecosystems: a systems approach. In: Sherman K, Alexander LM, Gold BD (eds) Large marine ecosystems: patterns, processes and yields. American Association for the Advancement of Science, Washington, District of Columbia, pp 188–205

394

K.A. Rose and S.E. Sable

Bogstad B, Hauge KH, Ultang O (1997) MULTSPEC – a multi-species model for fish and marine mammals in the Barents Sea. Journal of Northwest Atlantic Fisheries Science 22:317–341 Brander KM, Mohn RK (1991) Is the whole always less than the sum of the parts? ICES Marine Science Symposia 193:117–119 Carpenter SR, Kitchell JF (1993) The trophic cascade in lakes. Cambridge University Press, Cambridge Caswell H (2001) Matrix population models: construction, analysis, and interpretation. Sinauer Associates Inc., Sunderland, Massachusetts Clark ME, Rose KA (1997a) Individual-based model of sympatric populations of stream resident rainbow trout and brook char: model description, corroboration, and effects of sympatry and spawning season duration. Ecological Modelling 94:157–175 Clark ME, Rose KA (1997b) An individual-based modelling analysis of management strategies for enhancing brook char populations in southern Appalachian streams. North American Journal of Fisheries Management 17:54–76 Clark ME, Rose KA (1997c) Factors affecting competitive exclusion of brook char by rainbow trout in southern Appalachian streams: implications of an individual-based model. Transactions of the American Fisheries Society 126:1–20 Cohen EB, Grosslein MD, Sissenwine MP, Steimle F, Wright WR (1982) Energy budget of Georges Bank. In: Mercer MC (ed) Multispecies approaches to fisheries management advice. Canadian Special Publication for Fisheries and Aquatic Sciences 59:95–107 Collie JS, Spencer PD (1994) Modeling predator-prey dynamics in a fluctuating environment. Canadian Journal of Fisheries and Aquatic Sciences 51:2665–2672 Cury P, Bakun A, Crawford RJM, Jarre A, Quinones RA, Shannon LJ, Verheye HM (2000) Small pelagics in upwelling systems: patterns of interactions and structural changes in ‘‘wasp-waist’’ ecosystems. ICES Journal of Marine Science 57:603–618 DeAngelis DL, Mooij WJ (2005) Individual-based modeling of ecological and evolutionary processes. Annual Review of Ecology, Evolution, and Systematics 36:147–168 DeAngelis DL, Rose KA (1992) Which individual-based approach is most appropriate for a given problem? In: DeAngelis DL, Gross LJ (eds) Individual-based models and approaches in ecology. Routledge, Chapman and Hall, New York, pp 67–87 Doll WE (1993) A post-modern perspective on curriculum. Teachers College Press, New York Ebenman B, Persson L (1988) Size-structured populations: ecology and evolution. Springer, New York Fogarty MJ, Murawski SA (1998) Large-scale disturbance and the structure of marine systems: fishery impacts on Georges Bank. Ecological Applications 8:S6–S22 Fogarty MJ, Sissenwine MP, Grosslein MD (1987) Fish population dynamics. In: Backus RH, Bourne DW (eds) Georges bank. MIT Press, Cambridge, Massachusetts, pp 493–507 Frank KT, Petrie B, Choi JS, Leggett WC (2005) Trophic cascades in a formerly coddominated ecosystem. Science 308:1621–1623 Grigg RW, Polovina JJ, Atkinson MJ (1984) Model of a coral ref ecosystem, III. Resource limitation, community regulation, fisheries yield and resource management. Coral Reefs 3:23–27 Hilborn R, Gunderson D (1996) Chaos and paradigms for fisheries management. Marine Policy 20:87–89 Hollowed AB, Bax N, Beamish R, Collie J, Fogarty M, Livingston P, Pope J, Rice JC (2000) Are multispecies models an improvement on single-species models for measuring fishing impacts on marine ecosystems? ICES Journal of Marine Science 57:707–719 Horbowy J (1996) The dynamics of Baltic fish stocks on the basis of a multispecies stockproduction model. Canadian Journal of Fisheries and Aquatic Sciences 53:2115–2125 Jackson LJ (1996) A simulation model of PCB dynamics in the Lake Ontario food web. Ecological Modelling 93:43–56

12

Multispecies Modeling of Fish Populations

395

Jarre A, Muck P, Pauly D (1991) Two approaches for modeling fish stock interactions in the Peruvian upwelling ecosystem. ICES Special Symposia 193:171–184 Jones ML, Koonce, JF, O’Gorman R (1993) Sustainability of hatchery-dependent salmonine fisheries in Lake Ontario: the conflict between predator demand and prey supply. Transactions of the American Fisheries Society 122:1002–1018 Kendall AW, Schumacher JD, Kim S (1996) Walleye pollock recruitment in the Shelikof Strait: applied fisheries oceanography. Fisheries Oceanography 5(Suppl 1):4–18 Koen-Alonso M, Yodzis P (2005) Multispecies modelling of some components of the marine community of northern and central Patagonia, Argentina. Canadian Journal of Fisheries and Aquatic Sciences 62:1490–1512 Laevastu T, Larkins HA (1981) Marine fisheries ecosystem: its quantitative evaluation and management. Fishing News Books Ltd., Farnham, England Latour RJ, Brush MJ, Bonzek CF (2003) Toward ecosystem-based fisheries management: strategies for multispecies modeling and associated data requirements. Fisheries 28:10–22 Link J (2002) Does food web theory work for marine ecosystems? Marine Ecology Progress Series 230:1–9 Matsuda H, Katsukawa T (2002) Fisheries management based on ecosystem dynamics and feedback control. Fisheries Oceanography 11:366–370 May RM, Beddington JR, Clark CW, Holt SJ, Laws RM (1979) Management of multispecies fisheries. Science 205:267–275 McDermot D, Rose KA (1999) An individual-based model of lake fish communities: application to piscivore stocking in Lake Mendota. Ecological Modelling 125:67–102 Micheli F (1999) Eutrophication, fisheries, and consumer-resource dynamics in marine pelagic ecosystems. Science 285:1396–1398 Murawski SA (1984) Mixed-species yield-per-recruitment analyses accounting for technical interactions. Canadian Journal of Fisheries and Aquatic Sciences 41:897–916 Murawski SA, Lange AM, Iodine JS (1991) An analysis of technological interactions among Gulf of Maine mixed-species fisheries. ICES Marine Science Symposia 193:237–252 NMFS (National Marine Fisheries Service) (1999) Ecosystem-based fishery management. A report to Congress by the Ecosystems Principles Advisory Panel. United States Department of Commerce, Silver Spring, Maryland Pace ML, Glasser JE, Pomeroy LR (1984) A simulation analysis of continental shelf food webs. Marine Biology 82:47–63 Paine RT, Tegner MJ, Johnson EA (1998) Compounded perturbations yield ecological surprises. Ecosystems 1:535–545 Patten BC (1975) A reservoir cove ecosystem model. Transactions of the American Fisheries Society 104:569–619 Patten BC et al. (1975) Total ecosystem model for a cove in Lake Texoma. In: PattenBC (ed) Systems analysis and simulation in ecology, Volume III. Academic Press, New York, pp 205–421 Ploskey GP, Jenkins RM (1982) Biomass model for reservoir fish and fish-food interactions, with implications for management. North American Journal of Fisheries Management 2:105–121 Polovina JJ (1984) Model of a coral reef ecosystem, I. The ECOPATH model and its application to French Frogate Shoals. Coral Reefs 3:1–11 Pope JG (1976) The effect of biological interaction on the theory of mixed fisheries. International Commission for the Northwest Atlantic Fisheries Selected Papers 1:157–162 Pope JG (1989) Multispecies extensions to age-structured assessment models. American Fisheries Society Symposium 6:102–111 Quinn TJ, Deriso RB (1999) Quantitative fish dynamics. Oxford University Press, Oxford Rose, KA, Cowan JH (2003) Data, models, and decisions in US marine fisheries management: lessons for ecologists. Annual Review of Ecology, Evolution, and Systematics 34:127–151

396

K.A. Rose and S.E. Sable

Rose KA, Cowan JH, Winemiller KO, Myers RA, Hilborn R (2001) Compensatory densitydependence in fish populations: importance, controversy, understanding, and prognosis. Fish and Fisheries 2:293–327 Rose KA, Rutherford ES, McDermott D, Forney JL, Mills EL (1999) Individual-based model of walleye and yellow perch populations in Oneida Lake. Ecological Monographs 69:127–154 Rose KA, Tyler JA, SinghDermot D, Rutherford ES (1996) Multispecies modeling of fish populations. In: Megrey BA, Moksness E (eds) Computers in fisheries research. Chapman and Hall, New York, pp 194–222 Runge JA, Franks PJS, Gentleman WC, Megrey BA, Rose KA,Werner FE, Zakardjian B (2004) Diagnosis and prediction of variability in secondary production and fish recruitment processes: developments in physical-biological modelling. In: Robinson AR, Brink K (eds) The global coastal ocean: multi-scale interdisciplinary processes, Volume 13, The Sea. Harvard University Press, Cambridge, Massachusetts, pp 413–473 Rutherford ES, Rose KA, McDermot D, Mills EL, Forney JL Mayer CM, Rudstam LG (1999) Individual-based model simulations of zebra mussel (Dreissena polymorpha) – induced energy shunt on walleye (Stizostedion vitreum) and yellow perch (Perca flavescens) populations in Oneida Lake, NY. Canadian Journal of Fisheries and Aquatic Sciences 56:2148–2160 Sable SE, Rose KA (2008) A Comparison of individual-based and matrix projection models for simulating yellow perch population dynamics in Oneida Lake, New York, USA. Ecological Modelling 215: 105–121 Sable SE, Rose KA (in press) Simulating predator-prey dynamics of walleye and yellow perch in Oneida Lake: a comparison of structured community models. In: Mills EL, Rudstam LG, Jackson JR, Stewart DJ (eds) Oneida Lake: long-term dynamics of a managed ecosystem and its fisheries. American Fisheries Society, Bethesda, Maryland Shin YJ, Cury P (2001) Exploring fish community dynamics through size-dependent trophic interactions using a spatialized individual-based model. Aquatic Living Resources 14:65–80 Shin YJ, Cury P (2004) Using an individual-based model of fish assemblages to study the response of size spectra to changes in fishing. Canadian Journal of Fisheries and Aquatic Sciences 61:414–431 Sparre P (1991) Introduction to multispecies virtual population analysis. ICES Marine Science Symposia 193:12–21 Spencer PD, Collie JS (1995) A simple predator-prey model of exploited fish populations incorporating alternative prey. ICES Journal of Marine Science 53:615–628 Steele JH, Henderson EW (1981) A simple plankton model. American Naturalist 117:676–691 Strange EM, Moyle PB, Foin TC (1993) Interactions between stochastic and deterministic processes in stream fish community assembly. Environmental Biology of Fishes 36:1–15 Thomson RB, Butterworth DS, Boyd IL, Croxall JP (2000) Modeling the consequence of Antarctic krill harvesting on Antarctic fur seals. Ecological Applications 10:1806–1819 Tjelmeland S, Bogstad B (1998) MULTSPEC – a review of a multispecies modelling project for the Barents Sea. Fisheries Research 37:127–142 Tsou TS, Collie JS (2001) Estimating predation mortality in the Georges Bank fish community. Canadian Journal of Fisheries and Aquatic Sciences 58:908–922 Tyler JA, Rose, KA (1994) Individual variability and spatial heterogeneity in fish population models. Reviews in Fish Biology and Fisheries 4:91–123 Van Nes EH, Lammens EHRR, Scheffer M (2002) PISCATOR, an individual-based model to analyze the dynamics of lake fish communities. Ecological Modelling 152:261–278 Vinther M, Lewy P, Thomsen L, Petersen U (2001) Specification and documentation of the 4 M package containing Multi-species, Multi-fleet and Multi-area models. Danish Institute for Fisheries Research, Charlottenlund Walsh JJ (1981) A carbon budget for overfishing off Peru. Nature 290:300–304

12

Multispecies Modeling of Fish Populations

397

Walters C, Christensen V, Pauly D (1997) Structuring dynamic models of exploited ecosystems from trophic mass-balance assessments. Reviews in Fish Biology and Fisheries 7:139–172 Whipple SJ, Link JS, Garrison LP, Fogarty MJ (2000) Models of predation and fishing mortality in aquatic ecosystems. Fish and Fisheries 1:22–40 Wilson JA, Acheson JM, Metcaffe M, Kieban P (1994) Chaos, complexity and community management of fisheries. Marine Policy 18: 291–305 Wilson JA, French J, Kleban P, McKay SR, Townsend R (1991a) Chaotic dynamics in a multiple species fishery: a model of community predation. Ecological Modelling 58:303–322 Wilson JA, Kelban P, McKay SR, Townsend RE (1991b) Management of multispecies fisheries with chaotic population dynamics. ICES Marine Science Symposia 193:287–300 Yodzis P (1994) Predator-prey theory and the management of multispecies fisheries. Ecological Applications 4:51–58 Yodzis P (1998) Local trophicdynamics and the interaction of marine mammals and fisheries in the Benguela ecosystem. Journal of Animal Ecology 67:635–658

Chapter 13

Computers and the Future of Fisheries Carl J. Walters

13.1 Introduction The world’s fisheries are a shambles. Everywhere we see signs of massive overexploitation, breakdown of regulatory and enforcement systems, and woefully inadequate investment in assessment and science. The credibility of fisheries science has been questioned with studies that have revealed severe overestimates in abundance and productivity in historical assessments for several major fisheries (Hutchings and Myers, 1994; McGuire, 1991; Parma, 1993; Pauly, 1994; Walters, 1996), indicating that we have contributed directly to the overcapitalization that we have traditionally blamed on greed and managerial stupidity. And all this has happened while whole new worlds of information gathering and analysis have been opened to us through computer technology. I am reminded of an old poster that is still displayed prominently on my office wall, a gift from participants in a 1979 Sea Lamprey International Symposium, displaying the adage ‘‘To err is human. To really foul things up requires a computer!’’ In those days the adage was referring to the growing pains of an information management industry for such human affairs as banking; little did we know at the time how well it would apply to models that were appearing at the time for improving fisheries assessments. Everyone knows that fisheries are complex systems, involving interactions among fish, people, and aquatic ecosystems. But we still seem determined to pretend that fisheries biology is somehow at the heart of this complexity, and that our skills as biologists are the most important ones. Most fisheries students learn very quickly how wrong this pretense is when they take their first real job in fisheries, and are plunged into the world of trying to make useful assessments and quantitative predictions with messy data; most end up spending far more time handling spreadsheets than they do handling fish, and wishing that they had better understood the gibberish in their statistics and computer courses. What they are learning at this time is something very fundamental: many fisheries processes and problems are essentially statistical rather than C.J. Walters (*) Fisheries Centre, University of British Columbia, Vancouver, B.C., V6T1Z4, Canada

B.A. Megrey, E. Moksness (eds.), Computers in Fisheries Research, 2nd ed., DOI 10.1007/978-1-4020-8636-6_13, Ó Springer ScienceþBusiness Media B.V. 2009

399

400

C.J. Walters

biological. After all, what is a ‘‘population process’’ other than a statistical summation of the effects of things that happen to or are exhibited by many individual animals? How could we hope to understand how such simulations behave in space and time just by knowing much about the typical animal that engages in them, especially when we are also taught early on that individuals vary enormously? Computers are opening up whole new ways for us to monitor and represent the complex statistical dynamics of fishing systems, and we must make every effort to capitalize on this opportunity. Hopefully books like this one will hasten the difficult educational process that will make that possible. This chapter reviews just a few of the opportunities and pitfalls that computers have brought to fisheries science. Emphasis in the following sections is on those computer applications most directly and immediately useful in managing fisheries better, rather than on applications that will give us more ‘‘fundamental’’ understanding in the long run. I believe in fact that working harder to answer direct and well-focused applied questions is the quickest way for us to come to grips with the fundamental questions as well.

13.2 Opening New Windows for Measuring Fishery Dynamics Here are a few of the things that we can do today with computers that would have been unthinkable or outrageously costly even a few decades ago (chapter numbers refer to topics discussed in this volume): (a) Equip every vessel in a fishing fleet with transponders that accurately report positions every few minutes or hours to a central computer data system that can plot fishing distributions and pinpoint movements that may represent illegal fishing activities. (b) Have fishermen report their catches and oceanographic observations, accurately georeferenced, through the same transponder systems so that every vessel in a sense becomes part of a survey or ‘‘test fishing’’ operation (Chapter 7). (c) Access massive oceanographic, meteorological, and biophysical (‘‘habitat’’) data bases for comparison with direct information from fishing vessels and biological surveys (Chapters 4, 5, 6, 9 and 11). (d) Tag large numbers of fish with archival computer tags (or transmitting tags linked to our personal computers) that let us see how fish use the world in remarkable detail. (e) Compare complex atomic ratio, scale pattern, or DNA segment ‘‘signatures’’ for large numbers of individual fish to reference signatures for particular stocks/areas of fish origin, to assess stock composition from fisheries where many stocks are taken together (Chapter 9). (f) Use image processing methods to turn complex sonar, laser, and other physical signals into abundance and distribution maps (Chapter 9).

13

Computers and the Future of Fisheries

401

(g) Keep careful and accurate records of past management activities and measured responses of fish and fishermen to these activities. (h) Run many population simulations with different parameter values, to determine the set of parameter combinations that are consistent with historical data (i.e. how large the uncertainty is about the parameter values). (i) Simulate dispersal and migration patterns of thousands of small bins or packets of fish, to represent the spatial dynamics of fish relative to where and when fishing takes place (Chapters 10 and 12). (j) Use information on ecological interactions (predation, competition, etc.) to construct multispecies and ecosystem models that can help in design of balanced policies for ecosystem management (i.e. stop treating fish stocks as independent of one another in harvest management planning) (Chapter 12). (k) Present spatial and ecosystem simulations to fishery managers in the format of easily used simulation games, where they can do things like move fishing areas and times around on their PC screen map of the fishery, with a mouse (Chapters 4, 8, 10 and 11). (l) Devise complex empirical ‘‘rules’’ for making forecasts and carrying out regulatory actions to achieve harvest goals (Chapter 3). Consider the opportunities for assessment that these things have created. At the heart of all of them is the notion that we can now gather enough information to begin to discard the old and very misleading mental model of a fish stock as a ‘‘dynamic pool’’ distributed like molecules in a chemical reactor vat and behaving with comparable simplicity. This model has got us into trouble over and over again in fisheries assessment. While lulling us into a sense of confidence that we are studying the right processes (growth, recruitment, mortality), it has left us largely unable to use information effectively and to prescribe wise policies at the scales in space and time where policy must actually be implemented. On the information side, we are encouraged by the pool model to view fishing as somehow distributed simply in space, and to use indices such as aggregate catch per effort to describe changes in relative abundance. On the prescriptive side, we are able to provide little more in the way of policy advice than crude estimates of sustainable yield, population growth/decline rates, and likely responses to management measures for varying the sizes of fish captured. What is fundamentally missing in the pool model is something that much concerns every biologist and fisherman: spatial structure and dynamics. Fish are not randomly distributed, and their ontogeny almost always involves a complex trajectory of dispersal and migration through a range of habitats and ecological interactions. It drives us to distraction that fishermen know such distributions and trajectories very well, and exploit them in complex ways to end up providing us with catch and catch rate data that tell us very little about what populations are doing when interpreted in crude (nonspatial) ways. Sometimes catch rates decline very rapidly as harvesting develops, as natural aggregations of fish are removed, without there really being much impact of fishing on the stock. In other cases, aggregations form fast enough in the face of harvest removals, and

402

C.J. Walters

the range used by fish contracts after such removals, to make us think from crude statistics that everything is well when in fact the stock is rapidly collapsing. In still other cases, fisheries progressively ‘‘mine’’ out local populations further and further from ports, without this sequential depletion being evident at all in traditional statistics. There are two information situations that should frighten every fisheries scientist. The first is where collapse is so obvious in the data that any fool can see it, and the frightening questions in this case are about how to reverse the trend and deal with socioeconomic hardship. The second is where things look really good (catches stable or growing, catch per effort stable), and fishery managers are smiling complacently about what a good job they are doing. Many are the times that I have sat in management planning meetings with such smiling people, while wanting to scream in frustration about the lack of even crude spatial statistics to help determine whether aggregations are shrinking or fishing grounds are being sequentially depleted. The reader might have noticed that these two information situations cover most of the world’s fisheries; the few situations where we perhaps need not be frightened are ones where complex spatial structure (many lakes, streams, reefs, etc.) has offered many opportunities to test alternative management approaches without putting the whole system at risk. Many fisheries today are moving to quota management systems so as to improve economic performance, simplify the manager’s job, and presumably foster attitudes toward better resource husbandry by fishermen. However, these systems radically increase the demand placed on assessment science to provide more precise stock size estimates so as to avoid depensatory effects of quota removals (Pearse and Walters, 1992). In most fisheries we cannot even begin to meet this demand with the precision that managers and fishermen seem to expect, and I personally am coming to think of quota management as potentially one of the most destructive innovations in the history of fisheries development. The dangers of quota management are not lost on fishermen and the public, and we are beginning to see more frequent calls to design regulatory policies that will absolutely close the door on overfishing by complementing quota management with space/time harvesting restrictions that will place a firm upper bound on fishing mortality rates (i.e. directly limit the proportion of fish that are exposed to harvest). The design of such ‘‘refuge’’ systems requires fairly detailed information about how fish are distributed and how they move about, and correspondingly detailed models for comparing alternative refuge sizes and timings. We have very little experience in fisheries with the development of such regulatory tactics models, except in the very special case of gauntlet fisheries for migrating Pacific salmon. I predict that development of detailed regulatory models, and the data needed to make them work, is going to be one of the most exciting and important areas of fisheries research over the next decade or two.

13

Computers and the Future of Fisheries

403

A cursory look at recent literature on fisheries modeling and assessment methods might give the impression that we are getting better at both modeling how fish populations behave and at linking data to models through elaborate estimation procedures that link diverse sources of data. Terms such as stock synthesis (Methot, 1990) and auxiliary information (Deriso et al., 1985; Fournier and Warburton, 1989; this volume Chapter 11) have come to connote the promise of assessment methodology that can keep up with the demands of quota management. To some degree this promise is being fulfilled, but the situation is not nearly as blight as it may appear. Many of the recent advances in modeling and data analysis are not in fact aimed at or capable of producing more precise assessments at all; what they are doing instead is providing more accurate and honest assessments of how grossly uncertain we really are about abundance and production. We are coming to question many assumptions of traditional models, ranging from whether production dynamics are stationary to whether some types of data carry any information at all. Techniques such as Bayesian estimation are allowing us to express the uncertainty represented by such questions in terms of probability distributions, and there has been a nasty tendency recently for such distributions to get broader rather than narrower. Such admissions of deep and persistent uncertainty will almost certainly help to fuel the demand noted in the previous paragraph for regulatory models and systems that directly reduce the risk of overharvest.

13.3 Designing Robust Policies for Living with Uncertainty Although computer data acquisition and modeling systems may provide better stock assessments and understanding of fish-environment relations, it is likely that fisheries are always going to be managed under considerable uncertainty. This is particularly so in relation to impacts of environmental and climate change; even if we come to understand how such factors influence fish, there is little prospect for predicting accurately how the factors themselves will change. Over the past 20 yrs, computer simulation and optimization methods have allowed much progress in the design of feedback and adaptive policies for coping with unpredictable change; this policy design work has relied upon or assumed that we can describe and predict statistical patterns of variability even if we cannot say what causes that variability. Indeed, policy designs based on such statistical descriptions are robust in the sense that they are entirely invariant to causation of the variation (we get the same policy design answers no matter how we comfort ourselves through attempts to explain how the variation arises). A particularly interesting recent finding has been about how classic, fixed exploitation rate (or fixed fishing rate F) policies may provide a robust way to deal with the effects of climate change on biological carrying capacities (Walters and Parma, 1996; Parma and Deriso, 1990). We already know that such policies

404

C.J. Walters

are near optimum for dealing with risk averse harvest management objectives (Deriso, 1985; Hilborn and Walters, 1992), but they may have another really important bonus as well even in maximizing long term harvest objectives that do not involve risk aversion. Basically, it appears that successful implementation of a fixed harvest rate strategy should allow quite close tracking of changes in optimum population size associated with changes in carrying capacity. Figure 13.1 shows a 100-yr simulation of a population subject to wide swings in carrying capacity (as measured by equilibrium unfished stock size). Spawning stock size under an optimum fixed exploitation rate policy (calculated from the slope at low stock sizes of the population’s mean stock-recruit relation) moves up and down with the carrying capacity. But the really interesting thing is how this spawning stock size tracks the theoretical optimum spawning stock size, where this theoretical optimum is calculated by dynamic programming for each simulated year assuming perfect knowledge as of that year of all future carrying capacity changes. The tracking is not perfect, since the optimum policy anticipates big carrying capacity changes and begins to adjust stock size ahead of time if necessary to capitalize on such changes when they arrive. However, at least in cases such as Fig. 13.1, where the environmental change is not too rapid, the

Fig. 13.1 Simulation showing that a fixed harvest rate strategy can result in population sizes that closely track the theoretical optimum for responding to changes in carrying capacity caused by climate change. The carrying capacity ( ) represents simulated change in unfished stock size due to climate effects (generated by simple autoregressive model). The fixed U line (. . . . .) represents stock size that would be achieved by taking the best constant fraction (U) of the stock every year. The optimum escapement line (- - - - -) represents stock size target that a manager with perfect knowledge of future climate change would try to achieve each year (computed using dynamic programming)

13

Computers and the Future of Fisheries

405

fixed exploitation policy gives total long term harvests that are 90% or higher of the theoretical optimum. This finding is not at all sensitive to the population model used in the calculation, except that the impact of the environmental change must be mainly on parameters that determine maximum population size (carrying capacity) rather than on the population’s intrinsic rate of increase (slope of recruitment curve) when abundance is low. When intrinsic rate of increase is also variable with environmental conditions, the harvest rate should be varied over time to track such changes. Such simulations do not, of course, tell us how to achieve a desired fixed harvest rate in terms of annual harvest regulation. In quota management systems, this will presumably involve setting the quota for each year to the target harvest rate times an estimate of stock size. In other cases like Pacific salmon, we can directly regulate risk of harvest through area/time closures to fishing. Much work needs to be done using simulations and careful statistical analyses of stock size estimation to determine whether target harvest rates under quota management should be adjusted downward to reflect uncertainty in stock size estimates. Preliminary indications from simulations of how quota setting errors may affect long term management performance are that target harvest rates should be adjusted downward by a factor of roughly exp(CV2), where CV is the coefficient of variation of the stock size estimate (Walters and Pearse, 1994).

13.4 Keeping the Bad Guys at Bay: Seeing How Fisheries Really Operate Illegal fishing activities are rapidly becoming one of the biggest threats to sustainability of the world’s fisheries. Historically, fisheries scientists have not worried much about poaching, and it was not a big problem when fishermen had free run of the seas, lakes, and streams. But factors ranging from overfishing to development of Asian markets are now making fish more valuable, and our very efforts to create sustainable fisheries through quota management and harvest regulation are creating incentives for development of ‘‘underground’’ fishing and marketing systems. At the same time, there is political pressure to cut back on public subsidies to fisheries, including such management costs as enforcing regulations, and to place this burden on fishing industries that in many cases cannot afford traditional, manpower-intensive enforcement systems. Field enforcement and management staff in almost every public agency now have horror stories to tell about well-organized illegal fishing activities, complex schemes for transporting fish to safe marketing locations, and routine violations of regulations and quota shares even by legally licensed fishermen. It does not take much misreporting of catches to really foul up our existing stock assessment methods and models. For instance, we very nearly destroyed

406

C.J. Walters

British Columbia’s largest salmon stock in 1994. The Adams River dominant sockeye (Oncorhynchus nerka), because of an overestimate of its abundance at one point along the gauntlet of fisheries that it faces during its spawning migration. Very high catches were reported in one fishing area (the Johnstone Strait), and this led biologists to revise run size estimates upwards and to open the next fishing area (Fraser River mouth); luckily statistics from this next area showed that the stock had in fact already been overfished, and the area was closed in time to prevent an ecological disaster. The computer models for run reconstruction and estimation had apparently failed badly for the 1994 situation. Partly this was due to an unusual migration routing by the fish, but there are also suspicions that the very high catches reported in the Johnstone Strait did not in fact actually happen in the legal fishery, and instead represent fish caught illegally in various closed areas and times along the migration. In short, what looked to biologists like high abundance may actually have been a symptom of large scale illegal fishing. Computers can help in two dramatic ways to substantially reduce illegal fishing. The most obvious way is to require all fishing vessels to carry location devices, for which computers are necessary for interpreting transponder signals and providing maps of vessel distributions. But another thing we can do is to feed the great mass of information gathered this way into expert system software, programmed to look for and warn us about patterns of vessel movement that likely represent either illegal fishing or transport of fish. Fishermen are understandably not very happy about this concept (Big Brother is watching), but they may eventually come to support it if fisheries agencies can convince them of three key points: (a) the seas and the fish are still considered public property, and the public has every right to monitor how these properties are used; (b) their legal livelihoods really are being threatened by the bad guys; and (c) today’s fishing technologies and transportation systems are so quick and efficient today that traditional enforcement methods cannot offer a credible threat of catching them in the act.

13.5 Exploring Options and Opportunities: Integrating Assessment and Management In recent years there has developed a dangerous gap between the quantitative, computer intensive world of fisheries assessment and the world of the practicing fishery managers who work directly with industry and political decision makers. Many fishery managers understand neither the technology or limitations of the analysis results that they convey in various formats to fishermen and politicians, and there is often little onus on them to even keep up with recent literature pointing out alternative management approaches and difficulties. To some extent the misunderstandings that this state of affairs has created are

13

Computers and the Future of Fisheries

407

beginning to be circumvented through development of direct cooperative working arrangements between fishing industry representatives and assessment scientists. It appears that one of the most effective ways to promote better communication and mutual education among these management players is to literally place them in game playing situations (Walters, 1994) where computer models are used to represent at least some of the possible impacts of alternative management strategies. Such games can be extremely useful tools for identifying imaginative new policy options and for understanding basic dynamic principles and limits, even if the models used in them are quite unrealistic. However, today we do not have to be content with crude models; spatial simulations and graphical visualization techniques can be used to make fisheries dynamics literally come alive on computer screens. Ideas that we used to try to explain with complex graphs, tables, or equations can now be conveyed in pictures that not only inform, but also help to stimulate new ideas. I have been treated to a taste of the potential power of management gaming entirely by accident, during a workshop with rock lobster fishermen and biologists in South Australia. We were trying to develop a messy spatial model for the stock (along the lines of Walters et al., 1993), and at one point in the workshop there was nothing for the fishermen to do while the biologist participants were off to do data analysis. A fisherman suggested that we let them play with an early version of the model, and I agreed provided they were careful not to take this version seriously. So they went away to play, and returned to my surprise within the hour with two things. First was a set of questions about why the model made some predictions that they found surprising (consequences of basic yield per recruit theory when harvest rate is reduced). Second was a whole set of very exciting ideas about how to use reduced fishing seasons, changes in timing of fishing, and spatial refuges to maintain their economic performance while substantially reducing exploitation rates (and hence increasing annual egg production and reducing risk of recruitment overfishing). I frankly found this outcome somewhat embarrassing, because the options that the fishermen identified on their own were ones that we (the biologists) should have been suggesting to them in the first place. If this accidental interaction is any indication of what can be obtained from more carefully planned and structured interactions between management players, the future for game playing approaches is indeed bright.

13.6 Pitfalls: Megamodels, Megainformation, and Megamistakes There is a real danger that we will use new information gathering and modeling capabilities to grossly overparameterize models that are used for estimation of basic quantities such as optimum harvest rates and sustainable yields. Most biologists still assume that more realistic and detailed models are always

408

C.J. Walters

preferable to simple ones when reasonable data are available to estimate parameters for the realistic ones. The basic fallacy of this argument was first demonstrated clearly (Hilborn, 1979; Ludwig and Walters, 1985), by generating fake population and harvest data from realistic models and then showing that these simulated populations could often be managed better by estimating management parameters (e.g. MSY and optimum effort) for them using simpler, deliberately incorrect models. These simulations showed that the usefulness of a model for management analysis depends in a complicated way on how informative the data are for estimating its parameters, and good estimation performance can be more important than the biases introduced by deliberate simplifications in the model structure. Figure 13.2 shows a simple way to understand why it is often wise to deliberately choose the ‘‘wrong’’ model. When a model is used to estimate a parameter of management relevance, such as MSY, there is a cost associated with error in this estimate when it is used by managers (missed harvest opportunity, or overharvest); this cost is zero only if the data happen to give exactly the right parameter estimates. For any model and data quality (types of available, precision), there is a statistical distribution of possible error in the parameter estimates. For more realistic models, the estimator is usually unbiased (distribution centered on correct value), but can

Fig. 13.2 Statistical variation in estimates of policy parameters such as maximum sustainable yield depends on the model used for estimation, and there is a cost of making estimation errors. Estimates can be quite variable (R case) when realistic models are used with uninformative data, resulting in high expected cost (high odds of estimates being far enough off to cause high cost). Estimates from very simple equilibrium models (E case) may be quite precise for the same data, but dangerously biased. An ideal population model for assessment (S case) often is a deliberately oversimplified model that is not too badly biased (and in a safe direction), but gives estimates that are less variable than would be achieved with a more realistic model

13

Computers and the Future of Fisheries

409

be quite inaccurate due to overparameterization (trying to estimate too many parameters from the data, so that each cannot be estimated accurately because its effects are confounded with effects of other parameters being estimated). Simpler models typically give more precise (less variable), but often biased estimates. The classic example of a very precise but biased model for MSY estimation is the Gulland or Fox equilibrium regression of cpue on effort; such equilibrium models give very precise estimates, but these are very often dangerously biased upward. An ideal model for estimation is one that has relatively low bias in a safe direction (e.g. underestimates MSY), but few enough parameters to allow estimation with reasonable accuracy. The best such balance cannot be predicted a priori just by looking at the data available or the biology of the organism involved; it must be discovered through reference model simulation methods that simulate data close to the actual case in point, and trial and error testing of alternative assessment models using this reference model (Hilborn and Walters, 1992). Much more such estimation testing work needs to be done in fisheries assessment and this will continue to be a major area for fisheries computer applications. Many of the important advances in computer data gathering and analysis that are opening to us now are mainly going to be valuable in measuring relative abundance and stock size trend more precisely, by accounting better for the spatial structure of abundance patterns. However, we have found in simulation studies that inaccuracies in policy parameter estimates from realistic models do not arise just because of inaccurate abundance data; indeed, simpler models can often be shown to perform better than realistic ones even when the simulated assessment system is given absolutely accurate (exact) abundance information. Rather, parameter inaccuracies are much more importantly affected by lack of informative contrasts in the historical data. For example, if stock size has not been reduced enough to provide direct empirical evidence of the reduction needed to impair average recruitment, any model for estimating that reduction is necessarily going to provide an imprecise estimate no matter how elegantly or realistically it may represent the details of the recruitment process. In short, it is not just more data of all sorts that we need for better fisheries modeling; we also need to make the best use possible of information from those unhappy circumstances where such parameters as the recruitment curve slope at low stock size have been discovered the hard way, through fishery collapse. Perhaps the most fundamental issue and debate in fisheries today is about the relative importance of environmental factors versus factors related to stock size in causing fisheries collapses and policy failures. Major declines in groundfish recruitment off the east coast of North America appear to be related to declines in spawning stock, but could also be due at least in part to climatic changes in recent decades (Cushing, 1982; Mann, 1992; Myers et al., 1993). Declines in survival rates of hatchery salmon populations have been attributed to environmental factors by supporters of hatchery programs, and to overstocking (natural carrying capacity effects) by detractors of the programs (Emlen et al., 1990; Walters, 1993). These ‘‘Thompson-Burkenroad debates’’

410

C.J. Walters

must be resolved somehow if we are to proceed wisely in programs to rebuild and maintain natural production systems (Walters and Collie, 1988). Will high technology environmental monitoring and computer simulation methods allow us to resolve the debates by finally understanding how environmental factors really have (and will) influenced population and survival declines? Better monitoring and modeling technologies will certainly allow us to construct more elaborate and detailed mechanistic models and hypotheses about how fish relate to their environment. But the pitfall here is that just because we can build more credible models is no guarantee at all that we will be able to critically and scientifically test the assumptions underlying these models. In particular, being able to relate survival and growth during some life stages, places, and times precisely to environmental factors does not in any way preclude the possibility that compensatory responses occur at other stages to over-ride such relationships. That is, lovely mechanistic models can be dangerously incomplete, yet lull us into believing them through their very realism and good agreement with available data. In the end, no complex computer modeling or data gathering system can substitute for the fundamental scientific requirement of trying to challenge and reject our models through real field experience and experiments. Fisheries managers and funding agencies are not being told about this pitfall by the large community of oceanographers and fisheries scientists who have flocked to the environmental effects banner for funding. We may be able to mesmerize the managers with our elaborate models and data systems, and give them excuses not to take strong regulatory action when it is needed, but ultimately nature is not going to forgive us for this game. A simple way to summarize the previous paragraph is to warn that we cannot substitute technology for sound scientific method. But one might counter this with the argument that when computerized measurement and analysis methods (along with careful biological process research) allow us to see how fish interact with their environment in greater detail, key patterns and relationships will just leap out at us and be so obvious as to not require critical testing. Indeed, this may well be the case with some relatively simple processes like movement, and how migration/distribution patterns are shaped by response to environmental factors. But it is dangerously misleading when applied to the most crucial and complex process in fisheries science, recruitment. There is a long tradition in recruitment studies of finding spurious correlations that break down the year after you publish them, and having more data and sophisticated analysis methods could just speed this tradition along. In fact for most fish stocks we cannot do even the simplest possible test of elaborate recruitment-environment models, namely to see if they can reproduce historical recruitment data. For example, years ago, we completed an individual-based modeling exercise on groundfish stocks in the Hecate Strait B.C. We began the analysis by using environmental data bases and a fairly detailed hydrodynamic model to reconstruct (hindcast) historical current and thermal regimes in the Strait. Then we introduced simulated larvae into these regimes for 1950–1990 using a Walters et al. (1992) model. We found only very weak correlations between larval

13

Computers and the Future of Fisheries

411

distribution/survival patterns predicted by this model and actual recruitment histories estimated from fisheries statistics. We tried to argue our way around this ‘‘failure’’ by invoking the possibility of compensatory mortality at later juvenile stages, but a wise reviewer pointed out that we are not even entitled to this excuse. The reviewer pointed out that as is the case for most fish populations, we simply did not have detailed enough historical data on spawning times and locations to even place larvae properly into our reconstructed physical fields. So our environmental pattern analysis and process modeling effort could not even get past the first life history stage leading to recruitment in terms of the most basic validity check (past data) imaginable. And the really hard parts come later, in dealing with juvenile fish that have much more complex behaviors and reactions to environmental variables such as predation risk. We could, of course, just keep blundering along with the analysis while starting to gather the more detailed spawning data needed to check the larval stage calculations; but I strongly suspect that the end of this would be several lifetimes of finding new data needs, without ever a credible and well-tested result of value in fisheries recruitment forecasting or to help resolve Ttompson-Burkenroad debates. The pitfalls discussed in this section are not just ones for computer modeling freaks like myself to worry about. There is no reason to expect that the eyeball and armwave school of intuitive fisheries analysis is going to find more robust ways to analyze data so as to avoid fundamental efforts of overparameterization, confusing availability of details with overall understanding, and basing policy on plausible hypotheses that fit the facts well but are dangerously misleading. Computers are offering all of us the chance to turn fisheries from a discipline of guesswork, dogma, and armwaving into one that really deserves to be called a science; we must all learn to use them wisely.

References Cushing DH (1982) Climate and fisheries. Academic Press, London. 373p. Deriso RB (1985) Risk averse harvesting strategies. In: Mangel M (ed.) Resource Management, Proceedings of the Second Ralf Yorque Workshop. Lecture Notes in Biomathematics No. 61, Springer-Verlag, Berlin. pp 65–73. Deriso RB, Quinn TJ, Neal PR (1985) Catch-age analysis with auxiliary information. Canadian Journal of Fisheries and Aquatic Sciences 42:815–842. Emlen JM, Reisenbichler RR, McGie AM, Nickelson TE (1990) Density-dependence at sea for coho salmon (Oncorhynchus kisutch). Canadian Journal of Fisheries and Aquatic Sciences 47:1765–1772. Fournier DA, Warburton AR (1989) Evaluating fisheries management models by simulated adaptive control-introducing the composite model. Canadian Journal of Fisheries and Aquatic Sciences 46:1002–1012. Hilborn R (1979) Comparison of fisheries control systems that utilize catch and effort data. Journal of the Fisheries Research Board of Canada 36:1477–1489. Hilborn R, Walters C (1992) Quantitative fisheries stock assessment and management: choice, dynamics, and uncertainty. Chapman and Hall, New York, NY.

412

C.J. Walters

Hutchings J, Myers RA (1994) What can be learned from the collapse of a renewable resource? Atlantic cod, Gadus morhua, of Newfoundland and Labrador. Canadian Journal of Fisheries and Aquatic Sciences 51:2126–2146. Ludwig D, Walters C (1985) Are age-structured models appropriate for catch-effort data? Canadian Journal of Fisheries and Aquatic Sciences 42:1066–1072. Mann KH (1992) Physical oceanography, food chains, and fish stocks: a review. ICES Journal of Marine Science 50:105–119. McGuire TR (1991) Science and the destruction of a shrimp fleet. Marine Anthropological Studies 4:32–55. Methot RD (1990) Synthesis model: an adaptable framework for analysis of diverse stock assessment data. International North Pacific Fisheries Commission Bulletin 50:259–277. Myers RA, Drinkwater KF, Barrowman NJ, Baird JW (1993) Salinity and recruitment of Atlantic cod (Gadus morhua) in the Newfoundland region. Canadian Journal of Fisheries and Aquatic Sciences 50:1599–1609. Parma A (1993) Retrospective catch-at-age analysis of Pacific halibut: implications on assessment of harvesting policies. Proceedings of the International Symposium on Management Strategies for Exploited Fish Populations, Alaska Sea Grant College Program, AK-SG93-02, pp 247–265. Parma AM, Deriso RB (1990) Experimental harvesting of cyclic stocks in the face of alternative recruitment hypotheses. Canadian Journal of Fisheries and Aquatic Sciences 47:595–610. Pauly D (1994) On the sex of fish and the gender of scientists. Fish and Fisheries Series 14, Chapman and Hall, UK. Pearse PH, Walters C (1992) Harvesting regulation under quota management systems for ocean fisheries: decision making in the face of natural variability, weak information, risks and conflicting incentives. Marine Policy 16:167–182. Walters CJ (1993) Where have all the coho gone? In: Berg L, Delaney P (eds.) Proceedings of the Coho Workshop, Nanaimo B.C., May 26–28, 1992. Dept. Fisheries and Oceans, Communications Directorate, Vancouver, B.C., pp 1–9. Walters CJ (1994) Using gaming procedures in the design of adaptive management policies. Canadian Journal of Fisheries and Aquatic Sciences 51:2705–2714. Walters CJ (1996) Lessons for fisheries assessment from the northern cod collapse. Reviews in Fish Biology and Fisheries 6:125–137. Walters CJ, Collie JS (1988) Is research on environmental factors useful to fisheries management? Canadian Journal of Fisheries and Aquatic Sciences 45:1848–1854. Walters CJ, Hall N, Brown R, Chubb C (1993) Spatial model for the population dynamics and exploitation of the Western Australian rock lobster, Panulirus cygnus. Canadian Journal of Fisheries and Aquatic Sciences 50:1650–1662. Walters CJ, Hannah CG, Thomson K (1992) A microcomputer model for simulating effects of physical transport processes on fish larvae. Fisheries Oceanography 1:11–19. Walters CJ, Parma AM (1996) Fixed exploitation rate policies for responding to climate impacts on recruitment. Canadian Journal of Fisheries and Aquatic Sciences 53:148–158. Walters CJ, Pearse P (1994) Uncertainty and options for insuring sustainability of quota management systems. Reviews in Fish Biology and Fisheries 6:21–42.

Species Index

A Anchovy (Engraulis japonicus), 80, 111, 148, 158, 171–173, 210, 215, 217, 234, 237, 247, 254–255, 257, 263, 377–379 Antarctic fur seals, 379 Antarctic krill (Euphausia superba), 158, 174 Argopecten purpuratus, 261 Auxis, 265 B Bay anchovy (Anchoa mitchilli), 173 Brown trout (Salmo trutta), 380 Bowhead whales, 178 C Calanus finnmarchius, 166–167, 169, 294 Cape anchovy (Engraulis capensis), 148, 158, 172 Cape gannet, 256 Cape hake, 255 Cephalopods, 18, 134, 261, 377 Cetaceans, 170, 234, 276 Chaetognaths, 168–169, 174 Chilean hake, 254, 257 Chinook (Oncorhynchus tshawytscha), 41 Chokka Squid, 237 Chondrichthyans, 234 Chub mackerel, 80, 234, 379 Cod (Gadus morhua), 148, 151, 158, 216, 246, 249, 257–258, 263, 268, 306, 379, 380, 383 Coscinodiscus wailesii, 167 D Doliolids 169 Dolphin, 277 Dorado, 265 Dragonflies, 296

E Emperor (Lethrinus), 108–109 F Flying fish, 277 G Giant scallop larvae (Placoplecten magellanicus), 179 Goby, 255–256 H Hake (Merluccius merluccius), 158, 170, 234, 247, 254–258, 377–378 Herring (Clupea harengus), 80, 82, 124, 126, 148, 152–155, 158, 160, 172, 175, 177, 195, 199, 207, 211, 213, 216–218, 234, 306–307, 378, 380 Horse mackerel, 234, 237, 254–255, 257, 263 J Jellyfish, 244, 251, 255–256 L Limacina pteropods, 169 Loligo opalescens, 162 M Manila clam (Tapes philippinarum), 246, 257, 260 Marlin, 247, 262 M. capensis, 237 Meganyciphanes norvegica, 167 Mesopelagic fish, 130, 234 Meyenaster gelatinosus, 261 Mnemiopsis leidyi, 245, 254 M. paradoxus, 237 Mulinia sp, 261

413

414 N Neocalanus spp, 173 Northern anchovy (Engraulis mordax), 148, 158 Northern shrimp (Pandalus borealis), 173 Norway lobster (Nephrops norvegicus), 230 O Oithona spp., 167, 169 Oncaea spp., 167 P Pacific ocean perch (Sebastes alutus), 343, 363 Pacific salmon, 402, 405 Pacific sardine (Sardinops sagax), 148 Pelagic goby (Sufflogobius bibarbatus), 255 Pepino (S. fuscus), 239 Pink salmon (Oncorhynchus gorbuscha), 171 Plaice (Pleuronectes platessa), 148 Pleurobrachia, 169 Pseudocalanus sp 169 R Radiolaria, 169 Rainbow trout (Oncorhynchus mykiss), 380–381 Red drum, 178 Rhizosolenia, 169 Rock lobster, 407 Round herring, 234 S Sardine (Sardina pilchardus), 111, 148, 173, 210, 215, 234, 237, 247, 255, 257, 263, 377, 379 Seabird, 237

Species Index Sea cucumber (Isostichopus badionotus), 109–111 Seals, 234, 244, 251, 378–379 Sea scallops (Placopecten magellanicus), 170 Shark, 137, 263 Silver hake (Merluccius bilinearis), 258 Silver perch, 178 Snoek, 237 Sockeye salmon (406), 158 sole (Solea solea), 106–107, 200, 218 Sperm whales, 178, 377 Spotted sea trout, 178 Striped bass (Morone saxatilis), 277 Striped mullet (Mugil cephalus), 173 T Tahoe sucker (Catostomus sp.), 380 Thysanoessa inermis, 167 Trichodesmium, 169 W Walleye (Stizostedion vitreum), 158, 172, 322, 329, 356, 381, 384–390 Walleye pollock (Theragra chalcogramma), 158, 172, 322, 329, 356 White perch (Morone americana), 277 X Xantochorus cassidiformis, 261 Y Yellow perch (Perca flavescens), 381, 384–390 Z Zebra mussels (Dreissena polymorpha), 160

Subject Index

A Acoustic bottom-classifier system, 175 Acoustic Doppler current profiler (ADCP), 175 Acoustic SONAR system, 100 Acoustic SONAR techniques, 100 Acoustic survey, 145, 151, 152, 172, 173, 174, 178 Acoustic-lens-based sonars, 161 Active shutterglasses, 327, 329 AD Model Builder (ADMB), 339, 340, 341, 347, 348, 350, 355, 357, 363 Adams River, 406 ADAPT, 12, 22, 23, 50, 71, 78, 84, 96, 97, 104, 113, 171, 177, 179, 198, 200, 203, 206, 207, 220, 230, 235, 237, 240, 241, 248, 258, 260, 264, 275, 338, 366, 380, 382, 403 ADCP (acoustic Doppler current profiler), 175 ADIAC project (Automatic Diatom Identification and Classification), 300 ADMB-RE, 355 ADOL-C, 340, 351 Age determination, 20, 306 Age/size structured, 384 Age-structured models, 361, 380 Age-structured, 354, 359, 361, 374, 379, 380, 383 Aircraft Platform, 137 Alternative management policies, 105, 106, 359, 402, 406, 407 Analysis of management policies, 105 Anderson and Ursin, 383 Anisotropy, 197, 198, 215 Annual growth, 23, 24, 306, 386 Apple Macintosh, 1, 3, 8, 299 Application Program Interfaces (APIs), 333 Aquaculture, 36, 37, 40, 51, 117, 244, 253

Aquatic Science and Fisheries Abstracts (ASFA), 35, 36, 37, 52, 81, 84 Arc Marine (the ArcGIS Marine Data Model), 99 Archival computer tags, 400 Area/time closures of fishing, 405 Artificial Intelligence (AI), 69, 71, 73, 75, 77, 79, 81, 83, 85, 87 A-SCALA, 348 AskTM, 37, 38, 64 ASPIC, 351 Atlantic Meridional Transect (AMT), 167 Automatic Mass Balance Procedure (AMBP), 229 Auxiliary data, see Population Dynamic model B BASIC, 2, 9, 20, 22, 29, 31, 33, 39, 40, 41, 42, 43, 53, 63, 72, 73, 77, 95, 100, 111, 131, 150, 152, 154, 156, 162, 163, 164, 166, 168, 192, 228, 240, 241, 242, 301, 303, 306, 320, 322, 327, 328, 337, 339, 341, 347, 351, 381, 404, 407, 408, 411 Bathymetric, 161, 319 Bayes’ theory, 358 Bayesian posteriors, 358 Bayesian statistics, 19, 338 Bees, 296 BEI (Bergen Echo Integrator), 155, 162, 165 Beverton Holt model, 359, 383 Biological Abstracts, 38 Biosis, 34, 37, 40 Bloglines, 48 Boolean searching, 41 Bootstrapping, 348 British Columbia, 102, 160, 249, 250, 268, 343, 363, 399, 406

415

416 Budget models, 376, 377 Budget, 57, 59, 85, 244, 250, 376, 377, 382 BUGS, 19, 345, 346, 357, 358, 364 C Carrying capacity, 106, 245, 379, 404, 405, 409 CASAL, 340, 348, 350, 351 Catch at age data, 353 CD-ROM, 12 CEDA (Catch Effort Data Analysis), 351 Centre for Earth Observation, 104 Chaotic series, 112, 379 Cite Seer, 34 C language, 50 Classification of GIS functions, 96 Climate change, 16, 29, 95, 113, 213, 269, 282, 391, 403, 404 Coastal zone, 117 Coastal zone Management, 117 co-kriging, 201, 202, 219 Coleraine, 347, 350, 351 Combined acoustic and egg production survey, 172 Comprehensive R Archive Network (CRAN), 342 Computer-Aided Design (CAD), 94 Computer-aided microscopy, 309 Computer Architecture, 7 Computer Language, 340 Computer Hardware, 7, 19, 93, 162, 325, 333, 334 Constrained variogram, 211 Continuous Plankton Recorder, 150 Continuous Underway Fish Egg Sampler (CUFES), 148, 167 COPEMED, 102 Copyright, 32, 47, 53, 54, 55, 56, 58, 64, 100, 101 Coral reef ecosystem, 377 Coupled logistic population models, 379 Coupled single-species models, 376, 377 Cross-variogram, 201, 202, 211 Current ContentsTM, 48 D D2-variogram, 192, 212 Daily growth, 306, 307 Data loggers, 94 Data Sources, 2, 16, 101, 117, 122, 364 Databases, 4, 9, 18, 22, 23, 28, 34, 36, 37, 38, 40, 42, 44, 45, 46, 48, 52, 53, 60, 97, 98, 100, 117, 139, 359

Subject Index DBMS (Database Management System), 86, 105, 117, 402, 405 Decision theory, 313 DeGray Reservoir, Arkansas, 382 Delay difference models, 311 Delta rule, 13, 72, 78, 80, 206, 207, 209, 313, 340, 355, 361, 362, 364 delta-GLM abundance indices, 364 Dempster-Shafer theory, 73 Depletion methods, 95, 261, 268, 269, 276, 402 Desk-top printers, 296 Digital cartography, 95 Digitisers, 94 DLP (Digital Light Processing), 330 Dual-beam, 125, 157 The Dual-Frequency Identification Sonar (DIDSON), 161 Dynamic programming, 404 E Echo integrator, 155, 162, 165 Echo sounder, see Scientific Acoustic Systems Echogram, 126, 130, 135, 151, 154, 155, 156, 163, 164, 165, 166, 171, 174, 212, 213 Echo-integration survey data, 345 Ecopath and Ecosim modelling tool (EwE), 226 ECOPATH, 19, 102, 225, 226, 227, 228, 229, 230, 231, 233, 235, 237, 239, 240, 241, 242, 243, 244, 245, 247, 249, 250, 251, 252, 253, 254, 255, 256, 257, 259, 260, 261, 263, 265, 267, 268, 269, 271, 273, 275, 276, 277, 279, 281, 283, 375, 376, 377, 383 Ecopath (Ecological Pathways Model), 19, 102, 225, 226, 227, 228, 229, 230, 231, 233, 235, 237, 239, 240, 241, 242, 243, 244, 245, 247, 249, 250, 251, 252, 253, 254, 255, 256, 257, 259, 260, 261, 263, 265, 267, 268, 269, 271, 273, 275, 276, 277, 279, 281, 283, 375, 376, 377, 383 Ecoranger, 241, 242 Ecosim, 19, 102, 225, 226, 227, 229, 231, 233, 234, 235, 237, 239, 240, 241, 242, 243, 244, 245, 247, 249, 251, 253, 255, 257, 259, 260, 261, 262, 263, 264, 265, 267, 269, 271, 273, 274, 275, 276, 277, 278, 279, 280, 281, 283, 375 Ecospace, 226, 237, 238, 239, 240, 244, 249, 250, 270, 271, 272, 273, 282 Ecosystem management, 80, 282, 401 Ecosystem models, 15, 29, 253, 269, 282, 376, 401 Edgetech 272-TD, 159

Subject Index Effort data, 107, 351 Egg and larval survey, 145, 146, 147 El Nin˜o, 323 Electronic chart display, 100 Electrostatic plotters, 94 Energy flux budget, 377 EnhanceFish, 351 Environmental change, 242, 248, 264, 360, 404, 405 Environmental variability, 266 Estimation of variance, 194 Ethernet, 163 European Space Agency, 96, 100, 104, 217, 218, 300, 351 EVA, 195, 220 Expendable Bathy-Photometer (XBP), 178 Experimental variogram, 195, 197, 198, 199 Exploitation rates, 232, 407 Extended survivors analysis (XSA), 351 F Factorial Correspondence Analysis (FCA), 228 FAO (Food and Agriculture Organisation of the UN), 18, 35, 52, 95, 102, 104, 105, 225, 347, 351 FAO BEAM, 351 Feedback and adaptiver policies, 403 Feed-back-estimation-control algorithms, 338 FFT (Fast Fourier Transform), 301, 308 Fin rays, 306 Fish and Fisheries Worldwide, 36, 37, 52 Fish scales, 306 Fisheries assessment, 122, 192, 366, 399, 401, 406, 409 Fisheries GIS, 94, 97, 98, 101, 102, 103, 105, 112 Fisheries management and regulation, 138, 220, 225, 373, 405 Fisheries stock assessment, 340, 352, 353, 358, 384 Fishery Data, 86, 339, 341, 347, 362, 363 Fishery independent survey data, 106, 151, 162, 191, 204, 207, 214, 215, 216, 217, 218 Fishery population dynamics, 20, 40, 99, 221, 249, 262, 264, 337, 350, 364–366, 377, 383–384, 400–403, 407 Fishing in Balance (FIB), 274, 275 Fitting variogram models, 195, 198, 217, 218 Fixed exploitation policy, 405 Fixed exploitation rate, 403, 404 Flat-field illumination, 309 Flow Cytometer and Microscope (FlowCAM), 179

417 FORECAST MASTER, 11, 12, 34, 43, 52, 80, 361, 383, 401 Forth, 45 FORTRAN, 19, 219, 339, 340, 341, 342 Fourier transform, 299, 301, 310 Fox model, 409 Fraser River, 160, 406 FT (Fourier Transform), 299, 301, 310 Fuzzy logic, 80, 86 Fuzzy set theory, 69, 73 G GA (genetic algorithms), 69, 79, 81, 82–85 Gadget, 58, 351, 366 Game playing, 407 General Electric RTV-602, 161 Genetic algorithms (GA), 69, 79, 81, 82–85 GeoCrust 2.0, 107 Geographical Information Systems (GIS), 93 GEOMAP, 93 GEOMAP, 93 Geometric anisotropy, 197 Geo-referencing, 101 George’s Bank, 158, 169, 170, 324, 377, 378, 379, 382, 383, 391 GEOS-3 altimeter, 132 Geosat altimeter, 132 Geostatistical analysis, 146, 148 Geowall, 330, 331, 334 Gestalt perception, 296 GIS (Geographical Information Systems), 93–118, 176 GISFish, 102, 105 Global index of collocation (GIC), 208 GLOBEC (Global Ocean Ecosystem Dynamics), 219, 335 Goodness-of-fit criterion, 235 Google ScholarTM, 34, 38, 39 GoogleTM, 17, 32, 34, 37, 38, 39, 40, 64 GPS (Global Positioning System), 16, 24, 94, 107, 176 Grey literature, 52, 53, 115 GRID, 93 Growth information, 306 Growth-rate measurements, 306 Gslib, 218, 219 GUI (Graphical User Interface), 163, 344 Gulland’s method, 409 H Harvest regulation, 405 Harvest strategies, 404

418 Harvesting rules, 401 Hierarchical Cluster Analysis, 228 High-Intake, Defined Excitation Bathyphotometer (HIDEX-BP), 178 Holistic models, 376, 382 Hybrid coordinate ocean model (HYCOM), 134 Hydrodynamic model, 322, 380, 410 I IA (Image Analysis), 20, 213, 297–301, 306, 309–314 IBM, 317, 329, 374, 381, 384, 385, 392, 410 ICES (International Council for the Exploration of the Sea), 36, 191 Identifying phytoplankton, 299 Identifying species from Echo Sounders, 309 IdentifyIt TM, 302 IDRISI Kilimanjaro, 110 Illegal fishing, 400, 405, 406 3-D imagery, 320, 324 Image processing, 2, 168, 170 IMAX 3D, 320, 331 IMMERSADESK, 329 Individual-based modeling, 392, 410 Inertiogram, 210 IngentaTM, 48 INGRES, 163 Institutional repositories (IR), 59 Integrated acoustic and trawl survey, 172 Integrated Data Viewer, 332 Integrated expert systems, 79 Isatis, 218, 219 ISI Science Citation Index, 34 ISIS, 351 Isosurfaces, 321, 323, 332 IWC (International Whaling Commission), 51 J Johnstone Strait, 406 JSTOR, 63 K Kriging, 194, 195, 197, 200, 201, 202, 203, 204, 205, 206, 217, 218, 219, 220 L LAN (Local Area Network), 7, 24, 163 see also Computer Hardware Laser Optical Plankton Counter (LOPC), 148, 167 Least Squares Procedures, 199

Subject Index Length frequency analysis, 350, 351 Length frequency data, 350, 351, 353 LFDA (Length Frequency Distribution Analysis), 351 Lidar, 123, 131, 135, 136, 137, 138, 139, 140, 170, 171, 173 Likelihood profile, 340, 355 Linear estimation (linear kriging), 194 Linear regression, 358 Linnaeus II, 301, 302 LISTSERV1, 27, 48 Lotka-Volterra, 233, 235, 377, 378 LUT (Look-Up Tables), 298 M Macintosh, see Computer Hardware Management strategy evaluation (MSE), 338, 359, 360 MAPS (The Multiple Acoustic Profiling System), 158 Marine Explorer, 98 Marine GIS, 97, 102 Marine Protected area (MPA), 271 Mark recapture data, 149 Mark repacture, 146 Markov chain Monte Carlo (MCMC), 29, 348 MATHCAD, 345 Matlab, 19, 219, 345, 364 Matrix projection models, 374, 380, 383, 384, 386, 387 Maximum likelihood, 344 Meta database, 100, 101, 114, 154 Meta-analysis, 70, 358, 359, 375 MFLOPS (Million Floating Point Operations Per Second), 163 Microwave radar altimeter, 132 Microwave scatterometers, 131, 132 MINIMization package, 150, 235 MIPS (Million Instructions Per Second), 4, 163 Mixed Trophic Impact (MTI), 231, 232, 274 Model complexity, 243, 360 Model variogram, 195, 199 Momentum factor, 374 Monte-Carlo testing, 241, 338, 347 MS EXCELTM, 346 MSVPA (Multispecies Virtual Population Analysis), 279, 281, 383 MS-WindowsTM, 7 MSY (Maximum Sustainable Yield), 268, 377, 408

Subject Index The multiple acoustic profiling system (MAPS), 158 Multispecies biomass-based models (PROBUB and DYNUMES), 382 Multispecies example of an individual based model, 374, 410 Multibeam sonars, 128, 152, 159, 166 MULTIFAN-CL, 348, 350 Multiple Opening and Closing Net and Environmental Sensing System (MOCNESS), 167, 169 Multiple-frequency echo integration, 125, 154, 156, 157, 164, 165, 172, 345 Multispecies Assessment Working Group of ICES, 383 Multispecies biomass based models, 382 Multispecies interactions, 374 Multispecies VPA, 279, 281, 351, 383 N National Information Services Corporation (NISC), 36 Nautical charts, 117 NETLIB, 340 NetVibes, 48 NETWORK, 2, 7, 10, 14, 16, 20, 22, 23, 24, 25, 28, 32, 36, 69, 72, 79, 81, 82, 84, 96, 104, 162, 163, 226, 231, 244, 250, 251, 252, 259, 281, 294, 309, 318, 320, 326, 342, 363 Neural Network Models, 20, 69, 79, 81, 82, 84, 168, 169, 294, 309 Neural Networks, 20, 69, 79, 81, 82, 84, 168, 169, 294, 309 NEXUS, 296Nodes, 7, 8, 16, 205 Noise, 86, 137, 163, 165, 298, 303, 313 Non-stationary geostatistics, 220 Non-stationary, 219, 220 Nugget, 196, 197, 198, 199, 205, 206, 216, 217 Numerical simulation, 373 O Observation error, 3, 18, 19, 52, 57, 86, 87, 111, 122, 125, 128, 129, 132, 135, 137, 138, 145, 147, 159, 169, 173, 174, 177, 318, 356, 362, 400 Oneida Lake IBM, 385 OpenBUGS, 345, 364 Operations involving computers, 145 Optical Plankton Counter (OPC), 148, 166, 167 Ordinary kriging, 201 OSMOSE model, 281, 382

419 OSMOSE, 281, 381, 382 Otolith microstructure, 306, 307 Otolith, 82, 301, 303, 304, 306, 307, 308, 314 P Parallel architecture, 334 Parallel processing, 7, 363 ParFish, 351 Passive polarization, 327, 330 Pattern Recognition, 20, 293, 294, 296, 297, 309, 311, 313, 314 PBS Mapping, 342, 343 PBS Modelling, 342, 344, 345 PC/XT/AT, 9 Peer-reviewed journals, 50, 51, 52 Pella-Tomlinson model, 339 Penrose staircase, 312 Penrose, 312, 313 Pentium, 3, 4 Perception systems, 94, 296, 318 Peruvian coastal ecosystem, 377 Peruvian upwelling system, 377 Photogrammetry, 95 Photography, 307 PhotoshopTM, 298 Planning systems, 26, 94, 163, 268, 401, 402 Poaching, 405 Polar orbiting, 17 Population Dynamic Model, 86, 347, 366, 384 Population model for assessment, 387 Population simulations, 401 PowerPC, 6, 196, 276, 307Predator-prey interactions, 385 Predator-prey models, 387 Pre-season forecasts, 383, 401 Process error, 354 PubMed Central, 59 Q Quantitative Fisheries Research Surveys, 145, 146, 147, 149, 151, 153, 155, 157, 159, 161, 163, 165, 167, 169, 171, 173, 175, 177, 179 Quick navigation, 324 Quota management systems, 402, 405 R Raster data format, 113 Raster format, 106 Raster-based GIS (OSU-MAP), 110

420 Reality Modeling Language (VRML), 332 Recruitment overfishing, 407 Recruitment-environment models, 410 Red-blue anaglyphs, 328 Regional variogram, 195 Regression model, 388 Regulatory models, 402, 403 Remote Environmental Measuring Units (REMUS), 161 Remote sensing, 16, 20, 94, 100, 117, 121, 122, 123, 125, 127, 129, 131, 133, 134, 135, 137, 138, 139, 140, 141 RESON SeaBat 8101, 159 RISC (Reduced Instruction Set Computer), 163 RISC-based CPU, 163 Risk averse harvest management objectives, 404 RoxAnn, 175 RS (Remote Sensing), 16, 20, 94, 100, 117, 121–140 R, S, and S-PLUS, 341 RSS (real simple syndication or rich site summary), 48 S S-PLUS, Gauss, MATLAB, 364 Scales, 15, 29, 62, 81, 112, 139, 140, 152, 164, 168, 193, 212, 214, 215, 220, 221, 306, 318, 321, 334, 359, 360, 377, 401 Scanmar system 400, 151 Scanners, 23, 94 Scanning multichannel microwave radiometer (SMMR), 131, 133 ScienceDirect, 59 Scientific Acoustic Systems, 107, 150, 151, 172 Scilab, 19, 345 Scirus, 38, 39 ScopusTM, 34, 35, 38 Sea Benthic Exploration Device (SeaBED), 171 Sea surface temperature (SST), 131 Seasat Radar Altimeter (ALT), 132 Seasat satellite, 133 SEM (Scanning Electronic Microscope), 306 Shadowed Image Particle Profiling and Evaluation Recorder (SIPPER), 179 signal-to-noise (SNR), 137 Simple equilibrium models, 408 Simple kriging, 41, 63, 211, 194, 195, 197, 200, 201, 202, 203, 204, 205, 206, 217, 218, 219, 220, 241, 313, 321, 330 SIMRAD BI500, 151 Simrad EK500, 155

Subject Index Simrad Integrated Trawl Instrumentation (ITI), 151 Simrad ME70, 15, 127, 159 Simrad MS70, 127, 159 Simrad PS18 Parametric Sub-bottom Profiler, 177 Simrad SA950 sonar, 15, 16, 20, 94, 100, 123, 127, 128, 131, 138, 152, 154, 156, 159, 160, 161, 162, 163, 170, 177, 178, 179, 309, 314, 400 Simrad SM2000, 159 SIMRAD Subsea AS, 151 Simulated Annealing, 20, 82, 84, 85 Simulation games, 401 Simulation models, 86, 338, 366 Single-beam transducer, 158 Single-species models, 242, 376, 377 SMAST Video Survey Pyramid, 170 Sonar, see Scientific Acoustic Systems Southern Oscillation (ENSO), 248, 264 Spaceborne instruments, 131 SPARC site, 64 Spatial allocation, 106, 115, 131, 152, 154, 195, 207, 296 Spatial analyses, 117 Spatial autocorrelation, 96 Spatial dynamics, 249, 281, 282, 401 Spatial model, 113, 239, 282, 363, 381, 407 Spatial stock structure, 103, 109, 198, 200, 208, 209, 212, 215, 217, 237, 250, 359 Spatial structure, 193, 197, 198, 199, 210, 211, 212, 215, 216, 217, 218, 220, 222, 322, 323, 350, 351, 359, 401, 402, 409 Spatial visualisation, 94, 97, 105, 114, 131 Spawner-recruit relationships, 379, 380 Split-beam modes, 49, 57, 148, 157, 267, 317, 320, 326, 331, 333 Split-beam transducer, 124, 125, 126, 127, 128, 129, 152, 154, 156, 157, 158, 159, 160, 166, 177 SQL (Structured Query Language), 163 Stationary geostatistics, 94, 129, 151, 179, 191, 192, 193, 194, 195, 197, 200, 207, 210, 211, 218, 219, 220, 325, 403 STATLIB, 340 STD (Salinity-Temperature-Depth sonde), 154, 164, 165, 174 Stereo APIs, 333 Stereo-Immersive approaches, 326 Stochastic recruitment, 18, 71, 80, 85, 86, 149, 160, 215, 220, 264, 344, 353, 354, 358, 359, 365, 378, 379, 380, 381, 388, 401, 405, 407, 409, 410, 411

Subject Index Stochastic search algorithm, 83 Stock assessments, 342, 348, 351, 352, 356, 358, 359, 361, 403 Stock recruitment models, 358 Stock SYNTHESIS, 348, 403 Stock synthesis, 348, 403 Stock-assessment models, 340, 342, 346, 347, 348, 350, 353, 354, 358, 365, 409 Surplus production modes, 339 Sustainable fisheries, 98, 405 Sustainable yields, 407 SVD (Singular-Value Decomposition), 305 SYMAP, 93 Synthetic aperture radar (SAR), 131, 132 Synthetic Aperture Sonar (SAS), 129 T Target strength analyzer, 164 Target strength, 125, 127, 137, 140, 157, 164, 166, 174, 207, 309 Taxonomy of spatial variation, 299 TCT/IP (Transport Control Protocol/ Internet Protocol), 163 Technical interactions, 383 Teeth, 306 TEMAS, 351 Terrestrial and marine-based data, 93, 94, 95, 97, 100, 112, 113, 375 Time-series regression model, 16, 18, 237, 388 Total System Throughput (TST), 229 Transducers, 124 Transfer Efficiency (TE), 230 Transponders, 400 Transport control, 163 Trawl mensuration, see Scientific Acoustic Systems Trawl survey, 150, 151, 172, 191 Triangulated Irregular Networks (TINs), 96 Trophic Levels (TL), 229, 230, 231 TS (Target Strength), 125, 126, 164 The turning bands method, 203 TVG (Time-Varied Gain) Type II functional response, 248, 263, 375, 378, 380, 382, 385 U UDP/IP (User Datagram Protocol/Internet Protocol), 164 Uncertainty in parameter estimates, 229, 240, 242

421 Underwater vehicle as platform (AUV), 129 Universal kriging, 200 UNIX, 7, 163, 219, 345 V Variogram models, 195, 197–199, 216–218 Variowin, 218, 219 Vector format, 99, 132, 165, 169, 210, 211, 212, 319, 321, 323, 328, 329, 332, 333, 346 Vessel Monitoring Systems (VMS), 107 VIBES (Viability of exploited pelagics in the Benguela Ecosystem), 102 Video Plankton Recorder (VPR), 168 Virtual Population Analysis (VPA), 351 Visual circuitry, 294 Visualisation, 94, 97, 105, 114, 131 von Bertalanffy, 2, 344, 379, 381 VPA (Virtal Population Analysis), 281, 351, 383 VR-CAVE, 329 W WAN (Wide Area Network), 20 Web of Science1, 34, 35, 38, 40, 51, 52 Web of Social Science1, 34 Western Gulf of Alaska, 382 Westinghouse SM2000, 179 WinBUGS, 19, 345, 357, 358 Windows Live Academic, 27 Workstation, 7, 163, 296, 330, 332 World Wide Web (WWW), 20, 21 xWindow System, 163, 165 Y Yahoo1, 27, 37, 38 Yellow perch-walleye model, 385, 386 Yield, 37, 86, 106, 111, 146, 147, 150, 151, 154, 228, 253, 268, 304, 318, 321, 337, 342, 344, 346, 351, 377, 378, 383, 401, 407, 408 Yield-per-recruit, 342, 344 YOY (Young Of the Year), 380, 382, 385, 386, 387, 388, 389 Z ZeusTM, 297, 301, 304, 308, 311, 314 Zonal anisotropy, 197, 198 Zooplankton Visualization System (ZOOVIS), 179