Tab B, No. 10a
rev. 9/5/2007
REPORT
of the
ECOSYSTEM MODELING WORKSHOP
St. Petersburg, Florida
May 8-10, 2007
A workshop to evaluate whether existing ecosystem models can provide useful advice to the Council
about key ecosystem management questions, and if not, then what will be needed to make such models
useful
Gulf of Mexico Fishery Management Council
2203 North Lois Avenue, Suite 1100
Tampa, Florida 33607
813-348-1630
813-348-1711 (fax)
888-833-1844 Toll Free
gulfcouncil@gulfcouncil.org
http://www.gulfcouncil.org
National Oceanic & Atmospheric Administration
National Marine Fisheries Service
Southeast Regional Office
263 13th Avenue South
St. Petersburg, Florida 33701
727-824-5308
727-824-5305 (fax)
http://sero.nmfs.noaa.gov
This is a publication of the Gulf of Mexico Fishery Management Council Pursuant to National Oceanic and
Atmospheric Administration Award No. NA04NMF4410325.
This page left intentionally blank
Table of Contents
1
ABBREVIATIONS USED IN THIS DOCUMENT .......................................................... III
2
EXECUTIVE SUMMARY .................................................................................................... 1
3
INTRODUCTION .................................................................................................................. 2
3.1 BACKGROUND .................................................................................................................... 2
3.2 WORKSHOP OBJECTIVES ..................................................................................................... 3
3.3 CONDUCT OF AEAM WORKSHOP PROCESS ........................................................................ 3
3.4 POLICY SCENARIO TESTS AND MODEL IMPROVEMENTS DURING WORKSHOP .................... 5
3.5 POLICY RECOMMENDATIONS TO THE GULF COUNCIL ......................................................... 6
3.5.1 Shrimp Trawl Bycatch................................................................................................. 6
3.5.2 Harmful Algal Blooms (Red Tide) .............................................................................. 6
3.5.3 Hypoxic Zone .............................................................................................................. 7
3.5.4 Marine Protected Areas .............................................................................................. 8
3.5.5 Nutrient Loading Impacts on Stock Recruitment and Productivity ............................ 8
4
RECOMMENDATIONS FOR FURTHER ECOSYSTEM MODEL DEVELOPMENT
9
4.1
4.2
4.3
PROCESS ............................................................................................................................. 9
SPECIFIC DATA NEEDS FOR FOLLOW-UP WORKSHOPS ........................................................ 10
SUPPORT FOR ALTERNATIVE MODELING APPROACHES ...................................................... 10
5
WORKSHOP AGENDA...................................................................................................... 12
6
LIST OF ATTENDEES ....................................................................................................... 13
7
LIST OF PREPARERS ....................................................................................................... 14
8
REFERENCES ..................................................................................................................... 16
9 APPENDIX 1 - MODELING A CLUPEID DOMINATED ECOSYSTEM ON THE
WEST FLORIDA SHELF, SUBJECTED TO BOTH RED TIDE AND OTHER
PISCIVORE STRESSES, BY JOHN J. WALSH, COLLEGE OF MARINE SCIENCE,
UNIVERSITY OF SOUTH FLORIDA .................................................................................... 17
10 APPENDIX 2. A SPATIAL ECOSYSTEM MODEL FOR ATLANTIC COAST
MULTISPECIES FISHERIES ASSESSMENTS BY JERALD S. AULT, UNIVERSITY OF
MIAMI RSMAS .......................................................................................................................... 18
11 APPENDIX 3. QUANTITATIVE METHODS FOR FUNCTIONAL GROUP
ASSIGNMENT BY ERNST B. PEEBLES, UNIVERSITY OF SOUTH FLORIDA,
COLLEGE OF MARINE SCIENCE, AND DAVID D. CHAGARIS, FLORIDA FISH AND
WILDLIFE CONSERVATION COMMISSION, FISH AND WILDLIFE RESEARCH
INSTITUTE ................................................................................................................................. 18
i
12 APPENDIX 4. REPORT ON A FOOD HABITS OF FISHES BIBLIOGRAPHY
AND A PROPOSAL FOR A TROPHIC DATABASE FOR THE GULF OF MEXICO BY
JAMES SIMONS, TEXAS PARKS AND WILDLIFE DEPARTMENT .............................. 21
13 APPENDIX 5 - AN ECOSIM MODEL FOR EXPLORING ECOSYSTEM
MANAGEMENT OPTIONS FOR THE GULF OF MEXICO: IMPLICATIONS OF
INCLUDING MULTISTANZA LIFE HISTORY MODELS FOR POLICY
PREDICTIONS BY CARL WALTERS, UNIVERSITY OF BRITISH COLUMBIA, AND
BEHZAD MAHMOUDI, FLORIDA MARINE RESEARCH INSTITUTE......................... 23
ii
1
Abbreviations Used in This Document
AEAM
BRD
EwE
FIM
FWRI
LSU
MDS
MPA
NMFS
NOAA
PNAS
RSMAS
SEAMAP
SEDAR
SEFSC
SERO
SSC
iii
Adaptive Environmental Assessment and Management
Bycatch reduction device
Ecopath with Ecosim
Florida Fish and Wildlife Research Institute
Louisiana State University
Multidimensional scaling
Marine protected area
National Marine Fisheries Service (NOAA Fisheries)
National Oceanic & Atmospheric Administration
Proceedings of the National Academy of Sciences
Rosenstiel School of Marine and Atmospheric Science (Univ. of Miami)
Southeast Area Monitoring and Assessment Program
Southeast Data, Assessment and Review
Southeast Fisheries Science Center (of NMFS)
Southeast Regional Office (of NMFS)
Scientific and Statistical Committee
2
Executive Summary
On May 8-10, 2007, the Ecosystem Modeling Workshop convened at FWRI, with the objective
of demonstrating the feasibility of using ecosystem modeling as a tool to address fishery
management issues. Over the next three days, the workshop heard presentations from several of
the participants on a range of ecosystem research projects, including modeling a clupeid
dominated ecosystem Walsh, Appendix 1), a description of a spatial ecosystem model for
Atlantic coast multispecies fisheries assessments (Ault, Appendix 2), and an evaluation of
methods for assigning fishes to functional groups with similar trophic ecologies (Peebles and
Chagaris, Appendix 3).
In this workshop, the expert group used an Ecopath/Ecosim (EwE) ecosystem model developed
for the SSC by Walters, et al. (2006) (Appendix 5) as the strawman starting model as the basis
for examining a number of fishery related ecosystem issues. This model simulates agestructured population dynamics of a collection of important fish species, and biomass dynamics
of an additional 32 ecosystem functional groups ranging from phytoplankton to commercial
shrimp to a variety of fish species. The issues examined and the conclusions reached through
this modeling approach are:
Shrimp Trawl Bycatch - Management should implement experimental comparisons of areas with
historical shrimp fishing methods, areas with bycatch reduction, and areas closed to trawling.
Harmful Algal Blooms (Red Tide) - Toxic algae blooms will continue to occur, some may be
independent of anthropogenic nutrient loading, and will negatively impact overall ecosystem
production and carrying capacity (planktivores, predators)
Hypoxic Zone - Increases in hypoxic area have apparently contributed to an overall shift from
benthic to pelagic production until the most recent six years.
Marine Protected Areas - Existing MPAs on at least the northern part of the Florida Shelf will
not be an effective tool for regulation of fishing impacts; much larger, cross-shelf MPAs would
be needed to protect a range of species from fishing suffered during life-cycle offshore
movement.
Nutrient Loading Impacts on Stock Recruitment and Productivity - Changes in productivity
make it dangerous to base management policies on historical reference points calculated from
single species models that do not account for such changes
1
3
Introduction
This report summarizes results from an Adaptive Environmental Assessment and Management
(AEAM) workshop sponsored by the Ecosystem SSC, aimed at evaluating whether existing
ecosystem models can provide useful advice to the Gulf of Mexico Fishery Management Council
and to advise the Council as to whether to invest further in development of ecosystem models.
The basic approach in AEAM workshops is similar to the SEDAR process, namely to convene a
group of scientific experts to use mathematical models as a tool to help focus discussions of
available scientific data and to make specific predictions about the efficacy of various policy
alternatives. The AEAM process begins with discussion of policy questions to be addressed by
the expert group, then proceeds to use of an initial “strawman” model developed before the
workshop to provide initial answers to the policy questions, then to a process of critical analysis
and development of improved models and data analyses based on questionable predictions from
the strawman model. AEAM workshop discussions are facilitated by a modeling team,
consisting in this case of Drs. Walters and Villy Christensen, University of British Columbia.,
who are familiar enough with the modeling software used for strawman model development
(Ecopath/Ecosim in this case) to make at least some changes and improvements in model
structure and data as suggested by workshop participants.
Here we review the conduct of the AEAM workshop process, discuss scenarios tested with the
strawman model and model improvements developed during the workshop, review main policy
recommendations to the Council from the initial simulation tests, and recommend continuation
of the AEAM process to improve the model for further use by the Council. The
recommendations presented below are somewhat controversial and do not represent a consensus
of the entire Ecosystem SSC (only a few members were present at the workshop); they are
subject to change after further scientific review and after further discussion by the full SSC.
3.1
Background
In 2004, the Gulf Council was one of four regional fishery management councils selected by
NOAA Fisheries to be funded for a pilot project to begin the process of developing an ecosystem
based approach to fisheries management. An Ecosystem Scientific and Statistical Committee
(SSC) was formed to assist the Gulf Council in developing this project. As part of the
development, the Ecosystem SSC proposed holding a three day modeling workshop at the
Florida Fish and Wildlife Research Institute (FWRI) in St. Petersburg, Florida to demonstrate the
feasibility of using ecosystem modeling as a tool to address fishery management issues, and to
expose the capabilities and gaps in ecosystem model applications.
While there are a number of modeling approaches that could be used to address fishery issues,
the Ecosystem SSC decided to focus on a specific model, Ecopath with Ecosim, for the
workshop. This decision was made because of familiarity of the model by several of the
Ecosystem SSC members and because preliminary work has already been done towards
developing an Ecosim model of the Gulf of Mexico (Appendix 5).
2
On May 8-10, 2007, the Ecosystem Modeling Workshop convened at FWRI, with the Ecosystem
SSC, several invited ecosystem modeling experts, plus several guests who came from several
universities and research labs. A list of attendees is included in this report.
3.2
Workshop Objectives
Several research groups have been developing multispecies and ecosystem models aimed at
assisting in development of fishery ecosystem policies for the Gulf of Mexico. These models
have mainly focused on specific interactions and areas within the Gulf. By bringing these groups
together in a workshop setting, it was hoped to meet the following four broad objectives:
To determine what models now predict about the efficacy or impact of specific policy options
that the Council is now facing, particularly concerning bycatch reduction in the shrimp fishery,
impact of multispecies management tools such as closed areas and seasonal closures, and
impacts of hypoxic areas and toxic algae blooms on demersal and pelagic production;
To evaluate whether the Council should take these predictions seriously, what particular data
gaps limit the credibility of the predictions, and whether these gaps can be filled through
monitoring and management field trials (experiments);
To develop a Gulf-scale demonstration ecosystem model for major stocks, trophic interactions,
and habitat limiting factors, utilizing data from existing models and assessments and capable of
use for exploring policy alternatives particularly related to habitat protection and regulation of
shrimp and forage fish fisheries;
To plan a program of collaboration among the research groups to further develop the Gulf-scale
model and to fill information gaps identified during its development and testing.
All of these objectives were met to at least some extent. Not all modeling groups from the GOM
were able to attend, and in particular there were no representatives from important modeling
teams at LSU and at the NMFS Panama City laboratory. A demonstration model was developed
(objective 3) and used as a starting point for meeting objectives 1 and 2. Objective 4 was met by
a proposal to the Council to continue the AEAM model development process in further
workshops.
3.3
Conduct of AEAM Workshop Process
The workshop was organized using the Adaptive Environmental Assessment and Management
(AEAM) workshop process. This process was developed during the 1970s as a means to
organize knowledge and model development for complex ecological management problems, and
has been used on over 60 case examples. Its development and methodology are fully described
in Holling (1978) and Walters (1986). An excellent introductory guide to AEAM for project
leaders, participants, and administrators has been prepared by Brian Nyberg, B.C. Ministry of
Forests, and this guide is available on the web at
http://www.for.gov.bc.ca/hfp/amhome/INTROGD/Toc.htm. The workshop process proceeds in
the following basic steps:
Clearly identify policy options to be discussed and compared using models developed before and
during the workshop;
Identify precise performance measures or quantitative indicators to be used in comparing the
3
policy options;
Define and improve a model for linking the policy choices to performance indicators, i.e. state
variables and relationships needed to predict response of indicators to the specific options;
Conduct policy tests and perform iterative improvements in model structure based on credibility
of the policy predictions as the model is changed and improved;
Develop specific recommendations for further model development, data acquisition, and
experimental management in the case of uncertainties that cannot likely be resolved just through
further modeling and data analysis.
Typically step 3 is greatly facilitated by developing a strawman starting model before the
workshop, as a means to focus initial discussion and catalyze further development.
The AEAM model development process is very similar to the SEDAR process, except that the
first step in AEAM is a model development process rather than a data workshop. This reversal is
required in situations where it is unclear at the outset what data will actually be needed and
useful for model parameterization and testing, so that participants in an initial data workshop
would be unable to decide on priorities for data gathering without guidance about what model
structure would eventually be used for the data analysis. Further, AEAM typically involves
more iterations back and forth between model development (assessment), data gathering, and
review meetings than would typically be used in SEDAR. This increased complication reflects
the much more ambiguous and complex needs for ecosystem model development as opposed to
single species assessment model development.
In this workshop, the expert group used an Ecopath/Ecosim (EwE) ecosystem model developed
for the SSC by Walters, et al. (2006) (Appendix 5) as the strawman starting model. This model
simulates age-structured population dynamics of a collection of important fish species (snook,
red drum, spotted sea trout, ladyfish, mullet, mackerel, grouper, red snapper, menhaden), and
biomass dynamics of an additional 32 ecosystem functional groups ranging from phytoplankton
to commercial shrimp to a variety of fish species like jacks for which development of agestructured population models was not practical before the workshop. The model “calibrated”
before the workshop by statistical fitting to time series catch, relative abundance, and total
mortality rate data from single species stock assessments, SEAMAP fishery independent trend
data, and Florida FIM fishery independent trend data for as many of the species as possible; this
calibration process had demonstrated that the strawman model could at least provide realistic
hindcasts of abundance changes for most major fish species. The full model and all parameter
estimates and data used in model calibration have been made available to workshop participants
in the form of an Ecopath database (Access database, Gulf simple.mdb, available at
www.gulfcouncil.org).
In addition to the EwE model, presentations were made by workshop participants on a number of
other modeling approaches to specific ecological concerns, ranging from mechanisms that may
cause toxic algae blooms (John Walsh, Appendix 1) to complex spatial models for linking fish
population responses to biophysical ecosystem characteristics (Jerry Ault, Appendix 2) to recent
development of methods for aggregating species into model functional groups based on diet
(Ernst Peebles and Dave Chagaris, Appendix 3) and progress in development of a Gulf-wide
database on diets (James Simons, Appendix 4).
4
3.4
Policy Scenario Tests and Model Improvements During Workshop
Initial definition of policies to be tested went smoothly, with no additional tests suggested
beyond those planned before the workshop. Policies to be examined (simulated) with the
strawman model included:
shrimp trawl bycatch reduction and complete elimination of trawling
impact of historical and future occurrence of hypoxic areas off the Mississippi River mouth
impact of historical and possible future patterns of occurrence of toxic algae blooms
effects of marine protected areas on grouper populations of the West Florida shelf (using an
alternative multispecies modeling approach, FISHMOD)
impacts on existing fisheries of future development of fisheries for small pelagics to provide feed
for developing aquaculture of species like cobia and bluefin tuna.
Unfortunately, there was insufficient time to examine issue 5), but simulation runs and
evaluations were done for the other four issues.
Initial demonstration runs of the strawman model elicited largely positive comments (and some
surprise among the scientists) about how well the model has been able to hindcast abundance
changes in major fish stocks, and considerable discussion about whether the model reliably
represents impacts of trophic interactions (predation mortality rate changes) given the relatively
crude abundance, feeding rate, and diet composition data used to parameterize it.
Policy tests with the strawman model triggered intense discussion and debate, as is typical in
AEAM exercises. The first policy tested was closure of the shrimp fishery, and the strawman
model predicted only modest recovery of red snapper recruitment under this policy. Discussion
about the prediction centered on a “counterintuitive” dynamic response, namely a gross increase
in catfish (Arias) populations following elimination of bycatch mortality, which in turn led to
competition between simulated juvenile red snapper and catfish. The simulated catfish response
was found after discussion to be due to questionable assumptions about its bycatch and natural
mortality rates, and entry of more realistic assumptions into the model database led to predictions
of stronger increase in red snapper recruitment following bycatch reduction (via trawl closures or
BRDs). However, other model runs and fitting tests by Dr. Christensen showed that the same
large response by other competitors/predators on juvenile red snapper might occur (so as to
prevent red snapper recovery) as had been initially predicted for catfish, but might involve other
bycatch-impacted species like Atlantic croaker or pigfish. These discussions and model
revisions demonstrated that the AEAM process is indeed capable of assisting to both develop
policy scenarios and to provide improved model/parameter estimates within an “interactive”
modeling and data analysis process.
The original workshop plan and agenda called for spending about ½ the total workshop time on
such interactive policy testing and model improvement exercises, making use of the knowledge
of invited experts, particularly from various NMFS laboratories. Unfortunately, presentations on
other modeling approaches and discussion of basic model structure issues took much more time
than expected, so we were only able to make a few limited improvements in the strawman model
(like fixing the catfish dynamics) and to enter a few new “forcing” datasets (for red tides,
nutrient loadings from the Mississippi River) for examining policy issues besides changes in
fisheries management.
5
3.5
Policy Recommendations to the Gulf Council
Despite very considerable uncertainty about specific data and quantitative predictions from the
EwE model, there was consensus among the workshop participants about several policy
recommendations following from the test simulations. These are presented in the following
subsections. They illustrate the sort of policy predictions that are now possible with relatively
simple ecosystem models, and it is hoped that they will be of some use to the Council even
without precise quantification and numerical prediction results.
3.5.1 Shrimp Trawl Bycatch
Management should implement experimental comparisons of areas with historical shrimp fishing
methods, areas with bycatch reduction, and areas closed to trawling.
As noted above, simulation runs with either shrimp trawl closure or fleet-wide requirement for
BRDs (assuming these to be 100% effective at eliminating fish bycatch) led to highly variable
predictions about increases in red snapper recruitment, with some scenarios showing only
transient improvement followed by reduced recruitment later due to increases in
competitor/predator species (see also Walters et al 2006). It is unlikely that improvements in
model inputs for bycatch rates, diet compositions, and population dynamics characteristics of
bycatch species (like Atlantic croaker) based on historical trend data will substantially reduce
uncertainty about such competitor/predator responses. That is, we are not likely to be able to
provide definitive predictions about the efficacy of bycatch reduction measures for enhancing
recruitment of stocks like red snapper that are impacted as juveniles by bycatch mortality.
Based on the widely varying predictions from the model about changes in red snapper juvenile
mortality rates under bycatch reduction, the Council should have little confidence in predictions
that bycatch reduction will lead to reduced mortality rates and enhanced recruitment, as have
come for example from some SEDAR assessment model configurations. Given such high
uncertainty, an experimental approach to large-scale testing of bycatch reduction can be
considered a precautionary management approach.
It is also highly probable that several fish species now impacted by bycatch mortality will
increase following bycatch reduction, and these species will cause increased predation mortality
(increase in natural mortality rate M) of shrimp. This in turn will probably lead to at least some
reduction in total shrimp production and harvest. Ecosim scenarios showed this reduction to
most likely be in the range 10-30%.
3.5.2 Harmful Algal Blooms (Red Tide)
Toxic algae blooms will continue to occur, some may be independent of anthropogenic nutrient
loading, and will negatively impact overall ecosystem production and carrying capacity
6
(planktivores, predators)
John Walsh made an interesting case to the workshop that at least those toxic algae blooms that
have occurred along the West Florida shelf involve a complicated algae succession triggered by
inputs of iron from dust storms over the Sahara desert. His models explain why the blooms are
not clearly related in time or space to obvious sources of anthropogenic nutrients. If he is right,
then management efforts aimed at reducing toxic algae blooms through reduction in nutrient
loading (pollution management on land) will not be beneficial, and fishery management will
have to live with continuing mortality events.
When the historical pattern of toxic algae blooms was entered into the Ecosim model as a
“fishery” impacting mortality rates of a few key species (pinfish, scaled sardine) that are known
to have suffered local declines in abundance following blooms, the model predicts an overall
decrease in production at higher trophic levels (piscivores like groupers and mackerel) of around
10% due to loss of prey production, and this loss would be a maximum estimate since it assumes
the prey populations to be impacted all around the Gulf and not just at hot spots as observed
historically. So our basic conclusion is that while toxic algae blooms may be highly visible to
the public (dead fish, local decreases in abundance of species like pinfish), they are unlikely to
be a major issue in setting abundance goals and management policies for most Gulf fish stocks.
3.5.3 Hypoxic Zone
Increases in hypoxic area have apparently contributed to an overall shift from benthic to pelagic
production until the most recent six years.
A time series estimate of benthic habitat made available to fish each year due to the river-mouth
hypoxic zone was included in some Ecosim simulation tests during the workshop. The hypoxia
area time series was derived from data and model results in Scavia et al. 2003, and represents an
example of linking two types of models (a detailed model for dynamics of the hypoxic zone and
an overall ecosystem model with capability to represent impacts of changing habitat area on
production and predator-prey interactions). Including the hypoxia factor as a cause of reduced
feeding by benthic fishes on menhaden caused an immediate improvement in model fits to
menhaden stock assessment estimates provided by D. Vaughan (2004). Other simulated
planktivores (scaled sardine, bay anchovy) also showed positive responses in the model.
These results support arguments by Chesney and Baltz (2001) that increased nutrient loading,
increased hypoxic area, and fishing impacts on demersal predator stocks may all have
contributed to a general shift in community structure toward pelagic from benthic production.
However, Vaughan and Cowan (pers comm.) point out that recent declines in menhaden and bay
anchovy abundance are inconsistent with the hypothesis that the apparent shift in community
structure is persistent or irreversible.
The potential importance of this finding to the Council is that it helps to interpret supposed
warnings or criticisms of Gulf fishery management based on simple trophic level indices
calculated from catch data by Pauly and Palomares. Further criticisms of such calculations are
7
forthcoming from LSU and U. Washington scientists (K. de Mutsert, J. Cowan, T. Essington, R.
Hilborn, ms in review, PNAS).
3.5.4 Marine Protected Areas
Existing MPAs on at least the northern part of the Florida Shelf will not be an effective tool for
regulation of fishing impacts; much larger, cross-shelf MPAs would be needed to protect a range
of species from fishing suffered during life-cycle offshore movement.
Over the past three years, the Marfin program has provided a grant to Florida FWRI for
development of Fishmod, a detailed spatial multistock population dynamics and fishing effort
distribution model. As a demonstration case for this model, Walters and Mahmoudi have been
developing an analysis of red and gag grouper on the West Florida Shelf. A specific aim of the
model is to compare the efficacy of policies ranging from MPAs to size limits and seasonal
closures for protection and restoration of grouper stocks. Fishmod gives very good fits to stock
assessment data for gags and red grouper, closely matching observed monthly catch data and
depth distributions of catch for the last 20 yrs.
Demonstration runs of Fishmod during the workshop included policy gaming trials comparing
simulated stock and fishing mortality rate trends for current management, implementation of a
few offshore MPAs (eg Madison-Swanson) aimed at protecting spawning stocks, and larger
MPAs extending from shore to the continental shelf break and intended to protect fish
throughout their offshore ontogenetic migrations from nursery areas and seasonal onshoreoffshore movements. The basic result from these simulation trials is that the offshore MPAs are
likely to have almost no impact on abundance or fishing rates, since effort displaced from the
protected areas will simply target younger fish inshore of them, and the fish protected during
spawning times will be caught at other times of year during seasonal migrations. Only the very
large onshore-offshore protected areas had impacts on fishing rates comparable to those
achievable through extensive seasonal closures and/or larger size limits. Similar results have
been obtained in extensive gaming trials with Fishmod by FWRI and NMFS Panama City staff
who have participated in the model development, calibration, and testing.
Note that this prediction about lack of efficacy of small MPAs for management of fishing
mortality on groupers is not meant to be a comment on possible benefits of such MPAs for other
species or aspects of marine biodiversity. However, it is well in line with other recent modeling
work questioning the benefits of small MPAs for ecosystem protection in general.
3.5.5 Nutrient Loading Impacts on Stock Recruitment and Productivity
Changes in productivity make it dangerous to base management policies on historical reference
points calculated from single species models that do not account for such changes
Some discussions of the Gulf of Mexico ecosystem point to increases in nutrient loading from
the Mississippi as a possible cause of sustained high fish and shrimp production. However,
8
nutrient loading historical data in Scavia et al. 2001 indicate that loading peaked in the mid1980s, coincident with peaks in production of both shrimp and menhaden, and has in fact
decreased on average since then. Ecosystem models uniformly predict that such long term
production trends or cycles will drive corresponding peaks in production of fish at all trophic
levels, especially for stocks whose juveniles are dependent on estuarine and near-coast rearing
environments. Driving the Ecosim model with nutrient loading time series did not generally
improve model fits to historical data during the workshop, but did help to explain a few peaks in
production during the 1980s.
However, there is a broader implication of having a peak in historical nutrient loading and Gulfscale productivity during the 1980s. That is for stock assessment models in general for Gulf
species (e.g. snapper, mackerel, groupers, menhaden, etc.), where model fitting to historical data
has typically been done under the assumption of stationary stock-recruitment relationships with
any interannual variations explained by estimating annual recruitment “anomalies”. Policy
reference points derived from such models (e.g. FMSY, unfished biomass, ratio of current to
unfished biomass) could be biased upward if relatively low nutrient loadings persist into the
future, compared to the loadings that occurred during the times over which the models have been
fitted. That is, it is reasonable to expect that stock rebuilding plans for species like red snapper
will not achieve stock sizes predicted from models that are based on more productive historical
periods. Likewise, future production for shrimp and menhaden, species closely linked to coastal
nutrient delivery, cannot be reliably predicted from historical production , especially using data
from the 1980s peak period.
4
Recommendations for Further Ecosystem Model Development
It was clear from discussions during the workshop that there is considerable interest among
cooperating scientists and agencies in pursuing development of the EwE ecosystem model, and
reasonable expectation that improvements in model input and calibration data will result in a
model that can more reliably predict both direction and magnitude of responses of key
populations (shrimp, menhaden, major fish stocks) to policy initiatives such as bycatch reduction
and future development of fisheries and habitat changes. It appears that the AEAM workshop
approach, which combines model scenario development and data analysis tasks, can be used in
the SEDAR framework, through a workshop sequence similar to SEDAR.
4.1
Process
It is recommended that an AEAM-SEDAR workshop process be funded by the Council, with at
least two workshops over the remainder of 2007. The first workshop would build upon the May
8-10 initial one, and would focus on entry of better data into the model and careful review of
data sources and parameter estimates. The second workshop would involve a wider audience of
management as well as scientific staff, and would focus on development of specific policy tests
and analysis of uncertainty about model predictions of efficacy of various policy alternatives.
The Council should also consider a third workshop comparable to the SEDAR review
9
workshops. This workshop would be aimed at bringing in a broader group of scientific experts to
examine model data sources, model assumptions, and statistical fitting methods along with
revised policy recommendations arising from the second workshop.
It must be emphasized that ecosystem model development is a much more complex process than
single-species assessment, requiring a much wider range of data and expertise, and much more
complex policy questions, than have typically been addressed in the three-step SEDAR process
(data, assessment, review). Further, it is never going to be clear when an ecosystem model
development process has “succeeded” or been “completed”; models may at any point of
development be able to provide useful predictions about some policy questions while still being
incapable of dealing with others, and no ecosystem model can ever be considered a complete
representation of an ecosystem (there will always be new questions, surprising changes, etc.).
Indeed, the Council may wish to consider development of an ongoing, permanent ecosystem
model development process aimed at long-term model testing and improvement as new
information become available, just as is done for single-species assessments.
4.2
Specific data needs for follow-up workshops
There are a number of key data needs for further model development and calibration/testing.
People should be invited to the next AEAM workshop who can provide at least the following
specific information, as identified at the first workshop:
Improved estimates of species/size composition of shrimp bycatch, preferably time series
estimates as have been developed for red snapper assessments
The longest possible time series of abundance indices (most likely mainly from SEAMAP) for
non-targeted and bycatch fish species, such as rays, croakers, pigfish, and other small demersal
species.
Diet composition data gathered as widely over the Gulf as possible, for as many species as
possible, with particular emphasis on species that overlap spatially with menhaden, scaled
sardine, and shrimp (particularly species that are abundant in shrimp bycatch).
Longest possible time series of nutrient loading to the Gulf from the Mississippi and other
nutrient loading sources around the Gulf.
Longest possible time series of chlorophyll/phytoplankton biomass and zooplankton biomass
from areas near the Mississippi River plume.
The last two points in particular were raised as model driver and calibration needs during the
workshop, but for which none of the workshop participants was able to provide immediate
information or knew of how to access such information.
4.3
Support for alternative modeling approaches
While the EwE ecosystem modeling approach holds promise as a tool for representation of
policy issues related to large scale fishery production and trophic/habitat interactions that may
affect future fishery production, there is a clear need to pursue other modeling approaches that
can provide more detailed “mechanistic” insight into some key processes and interactions. The
workshop demonstrated that results of such detailed approaches can be linked into simpler
10
overall models like EwE to assist in broad policy formulation, e.g. results from detailed hypoxic
area and toxic algae bloom modeling were used to drive Ecosim scenarios discussed above.
Likewise, predictions of detailed models for spatial distributions of individuals, bioenergetics of
growth, and predation mortality risk (like those being developed by Ault at Rosensteil and Rose
at LSU) can be used to effectively assimilate a much wider range of monitoring data (e.g.
oceanographic information on currents, nutrient chemistry, temperature) into model predictions,
and to inform the simpler aggregate models. There is an old saying attributed to Richard Levin
that “truth often lies at the intersection of competing lies”; the various modeling approaches
should be viewed as complementary rather than competing, and the Council should seek advice
from all of them.
11
5
Workshop Agenda
GULF OF MEXICO FISHERY MANAGEMENT COUNCIL
ECOSYSTEM SCIENTIFIC AND STATISTICAL COMMITTEE
FLORIDA FISH AND WILDLIFE RESEARCH INSTITUTE
ST. PETERSBURG, FLORIDA
MAY 8 – 10, 2007
Workshop leader: Dr. Carl Walters
Ecosim/Ecopath consultant: Dr. Villy Christensen
This agenda is arranged in the format of Adaptive Environmental Assessment and Management
(AEAM) workshops, rather than the usual mini-conference format of many formal presentations
with simple question and discussion sessions. The idea is to promote discussion, evaluation, and
model development rather than simple information sharing.
Tuesday May 8
9:00 am – 5:00 pm
Welcome and Introductions
Description of Workshop Format and End Products
Review of key ecosystem policy questions that Council would like addressed by models (1 hr)
Predictions for each of the modeling exercises of impacts of policy change, including at least
predictions about shrimp fishery bycatch reduction, menhaden fishery restrictions (3 hr)
Analysis of differences among models in predicted policy impacts, discussion of causes for those
differences, e.g. data vs. model structure (3 hr)
Individual presentations to be given on an ad hoc basis as appropriate to the discussion. Refer to
end of agenda for a list of expected presentations.
Wednesay, May 9
8:00 am – 5:00 pm
Identification of key needs for improved data and management policy tests to resolve differences
among models (2 hr)
12
Development of initial state and policy variable list for Gulf-scale model (2 hr)
Parameterization and initial testing for Gulf-scale model (4 hr)
Individual presentations to be given on an ad hoc basis as appropriate to the discussion. Refer to
end of agenda for a list of expected presentations.
Thursday, March 10
8:00 am – 3:00 pm
Development and presentation of revised predictions from the group models based on model
comparisons and discussions of parameter values for Gulf-scale model (2 hr)
Conduct and documentation of initial policy tests for Gulf-scale model (2 hr)
Planning session to develop responsibilities and timetable for further development and utilization
of Gulf-scale model (3 hr)
Discussion of workshop final report to Council, writing assignments, and timetable for drafting
report.
Where do we go from here? Recommendations for future direction of the Ecosystem SSC.
Other business
Individual presentations to be given on an ad hoc basis as appropriate to the discussion. Refer to
end of agenda for a list of expected presentations.
Presentations listed in alphabetical order by author’s last name (this list is tentative and subject to
change)
Ernst Peebles - Trophic studies for the central Florida west coast
John Walsh - Modeling of a clupeid dominated system (post 2005 red tide) for west Florida shelf
6
List of Attendees
Ecosystem SSC members:
Joseph Powers, Chairman
Behzad Mahmoudi, Vice-Chairman
Vernon Asper
Felicia Coleman
Stephen Holiman
13
- Louisiana State University
- Florida FWRI
- University of Southern Mississippi
- Florida State University
- NMFS/SERO
James Simons
Carl Walters
- Texas Parks and Wildlife Dept.
- Mote Marine Laboratory
Invited Participants:
Jerry Ault
Joan Browder
Villy Christensen
George Guillen
Rick Hart
Tom Minello
Jim Nance
Ernst Peebles
Doug Vaughan
John Walsh
- RSMAS/University of Miami
- NOAA/NMFS/SEFSC - Miami Lab
- University of British Columbia
- University of Houston Clear Lake
- NOAA/NMFS/SEFSC - Galveston Laboratory
- NOAA/NMFS/SEFSC - Galveston Laboratory
- NOAA/NMFS/SEFSC - Galveston Laboratory
- University of South Florida
- NOAA/NMFS/SEFSC - Beaufort Laboratory
- University of South Florida
Council Staff:
Steven Atran
Karen Hoak
- technical staff - Tampa
- administrative staff - Tampa
Additional Attendees:
Alex Barth
Zy Biesinger
Dave Chagaris
Bonnie Coggins
Raychelle Daniel
Matt Lauretta
Jason Lenes
Mark Rogers
Brooke Shipley
Glen Sutton
Jim Thorson
Robert C. Wakeford
7
- University of South Florida
- graduate student - University of Florida
- University of South Florida
- graduate student - Virginia Tech
- Ocean Conservancy
- graduate student - University of Florida
- University of South Florida
- graduate student - University of Florida
- graduate student - Louisiana State University
- Texas Parks and Wildlife Department
- graduate student - Virginia Tech
- MRAG Americas
List of Preparers
All of the persons listed in the list of attendees contributed to the workshop, and therefore
contributed to the preparation of this report. However, the primary editor for writing and
compiling the report was:
Carl Walters, University of British Columbia
Additional writing, editing and formatting was provided by the following:
14
Behzad Mahmoudi, Florida Fish and Wildlife Research Institute
Steven Atran, Gulf of Mexico Fishery Management Council
The appendices contain reports written by various participants in the workshop describing
presentations given by them. Credit for the authors is given ion each appendix.
15
8
References
Chesney, E., and Baltz, D. 2001. The effects of hypoxia on the northern Gulf of Mexico coastal
ecosystem: a fisheries perspective. In Coastal hypoxia: consequences for living resources and
ecosystems, Coastal and Estuary Studies p. 321-354. Am. Geophysical Union.
Scavia, D., Rabalais, N., Turner, R.E., Justic, D., and Wiseman, W.J. 2003. Predicting the
response of Gulf of Mexico hypoxia to variations in Mississippi River nitrogen load. Limnol.
Oceanogr. 48:951-956.
Holling, C.S. (Ed.) 1978. Adaptive environmental assessment and management. The Blackburn
Press (http://blackburnpress.stores.yahoo.net/noname1.html).
Vaughan, S.D., Shertzer, K.W., and Smith, J.W. 2006. Gulf menhaden (Brevoortia patronus) in
the U.S. Gulf of Mexico: Fishery characteristics and biological reference points for management.
Fisheries Research 83(2007) 263-275.
Stock status of the gulf menhaden Brevoortia patronus off the U.S. Gulf of Mexico Coast.
Prepared for Gulf States Marine Fisheries Commission, Gulf menhaden advisory committee,
March 15, 2004.
Walters, C. 1986. Adaptive management of renewable resources. The Blackburn Press
(http://blackburnpress.stores.yahoo.net/admanofrenre.html).
Walters, C., Martell, S., and Mahmoudi, B. 2006. An Ecosim model for exploring ecosystem
management options for the Gulf of Mexico: implications of including multistanza life history
models for policy predictions. Ms submitted, Mote Symposium 6, Sarasota FL.
16
9
Appendix 1 - Modeling a clupeid dominated ecosystem on the West Florida shelf,
subjected to both red tide and other piscivore stresses, by John J. Walsh, College of
Marine Science, University of South Florida
Independent data from the Gulf of Mexico were used to develop and test the hypothesis that the
same sequence of physical and ecological events each year allows the ichthyotoxic dinoflagellate
Karenia brevis to dominate the coastal phytoplankton community(diatoms, diazotrophs,
microflagellates, non-toxic dinoflagellates, K. brevis) in a series of three-dimensional coupled
circulation/plankton dynamics models (Walsh et al., 2006). A phosphorus-rich nutrient supply
initiates phytoplankton succession, once deposition events of Saharan iron-rich dust allow
Trichodesmium blooms to utilize ubiquitous dissolved nitrogen gas within otherwise nitrogenpoor sea water. They and the co-occurring K. brevis are positioned within the bottom Ekman
layers, as a consequence of their similar diel vertical migration patterns on the middle shelf.
Upon onshore upwelling of these near-bottom seed populations to CDOM-rich surface waters of
coastal regions, light-inhibition of the small red tide of ~1 ug chl l-1 of ichthytoxic K. brevis is
alleviated. Thence, dead fish serve as a supplementary nutrient source, yielding large, selfshaded red tides of ~10 ug chl l-1.The source of phosphorus is mainly of fossil origin off west
Florida, where past nutrient additions from the eutrophied Lake Okeechobee had minimal
impact. In contrast, the P-sources are of mainly anthropogenic origin off Texas, since both the
nutrient loadings of Mississippi River and the spatial extent of the downstream red tides have
increased over the last 100 years. During the past century and particularly within the last decade,
previously cryptic Karenia spp. have caused toxic red tides in similar coastal habitats of other
western boundary currents off Japan, China, New Zealand, Australia, and South Africa,
downstream of the Gobi, Simpson, Great Western, and Kalahari Deserts, in a global response to
both desertification and eutrophication.
Fish kills have been described in the Gulf of Mexico since at least 1529-1534 off Texas, 1792 off
Mexico, and off Florida in 1844. At summer temperatures, half of the particulate biomass of
Florida fish decay to dissolved inorganic pools of ammonium and phosphorus within one day.
Based upon observed stocks of dead fish, such a nutrient source could fuel a large red tide of ~40
ug chl l-1 of K. brevis. Accordingly, the next version of our coupled biophysical model will
include water motion and mixing, light, multiple nutrients and phytoplankton functional groups
to drive one of microzooplanton (protozoans), large zooplankton (copepods), and larval
ichthyoplankton stages of epipelagic nekton (clupeids), harvested by apex predators (serranid
and scombrid piscivores), using two IF statements to specify where and when toxic
dinoflagellates and dead fish are present on the West Florida shelf.
Walsh, J.J., J.K. Jolliff, B.P. Darrow, J.M. Lenes, S.P. Milroy, D. Remsen, D.A. Dieterle, K.L.
Carder, F.R. Chen, G.A. Vargo, R.H. Weisberg, K.A. Fanning, F.E. Muller-Karger, E. Shinn,
K.A. Steidinger, C.A. Heil, J.S. Prospero, T.N. Lee, G.J. Kirkpatrick, T.E. Whitledge, D.A.
Stockwell, C.R. Tomas, T.A. Villareal, A.E. Jochens, and P.S. Bontempi. 2006. Red tides in the
Gulf of Mexico: where, when, and why. J. Geophys. Res. 111, C11003,
doi:10.1029/2004JC002813.
17
10 Appendix 2. A Spatial Ecosystem Model for Atlantic Coast Multispecies Fisheries
Assessments by Jerald S. Ault, University of Miami RSMAS
A spatial dynamic fisheries ecosystem model was adapted from pilot research in the south
Florida coastal ecosystem (Ault et al. 1999, 2003; Wang et al. 2004; Humston et al. 2004) and
applied as an Atlantic coast spatial fishery ecosystem model designed to allow evaluation of how
fishery management actions may impact fishery yields and stock productivity goals for a
particular target species when the species complex is directly coupled to spatial patterns of
fishing intensity, other predator and prey populations, ocean physics, and environmental
changes. In this study we: (1) used and improved the prototype regional (Maine to Florida)
spatial age-structured multispecies fishery ecosystem model; (2) assembled, assimilated and
mapped a broad range of new spatially-explicit databases on processes relevant to the modelbuilding, i.e.,: physical-biological (e.g., ocean currents, water temperatures, salinity, bathymetry,
chlorophyll, river flows, etc.), and population-dynamic (e.g., spatial abundance, biomass and
fisheries catches by age/size for menhaden, bluefish, striped bass, and weakfish; diets;
bioenergetics relationships, spatial fishing intensities by fleets, recruitment, etc.); (3)
parameterized the prototype model for a principal ecosystem trophic connection of two target
species (Atlantic menhaden and bluefish); (4) ran some management scenarios to illustrate model
dynamics and performance capabilities; and, (5) used advanced scientific visualization and GIS
tools to facilitate presentation of results concerning policy alternatives and environmental
impacts to assessment biologists, managers and decision-makers. As such, the model will assist
establishment of quantitative ecological measures of fishery management success for the Atlantic
States Marine Fisheries Commission and other regional US fishery management entities.
References
Ault, J.S., J. Luo, S.G. Smith, J.E. Serafy, J.D. Wang, R. Humston, and G.A. Diaz. (1999). A
spatial dynamic multistock production model. Canadian Journal of Fisheries and Aquatic
Sciences 56(S1): 4-25.
Ault, J.S., J. Luo, and J.D. Wang. (2003). A spatial ecosystem model to assess spotted seatrout
population risks from exploitation and environmental changes. Chapter 15, Pages 267-296 in
Biology of Spotted Seatrout. S.A. Bortone (ed.). CRC Press, Boca Raton, Florida.
Humston, R., D.B. Olson, and J.S. Ault. (2004). Behavioral assumptions in models of fish
movement and their influence on population dynamics. Transactions of the American Fisheries
Society 133: 1304-1328.
Wang, J.D., J. Luo, and J.S. Ault. (2003). Flows, salinity, and some implications for larval
transport in south Biscayne Bay, Florida. Bulletin of Marine Science 72(3): 695-723.
11 Appendix 3. Quantitative methods for functional group assignment by Ernst B.
Peebles, University of South Florida, College of Marine Science, and David D.
Chagaris, Florida Fish and Wildlife Conservation Commission, Fish and Wildlife
Research Institute
18
Two methods for assigning fishes to functional groups were reviewed and demonstrated: (1) diet
comparisons using Bray-Curtis similarity and (2) regular equivalence. Both methods use two- or
three-dimensional multidimensional scaling (MDS) plots to help visualize the grouping of
species that have similar trophic ecologies. The Bray-Curtis method measures similarity between
species based entirely on their diets, whereas the regular equivalence method takes into account
both the prey and the predators held in common. In the Bray-Curtis method, degree of diet
similarity can be varied to produce the desired number of functional groups. Clustering
algorithms are then used to delineate the resulting groups on the MDS plot (Clarke and Warwick
2001). In regular equivalence, a cluster optimization routine maximizes similarity within clusters
while minimizing similarity among clusters, thereby providing a statistical rationale for selecting
the total number of functional groups in the EwE model.
Regular equivalence reduces complexity in food webs by designating species that occupy the
same trophic position into “isotrophic” classes based on quantitative dietary data (Luczkovich et
al. 2002). Unlike structural equivalence models, regular equivalence will distribute species into
similar groups according to their ecological roles even if they do not consume the same prey or
are not eaten by the same predators. As such, the method can be used to compare species and
food webs across geographic locations and is capable of handling cycles and omnivory. UCINet
with NetDraw (Borgatti et al. 2002) is an application that helps visualize trophic level, trophic
group and trophic niche according to the regular equivalence approach. Likewise, PRIMER
software (Plymouth Marine Laboratory, Plymouth, UK) facilitates calculation and presentation
of Bray-Curtis diet comparisons. Both applications require diet data in the form prey-typespecific proportions of non-transformed bulk (biovolume or biomass). To produce meaningful
results, both approaches also require that the number of prey types in the diet database exceeds
the desired number of functional groups in the EwE model.
Both methods were demonstrated using the diet database published by Opitz (1996), and both
produced trophically distinctive groupings that agreed well with ecologists’ intuition. These two
methods can be used either to establish preliminary functional groups at the start of EwE model
construction or to identify misclassified species in existing EwE models.
References
Borgatti, S. P., M. G. Everett, and L. C. Freeman. 2002. UCINet for Windows: software for
social network analysis. Analytic Technologies. Harvard.
Clarke, K. R. and R. M Warwick. 2001. Change in Marine Communities: An Approach to
Statistical Analysis and Interpretation, 2nd ed., PRIMER-E Ltd., Plymouth, UK
Luczkovich, J. J. S.P. Borgatti, J.C. Johnson, and M.G. Everett. 2003. Defining and measuring
trophic role similarity in food webs using regular equivalence. Journal of Theoretical Biology
220: 303-321.
Opitz, S. 1996. Trophic Interactions in Caribbean Coral Reefs. International Center for Living
Aquatic Resources (ICLARM Contribution No. 1085), Makati City, Philippines
19
20
12 Appendix 4. Report on a Food Habits of Fishes Bibliography and a Proposal for a
Trophic Database for the Gulf of Mexico by James Simons, Texas Parks and Wildlife
Department
The need for detailed diet composition data for fishes and other functional groups, as widely as
possible and from as many species as possible, has been identified as a need for both theoretical
(Cohen et al, 1991, Link et al., 2005) and trophic fisheries models (Ulanowicz and Kay, 1991,
Walters et al., this report). A bibliography of food habits of estuarine and marine fishes (Simons
et al, unpublished manuscript) identifies 2,250 reports on the food habits of 530 species of fishes,
from 420 published studies. Some of the most frequently cited fishes include those that are of
interest to the GMFMC, including red snapper (33), red drum (50), gray snapper (26), gag (8),
red grouper (12) and goliath grouper (9). Other species of potential interest from a modeling
standpoint for which there are numerous diet studies include the Atlantic croaker (47), pigfish
(15), hardhead catfish (36) and gafftopsail catfish (25). (Numbers in parentheses indicate the
number of referenced studies in the food habits bibliography).
A proposal for an estuarine and marine trophic database for the Gulf of Mexico (Simons et al,
2006) is currently undergoing revision. This database project will extract diet data for fishes,
marine mammals, sea turtles, sea birds, crustaceans, cnidarians and ctenophores into a relational
database. The database will be both temporally and spatially explicit, bi-lingual
(Spanish/English) and web searchable. The database will be extremely useful for trophic
modeling exercises should the GMFMC continue to pursue trophically based models for
ecosystem-based management explorations. At present it is fairly easy to ask the question “What
does red snapper eat?” and find the literature and answer the question. It is much more difficult
to ask and answer the question “What eats baby snapper?”, as that requires a much more
exhaustive search through the literature of all potential predators of baby red snapper. The
proposed database would make such searches trivial.
Cohen, J.E., R.A. Beaver, S.H. Cousins, D.L. DeAngelis, L. Goldwasser, K.L. Heong, R. Holt,
A.J. Kohn, J.H. Lawton, N. Martinez, R. Omalley, L.M. Page, B.C. Patten, S.L. Pimm, G.A.
Polis, M. Rejmanek, T.W. Schoener, K. Schoenly, W.G. Sprules, J.M. Teal, R.E. Ulanowicz,
P.H. Warren, and P. Yodzis. 1993. Improving food webs. Ecology 74: 252-258.
Link, J., W.T. Stockhausen and E.T. Methratta. 2005. Food web theory in marine ecosystems
(pp. 98-113). In: A. Belgrano, U.M. Scharler, J. Dunne and R.E. Ulanowicz (eds). Aquatic food
Webs: an ecosystem approach. Oxford Universtiy Press, New York, NY. 262p.
Simons, J.D., T.J. Shirley, J. Wood. J. Lester, S. Glenn, L. Gonzalez, J. Ditty, J.E. Smith, K.
Withers, and M.E. Vega. 2006. A marine and estuarine trophic database for the Gulf of Mexico:
A proposal. Sixth William R. and Lenore Mote International Symposium, Life History in
Fisheries Ecology and Management, 14-16 November 2006, Sarasota, FL.
Simons, J.D., R.M. Darnell and M.E. Vega Cendejas. A bibliography of studies of food habits of
21
estuarine and marine fishes in the Gulf of Mexico.
Ulanowicz, R.E. and J.J. Kay. 1991. A package for the analysis of ecosystem flow networks.
Environmental Software 6: 131-142.
22
13 Appendix 5 - An Ecosim model for exploring ecosystem management options for the
Gulf of Mexico: implications of including multistanza life history models for policy
predictions by Carl Walters, University of British Columbia, and Behzad Mahmoudi,
Florida Marine Research Institute
[For resubmission to Bull Mar Sci, June 2007]
Carl Walters, Steven J. D. Martell, Villy Christensen
Fisheries Centre, University of B.C.
c.walters@fisheries.ubc.ca
s.martell@fisheries.ubc.ca
Behzad Mahmoudi
Florida Marine Research Institute
St. Petersburg, FL
Behzad.mahmoudi@fwc.state.fl.us
Abstract
An Ecopath-Ecosim ecosystem model under development for coastal areas of the Gulf of Mexico
simulates responses of 63 biomass pools to changes in fisheries and primary productivity. 10
key species are represented by detailed, multi-stanza population dynamics models (31 of the
biomass pools) that attempt to explicitly account for possible changes in recruitment rates due to
changes in bycatch rates and trophic interactions. Over a 1950-2004 historical reference period,
the model shows good simulated agreement with time series patterns estimated from stock
assessment and relative abundance index data for many of the species, and in particular offers an
explanation for apparent nonstationarity in natural mortality rates of menhaden (declining
apparent M over time). It makes one highly counterintuitive policy prediction about impacts of
management efforts aimed at reducing bycatch in the shrimp trawl fishery, namely that bycatch
reduction may cause negative impacts on productivity of several valued species (menhaden
Brevoortia patronus, red drum Sciaenops ocellatus, red snapper Lutjanus campechanus) by
allowing recovery of some benthic predators such as catfishes that have been impacted by
trawling but are also potentially important predators on juveniles of the valued species.
Recognition of this policy implication would have been impossible without explicit, multistanza
representation of juvenile life histories and trophic interactions, since the predicted changes in
predation regimes represent only very small overall biomass fluxes.
Introduction
Fisheries management Councils have been under considerable pressure to take account of
“ecosystem” effects in setting harvest policies, due to concerns ranging from impacts of bycatch
and habitat damage effects by some fishing activities to impacts of fishing on capabilities of
stocks to support other valued species. When harvest controls have been based only on reference
points from single species assessments, even including bycatch mortality effects, it is not that
23
such assessments ignore ecological interactions entirely; rather, typical single-species models
make very particular assumptions about how natural mortality and recruitment rates somehow
remain stable despite changes in ecological circumstances (e.g., changes in predation risk and
food availability). Discomfort about these very restrictive assumptions has led to investment in
development of models that account explicitly for at least some major trophic interaction effects.
As part of an evaluation of ecosystem modeling tools for comparing fisheries management
options in the Gulf of Mexico, the Ecosystem Scientific Committee of the Gulf of Mexico
Fisheries Management Council requested development of a demonstration Ecosim model using
the widely available Ecopath with Ecosim software (www.ecopath.org). This software facilitates
management of basic biomass and trophic interaction (food habits) data for whole ecosystems,
and checks for consistency of the data through the Ecopath “mass balance” process where
estimated total mortality rates for biomass components are checked against estimated total
predation and fishing loss rates calculated from predator abundances, diet compositions, and
historical fisheries data. Once a plausible (or at least physically possible without requiring
spontaneous creation of mass to satisfy input predator food demands and fishery removal rates)
mass-balanced biomass and trophic flow pattern has been found, the resulting ecosystem
state/flow estimate can be passed to Ecosim, a dynamic model that simulates temporal responses
to policy changes such as fleet reductions, along with impacts of changes in factors such as
nutrient loading and primary productivity.
The Florida Marine Research Institute (FMRI) had already contracted for development of an
Ecopath/Ecosim model of the West Florida Shelf (Mackinson, et al. 2001; Okey and Mahmoudi
2002; Okey et al. 2004), and that model had in turn been modified extensively to simulate effects
of changing nutrient loading and fishing effort on the fishes of Tampa Bay and other FIM study
areas. The Tampa Bay analysis demonstrated surprisingly good ability of Ecosim to fit historical
time series of abundances of a wide variety of fish species, and encouraged us to use that model
as a starting point for wider analysis of the Gulf of Mexico coastal ecosystem as a whole.
Starting with the Tampa Bay biomasses, feeding ecology (consumption rates, diet compositions),
and historical fishing patterns, we extended the model to GOM-scale by correcting biomasses to
averages over the larger area, adding a variety of species (such as red snapper, menhaden) that
are not abundant in Tampa Bay, and developing a model testing dataset that includes historical
fishing effort patterns for major fleets, estimates of abundance over time from stock assessments,
historical catches, and some historical information on changes in total mortality rates (Z). In
adding additional species and fisheries, we aimed to account for at least 90% of the total coastal
(excluding tuna) fish and invertebrate harvest for the GOM, and to account as well for bycatch
patterns and impacts for particular fisheries (especially shrimp trawling) that have been
suggested to have large impact on other fish stocks.
We view the development of a complex model like the current version of the GOM Ecosim
model as an ongoing process. That process will be facilitated in terms of including new data and
information from a variety of researchers using the EwE data management software. But the
really critical need is to continually challenge the model harshly by comparisons of its
predictions to historical and spatial data, and by demanding new policy predictions from it that
expose (through unrealistic predictions) weaknesses in the data and model structure. Below we
24
describe just the June 2006-version of the model, and compare its predictions to historical
relative abundance and stock assessment data for a variety of species for the reference period
1950-2004. We show that for one very particular fishery policy, regulation of bycatch by the
shrimp fishery, the ecosystem model does indeed make very different and disturbing predictions
from those obtained using single-species assessment models, and these predictions arise from the
particular capability of Ecosim to represent changes in predation mortality rates of pre-recruit
juvenile fishes.
Model Structure and Assumptions
The current model simulates biomass dynamics of 63 biomass “pools” (Table 1), ranging from
detritus and phytoplankton at the bottom of the food web to large piscivores like red snapper,
mackerel (Scomberomorus spp), and coastal sharks at the top of the web (no marine mammals
are included in the current model version). Overviews of how Ecosim represents dynamic
change in these pools over time can be found in Walters and Martell (2004) and Christensen and
Walters (2004). Dynamics are simulated on a monthly time step, using two alternative model
structures as described in the following subsections.
In this review we do not attempt to present all equations and parameter values used in the model;
all parameters used in the model are stored in an Access database that is freely available for
download at the following URL:
http://www.ecopath.org/index.php?name=Models&sub=model&modelID=128. The easiest way
to examine the data, test effects of alternative parameter estimates, and reproduce the results
described below is to use the Ecopath/Ecosim version 5 software, again freely available at
www.ecopath.org.
Biomass dynamics for aggregated biomass pools
First, some pools are represented only by total biomass per unit area, and are simulated by
solving differential equations for biomass rate of change of the form
(1)
dBi/dt=eQi(t)-Zi(t)Bi,
where Bi is biomass of pool i, e is a food conversion efficiency, Qi(t) is total food consumption
rate by the pool, and Zi(t) is instantaneous total mortality rate for the pool due to unexplained
causes plus predation plus fishing. Qi(t) is calculated as a sum of consumption rates of various
prey types, with preferences defined initially by Ecopath diet composition inputs. The
components of Qi(t) also form components of the total mortality rates Zj(t) of the prey types
eaten by each pool. Total mortality rate at any time is represented by the sum
(2)
Zi(t)=Moi+∑jQij(t)/Bi(t)+∑kqkiEk(t)
where Moi is an unexplained natural mortality rate, predation rates Qij(t) represent total
consumption rates of pool i by pool j predators, and fishing mortality rates qkiEk(t) by fishing
fleets k (including landed catches, bycatches, and dead discards) are represented as varying with
time-dependent fishing efforts Ek(t). (k=1…18 in the current GOM model, Table 2). Efforts are
25
scaled so that Ek(0)=1, i.e. are scaled to 1.0 at Ecopath base conditions, allowing estimation of
“catchabilities” qki as qki=Cki(0)/Bi(0) where Cki(0) is an Ecopath base catch of species i entered
for each fishing effort k. The qki can also be made density dependent (on Bi(t)) through a
function of the form qki(t)=qmax/(1+kBi(t)) where k is chosen so as to make qki equal the Ecopath
base value when Bi(t)=Bi(0); this option is particularly important for correctly predicting
historical catches of menhaden, for which catchability is obviously density-dependent (Vaughan,
et al. 2000).
Each component consumption/predation rate Qij(t) of prey type i by pool j is predicted with the
“foraging arena” consumption rate equation
(3)
Qij(t)=vijaijTi(t)Ti(t)Bi(t)Bj(t)/(2vij+aijTj(t)Bj(t)).
This rate equation can optionally also be modified to represent prey switching (change in aij)
effects, handling time per prey, and reduction in foraging time in direct response to increases in
predation risk; we have not used any of those advanced options in the GOM model to date. In the
basic foraging arena consumption equation, the rate constant vij represents “flow” of prey from
behaviorially (or locationally) invulnerable to vulnerable states (which can limit predation rates
to be far lower than would be predicted from simpler mass-action models, and create strong
“apparent competition” among predators through the denominator term of the equation), aij
represents rate of effective search or search efficiency for predator i on prey type j, and Ti(t)
represents the relative amount of time that individuals of type i (and j) spend foraging and hence
at risk to predation. Temporal adjustments in Ti(t) occur in the simulations when percapita
feeding rates (Q/B) drop below Ecopath input values; such adjustments lead to (1) density
dependence in predation risk, since animals spend less time at risk to predation when there are
fewer competitors, and (2) type II functional response forms (animals try to keep total food
intake constant, by increasing Ti, when any or all prey types become less for more abundant).
The aij parameters can be calculated from Ecopath diet input data, while the vij parameters must
be estimated for each pool by statistical fitting of the dynamics to time series data (low vij values
imply “bottom up control” of production and mortality, and strong compensatory responses of
each population to reductions in its own abundance; in time series settings, we look for two types
of effects indicating low vulnerability—lack of response of prey mortality rates to changes in
predator abundance, and strong predator compensatory responses, analogous to steep stockrecruitment curves).
Biomass pools represented using the differential equation approach (arranged in no particular
order of trophic position or importance to system function) are indicated by italicized pool names
in Table 1. Note that this list includes some important harvested species (shrimp, crabs, sharks,
pompano) along with a variety of “forage base” species and primay producers.
Age-structured dynamics for species with complex trophic ontogeny and size-age
dependent fishery impacts
The second basic simulation approach in Ecosim is to simulate monthly changes in numbers and
relative body weights of monthy age cohorts of species that undergo complex trophic and
fisheries impact ontogeny. For any species, Ecosim allows to split the species into an arbitrary
26
number of age (months) “stanzas” and to treat prey preferences and vulnerability to various
predators (and fisheries) as constant over the months of age included within each stanza. Stanza
age breaks can represent both ontogenetic shifts in habitat and diet, and also changes in
vulnerability to bycatch and retention fisheries. We used as many as five stanzas (Table 1, red
drum and snook Centropomus undecimalis) to represent species with very complex ontogeny and
fisheries selection patterns, and others with just two stanzas, e.g. menhaden. In one case (red
snapper), stanza breaks were chosen specifically to represent a particular age range (6-24 mo) for
which vulnerability to bycatch impact is known to be highest.
Monthly (age a) changes in numbers (Nia(t)) and relative weights (wia(t)) within each pool i
representing a life history stanza within a species “s” are predicted from the survival-growth
equations
(4)
Nia+1(t+1)=Nia(t)exp(-Zia(t))
Wia+1(t+1)=wia(t)+eiaqnia(t)-mswia(t)
Here, Zia(t) is the total mortality rate (sum of unexplained, predation, and fishing components as
for non-age structured pools), eia is a size and species-specific growth efficiency, qia(t) is relative
food consumption rate of age a individuals predicted from the foraging arena equation while
assuming predation rate proportional to wia(t)2/3, and ms is a metabolic rate for species s again
chosen so as to make growth match a vonBertalanffy curve with metabolic coefficient Ks. The
percapita predation rate qnia(t) is calculated from predicted total consumption Qi(t) as
qnia(t)=Qi(t)wia(t)/∑aNia(t)wia(t); in the foraging arena equation calculation of Qi(t), Bi(t) is
replaced by the relative total search area or volume Pi(t)=∑aNia(t)wia(t)2/3. Growth efficiencies
and search rate parameters for the foraging arena equation are calculated from vulnerability
estimates and from initial consumption rates Qi(0) entered in Ecopath for one leading stanza i.
Since a biomass-age pattern (and food consumption-age pattern proportional to w2/3) like Fig. 1
needs to be satisfied once stanza-specific base Z’s have been specified for every stanza,
Ecopath/Ecosim users are only allowed to enter initial biomass Bi(0) and food consumption per
biomass, Qi(0)/B(i(0) for one stanza (biomass pool i) for each multistanza species s. Then B and
QB are calculated for the other stanzas using the relative (per recruit) biomass and food
consumption rates totaled over ages in those stanzas.
Initial numbers entering the first stanza for multistanza species s each month are assumed to be
proportional to total egg production, and egg production is assumed proportional to body weight
minus a weight at maturity ws,mat. That is, Ni,1(t)=ks∑aNia(t)(wia(t)-ws,mat). The effective
fecundity parameter ks is calculated from initial numbers Ni,1(0), and these initial numbers are
calculated in turn from Ecopath input values of biomass for one “leading” stanza for each species
s, along with initial survivorships to age calculated from initial Ecopath input values of Zia(0).
For these calculations, relative body weights wia(0) are set initially to the vonBertalanffy
prediction along with the assumption that weight varies as the cube of length, as wia(0)=(1-eKsa 3
).
For typical species like red drum, a striking pattern appears when biomasses are calculated by
stanza (summing Nia(0)wia(0) over ages a within each stanza) using Ecopath initial survivorship
27
and growth estimates (Fig. 1). Such calculations generally predict very, very low biomasses for
younger stanzas compared to the biomass of older fish. Since predation components of Zi(t) are
calculated from predicted predation rates relative to biomass available, i.e. as M components
Mij(t)=Qij(t)/Bi(t) for each predator j, it only takes very small total predator consumptions Qij(t)
to cause very high M components for juvenile stanzas having low Bi(t). Since the prediction of
Qij(t) is parameterized from Ecopath predator abundance, consumption per biomass, and diet
composition (so Qij(0) is set to Qj(0)DCij where DCij is the Ecopath input proportion of predator
j food consumption that is prey i), it generally only takes a tiny diet fraction DCij for any
abundant predator j to cause a high predicted M for stanza i individuals. Put another way, the
modeled importance of predators to juvenile survival rates is critically sensitive to diet
composition assumptions involving very small diet proportions that are difficult or impossible to
estimate from the field diet data that are typically available. There is a critical research need to
develop methods for estimating predation impact (partitioning juvenile mortality rates) by means
other than predator diet composition data, e.g. by using tagging methods that permit
identification of mortality agent when tagged fish die or are eaten.
When incorporated into an Ecosim model, this population dynamics framework can produce
compensatory responses through three mechanisms. First, increases in abundance of a species
can result in ecosystem-scale reductions in its prey abundances and increases in its predator
abundances, leading to reduced growth rates and higher mortality rates at various stanzas. Such
ecosystem-scale effects are relatively uncommon in models that we have developed. Second,
changes in abundance of a species can lead directly to reduced feeding and body growth rates
through increased competition in restricted foraging areas/times whether or not ecosystem-scale
prey abundances are affected, resulting in reduced fecundity at high abundances. This
mechanism typically involves long time lags and tends to predict population cycles. Third and
most important, foraging time (Ti(t) in foraging arena equation) adjustments to increasing Ni(t)
for juvenile stanzas tend to cause increased predation mortality rates as juvenile numbers
increase. These rate changes lead to predicted overall recruitment relationships of Beverton-Holt
form (Walters and Korman, 1999). However, note that we do not explicitly build such forms
into the model, as is common practice in single-species models and some multispecies models
(e.g. MSFOR, Sparholt, 1995).
Fisheries definitions and estimated impacts
Model development to date has concentrated on ecological data and interactions, and the
complex fishery structure of the Gulf of Mexico has not been represented in great detail (Table
2). At present the model definitions include three basic types of fisheries.
First, there are four major fisheries that are cleanly defined by gear type and target species, and
for which there are usable historical effort data to use in setting relative efforts for model runs
that attempt to represent historical fishing impacts (shrimp trawl, menhaden purse seine, lobster
pot, crab pot). Only one of these, shrimp trawl, has major bycatch impacts on a wide variety of
species as summarized in Table 2 (these bycatch rates were estimated from FAO reviews
(Alverson et al. 1994), along with bycatch estimates used in various assessments (Ortiz et al.
2000).
28
Second, there is a “grab bag” of fisheries, mainly represented as having minor current impact,
defined by broad gear types such recreational line fishing and haul seining. No long time series
of data are available on efforts for these gear types, so they are represented as constant over time
in model fitting tests. They do include a few high-impact fisheries, such as recreational line
fishing for pompano, for which time series of species-specific fisheries impacts are not yet
available for stock assessment modeling.
Third, there is a set of species-specific fisheries for which stock assessment methods like stock
reduction analysis (e.g., Kimura et al. 1984, Walters et al. 2006) have allowed estimation of total
historical fishing mortality rates, summed over all gear types, and for which fisheries
management involves setting species-specific fishing mortality rate and catch regulations. For
these fisheries, we have not attempted to model the allocation of harvest among gear types, and
have opted instead to estimate relative fishing effort over time as Ek(t)=Fi(t)/Fi(0) where the Fi(t)
estimate are from stock assessment models for pool i representing dominant catch for the target
species of the fishery, and Fi(0) is the Ecopath base or reference fishing rate.
In using aggregate fishing mortality rates for several important fish species, we are failing to
represent one of the most serious policy issues facing fisheries management in the Gulf of
Mexico. For the last six fisheries listed in Table 2, there is growing recreational fishing demand,
and recreational fishing mortality is increasing relative to commercial fishing mortality. Since
recreational fishing effort is difficult to manage, regulatory efforts have favored use of minimum
and slot size limits to limit recreational fishing mortality. But such limits result in high and
growing discard rates, with associated discard mortality. In at least one case (groupers), recent
SEDAR assessments indicate that the recreational discard mortality now exceeds combined
landings from all gears (SEDAR 12, 2006); we include such estimated discard mortalities in the
Fk(t) estimates derived from stock assessment data, but have not attempted to predict how they
might grow in future.
Fitting and Comparison of the Model to Historical Time Series Data
Historical data sources
Basic abundance estimates (Ecopath inputs, Table 1) were obtained wherever possible from
Southeast Data, Assessment, and Review (SEDAR, http://www.sefsc.noaa.gov/sedar/) reports
and FMRI stock assessment reports (http://research.myfwc.com/features/default.asp?id=1035),
or by scaling survey catch rates from the Florida Fisheries Independent Monitoring (FIM) and
southeast SEAMAP databases using catchability estimates provided by FMRI researchers. For
some species (e.g., menhaden) we back calculated initial 1950 abundances by assuming catch
C=FB so B=C/F, then using NMFS catch statistics for the Gulf of Mexico to provide C and more
recent stock assessment results to provide estimates of likely F as of 1950.
Estimates of baseline total mortality rates (PB or Z) were obtained wherever possible by
examining age and size composition data (e.g. Vaughan et al. 2000 for menhaden), or predicted
for unfished species from rough assessments of body growth (von Bertalanffy K) and the Pauly
(1979) relationship between growth and M. Ecopath mass balance calculations help
considerably to bound these Z parameters, by comparing calculated baseline total removal rates
(ZiBi) for each pool i to calculated total removal rates due to catches (and discards) and
29
calculated predation mortality rates (eq. 2). Ecotrophic efficiencies (Table 1) summarize the
proportion of assumed Z for each pool accounted for by predation and catches, and for most
pools (with notable exceptions like shrimp) we were able to account for only a relatively small
proportion of the likely total Z.
Estimates of baseline food consumption rates (Table 1 Q/B estimates) were mainly obtained
from bioenergetics information summarized in FishBase (www.fishbase.org), and were typically
assumed to be 5x to 10x the “production” rates Z. Significantly, more complex QB estimates
based on bioenergetics models were not available for any of the species modeled. Again the
Ecopath mass balance procedure helps to bound QB estimates, since severe overestimates lead
(along with diet composition estimates) to food intake demands higher than could be satisfied by
assumed production rates of prey species.
The most difficult Ecopath baseline information to obtain was diet composition, especially for
younger stanzas of key species. Further, as noted above and in Fig. 1, for small juvenile fish it
only takes a tiny (e.g., 0.00001) diet proportion DCij for abundant predators j to cause very high
mortality rates on juveniles. In general we had to guess at diet compositions based on “expert
opinions” of FMRI scientists, unpublished diet reports that often gave only crude compositions
(like “fish”, “shrimp”, “other invertebrates”), and reasonable assumptions based on relative prey
and predator sizes along with general life history information on likely habitat overlap and
encounter patterns. There is a critical need if SEDAR-like processes are to be applied to
ecosystem models for GOM and other regions to develop regional databases of diet composition
data (J. Simons, NC Dept. Env. Nat. Res., Pers. Comm.).
Historical abundance trend and fishing effort/mortality rate data for model fitting and testing
were again extracted wherever possible from SEDAR and FMRI stock assessment reports, and
from FIM and SEAMAP summaries of mean survey catch rates. Historical catch statistics were
extracted from the NMFS catch database (http://www.st.nmfs.gov/st1/commercial/index.html
and .../recreational/index.html). In some cases (groupers, red snapper, lobster, mullet, and red
drum) for which SEDAR and FMRI assessments only covered recent years, we also ran
stochastic stock reduction analyses (SRA) using the software described by Walters et al. (2006)
to backcalculate stock size and fishing mortality rate trends for 1950-2004. Comparisons of
SRA and SEDAR stock trends have generally given similar results. While the absolute
abundance trends from single-species assessments and SRA were deliberately developed to
cover the whole Gulf of Mexico, it is questionable whether any of the relative abundance trend
data from sources like FIM and SeaMap are in fact representative of average coastal trends for
the Gulf of Mexico as a whole.
Fitting procedures
We used a four-step procedure to “calibrate” the model to historical time trend data. The first
step was an overall screening for possibly incorrect input parameters, and the next steps involved
formal model fitting procedures and sensitivity tests for effects of changes in model structure.
First, we examined each species for which there were long term trend data, to insure that (1) the
model gave reasonable fits to historical catch trends, i.e. gave reasonable temporal patterns of the
30
product Fi(t)Bi(t), and (2) gave temporal trends in Bi(t) similar to those estimated from stock
assessments based on catch and relative abundance trend data. This first step led to various
adjustments in both the Ecopath base inputs, particularly Bi(0) and Zi(0), and some of the
vulnerability parameters vij. When vulnerabilities vki of prey k to a predator i are set too low in
Ecosim, the model typically predicts “flatline” dynamics involving very strong compensatory
responses to increases in mortality rates due to fishing, and hence often too little decline in
responses to historical increases in fishing mortality. We manually adjusted the vki for some
stanzas of some species i to eliminate several obvious examples where the Ecosim vki default
values (equal to 2Mki(0)) were set too low.
Second, we used a maximum likelihood procedure to search for better fitting estimates of a
limited number (25) of vulnerability exchange rates vij to which the model likelihood was found
to be most sensitive. The likelihood function for fitting was taken to be (see Walters and Martell
2004 and Walters and Ludwig 1994 for details)
(5)
ln(L)=-∑d(nd/2)ln[∑t(ydt- ŷ t )2]
Here, the index d refers to time series data set (a time series of catches, relative abundances,
direct estimates of total mortality rate Z from age-size composition, relative stock sizes from
stock assessment model outputs), nd is the number of independent observations in the dth data set,
ydt is the observation of type d for year t, and ŷ t is the Ecosim prediction of that observation.
Ecosim predictions are of two types, absolute and relative. For absolute observations (catch,
total mortality rate, total biomass per area), ŷ t is just the Ecosim prediction of the quantity. For
relative observations (all abundance time series used in this analysis), ŷ t =qdBit where Bit is the
Ecosim predicted biomass and qd is the conditional maximum likelihood estimate of the
catchability (units scaling) parameter for the relative series. qd is given simply by the arithmetic
average of the ln(ydt/Bit) values for the series). Note that the likelihood function represented by
eq. 5 is “self weighting” over the data series, since it is derived by integrating over possible
values of the variance of each series d; this convention can be overridden in Ecosim to
deliberately place more or less weight on any data series, but we did not find it necessary to use
that capability to deal with contradictory or uninformative data sets.
Third, we estimated a time series of annual (1950-2004) apparent nutrient loading “anomalies”
for the Gulf of Mexico ecosystem as a whole (analogous to estimating process errors in single
species stock assessment). For each year, modeled primary production rates (phytoplankton,
microalgae, seagrass) were multiplied by an arbitrary time multiplier pt, where pt=1 by default.
The same nonlinear search procedure as used to vary vij values so as to maximize the likelihood
of the time series data (eq. 5) was then allowed to vary the pt values as primary production
“anomalies” assuming a prior variance of 0.1 for each pt. In earlier fitting tests for the Tampa
Bay version of the model, we had found that fitted pt values behaved similarly over time to pt
values calculated from annual nutrient loading estimates for the Bay. In this case, we found for
the Gulf as a whole that shared or ecosystem-scale pt values do not explain much of the time
series variation over multiple species and data types, though some model fitting trials did suggest
a slow decline in overall productivity (about 20%) since 1950, possibly due to loss of coastal
estuarine habitats around the mouth of the Mississippi River. However, other fitting trials, with
31
productivity described by a smooth spline function (rather than annual values) and with fewer
vulnerability parameters being estimated, resulted in a dome-shaped productivity signal with a
peak in the mid-1980 as might be expected from historical patterns of nutrient loading to the
Gulf from the Mississippi (Scavia et al. 2001).
Fourth, we tested effects of major changes in model structure by constructing a much simplified
model with 20 fewer biomass groups, by omitting a few species like snook that do not occur
gulf-wide and by aggregating various species of small fish into overall “forage fish” groups.
This model gave results and fits to data essentially identical to the more complex model, and will
likely be used in future model development and policy exploration.
Finally, to evaluate how sensitive the procedure for fitting vulnerabilities and forcing function
parameters is for evaluating the impact of shrimp trawling we evaluated combinations of fitting
procedures where we fitted vulnerabilities for 0 to 59 consumer pools (i.e. from none to all
consumers estimating one vulnerability parameter per consumer), and where we fitted primary
production anomalies using a spline function with from 0 to 40 splines as well as with annual
(54) data points. The consumers to include in each of the fitting-rounds were selected based on a
sensitivity test, where groups were included in order after how much impact they had on the
residuals. In total, we evaluated 63 parameter combinations for the fitting procedure, and for
each of these we conducted two simulations. In one, we ran the fitted model with the same time
series data, etc, as used elsewhere in this study, and in a second simulation, we performed a stop
to shrimp trawling from 1990 onwards.
Ability of the model to explain historical abundance trends
Despite obvious deficiencies in key input data such as diet compositions, the model is
surpisingly good at reproducing basic trends in relative abundance for most Gulf biomass pools
(Fig. 2). Note that simulated biomass, Z, and catch trajectories (solid lines in Fig. 2) are
calculated forward over time starting with 1950 Ecopath base values, and vary over time only in
response to input time series of fishing efforts and relative primary productivities (pt). That is,
the model is not readjusted or reset at each time step so as to agree with the data, so that if some
functional relationship or rate were badly wrong the equations could easily predict gross
divergence of the predictions from observed trends. In other words, Fig. 2 shows quite a harsh
test of the ability of the model to “stay on course” over long periods of time relative to the data,
given only “known” forcing changes associated with fisheries development and regulation (along
with modest effects of productivity changes). For a few species like catfish, we initialized the
model with lower biomasses representative of recent years (2000); in those cases, the model
predicts rapid biomass increase toward the higher levels that were likely present when fishing
mortality rates were lower in the 1950s (e.g. shrimp trawl bycatch of catfish). Resetting such
species to have higher initial biomasses did not substantially change the model predictions about
recent biomass levels or interaction patterns.
Most of the strong changes in Fig. 2 are associated with direct impacts of fishing. Several longlived fishes have declined considerably due to historical “overfishing” (red drum, red snapper,
groupers, mullet, lobsters). Two species (mullet, mackerel) have exhibited spectacular
recoveries associated with severe fishery reductions (e.g., Florida constitutional amendment in
32
1995 banning gillnet fishing for mullet). Perhaps the most interesting series from an ecosystem
perspective are for menhaden. The age composition data in Vaughan et al. (2000) indicate that Z
for Gulf menhaden has been decreasing over time, despite increases in fishing effort. The
Ecosim model explains this pattern as being due to decreases in the menhaden M over time due
to reductions in some key predator populations, particularly sharks (mainly due to directed shark
fishing, not bycatch of sharks in the menhaden fishery; see SEDAR 11 (2006), and catfish (Arias
felis, Bagre marinus) and red drum (Fig. 3). The reduction in catfish predation is predicted by
Ecosim to have been due to increases in bycatch/discard mortality in the shrimp trawl fishery.
A few species, particularly pinfish (Lagodon rhomboides) have shown large, abrupt declines
followed by rapid recoveries that are not explained by the Ecosim model’s trophic and fishery
interactions. Interestingly, we can explain the abrupt declines by including a pseudo-fishery in
the model named “red tide”, with high effort for this fishery causing high pinfish mortalities and
with “effort” estimated over time from relative cell counts of toxic algae along the southwest
coast of Florida (BEHZAD: ref please). Interestingly, other model predictions change very little
(as indicated by measures of goodness of fit to historical data) when the red tide pseudo-fishery
is turned on or off.
Preliminary Analysis of Critical Policy Issues for Multispecies
Management
Just because we can fit historical data as in Fig. 2-3 does not mean that the Ecosim model will
give correct predictions about impacts of substantial changes in Gulf of Mexico fishery
management. Beyond a few obvious trends due to high exploitation rates, good model fits could
be occurring for the wrong reasons, particularly in relation to issues of trophic support and
bycatch impacts. A much more important challenge for the model is to see whether it gives
credible predictions of response to major policy changes, particularly in relation to policies that
have major implications for trophic support (food supply) and predation regimes. In the
following subsections we describe model predictions for a few diagnostic policy tests related to
widely debated issues in the Gulf of Mexico.
Rebuilding stocks that have been historically overfished
There have been or likely will be severe restrictions of several fisheries for long lived species
that have historically been overfished. All of these species feed (at least as juveniles) on shrimp,
and several may be partly dependent on menhaden. So a basic policy question is whether the
intense shrimp and menhaden fisheries are now appropriating so much of the production of these
species as to reduce the carrying capacity of the system for long-lived species, or at least slow
rebuilding of stocks. To examine what Ecosim has to say about this question, we simply shut
down the fisheries for red drum, red snapper, and groupers as of 2005, and examined simulated
future stock trends.
For all three species (or species complex in grouper case), the model predicts stock recovery at
basically the same rates as predicted by single-species models, to near the unharvested biomass
levels predicted by such models. That is, Ecosim indicates that the shrimp and menhaden
fisheries will not in fact impair rebuilding programs for long-lived species. The model predicts
33
that as biomasses of the overfished species rebuild over time, fish will simply adjust foraging
times and diet compositions so as to feed more on the wide suite of abundant “forage species”
like scaled sardine (Harengula jaguana) and juvenile ladyfish (Elops saurus) that are not as yet
heavily exploited. Further, the model predicts that stock rebuilding will not substantially reduce
productivity of the shrimp and menhaden stocks, since the larger piscivores apparently only
account for a small proportion of the total mortality rate of shrimp and menhaden in the first
place and would not generate much higher M’s even if they were much more abundant.
Development of new fisheries for forage species
There is a possibility of large demand for small forage fish as feed for developing coastal
aquaculture opportunities, for example for bluefin tuna (Thunnus thynnus) and cobia
(Rachycentrodon canadum). We simulated development of a fishery targeting scaled sardine,
bay anchovy (Anchoa mitchelli), and silver perch (Bairdiella chrysoura) (all high biomass
groups, see Table 1), to a scale that would cause fishing mortality rates comparable to the current
menhaden fishery.
Not surprisingly, Ecosim predicts unequivocally that such a fishery would have devastating
impacts on most major piscivores in the system, from red drum to mackerels and red snapper.
Ecosim models have generally shown relatively weak “bottom up” effects of fisheries on
particular forage species compared to “top down” effects of reducing predator abundances on
prey mortality rates (Walters et al. 2005), but a forage fishery with such broad impacts would
leave few options for piscivores to find other food supplies.
Reduction of impacts of the shrimp trawl fishery
The Gulf shrimp trawl fishery has been an obvious target of demands for improved ecosystem
management through reduced bycatches and benthic habitat damage. Bycatch reduction in
particular is a central demand in plans for rebuilding at least one long-lived fish species (red
snapper, see SEDAR 7 recommendations). During SEDAR 7 assessment reviews, the wisdom of
bycatch reduction was questioned by a few reviewers when stock assessment models for red
snapper indicated positive recruitment anomalies over recent decades when shrimp bycatch
impacts have been greatest.
When we shut down the Gulf shrimp fishery in Ecosim, or at least eliminate its bycatch impacts
by creating a new fishery with very low assumed bycatch per unit trawl effort, the outcome is
very surprising (Fig. 4). Juvenile survival rates of several long-lived species (red drum, red
snapper, groupers) actually decline rather than increasing as expected, and there is also a
substantial decline in productivity of the menhaden stock. Modeled negative impacts are
particularly severe for red drum and menhaden. The basic cause of these negative impacts is
very simple: Ecosim indicates that shrimp trawling has had a very large negative impact on
abundances of some benthic predatory fish, particularly the catfishes. When bycatches are
reduced, these species increase several-fold in abundance, and cause high predation mortality on
a variety of juvenile fish (and older menhaden).
34
To evaluate how the predictions about shrimp trawl impact on adult red snapper are influenced
by the model fitting procedure we performed a sensitivity test where we fitted the model to time
series data with from 0 to 59 vulnerabilities estimated (one per consumer) and with from 0 to 40
spline points or 54 annual parameters estimated for system production anomalies. We ran two
simulations for each parameter combination, one of them with shrimp trawling stopped from
1990 onwards. We compared the biomass of adult red snapper in the two series of predictions,
see Figure 5. As the figure illustrates we do find that most simulations indicate that adult red
snapper would increase after a shrimp trawl ban, but only few of the combinations indicate a
substantial increase, while more indicate none or very limited. Indeed, if we sum up the endstates of the simulations, and express the relative change in biomass after a shrimp trawl ban, we
obtain the frequency distribution shown in Figure 6. The median of the frequency distributions is
1.6, indicating that half of the simulations gives an increase in adult red snapper of 1.6 or less.
The simulations thus indeed raise the question if an uncertain, potential increase in red snapper
biomass warrants a closure of the shrimp trawl fisheries.
We initially dismissed the results from the shrimp trawl stop scenario as obviously too extreme.
But on reflection, it warns us that abundances of many species in the current Gulf ecosystem
have developed in the face of massive shrimp trawling, and it is quite possible that some species
have even benefited from the impacts of that trawling. Catfish are particularly abundant in
coastal Florida where inshore trawling has been banned. They are long-lived and as mouth
brooders have very low fecundity, thus are likely to have declined severely in heavily trawled
areas. They certainly eat juvenile fishes when possible, and at least one Ecopath model has
attributed a high proportion of total menhaden natural mortality to them (J. Carlson, NMFS
Panama City, pers. comm.).
Reviews of changes in community structure in shrimp trawl areas since the 1930s (Chesney et al.
2000; Chesney and Baltz, 2001) do not indicate a large decline in catfish abundance, and catfish
tend to be distributed more inshore than juvenile red snapper (Muncy and Wingo, 1998). Thus it
would appear that we can reject the “catfish hypothesis” as a specific reason for failure of
bycatch reduction to result in red snapper recovery. Unfortunately, this rejection is not the point;
what the Ecosim model has warned us is not that catfish per se are important, but that there may
be species that have been impacted by trawl bycatch that will increase greatly so as to have
strong negative impacts on red snapper recovery. There is clearly a need for further, careful
analysis of historical bycatch and relative abundance trend data so as to identify other species
that might play the same role as catfish do in causing the Fig. 4 results.
Reduction of the menhaden fishery
Following from concerns about impact of the Atlantic menhaden fishery on capability of the
menhaden to support predators like striped bass (Morone saxatilis) and about bycatch of sharks
in Gulf menhaden purse seining, one possible ecosystem-scale management policy would be to
reduce or eliminate the menhaden fishery. The notion would be to maximize the value of
menhaden as forage for more valued species like mackerels and drum.
Ecosim runs indicate relatively little impact on other species and fisheries of shutting down the
menhaden fishery. It appears that relatively few species really depend on menhaden, so
35
increasing their abundance moderately by removing the fishery would not really change the total
forage fish “resource” for piscivores substantially. Incidentally, this same argument also holds
for striped mullet, Mugil cephalus, for which a major reason to shut down gill net fisheries in
Florida was concern about its importance as food for predatory fishes; in that case, we cannot see
in any available data (or Ecosim runs) that substantial increases in mullet abundance have
impacted growth or survival of their predators.
Discussion
The policy tests reveal a key point about ecosystem models like the one we are developing for
the Gulf of Mexico: such models cannot be simply classified as right or wrong, adequate or
inadequate in any general way. Instead, the credibility of the model predictions are highly
policy-specific. So long as we stick to simple predictions about single-species regulation, e.g.
rebuilding overfished stocks, the model is just as credible as its single-species counterparts, since
it gives essentially the same stock reconstructions and predictions (both types of models could of
course be badly wrong). It is when we step into a policy issue where species interactions, and
dynamics of poorly studied “bycatch” species (like catfish) are involved, that highly surprising
and less credible predictions begin to appear.
Certainly none should take the predictions above about benefits from shrimp trawling for other
fisheries at face value; these predictions are based on inference chains that begin with untested
assumptions about historical impact of trawling on abundance of benthic predators, move to
further untested assumptions about impact of those predators on juvenile (and adult in case of
menhaden) survival rates of several species based on very weak diet composition data and prey
preference assumptions for the benthic predators, and link both these uncertain effects with other
assumptions about population dynamics responses of long-lived species. The model could go
badly wrong at any point in such a chain, and there could be other such chains that are more
important but that have not yet even been recognized in the model formulation. Further, the
model obviously does not even account for all species in the Gulf of Mexico coastal ecosystem
(nor would it likely be practical to ever do so in terms of obtaining required parameter
estimates), and some currently “rare” species could show even stronger numerical responses to
policy changes like trawl closures than we predict for common and obvious species like catfish.
Simply by refitting the model with different protocols for deciding which vulnerability
parameters to include in the nonlinear estimation procedure, we were able to obtain good fits to
the red snapper data but a prediction that red snapper should recover strongly if the shrimp
fishery were reduced (Fig. 4).
In particular, the most critical predictions of change in juvenile mortality rates with changes in
predator abundance involve cases where abundant predators could be causing high mortality
rates of prey, without those rates being represented as high (or even practically measurable at all)
proportions of the predator diets. For such interactions, there may be no way to practically
determine whether the interaction is actually important other than to do some direct manipulation
of predator abundance, then see if there is some response in net juvenile survival and recruitment
rates. This raises a key point: to test the Ecosim predictions about changes in benthic predator
abundance due to trawling, and impact of benthic predators on other species, the best strategy
may be to develop spatial comparisons (trawled vs untrawled areas) rather than relying upon
36
time series model fitting. Indeed, it may be worthwhile considering the deliberate use of trawl
closed areas (MPAs) as a tool for investigating trawl effects on community structure, totally
independent of the value of such protected areas for marine habitat and biodiversity.
Given that it will likely never be possible to develop a “complete” model for any ecosystem as
large and complex as the Gulf of Mexico, so that the predictions of any practical model will be
highly suspect, should we simply abandon such modeling efforts entirely, in favor of intuitive
management prescriptions (e.g. trawling must be bad since it kills fish, so reduce its impact) or
large-scale management experiments? Here we think the answer is a resounding no. The models
have at least two key values, even when most of their parameters are highly uncertain. First,
they allow us to play “policy games” such as the shrimp trawl closure simulation, and thereby
screen policy options for possible efficacy and possible pitfalls such as a catfish population
explosion. Second, they can be used as the starting point for further, precisely focused research
and modeling on interactions that appear to have potential for producing surprising effects. The
“catfish hypothesis” has clearly indicated that it is be critical to understand more about
population dynamics, diet compositions, and impacts on other fish species of species like catfish
that have been impacted by trawling, before proceeding to implement policies for “obviously”
beneficial reduction of trawl bycatches Armed with the warning that Ecosim has provided, it
will now be relatively easy to do the more focused analysis of catfishes (and other predatory fish
that have been impacted by trawling) to determine whether they are indeed depressed in
abundance, do indeed eat enough of other valued species to be of concern, and will likely
respond in dangerous ways to bycatch reduction. Doubtless we will uncover more such critical
and well-defined uncertainties as we play further with the model, and as we challenge it with
more questions both about its input data, and about its predictions of response to other policy
options.
A really important benefit from the model development to date has been to provide a “straw
man” starting point for pulling together and documenting parameter estimates from a wider
variety of sources than have been accessible to us, vetting those data through careful review
processes like SEDAR, and systematically exploring policy options and sensitivity of predictions
to uncertain data in workshop processes again similar to the SEDAR process. The Gulf of
Mexico Fishery Management Council has supported one preliminary workshop aimed at starting
such an ongoing model development and policy testing process, and there is every reason to
believe that the process will continue.
References
Alverson, D. L., Freeber, M. H., Murawski, S. A., and Pope, J. G. 1994. A Global assessment of fisheries bycatch
and discards. FAO Fisheries Tech. Pap. 339, Rome, FAO 233 pp.
Chesney, E.J. D. M. Baltz and R.G. Thomas. 2000. Louisiana Estuarine and Coastal Fisheries and Habitats:
Perspectives from a Fish's Eye View. Ecological Applications 10, 350- 366.
Chesney, E., and Baltz, D. 2001. The effects of hypoxia on the northern Gulf of Mexico coastal ecosystem: a
fisheries perspective. In Coastal hypoxia: consequences for living resources and ecosystems, Coastal and
Estuary Studies p. 321-354. Am. Geophysical Union.
Christensen, V. and Walters, C.J. 2004. Ecopath with Ecosim: methods, capabilities and limitations. Ecol. Model.
172(2-4): 109-139.
Kimura, D. K., J. W. Balsiger, and D. H. Ito. 1984. Generalized stock reduction analysis. Can. J. Fish. Aquat. Sci.
37
41:1325-1333.Mackinson, S., Okey, T.A., and Mahmoudi, B. (eds). 2001. A preliminary model of the west
Florida shelf food web for use in ecosystem-based fisheries management and research. Florida Fish and
Wildlife Conservation Commission, Florida Marine Research Institute, St. Petersburg, FL.
Mahmoudi, B. 2005. The 2005 update of the stock assessment for striped mullet, Mugil cephalus, in Florida. Florida
Fish and Wildlife Research Institute, St. Petersburg, online
http://research.myfwc.com/features/view_article.asp?id=26636.
Muller, R., and Taylor, R. 2005. The 2006 stock assessment update of common snook, Centropomus undecimalis.
Florida Fish and Wildlife Research Institute, St. Petersburg, In-House Report IHR 2006-003.
Muncy, R.J. and W.M. Wingo. 1983. Sea catfish and gafftopsail catfish. Species profiles: life histories and
environmental requirements of coastal fishes and invertebrates (Gulf of Mexico). Fish and wildlife Service
TR EL-82-4.
Murphy, M. D. 2005. An assessment of red drum, Sciaenops ocellatus, in Florida: status of the stocks through 2003.
Report to the Florida Fish and Wildlife Conservation Commission, Division of Marine Fisheries dated 5
May 2005.
Okey, T.A., Vargo, G.A., Mackinson, S., Vasconcellos, M., Mahmoudi, B., and Meyer, C.A. 2004. Simulating
community effects of sea floor shading by plankton blooms over the West Florida shelf. Ecological
Modelling 172(2-4): 339-359.
Okey, T.A. and Mahmoudi, B. (Editors). 2002. An ecosystem model of the West Florida Shelf for use in fisheries
management and ecological research: Volume II. Model construction. Fish and Wildlife Conservation
Commission, Florida Marine Research Institute, St. Petersburg.
Ortiz, M., Legault, C.M., and Erhardt, N.M. 2000. An alternative method for estimating bycatch from the U.S.
shrimp trawl fishery in the Gulf of Mexico. Fish. Bull. 98:583-599.
Pauly, D. 1979. On the inter-relationships between natural mortality, growth parameters and mean environmental
temperature in 175 fish stocks. J. Cons. Cons. Int. Explor. Mer 39:175–192.
Porch, C.E. 2000. Status of the red drum stocks of the Gulf of Mexico, Version 2.1. Southeast Fisheries Science
Center, Miami Laboratory, Sustainable Fisheries Division Contribution: SFD-99/00-85. 67 p.
Scavia, D., Rabalais, N., Turner, R.E., Justic, D., and Wiseman, W.J. 2003. Predicting the response of Gulf of
Mexico hypoxia to variations in Mississippi River nitrogen load. Limnol. Oceanogr. 48:951-956.
SEDAR 5. 2005. Atlantic and Gulf of Mexico king Mackarel, complete stock assessment report. Available online at
http://www.sefsc.noaa.gov/sedar/.
SEDAR 7. 2005. Assessment summary report, Gulf of Mexico red snapper. Available online at
http://www.sefsc.noaa.gov/sedar/.
SEDAR 8. 2005. Stock assessment summary report for Southeast United States spiny lobster. Large coastal shark
complex assessment workshop report. Available online at http://www.sefsc.noaa.gov/sedar/.
SEDAR 11. 2006. Large coastal shark complex assessment workshop report. Available online at
http://www.sefsc.noaa.gov/sedar/.
SEDAR 12. 2006. Gulf of Mexico red grouper assessment. Availabe online at
http://www.sefsc.noaa.gov/sedar/Sedar_Workshops.jsp?WorkshopNum=12.
Sparholt, H., 1995. Using the MSVPA/MSFOR model to estimate the right-hand side of the Ricker curve for Baltic
cod. ICES J. Mar. Sci., 52:819-826
Vaughan, D.S., Smith, J.W., and Prager, M.H. 2000. Population characteristics of Gulf menhaden, Brevoortia
tyrannus. NOAA Tech. Rep. NMFS 149. 19p.
Walters, C.J., and D. Ludwig. 1994. Calculation of Bayes posterior probability distributions for key population
parameters. Can. J. Fish. Aquat. Sci. 51:713-722.
Walters, C.J. and J. Korman. 1999. Revisiting the Beverton-Holt recruitment model from a life history and
multispecies perspective. Rev. Fish Biol. Fisheries 9:187-202.
Walters, C.J., and Martell, S.J.D. 2004. Fisheries ecology and management. Princeton Univ. Press, Princeton N.J.
Walters, C., Christensen, V., Martell, S., Kitchell, J. 2005. Possible ecosystem impacts of applying MSY policies
from single species assessment. ICES J. Mar. Sci. 62:558-568.
Walters, C.J., Martell, S.J., and Korman, J.2006. A stochastic approach to stock reduction analysis. Can. J. Fish.
Aquat. Sci. 63:212-223.
38
N
w
B
1
0.5
0.01
0.005
0
Biomass
Numbers, body weight
Figures
0
0
50
100
Age (months)
Figure 1. Typical changes in numbers, body weight, and biomass with age for a multistanza
population in Ecosim. In this example, vonBertalanffy K=0.2 year-1 and the three stanzas
marked by dotted lines have the following total mortality rates Z: 2.0 year-1 for age 1-6
months, 1.0 year-1 for age 7-18 months, and 0.5 year-1 for ages 19+ months. Note that
total biomass, measured by area under biomass (B) curve, is very low for the 1-6 month
early juvenile stanza compared to the older stanzas. Also note extensive use of Florida
fishery independent monitoring (FIM) data in model development to date, rather than
longer term SEAMAP relative abundance series likely available for many species for
broader areas of the Gulf of Mexico.
39
Figure 2. Predicted (solid lines) changes in a variety of biomass (B), harvest (C), and total
mortality rate (Z) indices for the Gulf of Mexico Ecosim model, compared to time series
data (dots). All abundance time series treated as relative abundances (scaled to same
average as model predictions). X-axis for each graph is time, 1950-2004.
40
8
Model biomass
7
Assessment Egg trend
Biomass
6
5
4
3
2
1
0
1950
1960
6
1970
1980
1990
2000
1980
1990
2000
Model catch
Observed Catch
5
Catch
4
3
2
1
0
1950
Instantaneous Mortality rate
5
4
1960
1970
Model Z
Z static catch curves
Z dynamic catch curves
Model F
Model total predation
3
2
1
0
1950
1960
1970
1980
1990
2000
Figure 3. Predicted versus observed indicators of model performance for the Gulf of Mexico
menhaden stock. Note how growing fishery impact as measured by fishing mortality rate
F is predicted to have been accompanied by declining predation mortality rates, mainly
due to reduction in abundance of coastal sharks.
41
.
1.5
Historical assessment (SRA) biomass
Fitted ecosystem model biomass
Relative biomass (t/km2)
Alternate ecosystem model Fit
Alt. Fit, shrimp trawl off 1990
Fitted, shrimp trawl off 1990
1
0.5
0
1950
1960
1970
1980
1990
2000
Year
Figure 4. Predicted versus observed trends in red snapper adult abundance, under historical
fishery impacts and following simulated closure of the shrimp trawl fishery in 1990.
Divergent predictions of response to the trawl closure resulted from using different
protocols for choosing which predation vulnerability parameters to include in fitting the
model to historical data
42
Figure 5. Predicted response in biomass of adult red snapper following a simulated closure to
shrimp trawling from 1990 onwards as a function of number of vulnerability parameters
and of number of spline points included in the fitting. Line indicates model fit with
shrimp trawling allowed, and dots indicate trend with no shrimp trawling from 1990.
43
Figure 6. Frequency distribution for adult red snapper biomass expressed as biomass at the end
of the simulations with shrimp trawls banned relative to biomass for simulations with
continued shrimp trawling after 1990. The vertical, dotted line indicates the median
simulation with a relative biomass increase to 1.6 after trawl ban compared to the
biomass with shrimp trawling continued.
44
Tables
Table 1. Biomass pools included in Gulf of Mexico Ecosim model, along with Ecopath base
input estimates of biomass, production per biomass (or total mortality rateZ for
multistanza population components), total food consumption per biomass, and ecotrophic
efficiency defined as proportion of Z explained by modeled predation, and total catches
(landings plus bycatch and discards). Pools indicated by italicized names are represented
by differential equation model, others as stanzas within single-species age-structured
models. Note that low ecotrophic efficiencies for some groups do not imply lack of
predation mortality, but only that mortality is not explicitly explained by modeled
predator groups or fisheries.
Group name
Biomass
(t/km²)
Cons./
biom. (yr-1)
Ecotrophic
efficiency
0.000217
0.0185
0.2272
0.0984
0.02
Prod./
biom.
(Z, yr-1)
5
2
0.9
0.62
0.6
0-12 Snook
3-12 Snook
12-48 Snook
48-90 Snook
90+ Snook
25.5123
6.268
2.3628
1.4982
1.3
0.5008
0.008
0.0978
0.5751
0
0-3 Red Drum
3-8 Red Drum
8-18 Red Drum
18-36 Red Drum
36+ Red Drum
0.000181
0.00493
0.0323
0.1284
2
2
3.5
1.1
0.6
0.15
18.7423
6.699
2.9886
1.7166
0.95
0.5024
0.0335
0
0.1713
0.0002
0-3 Sea Trout
3-18 Sea Trout
18+ Sea Trout
0-3 Sand Trout
3-12 Sand Trout
12+ Sand Trout
0-6 Mullet
6-18 Mullet
18+ Mullet
Mackrel 0-3
Mackrel 3+
Ladyfish 0-10
Ladyfish 10+
Grouper 0
Grouper 1-3
Grouper 3+
Jacks
Bay Anchovy
Pin Fish
Spot
Silver Perch
Scaled Sardine
Mojarra
0.000091
0.026
0.22
1.97E-05
0.00252
0.1
0.0343
0.5224
2.8
3.68E-05
0.25
0.00979
0.089
0.0045
0.0246
0.52
0.2891
1.3653
0.75
0.8
1.7134
11
0.631
6
1.4
0.7
5
1.2
0.7
3
1.8
0.8
4
0.7
2.8
1.6
2
0.6
0.45
0.8
2.53
1.019
1.1
1.4
1.8
1.9
23.1667
4.0109
1.6
37.8657
8.7796
2.7
50.018
18.2253
8
73.1333
6
17.8409
6
33.1643
14.9354
6
2
14
8
12
9
12.106
15
0.7594
0.1337
0.3279
0.225
0.4279
0.2394
0.5123
0.3774
0.5428
0.0393
0.572
0.6864
0.1537
0
0.1357
0.4275
0.9
0.6
0.9507
0.8328
0.9
0.5487
0.8
45
Trend data
sources
Muller and Taylor
(2006)
Porch (2000),
Murphy (2005)
FIM sampling
Mahmoudi (2005)
SEDAR 5 (2004)
Stochastic SRA
FIM
FIM
FIM
FIM
SEAMAP
FIM
Threadfin Herring
0-12 Menhaden
12+ Menhaden
0.08
1.5336
6
1.31
2.3
1.9
12.5
14.5312
6
0.3655
0.5797
0.6973
Menidia
(silverside)
Catfish
Bumper
Caridan Shrimp
Shrimp
0.16
2.3
16
0.8815
0.15
0.15
4.2561
1
0.8
1.2
2.4
2.4
7.6
12
18
19.2
0.9377
0.8545
0.6
0.9233
Stone Crab
Blue Crab
Cyprinodontids
Poecilids
Pigfish
Gobies
Rays
Pompano
Lobster
red snapper 0-6
red snapper 6-24
red snapper older
0.1675
0.2
0.9
0.05
0.2072
0.179
4
0.1
0.025
0.00889
0.1929
0.55
2
2.4
2.5
2.5
0.8
1.5
0.3
1
0.8
3
2
0.6
7
8.5
10
10
4
7.5
1
8
5
61.3643
19.4157
8
0.4
0.584
0.0897
0.4511
0.75
0.75
0.2502
0.7
0.5
0
0.2591
0.1461
Atlantic croaker
LCsharks
0.6
12
1.5
0.08
10
1
0.8576
0.075
Benthic Inverts
Macro Zooplank.
Micro Zoolplank.
Infauna
Attach Microalgae
Sea Grass
Phytoplankton
Detritus
31.7911
10.734
7.6421
20
29.778
175.617
25
100
4.5
22
36
2
25
9.014
182.13
-
22
67
89
10
-
0.8
0.5
0.5
0.1
0.1821
0.0024
0.3237
0.1091
46
FIM
Vaughan et al
(2000)
FIM
FIM
FIM
NMFS catch,
effort stats.
FIM
FIM
FIM
FIM
SEDAR 8
Stochastic SRA,
SEDAR 7
Ortiz et al. (2000)
SEDAR 11
(2006)
Table 2. Fisheries included in initial Gulf of Mexico Ecosim model, and main biomass pools
impacted by each. Average fishing mortality rate estimates for 2000-2004 represent
Ecopath base input fishing rates times estimated relative fishing effort for the recent
period. Note that species-specific fisheries (last several table rows) include F
components due to both recreational and commercial line fishing; fisheries with those
general names are used only to impact species for which species-specific fishing rate
histories have not yet been assessed or entered into the Ecosim framework.
FISHERY
Main species impacted and estimated fishing rates (F) for 2000-2004
Gill/Cast net
Purse Seine
Haul Seine
Rec. Hook Line
Crab Traps
Cast Net
Bait Trawl
Comm. Line
Shrimp Trawl
6-18 mullet (0.21), 18+mullet (0.16), pompano (0.09)
bay anchovy (0.05), menhaden (0.63), sharks (0.002)
none with F>0.01
12-48 snook (0.08), ladyfish 10+ (0.10), catfish (0.06), pompano (0.43)
stone crab (0.21), blue crab (0.58)
mullet effect included in gillnet/castnet above, no others >0.001
all less than 0.01
all included in species-specific fisheries below
grouper 1-3 (0.08), jacks (0.69), pinfish (0.01), spot (0.5), silver perch
(0.88), scaled sardine (0.05), threadfin herring (0.13), menidia (0.31),
catfish (0.67), shrimp (0.8), pigfish (0.14), gobies (0.56), rays (0.08), red
snapper 6-24 (0.52), atlantic croaker (0.5)
lobster (0.4)
sharks (0.02)
48-90 snook (0.18)
18-36 red drum (0.37)
18+ sea trout (0.29), 12+ sand trout (0.21)
grouper 3+ (0.33)
24+ snapper (0.24)
mackarel 3+ (0.14)
Lobser pot
Shark Fishing
snook fishery
red drum fishery
trout fishery
grouper fishery
snapper fishery
mackarel fishery
F:\Ecosystem Modeling Workshop Draft Report.doc
47