Grantee Research Project Results
Final Report: Application of Individual-based Fish Models to Regional Decision-making
EPA Grant Number: R830886Title: Application of Individual-based Fish Models to Regional Decision-making
Investigators: Lamberson, Roland H. , Railsback, Steven F.
Institution: Humboldt State University
EPA Project Officer: Packard, Benjamin H
Project Period: May 1, 2003 through April 30, 2006 (Extended to November 30, 2006)
Project Amount: $418,710
RFA: Developing Regional-Scale Stressor-Response Models for Use in Environmental Decision-making (2002) RFA Text | Recipients Lists
Research Category: Ecological Indicators/Assessment/Restoration , Aquatic Ecosystems
Objective:
Individual-based models (IBMs) simulate stressor effects on populations as emerging from how individual organisms interact with their environment and each other. As stressor-response models, IBMs have many potential advantages: (1) complex, nonlinear interactions among stressors emerge naturally from the model instead of having to be written into the model; (2) population consequences of stressor effects that are known only at the individual level (e.g., from laboratory studies) can be predicted; (3) transient responses to time-varying stressors can be predicted; (4) IBMs are developed from a wide variety of information, not just calibrated to field data; (5) IBMs can be tested and validated in many ways; and (6) IBMs can be very general instead of site-specific, needing only new input data for application to new sites.
Prior to this research project, we developed inSTREAM, the individual-based Stream Trout Research and Environmental Assessment Model. As its name indicates, inSTREAM can support both environmental impact assessment and ecological research. The model’s original purpose was to address, for stream-dwelling trout, one of the most difficult general problems of impact assessment: understanding how alteration of habitat affects populations of animals that actively adapt to habitat change by moving. By its design, inSTREAM can predict how trout populations respond to changes in any of the inputs that drive the model: flow, temperature, turbidity, and channel morphology. inSTREAM can also predict how populations respond to changes in ecological conditions, such as food availability or mortality risk. Because inSTREAM provides a simplified, controllable, and completely observable virtual ecosystem, it is also a useful tool for addressing many basic research questions of ecology.
inSTREAM represents a stream fish population as a collection of virtual individuals living in a virtual stream environment. The environment is represented as a collection of cells, several square meters in area each. Cells have characteristics such as water depth and velocity (which vary with stream flow at a daily time step), temperature and turbidity (also daily inputs), and availability of food and cover for feeding and hiding. Mortality risks (due especially to predators, but also to environmental stressors such as extreme temperature) vary with characteristics of both habitat and trout (e.g., water depth and hiding cover availability; trout length and energy reserves). Food intake and growth depend on trout size, water velocity, food availability, and competition with larger trout. Each simulated day, the virtual trout choose which habitat cell to occupy as a tradeoff between mortality risks and growth. Then the trout experience the resulting growth and mortality and (when conditions are appropriate) spawn. Because of this design, effects of changes in habitat on the fish population emerge from trout physiology and behavior.
Our research objectives were to develop and demonstrate the usefulness of inSTREAM for regional decision-making by: (1) developing watershed-level methods for applying inSTREAM; (2) conducting demonstration applications; and (3) examining uncertainties and sensitivities in assessments using inSTREAM. We also updated the model and released it for public distribution.
Summary/Accomplishments (Outputs/Outcomes):
Model Release
We produced the first full public release of inSTREAM, which is available at our Web Site (below). The release includes the model software, example input files, and extensive documentation. The documentation report (being prepared as a U.S. Forest Service Technical Report) has four major chapters. The Introduction and Overview describes the objectives of inSTREAM, provides guidance for deciding whether it would be useful for a particular study, gives an overview of the model, and establishes terminology and conventions used throughout the model and its documentation. The Model Formulation chapter describes the model in full detail and identifies the literature and data used to support its assumptions, equations, and parameter values. A chapter on model applications then describes how to use inSTREAM; it discusses designing simulation experiments, field data collection methods, calibrating the model, and understanding the model’s sensitivities to parameters, initial conditions, and habitat input. Finally, the Software Guide provides detailed instructions on installing, running, and using the output from inSTREAM’s software.
Sensitivity and Uncertainty Analysis Methods for Individual-based Models
One of our first findings was that conventional methods for analyzing the sensitivity of models to parameter values are not feasible for inSTREAM and other large, management-oriented IBMs. The primary problem is that the high number of parameters and long run times combine to make factorial parameter perturbation analyses computationally infeasible. Yet, understanding parameter uncertainty and its effects is especially important for complex models used to support management decisions. We developed a new analysis process that is computationally feasible, while still addressing important sensitivity analysis objectives. The process has three phases:
- Individual-parameter sensitivity analyses. Each parameter is varied by itself to produce a sensitivity index—a measure of how much key model outputs change as the parameter is varied across a range of feasible values. Parameters with strong versus weak effects on primary model outputs are identified.
- Parameter interaction analysis. A small number of parameters with high sensitivity indices are varied in pairwise factorial experiments to determine how frequently and strongly the model’s sensitivity to one parameter depends on the value of another parameter.
- Analysis of the robustness of management results to parameter uncertainty. When an IBM is applied to a real management issue, several management alternatives can be simulated via alternative sets of input data. This analysis looks at how often the ranking of management alternatives by the model changes as uncertain parameters are varied.
Sensitivity of inSTREAM to Parameter Uncertainty
Our analyses of parameter sensitivity in inSTREAM produced few surprises and no evidence of extreme sensitivity or “error propagation.” As anticipated, the parameters with the strongest effects on inSTREAM’s predicted biomass of adult trout included those controlling food availability and mortality risk due to terrestrial predators (birds, otters, etc.; believed to be the most common cause of adult trout mortality). One unanticipated, but well-understood result was the highest sensitivity index being for a parameter controlling how risk-of-predation varies with water depth (fish are less visible and less accessible to terrestrial predators in deeper water). Many of the highest-sensitivity parameters represent physiological processes and have values from replicated laboratory experiments. Physiological processes found to strongly affect results are: the distance over which trout can detect food items (which limits feeding success); the maximum sustainable swimming speed (which limits the velocity of habitat trout can use and feeding success increases with maximum swimming speed); and respiration (trout growth is modeled as energy intake from feeding minus energy costs of respiration, which increases with swimming speed and temperature).
The model exhibited low sensitivity to most parameters. This result does not mean that inSTREAM includes many parameters that are unnecessary because they have no effect on results. Some of these parameters are necessary to represent one end of a relationship (e.g., for how the risk of being eaten by a larger fish varies with fish length) for which the model is sensitive to the other end. Other parameters represent processes (e.g., effects of extreme temperatures) that are not important at the study site we used, but likely would be important at other sites.
Robustness of Management Assessments to Parameter Uncertainty
The relative rank of management scenarios by inSTREAM was quite robust to parameter uncertainty. We simulated four scenarios for a hypothetical hydropower diversion (assumed to reduce flow) and a timber harvest (assumed to increase turbidity) at our study site. Using standard parameter values, the four scenarios produced average adult trout biomass values of 800 (± a sample standard deviation of 70, over 5 replicate simulations, 1390±60, 1190±160, and 1530±130). Note that the second and third scenarios produced results quite close to each other; the difference is not statistically significant at p = 0.05. Using Latin hypercube sampling to ensure wide distributions of parameter values, we executed 45 simulations of each scenario while perturbing the values of seven high-sensitivity parameters. The raw results correctly ranked all four scenarios in 56% of the parameter perturbations, and correctly identified both the best and worst scenarios in 73% of the parameter perturbations. When results were adjusted for the likelihood of each parameter’s value (e.g., giving less weight to results when parameter values were further from their standard value), the scenarios were ranked correctly in all cases.
Sensitivity of inSTREAM to Site Input
We analyzed how sensitive a primary output of inSTREAM (adult trout biomass) is to site-specific habitat input variables. We manipulated the values of three variables that are input for each habitat cell, using a synthesized study site (see the following section) with drainage area of 2600 ha. Simulated trout biomass was found to be highly sensitive to the availability of hiding cover and feeding cover (velocity shelters used by trout while feeding). However, feeding cover had a strong effect only at relatively low levels; once a threshold of 10% of stream area providing velocity shelter was reached, further increases had no effect. The availability of spawning gravel had no effect on trout biomass; inSTREAM does not represent how gravel quantity or quality affects spawning success in detail. Results of this analysis were incorporated in the model’s documentation as guidance on how important accurate values for these inputs are.
Methods for Synthesizing inSTREAM Input at Regional Scales
A main emphasis of our research was to address a significant challenge in applying detailed, site-specific models such as inSTREAM to regional decision support: reducing the need for field data collection. We looked for ways to apply inSTREAM throughout a watershed or region, capturing how habitat at the scale of several-m2 cells varies with stream size and elevation, but without extensive field data collection. The approach we developed uses several information sources and models: (1) Habitat inventories from the California Department of Fish and Game are used to statistically model how the frequency of major habitat types (pools, riffles, flatwater) varies with stream gradient within a region. (2) Measured channel cross-sections are used to parameterize stochastic models that synthesize cross-section shapes and sizes for each habitat type, as a function of watershed area. (3) A standard one-dimensional flow model (Hydrologic Engineering Center-River Analysis System [HEC-RAS]) is used to simulate depth and velocity in cells on each cross-section, the hydraulic input needed by inSTREAM. (4) Stochastic models fit to observations are used to synthesize other cell habitat variables: cover for hiding and feeding and spawning gravel availability.
Example Application: Regional Response to Temperature, Turbidity, and Flow Alteration
We used the input synthesis methods described above to create 99 sets of input to inSTREAM; these represent sites in the Bull Creek watershed (Humboldt County, CA; a tributary to the South Fork Eel River). The sites range in stream order between one and four, in gradient up to 8%, and in watershed area from approximately 200 to 10,000 ha. Flow, temperature, and turbidity are all higher in lower-elevation, higher-order streams. We simulated stressors typical in California watersheds: flow diversion for irrigation and temperature and turbidity increases resulting from forestry or other land uses that reduce shading and increase erosion. The experiment found simulated trout populations highly sensitive to temperature and turbidity in this watershed, with the strongest effects in the larger, lower-elevation (and naturally warmer) reaches. Diverting half the summer flow from the mainstem near its downstream end was predicted to cause a further strong reduction in trout populations, even without considering how this flow decrease could further increase temperature.
Example Application: Interactions of Hatchery and Wild Salmonids
Introduced species are a widespread ecological stressor. In California and the Pacific Northwest, stocking of hatchery-reared salmon and trout is controversial because these fish, even if belonging to a native species, have behavioral and genetic differences from wild fish and compete (in some situations) with endangered native salmonid populations. We used inSTREAM to address some of the most important, but difficult to study, questions about how hatchery salmonids affect wild populations: How do hatchery-raised fish behave differently from wild fish? How does stocking of hatchery trout affect wild trout? How would these hatchery-wild trout interactions change if hatchery trout were raised to behave more like wild trout? Hatchery trout were simulated as a separate species and given three potential behavioral differences from wild trout: (1) not considering predation risk in selecting their habitat cell, (2) not exploring to find better habitat, and (3) not considering food competition in selecting habitat. Simulation experiments compared hatchery-wild trout interactions to published observations under each combination of these behavioral differences and at several densities of stocked fish. All three behavioral differences contributed to observed interactions (e.g., hatchery fish aggregating in pools and having little effect on wild trout, except at high densities). As we made hatchery trout better adapted to natural conditions (e.g., as if hatcheries were operated to instill awareness of predation risk or food competition), we found stocking to have stronger effects on wild trout: emigration increased and growth and survival decreased.
Example Application: Effects of Flow and Temperature Regime Alteration on an Introduced Species Invasion
As an additional example of the application of inStream to management issues, we applied it to the study of competitive interactions between native Yellowstone cutthroat trout and rainbow trout (introduced a century ago) in the Idaho’s Teton River. Though rainbow trout have been in the Teton for over 100 years, until recently they have existed at very low densities and the native cutthroats have been the dominant trout species. Since the mid 1990s, the rainbows have become dominant and now are about 95% of the trout in the river. inSTREAM was applied to reaches on both the mainstem of the Teton River and to two tributaries, Fox Creek and Teton Creek. Results of these experiments demonstrate that the mostly likely cause for the dramatic increase in the rainbow population and the accompanying cutthroat decline is the alteration of the flow regime in the river, resulting from changing irrigation practices. Over the past decade, changes in the patterns of irrigation withdrawal and ground water recharge have resulted in a flattening of the hydrograph—increases in the low flow periods due to the influx of ground water and a decrease in the peak flow due to withdrawals. This has resulted in substantially greater reproductive success for the rainbow trout and much reduced survival for newly emerged cutthroats who cannot successfully compete with the young rainbows that emerged two months earlier.
Collaborative Studies and Management Applications
We devoted a significant effort to supporting applications of inSTREAM to management issues of several federal agencies. Our primary collaboration was with the U.S. Forest Service’s Pacific Southwest Research Station, Redwood Sciences Laboratory (RSL), on potential forest management and stream restoration effects on salmonids. For example, Bret Harvey of RSL conducted simulation experiments using inSTREAM to examine cumulative effects of habitat effects sometimes attributed to forestry: elevated stream temperature and turbidity, and sedimentation (filling of pools). These experiments tested the assumption that these stressors act independently, so their separate effects can be added to estimate total effects. This assumption was found sufficient at low stress levels, but strongly underestimated cumulative impacts at higher (but still realistic) stress levels.
A second collaboration (eventually funded separately) was with the Forest Service’s Rocky Mountain Research Station (RMRS) and addressed forest fire effects on trout populations. RMRS field studies indicated that trout populations in streams strongly affected by fire and subsequent debris flows were not strongly reduced in abundance but were reduced in age and size distributions. inSTREAM is being used to determine what specific effects of fire (increased temperature, increased food production, reduced hiding cover) explain these observations and determine the relative value of alternative restoration strategies.
Journal Articles on this Report : 2 Displayed | Download in RIS Format
Other project views: | All 27 publications | 5 publications in selected types | All 3 journal articles |
---|
Type | Citation | ||
---|---|---|---|
|
Grimm V, Revilla E, Berger U, Jeltsch F, Mooij WM, Railsback SF, Thulke H-H, Weiner J, Wiegand T, DeAngelis DL. Pattern-oriented modeling of agent-based complex systems: lessons from ecology. Science 2005;310(5750):987-991. |
R830886 (Final) |
|
|
Railsback SF, Lytinen SL, Jackson SK. Agent-based simulation platforms: review and development recommendations. Simulation 2006;82(9):609-623. |
R830886 (Final) |
Exit |
Supplemental Keywords:
media, water, watersheds, risk assessment, ecological effects, sensitive populations, animal, population, cumulative effects, ecosystem protection, regionalization, aquatic, habitat, public policy, decision making, conservation, scientific disciplines, biology, ecology, methods/techniques, modeling, geographic areas, northwest, western, Pacific coast, Pacific Northwest, California, CA,, RFA, Economic, Social, & Behavioral Science Research Program, Scientific Discipline, Water, Ecosystem Protection/Environmental Exposure & Risk, Water & Watershed, Monitoring/Modeling, Regional/Scaling, decision-making, Ecology and Ecosystems, Biology, Watersheds, Economics & Decision Making, risk assessment, ecosystem modeling, aquatic ecosystem, watershed, ecosystem assessment, Bayesian approach, decision analysis, decision making, environmental decision making, ecological variation, TMDL, regional scale impacts, water quality, assessment endpoint mechanistic research, ecological indicators, ecology assessment models, ecosystem stress, watershed assessment, ecological models, fish models, individual based models, stressor response model, decision support tool, environmental risk assessment, water monitoring, adaptive implementation modelingRelevant Websites:
http://www.humboldt.edu/~ecomodel Exit
Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.