Project Title
Ecosystem Diagnosis and Treatment (EDT) sensitivity analysis
Description
The goal of the EDT sensitivity analysis is to identify the best uses of EDT model output in freshwater management and habitat recovery planning. The specific objectives of the analysis are five-fold: (1) to understand how the model works, (2) to understand how variations in habitat input data affect model output, (3) to understand how variations in fish or stream reach input data affect model output, (4) to understand how assumptions that are quantified in the EDT rules and internal trajectory parameters affect model output, and (5) to estimate a confidence interval for EDT outputs that are used in habitat management decision-making.
The project is a multi-agency effort. The following examples describe only a part of the project. Our general approach follows Saltelli et al. (2000a) and Saltelli et al. (2000b).
What is the effect of uncertainty in habitat input parameters (Level IIs) on EDT output?
Step I: Create 1000s of random habitat data input tables (level IIs) with small, medium, and large amounts of variability that maintain an appropriate correlation structure both across habitat attributes and across reaches. We have made good progress in this step using the approach proposed by (Li and Hammond 1975, and Lurie and Goldberg 1998). The definitions of small, medium, and large errors are being developed by the larger collaborative working group.
Step II: Run the model with each of the hundreds of thousands of alternative input sets.
Step III: Explore the output. The output will be a distribution that can be statistically related to the distribution of the inputs. For example, the output can be described as a distribution of possible equilibrium population sizes (Neq) outputs. We can also investigate the impacts of potential habitat error on predicted diversity, capacity, restoration priorities, etc.
Capacity results from a pilot EDT Sensitivity Analysis using 500 Monte Carlo simulations of the on-line scenario builder and holding the trajectories constant between runs. All possible habitat inputs were randomly varied between –0.5 and + 0.5 of the original EDT input value. Note that differences between the historical (template) and current (patient) conditions limited the number and extent of variables that could be modified.
What is the effect of uncertainty in rule curve parameters on EDT output in the Lewis River basin?
Step I: Generate distributions for the rule curves that link Level IIs to Level IIIs (habitat data to fish survival) with small, medium, and large amounts of variability. Again, the definitions of small, medium, and large errors are being developed by the larger collaborative working group.
Step II: Run the model 100,000s of times, each time selecting rule parameters from the distribution for that particular parameter. To examine impacts of particular rules of interest, we can select from a distribution of parameters for each rule independently (using the mean parameter for the rest of the rules) and then for relevant combinations of rules.
Step III: Explore the output. Again, the output will be a distribution that can be statistically related to the distribution of the inputs. A similar approach would be used to estimate the effects of uncertainty in life history parameters, the parameters that affect fish movements, and the selection of monthly patterns.
|
