A Strategy for Parameter Sensitivity and Uncertainty

5 Parameter uncertainty and sensitivity analysis is especially important for large, complex 6 individual-based models intended to support management decisions. Yet these models are 7 difficult to analyze because they tend to have many parameters and long execution times. 8 We define a three-phase analysis strategy. Phase 1 examines model sensitivity to each pa9 rameter by itself. Phase 2 identifies interactions in model response to a limited number of 10 parameter pairs. Phase 3 examines how robust decision-support results are to parameter 11 uncertainty: several management alternatives are defined and simulated, then the analysis 12 looks at how often the model’s ranking of the alternatives changes as a limited number of 13 important parameters are perturbed. This strategy was applied to inSTREAM, an IBM that 14 simulates effects of river management on trout populations. The analysis found no evidence 15 of extreme sensitivity or “error propagation”; one parameter had effects that were stronger 16 than anticipated but easily explained. Decision-support results of inSTREAM were highly 17 robust to parameter uncertainty. Energetic parameters (for food intake and metabolism) 18 were especially important, a result also found in other sensitivity analyses of large IBMs. 19