Application of multi- and univariate techniques of sensitivity analysis to SKEBUB, a biomass-based fisheries ecosystem model, parameterized to georges bank

Abstract

Abstract Sensitivity analyses were performed on the biomass-based fisheries ecosystem simulation, SKEBUB, developed at the Northwest and Alaska Fisheries Center. This simulation contains an equilibration routine that forces it to an equilibrium state by adjusting the species biomasses. Sensitivity analyses were performed at this equilibrium state. Analyses indicated a multivariate GH'Biplot of Monte Carlo error analysis data to be more useful than a univariate fractional factorial analysis of variance, because the biplot simultaneously analyzes all output variables and accounts for input parameters perturbed over a range of values. The biplot identifies the maximum variability in the simulation, but in this application did not account for variability not described by the first two principal components. Response surfaces of sensitive output variables were generated to study nonlinearities in response. The biomass at equilibrium and the predation/biomass ratio were the most sensitive output variables, and their sensitivities were species-specific. Silver hake was identified as a keystone predator. The consumption/biomass ratio (an indicator of food limitation) was not a sensitive output variable in this simulation. Simulation results were most sensitive to the following input variables: growth, the food requirement for growth and maintenance, the availability to predation, percent contribution to other species diets, and the rate of prey switching by predators. Again the sensitivities of the different species to these parameters were species-specific, with the silver hake dominating factors causing increased predation. In general, variability in output variables was an order of magnitude more than in the input parameters. This increased variability would be reduced if simulation in the sensitivity analyses were required to reach an equilibrium state at least as stable as that of the baseline run. Response surfaces had limited applicability because of the many variables and interactions excluded; however, they did indicate the presence of multiple equilibria in the silver hake and flatfish groups. The particular equilibrium state which is chosen as the baseline for the sensitivity studies is anticipated to influence the outcome of the sensitivity analyses. Sensitivity analyses will thus be specific to both the simulation model and the data input to that model.