A sensitivity analysis of an individual-based trout model

Individual-based models (IBMs) are a relatively new way to perform population modeling of a particular species. In this type of modeling, population level dynamics emerge as the result of actions of individuals, as members of the population make choices regarding habitat due to influence by environmental factors as well as competition from other members of the species. There are very few, if any, sensitivity analyses of individual-based ecological models reported in the literature. This is in part due to the recent advent of this type of modeling, as well as the overall complexity of the models themselves. The intention here is to demonstrate a method of analyzing an IBM so that relevant information is provided to the modeler regarding the strong and weak points of the model in the face of the uncertainty in parameter values, while consuming a minimum of time and computational resources. In addition, this type of information will be useful to resource managers looking to support management decisions, as it will provide indications of how much confidence can be put into model output because the effects of parameter uncertainty on alternative management actions will be included in the ranking with regards to the efficacy of each proposed action. In this fashion, the model will aid resource managers to better understand the risk of making false predictions. An individual-based model, inSTREAM v.4.05 (Railsback et al, 2004), was used to model a population of northern California coastal cutthroat trout in Little Jones Creek, a third order tributary of the Smith River in northern California. In this model, trout make daily habitat selection choices to maximize their fitness. The decision rules that govern these choices are influenced by the daily conditions within the modeled stream, various mortality risks, as well as factors concerning respiration, growth and reproduction. The uncertainty in 90 parameters of the model was analyzed. Effects of parameter uncertainty were investigated in a three-phase process. The first phase ranked the parameters based on a sensitivity index which reflected how much the uncertainty in parameter value affected model output. In the second phase, the relative level of higher-order interactions amongst the parameters with the highest sensitivity index was assessed. The third phase tested the robustness of model output, by varying all of the most sensitive parameters simultaneously, when evaluating proposed alterations to the watershed. The model was found highly sensitive to some parameters, moderately sensitive to most, and not sensitive at all to a few. Although varying two high-sensitivity parameters at a time did generally have a non-linear effect on model output, the model was able to reproduce the same population trends under multiple parameter perturbation as it did when only calibrated parameter settings were used.