Parameter sensitivities, Monte Carlo filtering, and model forecasting under uncertainty

Complex models are often used to make predictions of environmental effects over a broad range of temporal and spatial scales. The data necessary to adequately estimate the parameters of these complex models are often not available. Monte Carlo filtering, the process of rejecting sets of model simulations that fail to meet prespecified criteria of model performance, is a useful procedure for objectively establishing parameter values and improving confidence in model predictions. This paper uses a foodweb model to examine the relationship between model sensitivities and Monte Carlo filtering. Results show that Monte Carlo filtering with a behavior definition that is closely related to the sensitivity structure of the model will produce substantial reductions in model forecasting uncertainty.