Improving ecological impact assessment by statistical data synthesis using process-based models

Population dynamic modelling often entails parameterizing quite sophisticated biological and ecological mechanisms. For models of moderate mechanistic complexity, this has traditionally been done in an "ad hoc" manner, with different parameters being estimated independently. The point estimates so obtained are then used for model simulation, perhaps with some further "ad hoc" adjustment based on comparison with any available data on population dynamics. Quantitative assessments of model adequacy and prediction uncertainty are not easily made using this approach. As an alternative, the paper investigates the practical feasibility of fitting a moderately complex population dynamic model directly and simultaneously to all the data available for parameterization of the model, and to all available data on the population dynamics of the target animal. This alternative approach allows us to combine all available quantitative information on the target species, to assess the viability of the model, the mutual consistency of model and different sources of data and to estimate the uncertainties that are associated with model-based predictions. The target organism in this study is the freshwater amphipod "Gammarus pulex" ("L."), which we model using a stage-structured population dynamic model, implemented via a set of delay differential equations describing the basic demography of the population. Target data include population dynamic data from two sites, information on basic physiological relationships and environmental temperature data. Fitting is performed by using a non-linear least squares approach supplemented with a bootstrapping method for avoiding small scale local minima in the least squares objective function. Variance estimation is performed by further bootstrapping. Interest in "Gammarus pulex" population dynamics in this case is primarily related to likely population level responses to chemical stressors, and for this we examine predicted 'recovery times' following exposure to a known toxicant. Copyright 2006 Royal Statistical Society.