Automated parameter optimization for Ecopath ecosystem models

Abstract

Ecopath is mass-balance modeling approach that is widely used for incorporating ecosystem considerations into fisheries science. Up to now, users of Ecopath software who are constructing a model of a given area must carefully adjust input biomass, diets, and other parameters until the Ecopath parameterization is mass-balanced, a slow process leading to non-unique solutions. We present a new computer-automated iterative technique for mass-balancing Ecopath models which has the advantages of (1) reducing the lengthy process of and opportunity for encoding errors of the manual approach; (2) standardizing results for the same set of starting conditions; and (3) allowing exploration of alternative solutions, with consideration of the estimated confidence of each input parameter. Users can select random and/or gradient descent model perturbation of biomass and/or diet parameters, specify an objective (cost) function for optimization of the search, and modify decision logic, including simulated annealing. An objective function is defined to help target mass-balance solutions with minimum change to original input parameters. A Monte Carlo mode allows exploration of sensitivity to different starting conditions and random perturbations. The new procedure is implemented in the current version of the freely available Ecopath with Ecosim software ( http://www.ecopath.org).