Using evolutionary computation to assess process models.
TURLEY, M.C.* and E.D.FORD
University of Washington, Seattle, WA 98195-1720 USA 1
Many process models based on ecological theory are difficult to assess with conventional optimization. Conventional techniques test a model on one ecological characteristic at a time. New optimization techniques, such as evolutionary computation, permit model assessment on multiple characteristics or criteria simultaneously. Evolutionary Computation (EC) is an optimization algorithm that borrows ideas from evolutionary theory. It combines random searching in the parameter space and exploitation of the best solutions to find the model parameters that best satisfy multiple criteria simultaneously. EC allows for criteria based on continuous, discontinuous, and possibly qualitative ecological characteristics. The flexibility of the parameter search and the multiple criteria assessment on the model work together to test how well a model explains or represents the ecological theory upon which it is based. We have developed an EC software package for multiple criteria assessment of ecological models that is available for general release. We used EC to compare alternative plant competition theories. An experiment was conducted to examine competition in a high-density stand of even-aged marigolds, Tagetes patula, for sunlight. Two competing models were developed to explain the competition over several periods of growth. Both models were assessed for their ability to satisfy multiple criteria based on characteristics of competition and to assess them with less uncertainty. The results showed that one model was better in its ability to satisfy the criteria. Much uncertainty remained in the model as was quantified by the parameter estimation uncertainty and the criteria uncertainty.
This abstract is being presented at: 10:30 AM in session: