The use of genetic algorithms and Bayesian classification to model species distributions

It is shown that both climate and land use variables are important for modelling the spatial distribution patterns of the sampled species and the algorithm is tested on an artificial “species” and is shown to perform well.

Abstract

This paper develops a method to model species' spatial distributions from environmental variables. The method is based on a search for an optimal identification of environmental niches to match observed species presence/absence data. The identification is based on Bayesian classification and the optimisation is based on a Genetic Algorithm (GA). The algorithm is tested on an artificial "species" and is shown to perform well. We apply the approach to a random sample of 100 plant species native to the British Isles. This enables an identification of the environmental variables that are most important for capturing the species' spatial distribution. We show that both climate and land use variables are important for modelling the spatial distribution patterns of the sampled species.