Some considerations on the use of ecological models to predict species' geographic distributions

Abstract

Peterson (2001) used Genetic Algorithm for Rule-set Prediction (GARP) models to predict distribution patterns from Breeding Bird Survey (BBS) data. Evaluations of these models should consider inherent limitations of BBS data: (1) BBS methods may not sample species and habitats equally; (2) using BBS data for both model development and testing may overlook poor fit of some models; and (3) BBS data may not provide the desired spatial resolution or capture temporal changes in species distributions. The predictive value of GARP models requires additional study, especially comparisons with distribution patterns from independent data sets. When employed at appropriate temporal and geographic scales, GARP models show considerable promise for conservation biology applications but provide limited inferences concerning processes responsible for the observed patterns.