An evaluation of methods for modelling species distributions

摘要: Aim  Various statistical techniques have been used to model species probabilities of occurrence in response environmental conditions. This paper provides a comprehensive assessment methods and investigates whether errors predictions are associated specific kinds geographical distributions species. Location  Portugal, Western Europe. Methods  Probabilities for 44 amphibians reptiles Portugal were modelled using seven modelling techniques: Gower metric, Ecological Niche Factor Analysis, classification trees, neural networks, generalized linear models, additive models spatial interpolators. Generalized constructed with without term accounting autocorrelation. Model performance was measured two methods: sensitivity Kappa index. Species grouped according their (area occupancy extent occurrence) (marginality tolerance) distributions. Two-way comparison tests performed detect significant interactions between groups. Results  Interaction groups both indicates that varied different Artificial networks generally better, immediately followed by including covariate Non-parametric preferred parametric approaches, especially when greater area occupancy, larger occurrence, lower marginality higher tolerance. Main conclusions  is first attempt relate Results indicate strong relationship the being modelled. Some but no method superior all circumstances. A suggestion made choice appropriate should be contingent on goals