An evaluation of alternative algorithms for fitting species distribution models using logistic regression

摘要: Logistic regression is being used increasingly to develop regional-scale predictive models of species distributions for use in regional conservation planning. These are usually developed using automated stepwise procedures select the explanatory variables include each model and fit functions relating these probability occurrence. Available fitting logistic differ terms a number factors, including basic modelling technique employed (generalised linear or generalised additive modelling), strategy determine complexity fitted functions, approach correct multiple testing. This study evaluates effect that factors has on accuracy models, fauna flora survey data from north-east New South Wales. The results suggest maximised by employing variable selection stringently guard against inclusion extraneous model, such as forwards with 5% significance level removal insignificant at stage process. Models were more accurate than those derived modelling. best controlling was less clear, it tended vary between biological groups examined. Small reptile modelled complex relationships (3 4 df), vascular plants diurnal birds simple (1 2 df). Correction testing Bonferroni correction factor did not improve models.