A systematic analysis of factors affecting the performance of climatic envelope models

Effective application of species distribution models requires some knowledge concerning the accuracy of model predictions. Yet very few studies have attempted to systematically analyze factors affecting the predictive power of distribution models. This study fills this gap for Climatic Envelope Models, which have been applied extensively for a variety of conservation and management purposes. We hypothesized that model predictions are influenced by properties of the data (both quantity and quality) and distribution properties of the modeled species. Hypotheses concerning the effects of both types of factors were tested by analyzing distribution patterns of 192 species of woody plants in Israel. Analyses were based on Monte Carlo simulations and standard statistical tests. The total number of observations had a strong positive effect on model performance; but on average, 50-75 observations were sufficient to obtain the maximal accuracy. Climatic bias (the degree of sampling bias with respect to climatic conditions) had a significant negative effect on predictive accuracy. Climatic completeness (the degree to which the climatic range occupied by the species is covered by the observations) had a negative effect on model performance - a result contradicting our original hypothesis. Among the species properties, commonness had a positive effect while niche width had a negative one. Niche position with respect to rainfall and temperature was also important in determining the accuracy of model predictions. The overall results are discussed with respect to trade-offs between commission and omission errors and the potential implications of scale dependency.