More complex distribution models or more representative data? Biodiversity Informatics 5

Abstract. ⎯ Distribution models for species are increasingly used to summarize species ’ geography in conservation analyses. These models use increasingly sophisticated modeling techniques, but often lack detailed examination of the quality of the biological occurrence data on which they are based. I analyze the results of the best comparative study of the performance of different modeling techniques, which used pseudo-absence data selected at random. I provide an example of variation in model accuracy depending on the type of absence information used, showing that good model predictions depend most critically on better biological data. Key words. ⎯ distribution models, model reliability, pseudo-absences, conservation usefulness. Recently, many efforts have focused on creation of models able to predict species ’ distributions from partial data. These distributional models use known distribution records of a species, as well as environmental and spatial explanatory variables, to build statistical functions for interpolating species’ distributions across the environmental spectrum (Guisan & Zimmermann 2000). Models may also extrapolate species ’ distributions to sets of environmental conditions outside those used to build the models (Peterson 2003). The reliability of these predictions depends on many factors, but the three main ones are (1) quality of data used for model calibration, (2) predictive power of the explanatory variables, and (3) the modeling technique chosen to produce predictions from such variables. In spite of recent theoretical opinions on the errors that may result from the first two sources (Soberón & Peterson 2005, Barry & Elith 2006, Araújo & Guisan 2006), little effort has been devoted to testing experimentally the effects of these factors on the reliability of model outputs. Instead, much effort has been devoted to comparisons of the different available modelin