The ongoing alteration of ecosystems through human activity (IPCC 2007) endangers vitally important resources and leads to the rapid extinction of species world wide. Research is needed to assess the ecological impacts and consequences of this change, but for large parts of the world detailed species inventories are difficult – if not impossible – to obtain. Attempts to protect individual organisms (Gibson et al. 2004) or whole ecosystems (Williams et al. 2003) therefore increasingly depend on ecological modeling techniques. One of the most widely applied methods are species distribution models (see Guisan and Zimmermann 2000).
Species distribution models (SDMs) are useful tools for the analysis of species-environment relationships: They attempt to generate detailed predictions of species distributions by math- ematically linking presence/absence data to a set of evironmental predictors (Guisan and Thuiller 2005, Schröder 2008). As such, SDMs enable researchers to explore various ques- tions in ecology, conservation and evolution. For example, they have been applied to study interspecific competition (Leathwick and Austin 2001), estimate species persistence in bio- logical reserves (Burns et al. 2003), project species distributions in the past (Peterson et al. 2004) or in future climates (Thuiller 2004), predict the likely success of new invasions (Pe- terson 2003, Thuiller 2003) or detect evolutionary processes in the species range dynamics (Peterson et al. 2003).
Particularly when applied to propose adequate conservation strategies, and subsequently convince both conservation planners and policy makers, modeling techniques need to be accurate and reliable. Studies addressing this matter found that the statistic methodology (Thuiller 2003, Segurado and Araujo 2004) and the available data basis (McPherson et al. 2004) interfere with model accuracy and model reliability. Similarly, ecological characteris- tics seem to influence model reliability in at least two ways: First, species traits affect the data available for the modeling process: Boone and Krohn (1999) report that extremelyrare species are harder to model. They showed that the avian fauna of a region with few endemic species, which are unlikely to occur in surveys, will statistically yield better results than the fauna in regions with more such species. Second, species can complicate the statis- tical identification of the species-environment relationship: Brotons et al. (2004) found that species inhabiting a wide variety of habitats may prevent model algorithms from accurately distinguishing between suitable habitats and the surrounding environment.
Venier et al. (1999) argue that distribution models will always perform better for some taxa than for others. Concordantly, recent studies comparing the model performance among var- ious statistical methods found it difficult to build accurate models for all species (Boone and Krohn 2002, Hernandez et al. 2006), regardless of the applied...