A common objective of correlative species distribution model is to be able to project the potential distribution of the target species into a new environmental space. This can be a new geographic space (e.g., invasive species) or projected future conditions.
One should be very careful in interpreting results if extrapolating to areas with conditions that fall outside the range of reference environmental variation. There are several methods to visualize this uncertainty. On this blog I have for example mentioned the multi-environmental similarity tool in Maxent (also implemented in R in amongst others the dismo package and as an addon for GRASS GIS), which allows you to create maps that provides the similarity of each point to a set of reference points (Elith et al. 2010) and thus provide a quick overview of areas with ‘novel’ conditions.
A disadvantage of this and other methods is that they only consider the ranges of the individual predictors, and ignore the correlation structure of the covariates used to build the model. In reality, it is not unlikely that at a given locations values of each univariate factor fall within the original range of values, but the combination of environmental conditions is new. Continue reading