Aim In this study, we (1) determine whether simple species distribution models based on regional data provide incomplete descriptions of potential distributions; (2) investigate whether underrepresented areas where potential distributions are estimated using only regional data are spatially and environmentally structured; and (3) examine why regional data may not adequately describe potential distributions.
Location Iberian Peninsula.
Methods We used a multidimensional envelope procedure to estimate the potential distributional areas of 73 species of Iberian diving beetles (Dytiscidae) using two data sets (Iberian data and data from the entire range). We used a Mann–Whitney U-test to compare the features (climate, number of database records and proportion of human-transformed land uses) of these underrepresented areas with those of the remaining Iberian territory.
Results By comparing species-richness estimates obtained by overlaying predicted species distributions modelled using either global or regional data, we found that some areas of species’ potential distributions are underrepresented when only regional data are used. Incomplete estimates of potential distributions when using only Iberian data may be partly attributable to limited survey efforts combined with unique local climates, but none of the considered factors by itself seems to fully explain this underrepresentation.
Main conclusions Our results show that species data from regional inventories may provide an incomplete description of the environmental limits of most species, resulting in a biased description of species’ niches. The results of distribution models based on partial information about the environmental niche of a species may be inaccurate. To minimize this error, we highlight the importance of considering all known populations of a given species or at least a sample of populations distributed across the whole range, to include environmental extremes of the distribution. We highlight some methodological and conceptual concerns that should be considered when attempting to infer potential distributions from occurrence data.