Worldclim 2.1 versus Worldclim 1.4: climatic niche and grid resolution affect between-version mismatches in habitat suitability models predictions across Europe

The influence of climate on the distribution of taxa has been extensively
investigated in the last two decades through Habitat Suitability Models
(HSMs). In this context, the Worldclim database represents an invaluable
data source as it provides worldwide climate surfaces for both historical
and future time horizons. Thousands of HSMs-based papers have been
published taking advantage of Worldclim 1.4, the first online version of
this repository. In 2017, Worldclim 2.1 was released. Here, we evaluated
spatially explicit prediction mismatch at continental scale, focusing on
Europe, between HSMs fitted using climate surfaces from the two Worldclim
versions (between-version differences). To this aim, we simulated
occurrence probability and presence-absence across Europe of four virtual
species (VS) with differing climate-occurrence relationships. For each VS,
we fitted HSMs upon uncorrelated bioclimatic variables derived from each
Worldclim version at three grid resolutions. For each factor combination,
HSMs attaining sufficient discrimination performance on spatially
independent test data were projected across Europe under current
conditions and various future scenarios, and importance scores of the
single variables were computed. HSMs failed in accurately retrieving the
simulated climate-occurrence relationships for the climate-tolerant VS and
the one occurring under a narrow combination of climatic conditions. Under
current climate, noticeable between-version prediction mismatch emerged
across most of Europe for these two VSs, whose simulated suitability
mainly depended upon diurnal or yearly variability in temperature;
differently, between-version differences were more clustered toward areas
showing extreme values, like mountainous massifs or southern regions, for
VSs responding to average temperature and precipitation trends. Under
future climate, the chosen emission scenarios and Global Climate Models
did not evidently influence between-version prediction discrepancies,
while grid resolution synergistically interacted with VSs’ niche
characteristics in determining extent of such differences. Our findings
could help in re-evaluating previous biodiversity-related works relying on
geographical predictions from Worldclim-based HSMs.