Dangers of using global bioclimatic datasets for ecological niche modeling. Limitations for future climate projections

• Three public high-resolution bioclimatic datasets are intercompared for niche modelling.
• Regional deficiencies of WorldClim lead to misleading distribution models.
• Model problems are not readily apparent in terms of predictive performance.
• Serious inconsistencies arise when projecting models into future climate conditions.

Global bioclimatic datasets are being widely used in ecological research to estimate the potential distribution of species using Climate Envelope Models (CEMs). These datasets are easily available and offer high resolution information for all land areas globally. However, they have not been tested rigorously in smaller regions, and their use in regional CEM studies may pose problems derived from their poor representation of local climate features. Moreover, these problems may be enhanced when using CEMs for future climate projections—a topic of current active research—due to the uncertainty derived from the future altered climate scenarios.In this paper we use distributional data of European beech (Fagus sylvatica) in Northern Iberian Peninsula to analyze the discrepancies of the CEMs (predictive skill, variable importance and consistency using different predictor subsets) resulting from three alternative public, high-resolution climate datasets: a benchmarking regional climate dataset developed for the area of study (UC), the University of Barcelona Atlas for the Iberian Peninsula (UAB) and the worldwide WorldClim bioclimatic dataset (WC). The same CEM techniques (multiple logistic regression and multivariate adaptive regression splines) were applied to the different datasets, showing that the quality of the baseline climate has a great impact on the resulting models, as manifested by the different contributions of the bioclimatic predictors to the resulting models. Artifactual bioclimatic variables were found in some datasets, representing topographical features and spatial gradients, rather than true climatic patterns, thus significantly contributing to the models, although not for the right reasons. This causes a misleading model interpretation and problems for extrapolation in future climate conditions, as evidenced analyzing the future projections obtained using state-of-the-art regional climate projections from the ENSEMBLES project.