Sensitivity of species climate envelope models to baseline climatology and effect on RCM-based future projection

Abstract

Climate Envelope Models (CEMs) are predictive tools widely used in ecological research to estimate the distribution of species by combining observations of their occurrence/abundance with bioclimatic indicators. In this contribution, we show that the resulting projections are highly sensitive to the quality of the baseline climate data, an aspect often overlooked in model criticism. Using distributional data of European beech in northern Spain (Cantabria region), we analyse the discrepancies in model performance and future projections using three public high-resolution climate datasets: WorldClim (WC), the University of Barcelona Atlas (UAB) and a new regional climate grid developed by Cantabria University (UC). We considered the future climate scenarios from several regional climate models (RCMs) of the EU-funded project ENSEMBLES. We demonstrate that the quality of the baseline climate used to derive the present and future bioclimatic indices has a great impact on the stability of the estimated CEMs, although commonly used performance metrics (AUC, Cohen’s kappa) failed to detect this in the cross-validation experiments. WC models lead to unreliable future projections, whereas UAB models performed better but were outperformed by UC, demonstrating the paramount importance of reliable climate input data.