Assessing the Effectiveness of Correlative Ecological Niche Model Temporal Projection through Floristic Data

SIMPLE SUMMARY: Climate change is the main threat for conservation in the 21st century. Reliable methodologies and tools for the evaluation of its impact are urgently needed. Correlative ecological niche models (ENMs) are effective tools for predicting the future distribution of species under climate change scenarios. Despite this, many alternative different methods have been proposed, and objective reasons for a proper selection are unclear. Therefore, a comparative study to evaluate the consistency of predictions of the main ENM algorithms was performed. To test the effectiveness of correlative ENM temporal projection, we compared predictions generated using historical data and projected to the modern climate with predictions generated using modern distribution and climate data. In total, 600 case studies were generated, by using 25 Italian endemic plant species, 12 algorithms and 2 alternative sets of environmental variables. As a result, we highlighted the similarity of eight algorithms and the poor performance of four. ABSTRACT: Correlative ecological niche modelling (ENM) is a method widely used to study the geographic distribution of species. In recent decades, it has become a leading approach for evaluating the most likely impacts of changing climate. When used to predict future distributions, ENM applications involve transferring models calibrated with modern environmental data to future conditions, usually derived from Global Climate Models (GCMs). The number of algorithms and software packages available to estimate distributions is quite high. To experimentally assess the effectiveness of correlative ENM temporal projection, we evaluated the transferability of models produced using 12 different algorithms on historical and modern data. In particular, we compared predictions generated using historical data and projected to the modern climate (simulating a “future” condition) with predictions generated using modern distribution and climate data. The models produced with the 12 ENM algorithms were evaluated in geographic (range size and coherence of predictions) and environmental space (Schoener’s D index). None of the algorithms shows an overall superior capability to correctly predict future distributions. On the contrary, a few algorithms revealed an inadequate predictive ability. Finally, we provide hints that can be used as guideline to plan further studies based on the adopted general workflow, useful for all studies involving future projections.