Methods for improving species distribution models in data-poor areas: example of sub-Antarctic benthic species on the Kerguelen Plateau

Species distribution models (SDMs) are essential tools to aid conservation biologists in evaluating the combined effects of environmental change and human activities on natural habitats and for the development of relevant conservation plans. However, modeling species distributions over vast and remote regions is often challenging due to poor and heterogeneous data sets, and this raises questions regarding the relevance of the modeling procedures. In recent years, there have been many methodological developments in SDM procedures using virtual species and broad data sets, but few solutions have been proposed to deal with poor or heterogeneous data. In the present work, we address this methodological challenge by studying the performance of different modeling procedures based on 4 real species, using presence-only data compiled from various oceanographic surveys on the Kerguelen Plateau (Southern Ocean). We followed a practical protocol to test for the reliability and performance of the models and to correct for limited and aggregated data, as well as accounting for spatial and temporal sampling biases. Our results show that producing reliable SDMs is feasible as long as the amount and quality of available data allow testing and correcting for these biases. However, we found that SDMs could be corrected for spatial and temporal heterogeneities in only 1 of the 4 species we examined, highlighting the need to consider all potential biases when modeling species distributions. Finally, we show that model reliability and performance also depend on the interaction between the incompleteness of the data and species niches, with the distribution of narrow-niche species being less sensitive to data gaps than species occupying wider niches.