sdmbench: R package for benchmarking species distribution models

Abstract

Species Distribution Modeling (SDM) is an emerging field in ecology. The effects of anthropogenic climate change, habitat destruction (deforestation, pollution) and poaching are observable in ecosystems around the world (Elith & Leathwick, 2009). SDMs have been used to address those challenges with notable success in estimating the effects of climate change on species distributions (MP Austin & Van Niel, 2011), natural reserve planning (Guisan et al., 2013) and predicting invasive species distributions (Descombes et al., 2016). Steady improvements in analytical tools (new machine learning algorithms and faster processors in particular) and larger amounts of data gathered recently opened up new possibilities in the field. Although researchers benefit from these new tools, they face challenges in method selection and evaluation. In the case of machine learning algorithms, this issue is particularly relevant. How to decide which algorithm to use, knowing that model performance can vary significantly between datasets? And, how do we compare models in a consistent manner, without introducing additional bias? How can we demonstrate the improvement of a new method over the current state-of-the-art?