Machine learning applied to global scale species distribution models (SDMs)

Abstract

Species Distribution Models (SDMs) have been widely applied in ecology to analyze the historical and future patterns of marine species' distributions. With the increasing impact of climate change in recent decades, understanding potential shifts in species distributions has become a crucial challenge. Research on alterations in spatial and temporal distributions has revealed an increasing focus on developing different statistical approaches for global-scale and long-term forecasts. One promising approach is Bayesian Additive Regression Trees (BART), a non-parametric machine learning tool based on a sum-of-trees model that could be useful for addressing ecological problems. The goal of this study is to apply BART on a global scale and use it to estimate and predict possible present and future habitats of marine species under different climate change scenarios. Here we show an application of BART focused on the functional group of marine turtles, analyzing their historical and future distributions both individually and as a taxonomic group, their relationship with environmental variables, and BART's capacity to predict long-term distributions at global scales. Furthermore, to assess the capabilities of BART, we conduct a simulation study under two distinct scenarios: 1) simulating a hypothetical cosmopolitan species distribution and 2) simulating a hypothetical persistent species distribution. Our results show that BART is a promising approach to predict the potential distribution of our target species, as well as their relationship with key environmental variables, on a global scale.