Spatial Downscaling of Alien Species Presences Using Machine Learning

The spatial resolution of available data often poses limitations in the analyses. For species distribution data, spatial resolution and spatial extent are typically inversely proportional. Based on the fundamental assumption that detectable relationships exist between information across spatial scales, spatial downscaling refers to the process and methodologies of using coarse resolution input to infer finer resolution output. Given the absence of fine scale alien species data, potential environmental explanatory covariates available at the resolution of the alien species, as well as at finer resolutions, can be used to infer alien species presences at finer resolutions. The method employing random forests machine learning worked well using a "difficult" dataset and thus may work even better in smoother ones - false presences and absences were also quantified. The method can be employed in any other species (i.e. not only alien) for example endemics or a particular charismatic or of medicinal interest species. The full R code used is also available.

The paper is available here (open access):

The full R code is available here: