Remote sensing for species distribution models: An illustration from a sentinel taxon of the world's driest ecosystem

In situ observed data are commonly used as species occurrence response variables in species distribution models. However, the use of remotely observed data from high-resolution multispectral remote-sensing images as a source of presence/absence data for species distribution models remains under-developed. Here, we describe an ensemble species distribution model of black microbial mats (Nostoc spp.) using presence/absence points derived from the unmixing of 4-m resolution WorldView-2 and WorldView-3 images in the Lake Fryxell basin region of Taylor Valley, Antarctica. Environmental and topographical characteristics such as soil moisture, snow, elevation, slope, and aspect were used as predictor variables in our models. We demonstrate that we can build and run ensemble species distribution models using both dependent and independent variables derived from remote-sensing data to generate spatially explicit habitat suitability maps. Snow and soil moisture were found to be the most important variables accounting for about 80% of the variation in the distribution of black mats throughout the Fryxell basin. This study highlights the potential contribution of high-resolution remote-sensing to species distribution modeling and informs new studies incorporating remotely derived species occurrences in species distribution models, especially in remote areas where access to in situ data is often limited.