CANCELLED - Incorporating scale-dependent geodiversity predictors to improve species distribution models for data poor species

Background/Question/Methods

Species distribution models (SDMs) are often used to estimate species’ ranges and designate their conservation status. Despite the utility of SDMs, many species are data deficient, especially in biodiversity hotspots, making it difficult to generate robust range estimates. Further, SDMs often rely solely on environmental means as predictors at single resolutions, yet species distributions are also a function of environmental heterogeneity and filtering acting at different spatial scales. Geodiversity variables capture the spatial variation in Earth’s environment, offering an opportunity to represent this heterogeneity and improve conservation assessments, but their utility for SDMs has not been tested. We test the ability of scale-dependent measures of geodiversity to improve SDMs for Colombian mammals, across varying levels of occurrence data availability. We computed a range of geodiversity variables using CHELSA bioclim variables and SRTM elevation, at varying spatial extents from focal occurrence records. We then compared performance of Maxent SDMs generated using the standard approach of single-pixel values of climate and elevation variables, to models including geodiversity measures for those variables. In accordance with environmental filtering theory, we expected that geodiversity of climate at broader scales and geodiversity of elevation at more local scales would best explain the variation in species distributions.

Results/Conclusions

Certain scales of geodiversity, such as standard deviation of elevation (SDelev), improve the ability of SDMs to explain distributions of data poor species. Permutation importance, a measure of variable contribution in Maxent, also increased with increased spatial resolution of SDelev, signaling that certain measures of geodiversity are more relevant at some scales than others, but not that SDelev was ubiquitously more important locally. For example, for the ornate tití monkey, a lowland species, the importance of SDelev increased with spatial scale, likely because a broadening of the calculation radius increasingly included montane areas, a strong limiting factor to the species’ dispersal and therefore their distribution. This indicates that some topographic geodiversity variables are more important at broader scales when the radius of calculation incorporates known limiting factors to the species’ distribution. By identifying a vetted set of variables that perform well for data-rich and poor species, we can help conservation groups generate SDMs that can be used for conservation planning. This study’s open workflow for identifying scales of geodiversity predictors and their SDM performance, will help practitioners generate more robust models of species ranges, and improve conservation assessments for data-rich and data-poor species alike.