Talk: Using SVMs to delineate parapatric ranges: An example with three-toed sloths (Bradypus)
Hybrid weighted support vector machine classifier
Abstract
Species distribution modeling (SDM) techniques are a common tool for estimating species ranges. These models typically rely only only on abiotic variables without accounting for biotic interactions, despite the fact that these interactions may impose important constraints on ranges. Distribution patterns in which closely-related parapatric species replace each other across geographic space are common in ecology. We sought to address whether incorporating biotic information into range estimates for three species of sloth (genus Bradypus) would improve distribution models for species demonstrating this parapatric pattern of distribution. We used support vector machines (SVMs) as masks to delineate the predicted boundaries between these three species’ ranges. We created two different kinds of SVMs: 1) spatial SVMs using only occurrence data, and 2) sp+env SVMs using occurrence data in conjunction with predicted habitat suitability from SDMs. We found that the sp+env SVM resulted in the most ecologically realistic distribution model, accounting for contact zones between species and the effects of climate.