Comparison of MaxEnt and boosted regression tree model performance in predicting the spatial distribution of threatened plant, Telephus spurge (Euphorbia telephioides)

A species distribution model (SDM) was developed to predict the presence and suitable habitat of the federally threatened plant, Euphorbia telephioides, in northwest Florida using data acquired from the U.S. Fish & Wildlife Service. I used two machine-learning models, MaxEnt and boosted regression trees (BRTs), as previous research has shown them to yield high predictability, especially with presence-only data and different types of predictor variables. Different methods were used to reduce effects of spatial autocorrelation and sampling bias in the model predictions since E. telephioides populations are strictly located along the coast. The 29 predictor variables were a combination of categorical, continuous, and distance-based variables. Both the MaxEnt and BRT models had high accuracy as measured by area under the curve (AUC), sensitivity, specificity, and true skill statistic (TSS), but the BRTs had a much lower deviance. The BRT models were also validated with the discovery of a new population in an area predicted as high probability of occurrence. This study demonstrates that machine-learning SDMs can be used by conservation organizations as cost-effective tools to find and protect new populations of threatened or endangered species.