Model Selection- SDM True Skill Statistics

Model performance characterised by True Skill Statistic (TSS) for each model run and for each species.TSS was then averaged to select an ensemble of models to use for all species. Models scoring an average above 0.7 were deemed good, and models were fitted using an ensemble of eleven different statistical techniques (BIOCLIM; boosted regression trees (BRT); Euclidean distance (EUC); generalised additive models (GAM);  generalised linear models (GLM); multivariate adaptive regression splines (MARS); Mahalanobis model (Mahal); maximum entropy (Maxent); MaxLike; multi-layer perceptron (MLP); random forests (RF); recursive partitioning (RPart);  support vector machine (SVM)) in <i>sdm</i> package (Naimi and Araújo, 2016).