Fig. 6: Comparison of predicted suitable habitat with soil type as a predictor and without soil type as a predictor.(a). the prediction map resulting from the best BRT model with soil type, the same results shown in Fig. 4. (b). the map from the best BRT without soil type. Variables are identical in both models except for omission of soil type in model B: annual temperature range, the mean temperature of the warmest quarter (July-September), precipitation of the warmest quarter (July-September), and precipitation of the coldest quarter (December-February). The best model without soil type fitted 2270 trees, with a 10-fold cross-validated AUC of .85. The parameters used for the boosted regression tree analysis were a tree complexity of 1, a learning rate of .0007, bag fraction of .7 and a step size of 15.