Models were calibrated using training and test data (75% and 25% randomly selected occurrence points respectively).

Area under the curve (AUC) was calculated as an average of 30 model replicate runs using subsample run type. Mean AUC and omission rate values were calculated both for test and training data. The mean omission rates are calculated at an arbitrarily chosen cumulative threshold of 10. All model omission results performed significantly better than random (p<0.0001). Models were calibrated using training and test data (75% and 25% randomly selected occurrence points respectively).