“…First, each model included variables that were not highly correlated ((|r|≤0.7) and that had a model contribution >5% and then step-wise optimized the β multiplier from to 15 at an increment of 0.5. Second, as MaxEnt calculates five models for each variable, known as features (i.e., linear (L), quadratic (Q), product (P), threshold (T), and hinge (H)) (Phillips et al, 2017), we selected feature sets by the lowest AIC among "L", "H", "LQ", "LQT", "LP", "HP", "LQP", and "LQTP", then used the optimized model to predict a habitat suitability map. The prediction was evaluated by threshold-independent (Area Under the Curve of the Receiver Operating Characteristic plot, AUC) and threshold-dependent omission rate.…”