3.1 Established models
As for LIG, PCA results show a 0.6 KMO value, and the second axis expresses the most significant AUC value (AUC = 0.98) among the three axes, accounting for 24.2% of the total variance, on which sevenBC variables were selected (Table S1, Figure. S2, and S3 in Appendix). As for 1970-2000, the first PCA axis accounts for 39.7% of the total variance with a 0.6 KMO value and a 0.99 AUC value (Figure S2 and S3 in Appendix), and five BC variables were chosen (Table S1). The optimal combination of feature class (FC) andregularization multiplier (RM) for all five models is presented in Table S2.
Model accuracy was assessed by 10-fold cross-validation; all models’ mean AUCTEST values are greater than required for a significant threshold (0.5) (Figure S4 in the Appendix), implying the demonstrated predictions are excellent.
Variable contributions to MaxEnt prediction models are illustrated in Figure S5. Regarding 1970-2000, var8 (33.5%), var16 (25.1%), and var5 (19.1%) play significant roles, accounting for 77.7% of the total contribution. As for SSP2, var8 (42.3%), var5 (21.3%), and var16 (15.9%) are highlighted, making 79.5% of the contribution. Concerning SSP5, 47.3% for var8, 21.9% for var5, and 14.2% for var16 are stressed for their 83.4% of the total contribution.