2.3 Optimization of model parameters and model building
The feature combination (FC) and regularization multiplier (RM) were adjusted by the ENMeval package in R 4.0.2 software (Yan et al., 2021). There are five FCs in the MaxEnt model: linear, quadratic, product, threshold, and hinge (Phillips et al., 2006). The MaxEnt default parameters are RM = 1, FC = LQHPT. To optimize the model, we set RM to 0.5-4, increased 0.5 each time, and adopted 8 regularization and 6 FCs, namely, L, LQ, H, LQH, LQHP, and LQHP (Li et al., 2020). The ENMeval data package was used to test the 48 parameter combinations and the model’s fitting and complexity according to the Delta in Akaike information criterion models Delta AICc and AUC.diff (Zhao et al, 2021).
Data of Satyrium species occurrence and each environmental variable were imported into the MaxEnt model, and MaxEnt software (http://www.cs.princeton.edu/, version 3.4.1) was employed for simulation. To validate the model’s overall predictive performance, 75% of occurrence records were used for training and the remaining 25% were used for testing (Fielding & Bell, 1997). An average of 10 repetitions was the final output result. In the end, the main environmental variables affecting Satyrium species’ distribution were evaluated based on the jackknife test results and the contribution rate of environmental variables. Values from the receiver operating characteristic (ROC) and true skill statistic (TSS) were applied to evaluate the model’s accuracy (Jiang et al., 2018; Xu et al., 2019). It has been shown that model simulation results are excellent when the values of the area under the curve (AUC) and TSS are > 0.9 (Janitza et al., 2013; McIntyre et al., 2017; Phillips et al., 2006).