2.3 Model Building, Optimization, and Evaluation
In accordance with the latest model optimization methodology proposed by
Cobos et al. (2019), adjustments were made to the parameters of the
MaxEnt model to assess the degree of alignment between the distribution
points of T. sinense and the model, as determined by the
corrected Akaike information criterion (AICc) (Yu et al., 2019). Optimal
model parameters, indicated by the lowest AICc value, were selected for
utilization within the MaxEnt software (Philips et al., 2017).
The distribution data of T. sinense and its corresponding
environmental variables spanning the study period were inputted into the
MaxEnt model. In this investigation, 25% of the 232 distribution sites
of T. sinense were earmarked for model evaluation, while the
remaining 75% constituted the training set for constructing a response
curve. 10 replicates were generated for each training partition, and the
resultant outcomes were averaged. Model results were generated in both
Logistic and ASC format files, with a multi-feature combination based on
optimization outcomes (Moreno et al., 2011).
To calibrate the model and verify its robustness, the receiver operating
characteristic (ROC) curve was analyzed, independent of the threshold.
The area under the curve (AUC) was computed to ascertain the accuracy of
the model. Model performance was categorized as failure (0.5-0.6), poor
(0.6-0.7), fair (0.7-0.8), good (0.8-0.9), or excellent (0.9-1.0), with
higher AUC values indicating superior model performance (Fithian et al.,
2015).
Classification of Fitness Levels and Area Statistics
The average output data from each period, following 10 simulations in
the MaxEnt model, were imported into ArcGIS software. These data were
then converted into raster layers and subsequently reclassified based on
the distributed probability (P) values. Employing natural breakpoint
classification, the distribution area of T. sinense was
categorized into four distinct levels: Non-suitable area (P <
0.2), Low suitable zone (0.2 ≤ P < 0.4), Intermediate suitable
zone (0.4 ≤ P < 0.6), High suitable zone (P ≥ 0.6) The
reclassified layers were processed using an ArcGIS grid table to
determine the area encompassed by each level (Zhuang et al., 2018).