2.3. MaxEnt modelling
MaxEnt (version 3.4.1) was used for modelling (http://
biodiversityinformatics.amnh.org/open_source/maxent/). In the
modelling, 25% of the occurrence points were randomly set as test
points, and the remaining 75% were training points (Kramerschadt et
al., 2013). Each model ran ten repetitions, and the average logical
output was used for the final prediction (Conley et al., 2014). To
account for the sampling bias (Kramer-Schadt et al., 2013), we created a
bias file and 10,000 background points were taken into the MaxEnt models
as “pseudo-absence” data.
The AUC of the receiver operating characteristic assesses the predictive
performance of the model; a high value (0–1) corresponds to a better
predictive model (Phillips et al., 2006). To evaluate the importance of
the environmental variables in modelling, the jackknife test and percent
contribution of variables were used as indicators of MaxEnt. Finally, we
followed the methods used by the previous researchers (Fekede, van Gils,
Huang, & Wang, 2019; Liu et al., 2020), which combined the AHS risk
prediction with the C.imicola suitability map to model the risk
areas for AHS occurrence in Mainland China. The result maps were
visualized using ArcGIS 10.2.