Mapping Suitable Habitats of Apodemus agrarius Based on MaxEnt–SHAP Framework

Abstract Apodemus agrarius is the primary reservoir host of hemorrhagic fever with renal syndrome (HFRS), a severe rodent-borne zoonosis caused by hanta viruses. Xi’an, situated in the Guanzhong Plain of northwestern China, has long been a hotspot for HFRS. Accurate characteristics of the spatial distribution of A. agrarius and its underlying environmental drivers is critical for HFRS control. We compiled 32 georeferenced occurrence records of A. agrarius from field surveys conducted from 2023 to 2025. These records were integrated with 24 environmental predictors—encompassing climate, topography, land cover, and elevation—at a 1-km spatial resolution. To minimize multicollinearity, we excluded highly correlated variables (|r| ≥ 0.8) and retained only those with variance inflation factors (VIF) < 5. A MaxEnt model was calibrated using 75% of the data and evaluate via 10-fold cross-validation. To enhance model interpretability beyond conventional “black-box” approaches, we employed the SHapley Additive exPlanations (SHAP) to quantify both variable importance and nonlinear response relationships. The MaxEnt model achieved high predictive accuracy on test dataset (AUC = 0.901, TSS = 0.708). SHAP analysis revealed that cropland and woodland positively contributed to habitat suitability, whereas urban/built-up areas exerted a strong negative effect. High-suitability habitats (predicted probability ≥ 0.6) covered 818.9 km² (8.10% of the total area), primarily concentrated in agricultural landscapes along the northern foothills of the Qinling Mountains and the margins of the Wei River. Notably, despite their limited spatial extent, these zones accounted for 63% of all observed A. agrarius occurrences. By integrating MaxEnt with SHAP interpretability, this study delivers the first high-resolution, ecologically transparent map of A. agrarius suitability in Xi’an. Our results underscore the critical influence of land cover patterns and climate seasonality on zoonotic reservoir distribution. The identified hotspots provide actionable targets for rodent control and early-warning systems in rapidly urbanizing regions under climate change.