Integrating MaxEnt and explainable AI to assess habitat suitability and conservation gaps of Przewalski's gazelle (Procapra przewalskii) in the Qinghai Lake region.

Habitat suitability assessment and conservation gap identification are critical for endangered species conservation. Using Przewalski's gazelle (Procapra przewalskii), an endangered ungulate endemic to China, we assessed habitat suitability, key environmental predictors, and conservation gaps in the Qinghai Lake region. Based on 1597 field occurrence records collected from 2020 to 2025, we integrated the Maximum Entropy (MaxEnt) model with explainable artificial intelligence (XAI), including SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDPs), and GAP analysis. The selected MaxEnt model showed high discriminatory ability within the surveyed region (mean test AUC = 0.972 ± 0.003). The Human Influence Index (HII), mean temperature and precipitation of the driest quarter, elevation, and vegetation type were key predictors associated with habitat suitability, while mean diurnal range and HII had the highest mean SHAP importance. Suitable habitats were concentrated around Qinghai Lake and in the northern Gonghe Basin. The Qinghai Lake Basin represented the current core distribution area, whereas the Gonghe Basin contained extensive currently predicted suitable habitat and may represent an important conservation opportunity. GAP analysis showed that 71.83% of suitable habitat and 52.34% of the current distribution range lay outside the protected-area boundaries assessed in this study. Fence occurrence in the Hubei population area was significantly associated with relatively high predicted habitat suitability and showed marked spatial clustering. These findings provide spatially explicit evidence for protected-area optimization and habitat management of Przewalski's gazelle.