Identification and Spatial Differentiation of High-Risk Areas for Brown Bear Incidents in Yushu Prefecture, China, Using Machine Learning and Remote Sensing.

The Sanjiangyuan Region is among China's most critical ecological function zones and serves as an important habitat for rare wildlife species such as brown bears and snow leopards. Driven by factors including climate change and intensified human activities, human-wildlife conflicts have become increasingly frequent on the Qinghai-Tibet Plateau, threatening the living space of both herders and wildlife. This study centers on the Yushu Tibetan Autonomous Prefecture in Qinghai Province, integrating multi-source remote sensing data with field survey data, and employs the Maximum Entropy Model (MaxEnt) MaxEnt model and the BIOMOD2 framework to simulate high-risk areas for brown bear incidents. Results indicate that the BIOMOD2 ensemble model (EMca) achieved the highest predictive accuracy, with the Random Forest (RF) model demonstrating strong robustness among individual models. Digital Elevation Model (DEM), Soil Surface Moisture (SSM), Fractional Vegetation Cover (FVC), and Human Footprint (HFP) were identified as the primary factors influencing the spatial distribution of brown bear incidents. High-risk areas exhibited significant clustering, mainly concentrated in the southern and southeastern regions of Qumalai, Nangchen, and Chindu; the eastern part of Zadoi County; and the central and southern parts of Yushu City, particularly within the elevation range of 4304-4544 m, where human activity intensity is relatively low. The core high-risk zone is located along the Tongtian River in southern Qumalai County, demonstrating strong spatial connectivity. By investigating the spatial distribution patterns and driving mechanisms of brown bear incidents in Yushu Prefecture, this study offers some references for government agencies to formulate strategies that promote harmonious coexistence between humans and nature.