Abstract: The complex geological structures,pronounced topographic relief,and intricate hydrological environment of the Eastern Tibetan region lead to frequent occurrences of geothermal hazards,posing significant challenges for deep engineering projects such as railway tunnels.To address this issue,this study builds upon the traditional geothermal gradient method to further investigate the spatial distribution of geothermal heat hazard susceptibility in eastern Tibet.Based on 169 measured geothermal data points collected from six tunnels and by incorporating identified hot spring sites as high-temperature geothermal sample points,a spatial dataset of 239 geothermal heat hazard samples was constructed.Using this dataset,environmental attributes such as elevation,slope,aspect,terrain relief,lithology,and distance to fault zones were analyzed.Susceptibility zoning models for geothermal heat hazards in the Eastern Tibetan Plateau were developed using Random Forest (RF) and Convolutional Neural Network (CNN) approaches.The accuracy of these models was validated and compared using ROC curves and confusion matrices.The results indicate that the CNN model achieved an accuracy of approximately 89% for geothermal heat hazard susceptibility zoning in eastern Tibet,outperforming the RF model.This study can help identify high-risk areas of geothermal heat hazards and provides essential references for railway alignment and design in geothermal hazard-prone regions.