Susceptibility assessment of glacier-related debris flow on the southeastern Tibetan Plateau using different hybrid machine learning models.

The southeastern Tibetan Plateau (SETP) is a construction area of several key infrastructure projects in China, such as the Sichuan-Tibet Railway and hydropower developments, which has historically faced the threat of glacier-related debris flows. However, a robust assessment of such debris flow susceptibility is a challenge due to the complex and variable climate, terrain and glacial environment. In this study, we used the hybrid models that combine statistical techniques (certainty factors, CF) with machine learning methods (logistic regression, LR; random forest, RF; extreme gradient boosting, XGBoost) to more accurately identify debris flow susceptible (DFS) areas. Topography, geology, and hydrological factors including glaciers and snow cover were used in these models to assess the DFS. Results show that 21 % to 42 % of the study area is very high susceptible to debris flows, particularly from Ranwu to Bomi and around Namcha Barwa. The hybrid models effectively enhance the accuracy of the DFS assessments. The CF-RF model showed the greatest improvement, with an 8.4 % increase in accuracy compared to the single model, the DFS spatial distribution of which aligns closely with field survey results. The glacial area ratio and annual snowmelt positively impact DFS accuracy, ranking 2nd and 9th in the factor importance, respectively. The results of this research could provide valuable assistance and guidance in mitigating glacier-related debris flow hazards.