Susceptibility analysis of glacier debris flow by investigating glacier changes based on remote sensing imagery and deep learning: A case study

Glacier debris flow is becoming increasingly severe as global warming intensifies. To minimize the catastrophic losses caused by glacier debris flow, it is essential to analyze the susceptibility of glacier debris flow. In this study, we employed remote sensing with deep learning methods to investigate glaciers changes, evaluating the susceptibility of glacier debris flow in Zelongnong ravine, Southeast Tibet. First, we utilized Landsat optical remote sensing imageries to obtain the semantic segmentation dataset and trained a deep learning model to automatically extract the glacier boundary in Zelongnong ravine. Second, by pre-processing the DEMs (Digital Elevation Model) and integrating them with the glacial boundaries, the volume of glacier ablation was measured. Eventually, according to the glacier ablation, the correction coefficient was determined, which modified the geomorphic information entropy theory, and further analyzed the susceptibility of glacier debris flow in Zelongnong ravine. The research results of the study present that the evaluation indices of the deep learning model that extracted glacier boundaries are over 90%. Moreover, the study results confirm the accuracy of the modified susceptibility evaluation method for glacier debris flows, and the susceptibility of glacier debris flows in Zelongnong ravine generally ranges between high and very high. This study reveals the feasibility and progress of using remote sensing and deep learning in glacier boundary extraction, providing a promising reference for the evaluation of the susceptibility and prediction of glacier debris flow in similar high mountainous areas as well.