Glacial debris flow susceptibility mapping based on combined models in the Parlung Tsangpo Basin, China

• Published in: Journal of mountain science 2024-04, Vol.21 (4), p.1231-1245
• Main Authors: Zhou, Yonghao, Hu, Xiewen, Xi, Chuanjie, Wen, Hong, Cao, Xichao, Jin, Tao, Zhou, Ruichen, Zhang, Yu, Gong, Xueqiang
• Format: Article
• Language: English
• Subjects: Debris flow; Detritus; Earth and Environmental Science; Earth Sciences; Ecology; Environment; Factor analysis; Flow mapping; Geography; Glacial drift; Gullies; Landforms; Landslides; Machine learning; Mapping; Mathematical models; Original Article; Prediction models; Statistical analysis; Statistical models; Susceptibility; Valleys
• ISSN: 1672-6316; 1993-0321; 1008-2786
• DOI: 10.1007/s11629-023-8500-0

Machine learning (ML)-based prediction models for mapping hazard (e.g., landslide and debris flow) susceptibility have been widely developed in recent research. However, in some specific areas, ML models have limited application because of the uncertainties in identifying negative samples. The Parlung Tsangpo Basin exemplifies a region prone to recurrent glacial debris flows (GDFs) and is characterized by a prominent landform featuring deep gullies. Considering the limitations of the ML model, we developed and compared two combined statistical models (FA-WE and FA-IC) based on factor analysis (FA), weight of evidence (WE), and the information content (IC) method. The final GDF susceptibility maps were generated by selecting 8 most important static factors and considering the influence of precipitation. The results show that the FA-IC model has the best performance. The areas with a very high susceptibility to GDFs are primarily located in the narrow valley section upstream, on both sides of the valley in the middle and downstream of the Parlung Tsangpo River, and in the narrow valley section of each tributary. These areas encompass 86 gullies and are characterized as “narrow and steep”.