Debris flow susceptibility mapping in alpine canyon region: a case study of Nujiang Prefecture

• Published in: Bulletin of engineering geology and the environment 2024-05, Vol.83 (5), p.169, Article 169
• Main Authors: Li, Yimin, Jiang, Wenxue, Feng, Xianjie, Lv, Shengbin, Yu, Wenxuan, Ma, Enhua
• Format: Article
• Language: English
• Subjects: Accuracy; Artificial neural networks; Canyons; Debris flow; Deep learning; Detritus; Disasters; Earth and Environmental Science; Earth Sciences; Flow mapping; Foundations; Geoecology/Natural Processes; Geoengineering; Geology; Geotechnical Engineering & Applied Earth Sciences; Gullies; Hydraulics; Learning algorithms; Lithology; Machine learning; Mapping; Mathematical models; Nature Conservation; Neural networks; Original Paper; Performance prediction; Rainfall; Regression analysis; Remote sensing; Sensitivity analysis; Statistical analysis; Statistical models; Support vector machines; Surface roughness; Topography; Watersheds
• ISSN: 1435-9529; 1435-9537
• DOI: 10.1007/s10064-024-03657-2

Accurate debris flow susceptibility mapping (DFSM) plays a crucial role in enabling government authorities to devise rational policies to mitigate the threats posed by debris flows to human life and property. Nujiang Prefecture, located in the alpine canyon region, is prone to frequent debris flows in China. Therefore, this study focuses on Nujiang Prefecture as the research area. Based on the characteristics of debris flow development, the occurrence mechanism, and the actual conditions of the study area, small watersheds are selected as mapping units. Fifteen influencing factors, including elevation, slope, aspect, relief, surface roughness, Melton ratio, NDVI, lithology, distance to faults, rainfall, SPI, TWI, STI, watershed aera, and gully density, are considered in the mapping process. We explored the predictive performance of three single models, namely, the statistical model certainly factor (CF), the machine learning model support vector machines (SVM), and the deep learning model convolutional neural network (CNN). Additionally, we investigated the coupling models CF-LR (statistical model coupled with machine learning model) and CNN-SVM (machine learning model coupled with deep learning model) in the mapping of debris flow sensitivity. The analysis and comparison of model performance were conducted using the area under the receiver operating characteristic curve (AUC) and the mean value (MV) and standard deviation (SD) of debris flow sensitivity values. The results demonstrate that all five models show promising performance in DFSM. Among them, the CNN-SVM coupled model (AUC = 0.933, MV = 0.211, SD = 0.199) outperforms the others, exhibiting the best predictive capability. These findings can serve as valuable references for debris flow prevention and control efforts.