Establishing a GIS-based evaluation method considering spatial heterogeneity for debris flow susceptibility mapping at the regional scale

• Published in: Natural hazards (Dordrecht) Vol. 114; no. 3; pp. 2709 - 2738
• Main Authors: Qin, Shengwu, Qiao, Shuangshuang, Yao, Jingyu, Zhang, Lingshuai, Liu, Xiaowei, Guo, Xu, Chen, Yang, Sun, Jingbo
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.12.2022; Springer Nature B.V
• Subjects: Artificial neural networks; Civil Engineering; Debris flow; Detritus; Disasters; Earth and Environmental Science; Earth Sciences; Environmental Management; Flow mapping; Frameworks; Geographic information systems; Geographical information systems; Geophysics/Geodesy; Geotechnical Engineering & Applied Earth Sciences; Heterogeneity; Hydrogeology; Information processing; Information systems; Machine learning; Mapping; Modelling; Natural Hazards; Neural networks; Original Paper; Patchiness; Regional analysis; Remote sensing; Spatial heterogeneity; Support vector machines; Regional-scale debris flow susceptibility; China; Jilin province; Spatial heterogeneity; GIS-based processing framework; Support vector machine; Convolutional neural networks
• ISSN: 0921-030X; 1573-0840
• DOI: 10.1007/s11069-022-05487-5

Susceptibility mapping is an effective means of preventing debris flow disasters. However, previous studies have failed to solve spatial heterogeneity well, especially at the regional scale. The main objective of this study is to solve the spatial heterogeneity of regional-scale debris flow susceptibility (DFS) mapping by establishing a geographic information system (GIS)-based processing framework. The framework was realized by integrating the determination factor (DFactor) model with machine learning models. The DFactor model established different combinations of evaluation factors in each local region and clarified the differing contributions of influencing factors to DFS. To test the feasibility of the framework, the support vector machine (SVM) and two-dimensional convolutional neural network (CNN) were integrated with the DFactor model (DFactor-SVM and DFactor-CNN) to evaluate DFS in Jilin Province, China. The individual models (SVM and CNN) were also used to map the DFS for comparison with the integrated models. For debris flow modeling, 868 debris flow samples were collected and randomly divided into two datasets: 70% of the samples were used for training and the result was used for verification. The results of the receiver operating characteristic curve showed that the integrated models performed better. The DFactor-CNN model had the highest predictive accuracy, followed by the DFactor-SVM, CNN and SVM models. In general, the GIS-based processing framework maximizes the contribution of the influencing factors to debris flows and enhances the prediction ability of models. Furthermore, it provides a reliable means to predict debris flows at the regional scale.