A two-layer heterogeneous ensemble model based on stacking for debris-flow susceptibility mapping: a case study of Yajiang County, China

To reduce the serious losses caused by debris flows, debris-flow susceptibility mapping (DFSM) is valuable and indispensable. The objective of this study was to build a high-quality regional debris flow prediction model through a heterogeneous ensemble technology, which also supports environmental sustainability by promoting safer land-use planning and ecosystem conservation. This technology aims to enhance the prediction accuracy by aggregating the advantages of different algorithms and counteracting their weaknesses and deviations. Thus, a two-layer heterogeneous ensemble model based on Stacking (ASC-Stacking) was proposed by combining artificial neural network (ANN), support vector machine (SVM), and C5.0 decision tree (C5.0 DT). Yajiang County, China was selected as the study area. Debris-flow inventory data were randomly divided into a training set (70%) and test set (30%). We considered 12 affecting factors and established four models (ANN, SVM, C5.0 DT, and ASC-Stacking). The prediction accuracy was validated by the receiver operating characteristic (ROC) curve. The results showed all four models were accurate and the ASC-Stacking model achieved the best performance. In conclusion, the idea of heterogeneous ensemble can further enhance the prediction ability and yield high-quality DFSM, thereby reducing the risk of regional debris flow and contributing to sustainable environmental management practices.