<scp>Assessment</scp> of Debris Flow Susceptibility Based on <scp>PCA</scp> ‐ <scp>IV</scp> Coupling Method and <scp>SHAP</scp> Model: A Case Study of Linan District

ABSTRACT The evaluation of debris flow susceptibility helps assess both the degree of susceptibility and the spatial distribution characteristics of debris flows in a region, providing a basis for disaster prevention and mitigation planning. However, traditional methods face challenges such as subjective weight calculation and insufficient model interpretability. This study focuses on Lin'an District and constructs a coupling model (PCA‐IV) that combines principal component analysis (PCA) with the information value (IV) method. By applying the SHAP interpretability framework, the reliability of the model's predictions is effectively enhanced. Based on 404 watershed units, nine evaluation indicators were selected, including basin average slope, Melton ratio, engineering geological lithological groups, basin fault density, and average rainfall during the flood season. PCA is used to eliminate multicollinearity among factors and optimize weight calculation, followed by integration with the IV model to classify and quantify the contribution of each indicator. The debris flow susceptibility in Lin'an District is categorized into four levels: Extremely high, High, Medium, and Low. The research results indicate that: (1) The PCA‐IV model achieved an AUC value of 0.89, improving prediction accuracy by 17.6% and 6.5% compared with the single models (IV and PCA), demonstrating good performance. (2) SHAP analysis reveals that rainfall (contributing 28.6%) and slope (23.4%) are the key driving factors in the study area, with a significant interaction between them. (3) The frequency ratio reliability validation shows that the zoning levels produced by the model align with the actual distribution patterns, indicating the model's reliability. This study innovatively combines statistical dimensionality reduction with mechanistic explanation. The proposed PCA‐IV‐SHAP coupling framework, based on watershed units, provides a novel method for debris flow risk assessment that integrates both mathematical reliability and engineering applicability. Compared with more complex algorithmic models, this approach offers clearer physical meaning and can provide valuable technical references for disaster prevention and mitigation in similar mountainous regions.