渐进式优化框架下的地质灾害易发性评价与可解释性分析OA

This study aims to establish a refined modeling framework for debris flow susceptibility at the township scale,focusing on the complex mountainous regions of the subtropical monsoon climate in southwest China.The study proposes a progressive optimization framework based on genetic algorithm(GA),categorical boosting(CatBoost),and Shapley additive explanations(SHAP),which innovatively integrates optimal watershed unit selection,high-quality negative sample set construction,and hyperparameter optimization strategies.In the preprocessing phase,the study constructed a database of debris flow influencing factors,designed five different threshold watershed units,and optimized the negative sample sampling strategy.During the model construction phase,the study employed extreme gradient boosting(XGBoost),light gradient boosting machine(LGBM),categorical boosting(CatBoost),and natural gradient boosting(NGBoost)algorithms as base models and integrated GA hyperparameter optimization methods for optimal testing.Finally,SHAP technology was used to quantitatively analyze the contribution of influencing factors,revealing the primary driving factors behind debris flow occurrence in the southwestern mountainous region.The results indicate that the 1 000 threshold watershed unit performs best among all designs.The CatBoost model performs best among all machine learning algorithms.After hyperparameter optimization,the GA-CatBoost model achieves the highest predictive performance,with accuracy,F1-score,and area under the curve(AUC)values of 0.860,0.880,and 0.910,respectively.SHAP analysis reveals that rock type,soil type,and normalized difference vegetation index(NDVI)are the most significant influencing factors for landslide occurrence in the study area.The findings of this study provide reliable technical support and decision-making basis for geological disaster risk assessment at the township level,and offer valuable references for geological disaster management and prevention efforts.