Scholarlink

<p>Strong earthquakes can trigger large landslides, generating loose material that increases the long-term risk of debris flows. The 2008 Wenchuan earthquake caused widespread slope instability, making the region a globally significant area for debris flow research under rainfall influence. This study proposes a multi-algorithm fusion model for rapid and accurate post-earthquake debris flow hazard identification. We created a dataset of 931 catchments in the Wenchuan seismic area, covering 30 variables across seven categories related to debris flow susceptibility. Key drivers of debris flow activity were identified through multicollinearity analysis and feature importance ranking. Three models—Linear Discriminant Analysis, Logistic Regression, and Random Forest—were developed and assessed using confusion matrices and Partial Dependence Plots to integrate geoscientific understanding. The results show that peak ground acceleration, distance to main faults, mean monsoon rainfall, mean normalized difference vegetation index, and coseismic landslide area are the primary controls on post-Wenchuan debris flows. Among the models, Random Forest performed best, achieving an Area Under the Curve (AUC) of 0.96 and a Matthews Correlation Coefficient of 0.80, outperforming traditional methods by over 10% and 50%, respectively. This study demonstrates the potential of machine learning to improve both the accuracy and interpretability of debris flow prediction, moving beyond black box approaches. The findings provide a scientific basis for regional risk assessment and long-term disaster mitigation planning.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A multi-algorithm fusion approach for identifying post-earthquake rainfall-induced debris flow catchments

  • 期刊论文

  • 发表日期：

    2025年10月14日

查看全文
  • Huaqiang Yin,
  • Wei Zhou,
  • Renwen Liu,
  • Yaping Zhou,
  • Ming Chen,
  • Zhangqiang Peng

摘要

Strong earthquakes can trigger large landslides, generating loose material that increases the long-term risk of debris flows. The 2008 Wenchuan earthquake caused widespread slope instability, making the region a globally significant area for debris flow research under rainfall influence. This study proposes a multi-algorithm fusion model for rapid and accurate post-earthquake debris flow hazard identification. We created a dataset of 931 catchments in the Wenchuan seismic area, covering 30 variables across seven categories related to debris flow susceptibility. Key drivers of debris flow activity were identified through multicollinearity analysis and feature importance ranking. Three models—Linear Discriminant Analysis, Logistic Regression, and Random Forest—were developed and assessed using confusion matrices and Partial Dependence Plots to integrate geoscientific understanding. The results show that peak ground acceleration, distance to main faults, mean monsoon rainfall, mean normalized difference vegetation index, and coseismic landslide area are the primary controls on post-Wenchuan debris flows. Among the models, Random Forest performed best, achieving an Area Under the Curve (AUC) of 0.96 and a Matthews Correlation Coefficient of 0.80, outperforming traditional methods by over 10% and 50%, respectively. This study demonstrates the potential of machine learning to improve both the accuracy and interpretability of debris flow prediction, moving beyond black box approaches. The findings provide a scientific basis for regional risk assessment and long-term disaster mitigation planning.