Scholarlink

<p>The long-term security of downstream ecosystems and communities is significantly impacted by the landslide-induced river blockages and the subsequent breach floods. Accurate risk level prediction of landslide dam is crucial for appropriate disaster prevention. This study employs five machine learning (ML) algorithms, i.e., support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM) and k nearest neighbor (KNN) to predict the risk level of landslide dams. We first establish a multi-factor landslide dam database and select the optimal algorithms for different missing factors. By applying the imputation database and sampling strategies, the landslide dam stability probability is evaluated. The longevity and breach peak flow of dams are then predicted for the dams with low stability probability. Results from five-fold cross-validation (CV) demonstrate that the oversampling-XGBoost stability model, RF longevity model and support vector regression (SVR) peak flow model outperform others. Finally, we comprehensively evaluate the risk level based on the above three dimensions and merge the prediction models to create a user-friendly application. The proposed models are validated by representative case studies, showing better generalization and accuracy than the conventional models, which enhance emergency planning for landslide dam hazards.</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

Machine learning-based risk level prediction of landslide dams considering stability, longevity and breach peak flow

  • 期刊论文

  • 发表日期：

    2025年4月8日

查看全文
  • Zhenyu Feng,
  • Jieyuan Zhang,
  • Congjiang Li,
  • Haimei Liao,
  • Jiawen Zhou,
  • Hongyu Luo

摘要

The long-term security of downstream ecosystems and communities is significantly impacted by the landslide-induced river blockages and the subsequent breach floods. Accurate risk level prediction of landslide dam is crucial for appropriate disaster prevention. This study employs five machine learning (ML) algorithms, i.e., support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM) and k nearest neighbor (KNN) to predict the risk level of landslide dams. We first establish a multi-factor landslide dam database and select the optimal algorithms for different missing factors. By applying the imputation database and sampling strategies, the landslide dam stability probability is evaluated. The longevity and breach peak flow of dams are then predicted for the dams with low stability probability. Results from five-fold cross-validation (CV) demonstrate that the oversampling-XGBoost stability model, RF longevity model and support vector regression (SVR) peak flow model outperform others. Finally, we comprehensively evaluate the risk level based on the above three dimensions and merge the prediction models to create a user-friendly application. The proposed models are validated by representative case studies, showing better generalization and accuracy than the conventional models, which enhance emergency planning for landslide dam hazards.