Scholarlink

<p>This study presents a machine learning-based geospatial approach for mapping typhoon-induced flood susceptibility in urban environments. The urban center of Thanh Hoa Province (Vietnam) during Typhoon Kajiki in August 2025 was selected as a case study. Light Gradient Boosting Machine (LightGBM) and deep neural networks (DNN) were employed to classify flood-prone areas using a set of topographical, urban morphological, hydrological, and infrastructure-accessibility factors. Both LightGBM and DNN demonstrated strong capabilities in flood susceptibility mapping, achieving classification accuracy rates of 91% and 89%, respectively. Notably, LightGBM demonstrated a robust performance, with an area under the receiver operating characteristic curve of 0.97 and a Cohen’s Kappa coefficient of 0.81. Furthermore, SHapley Additive exPlanations (SHAP) analysis supported assessments of the influential features and their spatial effects on susceptibility modeling. Based on the constructed flood susceptibility map, approximately 22% of the study area is classified as being at high or very high flood risk. Key public transport, healthcare, and educational facilities, as well as road sections within the city, have been identified at varying risk levels. In summary, the proposed framework provides a fast and low-cost approach to flood risk assessment by leveraging machine learning and publicly available remote sensing data. Its adaptive structure can be quickly updated as new information becomes available. This fact makes the framework a practical solution for flood risk management in the study area and other typhoon-prone urban regions of Vietnam.</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 Geospatial Modeling of Typhoon-Induced Urban Flood Susceptibility Using Gradient Boosting Machine, Deep Neural Network, and SHAP Interpretability

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • Nhat-Duc Hoang,
  • Van-Duc Tran

摘要

This study presents a machine learning-based geospatial approach for mapping typhoon-induced flood susceptibility in urban environments. The urban center of Thanh Hoa Province (Vietnam) during Typhoon Kajiki in August 2025 was selected as a case study. Light Gradient Boosting Machine (LightGBM) and deep neural networks (DNN) were employed to classify flood-prone areas using a set of topographical, urban morphological, hydrological, and infrastructure-accessibility factors. Both LightGBM and DNN demonstrated strong capabilities in flood susceptibility mapping, achieving classification accuracy rates of 91% and 89%, respectively. Notably, LightGBM demonstrated a robust performance, with an area under the receiver operating characteristic curve of 0.97 and a Cohen’s Kappa coefficient of 0.81. Furthermore, SHapley Additive exPlanations (SHAP) analysis supported assessments of the influential features and their spatial effects on susceptibility modeling. Based on the constructed flood susceptibility map, approximately 22% of the study area is classified as being at high or very high flood risk. Key public transport, healthcare, and educational facilities, as well as road sections within the city, have been identified at varying risk levels. In summary, the proposed framework provides a fast and low-cost approach to flood risk assessment by leveraging machine learning and publicly available remote sensing data. Its adaptive structure can be quickly updated as new information becomes available. This fact makes the framework a practical solution for flood risk management in the study area and other typhoon-prone urban regions of Vietnam.