Scholarlink

<p>Flood susceptibility mapping (FSM) is vital for urban flood risk management. This study first generates a flood inventory by using Sentinel-2 imagery and identifies key conditioning factors. Secondly, a novel FSM method is proposed, namely WHD-LightGBM, which integrates Whale Optimization Algorithm (WOA), Harris Hawk Optimization Algorithm (HHO), and Dung Beetle Optimization Algorithm (DBO) to tune Light Gradient Boosting Machine (LightGBM). Subsequently, based on the SHapley Additive exPlanations (SHAP) method and the partial dependency plot (PDP) method, a new interpretable approach, referred to as SHAP-PDP, is introduced to reveal how predictors influence FSM. Finally, WHD-LightGBM was evaluated on a Sentinel-2-derived flood inventory for Nanchang, China, and compared with seven benchmark models (LightGBM, Grid-LightGBM, Random Forest, XGBoost, WOA-LightGBM, HHO-LightGBM, and DBO-LightGBM). The findings indicate that the FSM based on WHD-LightGBM demonstrates high accuracy and stability (Accuracy = 0.8288, Precision = 0.8506, Specificity = 0.8456, F1-score = 0.8315, Recall = 0.8132); SHAP-PDP analysis identifies altitude, normalized difference vegetation index (NDVI), and rainfall as the most influential conditioning factors in the study area. This study provides a novel approach and methodology for city managers to effectively manage flood disasters.</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 hybrid LightGBM model for urban flood susceptibility mapping based on meta-heuristic algorithms

  • 期刊论文

  • 发表日期：

    2026年8月1日

查看全文
  • Haidong Yang,
  • Ting Zou

摘要

Flood susceptibility mapping (FSM) is vital for urban flood risk management. This study first generates a flood inventory by using Sentinel-2 imagery and identifies key conditioning factors. Secondly, a novel FSM method is proposed, namely WHD-LightGBM, which integrates Whale Optimization Algorithm (WOA), Harris Hawk Optimization Algorithm (HHO), and Dung Beetle Optimization Algorithm (DBO) to tune Light Gradient Boosting Machine (LightGBM). Subsequently, based on the SHapley Additive exPlanations (SHAP) method and the partial dependency plot (PDP) method, a new interpretable approach, referred to as SHAP-PDP, is introduced to reveal how predictors influence FSM. Finally, WHD-LightGBM was evaluated on a Sentinel-2-derived flood inventory for Nanchang, China, and compared with seven benchmark models (LightGBM, Grid-LightGBM, Random Forest, XGBoost, WOA-LightGBM, HHO-LightGBM, and DBO-LightGBM). The findings indicate that the FSM based on WHD-LightGBM demonstrates high accuracy and stability (Accuracy = 0.8288, Precision = 0.8506, Specificity = 0.8456, F1-score = 0.8315, Recall = 0.8132); SHAP-PDP analysis identifies altitude, normalized difference vegetation index (NDVI), and rainfall as the most influential conditioning factors in the study area. This study provides a novel approach and methodology for city managers to effectively manage flood disasters.