Scholarlink

<p>Floods are among the most destructive and common natural disasters, threatening the lives and livelihoods of millions worldwide as climate change worsens. Therefore, creating flood susceptibility maps in watersheds is crucial for hazard management and reducing flood damage. This study aimed to compare machine learning models for flood susceptibility mapping in the Polroud watershed, focusing on preventive management and sustainable development. Three machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), and Decision Tree (DT)—were used with 17 variables across three categories: physiographic, environmental-anthropogenic, and hydrological factors. These included elevation, slope, aspect, curvature, drainage network distance and density, Topographic Wetness Index (TWI), Curve Number (CN), Stream Power Index (SPI), Normalized Difference Vegetation Index (NDVI), Topographic Position Index (TPI), Terrain Ruggedness Index (TRI), land use, geology, precipitation, and proximity to roads. Model performance was assessed using Accuracy, Sensitivity, Specificity, Precision, F1-score, Kappa coefficient, and ROC-AUC. Results showed that the RF model achieved the highest effectiveness, with an accuracy of 0.857, a Kappa coefficient of 0.714, and an AUC of 0.906, in identifying flood-prone zones. Variable importance analysis indicated that distance from drainage networks (80%), elevation (67.41%), precipitation (45.82%), and slope (27.44%) were the most influential factors for flood occurrence. The flood susceptibility map generated by RF revealed that 10.40% of the Polroud watershed, covering 179.04 km, is at hazard of flooding, mainly affecting rural settlements. These findings can inform flood hazard management and regional development 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

Comparative assessment of machine learning models for flood susceptibility mapping

  • 期刊论文

  • 发表日期：

    2026年1月27日

查看全文
  • M. Heydari,
  • A. Fazlollahi,
  • S. P. Nainiva

摘要

Floods are among the most destructive and common natural disasters, threatening the lives and livelihoods of millions worldwide as climate change worsens. Therefore, creating flood susceptibility maps in watersheds is crucial for hazard management and reducing flood damage. This study aimed to compare machine learning models for flood susceptibility mapping in the Polroud watershed, focusing on preventive management and sustainable development. Three machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), and Decision Tree (DT)—were used with 17 variables across three categories: physiographic, environmental-anthropogenic, and hydrological factors. These included elevation, slope, aspect, curvature, drainage network distance and density, Topographic Wetness Index (TWI), Curve Number (CN), Stream Power Index (SPI), Normalized Difference Vegetation Index (NDVI), Topographic Position Index (TPI), Terrain Ruggedness Index (TRI), land use, geology, precipitation, and proximity to roads. Model performance was assessed using Accuracy, Sensitivity, Specificity, Precision, F1-score, Kappa coefficient, and ROC-AUC. Results showed that the RF model achieved the highest effectiveness, with an accuracy of 0.857, a Kappa coefficient of 0.714, and an AUC of 0.906, in identifying flood-prone zones. Variable importance analysis indicated that distance from drainage networks (80%), elevation (67.41%), precipitation (45.82%), and slope (27.44%) were the most influential factors for flood occurrence. The flood susceptibility map generated by RF revealed that 10.40% of the Polroud watershed, covering 179.04 km, is at hazard of flooding, mainly affecting rural settlements. These findings can inform flood hazard management and regional development planning.