Scholarlink

<p>Floods are among the most destructive natural disasters, threatening lives, infrastructure, and ecosystems. This study investigates flood susceptibility in Longnan, a hilly county of Jiangxi Province, China, using artificial intelligence (AI). We selected ten geographic and climatic factors—such as topography and precipitation—and identified five key contributors to flood risk: land use/cover (LULC), sediment transport index (STI), stream power index (SPI), slope (SLOP), and precipitation (PRE). Six AI models were evaluated, including Gradient Boosting (GB), AdaBoost (ADA), Random Forest (RF), Extra Trees (ET), Multi-Layer Perceptron (MLP), and Support Vector Classification (SVC). GB performed best, with an AUC of 0.92 and overall accuracy of 0.94. The GB model identified 45,509 raster cells (1.13% of the study area) as highly flood-prone, mostly alone the Taojiang River and in low-lying regions. Chenglong, Longnan, and Dujiang towns were found at highest risk. These results demonstrate AI’s potential to improve our understanding of flood dynamics and support targeted mitigation in complex hilly environments.</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

Factors’ feature optimization and flood susceptibility mapping in hilly regions: an Artificial Intelligence approach

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • Guangxu Liu,
  • Zhiwei Wan,
  • Haipei Liu,
  • Baolei Li,
  • Lihong Meng,
  • Zhen Hu,
  • Yingmin Liu

摘要

Floods are among the most destructive natural disasters, threatening lives, infrastructure, and ecosystems. This study investigates flood susceptibility in Longnan, a hilly county of Jiangxi Province, China, using artificial intelligence (AI). We selected ten geographic and climatic factors—such as topography and precipitation—and identified five key contributors to flood risk: land use/cover (LULC), sediment transport index (STI), stream power index (SPI), slope (SLOP), and precipitation (PRE). Six AI models were evaluated, including Gradient Boosting (GB), AdaBoost (ADA), Random Forest (RF), Extra Trees (ET), Multi-Layer Perceptron (MLP), and Support Vector Classification (SVC). GB performed best, with an AUC of 0.92 and overall accuracy of 0.94. The GB model identified 45,509 raster cells (1.13% of the study area) as highly flood-prone, mostly alone the Taojiang River and in low-lying regions. Chenglong, Longnan, and Dujiang towns were found at highest risk. These results demonstrate AI’s potential to improve our understanding of flood dynamics and support targeted mitigation in complex hilly environments.