Scholarlink

<p>Natural and quasi-natural riverine flood hazard is a recurrent phenomenon in the Himalayan Foreland Basin. The present study utilizes several geospatial datasets to assess flood susceptibility in the Jaldhaka River basin using robust machine learning models, including Gradient Boosting (GB), Support Vector Machine, Random Forest, and Adaptive Boosting (AdaBoost). A total of fourteen flood conditioning factors were considered through multi-collinearity tests, which yielded tolerance and Variance Inflation Factor values. A flood inventory map was then prepared using the Sentinel-1 (SAR) dataset and field verification. However, a total of 335 flooded and 330 non-flooded points are randomly selected based on the derived flood inventory map, which is further categorised into training and testing using 70% and 30% ratio of the total dataset, whereas the ROC-AUC curves of the model's outputs are assessed to evaluate the model's performance. Each model generated different flood susceptibility zones (very low, low, moderate, high, and very high) of the river basin. The results highlighted that the GB model has a higher capability (AUC = 0.945) to detect flood susceptibility zones compared to the other implemented models. The flood susceptibility maps of all four models indicate that the low-lying, flat terrain, mainly in the villages of Ghokashadanga, Mathabhanga, Dinhata, Gitaldaha, Mogolhat, and Kurigram, is classified as high to very high flood-susceptible zones. In this regard, proper land use planning, reduction of human intervention in the river corridor, and flood mitigation strategies should be implemented to reduce this type of geoenvironmental hazard.</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

Flood susceptibility mapping of a tropical transboundary river basin, Jaldhaka, using machine learning algorithms

  • 期刊论文

  • 发表日期：

    2025年11月17日

查看全文
  • Sudipa Sarkar,
  • Biswajit Bera

摘要

Natural and quasi-natural riverine flood hazard is a recurrent phenomenon in the Himalayan Foreland Basin. The present study utilizes several geospatial datasets to assess flood susceptibility in the Jaldhaka River basin using robust machine learning models, including Gradient Boosting (GB), Support Vector Machine, Random Forest, and Adaptive Boosting (AdaBoost). A total of fourteen flood conditioning factors were considered through multi-collinearity tests, which yielded tolerance and Variance Inflation Factor values. A flood inventory map was then prepared using the Sentinel-1 (SAR) dataset and field verification. However, a total of 335 flooded and 330 non-flooded points are randomly selected based on the derived flood inventory map, which is further categorised into training and testing using 70% and 30% ratio of the total dataset, whereas the ROC-AUC curves of the model's outputs are assessed to evaluate the model's performance. Each model generated different flood susceptibility zones (very low, low, moderate, high, and very high) of the river basin. The results highlighted that the GB model has a higher capability (AUC = 0.945) to detect flood susceptibility zones compared to the other implemented models. The flood susceptibility maps of all four models indicate that the low-lying, flat terrain, mainly in the villages of Ghokashadanga, Mathabhanga, Dinhata, Gitaldaha, Mogolhat, and Kurigram, is classified as high to very high flood-susceptible zones. In this regard, proper land use planning, reduction of human intervention in the river corridor, and flood mitigation strategies should be implemented to reduce this type of geoenvironmental hazard.