Scholarlink

<p>Floods, among the most dangerous natural phenomena, can result in substantial loss of life and property. Consequently, precise flood forecasting and effective management are crucial to reducing risks and mitigating their effects. This study focuses on developing a suitable model for predicting hourly flood flows in the northern region of Tehran. Five basins in the area—Kan, Darakeh, Darband, Golabdarreh and Darabad—were thoroughly examined. The Weather Research and Forecasting (WRF) model was initially utilized to simulate flood-producing rainfall events. Following this, machine learning (ML) algorithms such as Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANNs) were applied to predict flood flows in the region. The WRF model, with its downscaling capability, successfully simulated heavy rainfall events in the area with satisfactory accuracy, although it occasionally under- or overestimated the precipitation. The highest precipitation forecasting accuracy of the WRF model, with an average Nash-Sutcliffe Efficiency (NSE) of 0.30 for 22 selected events, was achieved at the Emamzadeh rain gauge station, whereas the lowest accuracy, with an average NSE of -1.73, was observed at the Sangan rain gauge station. Following this, hybrid rainfall-runoff models were developed using the WRF model outputs and the ML algorithms. These models demonstrated high accuracy in predicting flood discharge in all Five basins. Among the applied algorithms, XGBoost produced the most accurate results. WRF-XGBoost with an average NSE of 0.85 across all studied basins showed the highest accuracy in flood flow prediction compared to using Random Forest and ANNs with WRF.</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

Hybrid modeling of WRF and machine learning for enhanced flood forecasting

  • 期刊论文

  • 发表日期：

    2025年7月16日

查看全文
  • Alireza B. Dariane,
  • Hosein Esmaili

摘要

Floods, among the most dangerous natural phenomena, can result in substantial loss of life and property. Consequently, precise flood forecasting and effective management are crucial to reducing risks and mitigating their effects. This study focuses on developing a suitable model for predicting hourly flood flows in the northern region of Tehran. Five basins in the area—Kan, Darakeh, Darband, Golabdarreh and Darabad—were thoroughly examined. The Weather Research and Forecasting (WRF) model was initially utilized to simulate flood-producing rainfall events. Following this, machine learning (ML) algorithms such as Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANNs) were applied to predict flood flows in the region. The WRF model, with its downscaling capability, successfully simulated heavy rainfall events in the area with satisfactory accuracy, although it occasionally under- or overestimated the precipitation. The highest precipitation forecasting accuracy of the WRF model, with an average Nash-Sutcliffe Efficiency (NSE) of 0.30 for 22 selected events, was achieved at the Emamzadeh rain gauge station, whereas the lowest accuracy, with an average NSE of -1.73, was observed at the Sangan rain gauge station. Following this, hybrid rainfall-runoff models were developed using the WRF model outputs and the ML algorithms. These models demonstrated high accuracy in predicting flood discharge in all Five basins. Among the applied algorithms, XGBoost produced the most accurate results. WRF-XGBoost with an average NSE of 0.85 across all studied basins showed the highest accuracy in flood flow prediction compared to using Random Forest and ANNs with WRF.