Scholarlink

<p>River flow is a crucial component in water resource management, environmental protection, and disaster mitigation. Also, the Mississippi River, one of the largest and most significant rivers in North America, plays an essential role in shaping the region’s infrastructure and economy. Moreover, reliable flow forecasting is essential for several key purposes, including flood prevention, optimal water resource allocation, and ecosystem preservation, all of which are critical to promoting sustainable development and enhancing disaster resilience in the region. This study aims to predict the discharge of the Mississippi River at the Memphis station for the period from 1990 to 2024 by employing hybrid models that integrate short-term long-term memory (LSTM) with random forest (RF) and neural network (CNN), considering lag intervals of 3–15&#xa0;days. Various evaluation criteria were utilized to assess the accuracy of these models. The performance evaluation revealed that a three-day lag interval produced the most accurate results, with the CNN-LSTM model achieving the best performance, with NRMSE = 0.0165, at the Memphis station. Additionally, the RF-LSTM and CNN-RF-LSTM models demonstrated high accuracy in predicting daily discharge, with NRMSE = 0.0179, 0.0177, respectively. These findings have significant implications for water resource managers, providing enhanced efficiency in reducing labor costs and saving time in forecasting.</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

CNN-LSTM-RF integration for predicting Mississippi River discharge dynamics

  • 期刊论文

  • 发表日期：

    2025年10月11日

查看全文
  • Fariborz Ahmadzadeh Kaleybar,
  • Ahad Molavi

摘要

River flow is a crucial component in water resource management, environmental protection, and disaster mitigation. Also, the Mississippi River, one of the largest and most significant rivers in North America, plays an essential role in shaping the region’s infrastructure and economy. Moreover, reliable flow forecasting is essential for several key purposes, including flood prevention, optimal water resource allocation, and ecosystem preservation, all of which are critical to promoting sustainable development and enhancing disaster resilience in the region. This study aims to predict the discharge of the Mississippi River at the Memphis station for the period from 1990 to 2024 by employing hybrid models that integrate short-term long-term memory (LSTM) with random forest (RF) and neural network (CNN), considering lag intervals of 3–15 days. Various evaluation criteria were utilized to assess the accuracy of these models. The performance evaluation revealed that a three-day lag interval produced the most accurate results, with the CNN-LSTM model achieving the best performance, with NRMSE = 0.0165, at the Memphis station. Additionally, the RF-LSTM and CNN-RF-LSTM models demonstrated high accuracy in predicting daily discharge, with NRMSE = 0.0179, 0.0177, respectively. These findings have significant implications for water resource managers, providing enhanced efficiency in reducing labor costs and saving time in forecasting.