Scholarlink

<p>Accurate prediction of water levels is essential for flood risk mitigation, management of water resources, and disaster preparedness. This study presents a hybrid model that combines long short-term memory (LSTM) networks with a random forest (RF)-based residual correction scheme. The model is developed based on the available data of the water level and the discharge at the Triveni gauging station and rainfall data in three stations surrounding the Triveni station. The performance of the lead times of 1–30 days is assessed with Kling–Gupta efficiency (KGE), mean absolute error (MAE), and root mean squared error (RMSE). Findings indicate that the LSTM-RF combination model consistently gives better results than the standalone LSTM. For instance, at a 1-day lead time, KGE improved from 0.948 (LSTM) to 0.9854 (hybrid), and RMSE decreased from 0.1984 to 0.1089. After a 30-day lead time, the hybrid model had a better accuracy of 0.8774 compared to 0.7904. Graphical plots also demonstrate the increased ability of the proposed model to detect the short and medium-term changes in water levels. There are subtle variations in extreme peak events over extended lead times. This study highlights the potential of integrating hybrid deep learning with ensemble methods to improve hydrological forecasting and sets the stage for future advancements in peak flow predictions. The findings suggest that integrating these methodologies can lead to more accurate and reliable forecasts, particularly in regions with data scarcity.</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

Improved water-level predictions using hybrid long short-term memory networks and random forest techniques

  • 期刊论文

  • 发表日期：

    2026年7月21日

查看全文
  • Rahul Prakash,
  • Joseph Tripura

摘要

Accurate prediction of water levels is essential for flood risk mitigation, management of water resources, and disaster preparedness. This study presents a hybrid model that combines long short-term memory (LSTM) networks with a random forest (RF)-based residual correction scheme. The model is developed based on the available data of the water level and the discharge at the Triveni gauging station and rainfall data in three stations surrounding the Triveni station. The performance of the lead times of 1–30 days is assessed with Kling–Gupta efficiency (KGE), mean absolute error (MAE), and root mean squared error (RMSE). Findings indicate that the LSTM-RF combination model consistently gives better results than the standalone LSTM. For instance, at a 1-day lead time, KGE improved from 0.948 (LSTM) to 0.9854 (hybrid), and RMSE decreased from 0.1984 to 0.1089. After a 30-day lead time, the hybrid model had a better accuracy of 0.8774 compared to 0.7904. Graphical plots also demonstrate the increased ability of the proposed model to detect the short and medium-term changes in water levels. There are subtle variations in extreme peak events over extended lead times. This study highlights the potential of integrating hybrid deep learning with ensemble methods to improve hydrological forecasting and sets the stage for future advancements in peak flow predictions. The findings suggest that integrating these methodologies can lead to more accurate and reliable forecasts, particularly in regions with data scarcity.