Scholarlink

<p>Accurate forecasting of water discharge and water level is essential for managing water resources effectively, particularly in active river basins like the Mahanadi River in India. Traditional hydrological models often struggle with the complex temporal relationships inherent in hydrological data. This study evaluates LSTM-GRU-Attention (LSTM-GRU-ATT) models for predicting water discharge and water level on the Mahanadi River. The aim is to assess model performance across varying window sizes and measure their ability to capture temporal dependencies in river flow dynamics. LSTM-GRU-ATT models combine Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures with attention mechanisms. Models are trained and evaluated using historical data, with 1, 7, 14, and 30 days window sizes. Performance metrics, including Mean Squared Error (MSE), Mean Absolute Error (MAE), Nash-Sutcliffe Efficiency (NSE), and Percent Bias (PBIAS) are computed to evaluate model accuracy. For water discharge prediction, a window size of 14 yields the lowest MSE of 0.000394 and the highest NSE of 0.413. However, for water level prediction, a window size 30 achieves the lowest MSE of 0.0081 and the highest NSE of 0.9182. LSTM-GRU-ATT models significantly capture temporal dynamics and enhance accuracy in predicting water discharge and water level on the Mahanadi River. These findings underscore their potential utility in hydrological modelling for improved water resource management and flood 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

Hydrological Predictive Modeling for Indian River: Leveraging LSTM and GRU Attention Mechanisms

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Sagar Lachure,
  • Ashish Tiwari

摘要

Accurate forecasting of water discharge and water level is essential for managing water resources effectively, particularly in active river basins like the Mahanadi River in India. Traditional hydrological models often struggle with the complex temporal relationships inherent in hydrological data. This study evaluates LSTM-GRU-Attention (LSTM-GRU-ATT) models for predicting water discharge and water level on the Mahanadi River. The aim is to assess model performance across varying window sizes and measure their ability to capture temporal dependencies in river flow dynamics. LSTM-GRU-ATT models combine Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures with attention mechanisms. Models are trained and evaluated using historical data, with 1, 7, 14, and 30 days window sizes. Performance metrics, including Mean Squared Error (MSE), Mean Absolute Error (MAE), Nash-Sutcliffe Efficiency (NSE), and Percent Bias (PBIAS) are computed to evaluate model accuracy. For water discharge prediction, a window size of 14 yields the lowest MSE of 0.000394 and the highest NSE of 0.413. However, for water level prediction, a window size 30 achieves the lowest MSE of 0.0081 and the highest NSE of 0.9182. LSTM-GRU-ATT models significantly capture temporal dynamics and enhance accuracy in predicting water discharge and water level on the Mahanadi River. These findings underscore their potential utility in hydrological modelling for improved water resource management and flood forecasting.