Scholarlink

<p>Accurate streamflow prediction is of great significance in the management of water resources, yet conventional models fall short of capturing the intricate and non-linear behavior of hydrologic processes. This study provides an advanced method to daily streamflow prediction in the SOUMMAM watershed, northern Algeria, by combining historical flow data, evapotranspiration, and precipitation with certain temporal lags. The hybrid models applied the methods of Variational Mode Decomposition and Empirical Wavelet Transform methods to decompose complex streamflow patterns into simpler patterns before processing. The models evaluated include Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BLSTM), and Bidirectional Gated Recurrent Unit (BGRU), with and without hybrid configurations. Performance evaluation of the models was carried out with regard to multiple metrics at two observation stations (FERMATOU and BOU BIREK). The results indicated that the hybrid models performed remarkably better than the individual models with the VMD-EWT-CNN1 model registering outstanding results with R = 0.98, 0.94–0.95 and KGE = 0.94 at the two observation stations, representing a remarkable improvement. The further analysis with the use of Taylor diagrams and comparison of the time series demonstrated that hybrid models performed remarkably in the representation of low and high flow variability including the occurrence of the peak flow events. The study concludes that the use of hybrid models based on decomposition methods and with the use of CNN architectures has great promise in enhancing the prediction of streamflow with higher accuracies. The models may revolutionize operational hydrological forecasting and lead to more informed water resource management decisions.</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

Optimizing Daily Streamflow Forecasting Under Mediterranean Climate: A Novel Hybrid VMD-EWT and Deep Learning Framework

  • 期刊论文

  • 发表日期：

    2025年10月8日

查看全文
  • Noureddine Daif,
  • Aziz Hebal,
  • Salah Difi,
  • Salim Heddam,
  • Nadjem Bailek,
  • Bilel Zerouali

摘要

Accurate streamflow prediction is of great significance in the management of water resources, yet conventional models fall short of capturing the intricate and non-linear behavior of hydrologic processes. This study provides an advanced method to daily streamflow prediction in the SOUMMAM watershed, northern Algeria, by combining historical flow data, evapotranspiration, and precipitation with certain temporal lags. The hybrid models applied the methods of Variational Mode Decomposition and Empirical Wavelet Transform methods to decompose complex streamflow patterns into simpler patterns before processing. The models evaluated include Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BLSTM), and Bidirectional Gated Recurrent Unit (BGRU), with and without hybrid configurations. Performance evaluation of the models was carried out with regard to multiple metrics at two observation stations (FERMATOU and BOU BIREK). The results indicated that the hybrid models performed remarkably better than the individual models with the VMD-EWT-CNN1 model registering outstanding results with R = 0.98, 0.94–0.95 and KGE = 0.94 at the two observation stations, representing a remarkable improvement. The further analysis with the use of Taylor diagrams and comparison of the time series demonstrated that hybrid models performed remarkably in the representation of low and high flow variability including the occurrence of the peak flow events. The study concludes that the use of hybrid models based on decomposition methods and with the use of CNN architectures has great promise in enhancing the prediction of streamflow with higher accuracies. The models may revolutionize operational hydrological forecasting and lead to more informed water resource management decisions.