Scholarlink

This study presents a comparative analysis of machine learning and deep learning models for predicting Suspended Sediment Concentration (SSC) and solid flow (QS) in the Sebaou Basin, Northern Algeria. Using historical hydrological data collected from 1968 to 2008, three forecasting scenarios were investigated: S1—prediction of QS using water level; S2—prediction of SSC using water level; and S3—prediction of SSC using both water level and solid flow. Two models were employed: Memetic Programming (MP) and Long Short-Term Memory (LSTM). Model performance was assessed using several statistical metrics, including the correlation coefficient (R), Normalized Nash–Sutcliffe Efficiency (NNSE), Mean Square Error (MSE), and Mean Absolute Error (MAE). Across all experiments, the MP model consistently outperformed the LSTM network, exhibiting higher predictive accuracy, greater stability under cross-validation, and reduced sensitivity to data variability. The MP model achieved R values ranging from 0.6 to 0.99 and NNSE values between 0.3 and 0.98, compared to R values of 0.5 to 0.8 and NNSE values between 0.4 and 0.7 for LSTM. In Scenario S3, MP showed significant improvements, highlighting the benefits of multivariate input. This study demonstrates the effectiveness of MP in forecasting SSC, offering valuable insights for watershed management and environmental monitoring.

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

Predicting Solid Flow and Suspended Sediment Concentrations in a Semi-Arid Environment Using Machine Learning and Deep Learning Approaches

  • 章节

  • 发表日期：

    2025年

查看全文
  • Mohamed Nadjib Medfouni,
  • Mohamed Saber,
  • Khaled Korichi,
  • Nadir Marouf,
  • Emad Mabrouk

摘要

This study presents a comparative analysis of machine learning and deep learning models for predicting Suspended Sediment Concentration (SSC) and solid flow (QS) in the Sebaou Basin, Northern Algeria. Using historical hydrological data collected from 1968 to 2008, three forecasting scenarios were investigated: S1—prediction of QS using water level; S2—prediction of SSC using water level; and S3—prediction of SSC using both water level and solid flow. Two models were employed: Memetic Programming (MP) and Long Short-Term Memory (LSTM). Model performance was assessed using several statistical metrics, including the correlation coefficient (R), Normalized Nash–Sutcliffe Efficiency (NNSE), Mean Square Error (MSE), and Mean Absolute Error (MAE). Across all experiments, the MP model consistently outperformed the LSTM network, exhibiting higher predictive accuracy, greater stability under cross-validation, and reduced sensitivity to data variability. The MP model achieved R values ranging from 0.6 to 0.99 and NNSE values between 0.3 and 0.98, compared to R values of 0.5 to 0.8 and NNSE values between 0.4 and 0.7 for LSTM. In Scenario S3, MP showed significant improvements, highlighting the benefits of multivariate input. This study demonstrates the effectiveness of MP in forecasting SSC, offering valuable insights for watershed management and environmental monitoring.