Scholarlink

<p>Accurate prediction of sediment transport is crucial for the sustainable management of water resources and the protection of hydraulic infrastructure, particularly in regions with high hydrological variability. This study introduces an innovative hybrid model, RF M5 (CNN–LSTM), which combines Random Forests (RF), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) to simulate the spatio-temporal variability of sediment transport in the Cheliff Basin (43,750&#xa0;km², northern Algeria). The model was trained and validated using data from four hydrometric stations along the Cheliff River. Compared to conventional approaches (SVR, MLR, KNN), the hybrid model demonstrated clear superiority, achieving high R² and NSE values, low MAE, minimal PBIAS, and a robust KGE (KGE &gt; 0.6). Although RMSE was occasionally higher, reflecting sensitivity to flood events, the overall reliability of the model remained strong. In contrast, SVR and MLR showed poor performance (low or negative R²), while KNN proved moderately effective. Seasonal analysis revealed that maximum sediment transport occurred in spring (60%) and autumn (44%), with more than 80% of variability linked to fluctuations in water discharge. The estimated mean annual specific sediment yield was 2,587 t/km². These findings confirm the efficiency and reliability of the RF M5 (CNN–LSTM) model for sediment transport forecasting and hydrological planning in semi-arid Mediterranean catchments. They also open perspectives for integrating large-scale climate variables and testing the model in other contexts to strengthen its generalizability.</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

Hybrid machine learning model for sediment transport forecast in the cheliff Basin, Northern Algeria

  • 期刊论文

  • 发表日期：

    2026年1月9日

查看全文
  • Rachid Mokhtari,
  • Faiza Hallouz,
  • Brahim Habibi,
  • Mohamed Meddi

摘要

Accurate prediction of sediment transport is crucial for the sustainable management of water resources and the protection of hydraulic infrastructure, particularly in regions with high hydrological variability. This study introduces an innovative hybrid model, RF M5 (CNN–LSTM), which combines Random Forests (RF), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) to simulate the spatio-temporal variability of sediment transport in the Cheliff Basin (43,750 km², northern Algeria). The model was trained and validated using data from four hydrometric stations along the Cheliff River. Compared to conventional approaches (SVR, MLR, KNN), the hybrid model demonstrated clear superiority, achieving high R² and NSE values, low MAE, minimal PBIAS, and a robust KGE (KGE > 0.6). Although RMSE was occasionally higher, reflecting sensitivity to flood events, the overall reliability of the model remained strong. In contrast, SVR and MLR showed poor performance (low or negative R²), while KNN proved moderately effective. Seasonal analysis revealed that maximum sediment transport occurred in spring (60%) and autumn (44%), with more than 80% of variability linked to fluctuations in water discharge. The estimated mean annual specific sediment yield was 2,587 t/km². These findings confirm the efficiency and reliability of the RF M5 (CNN–LSTM) model for sediment transport forecasting and hydrological planning in semi-arid Mediterranean catchments. They also open perspectives for integrating large-scale climate variables and testing the model in other contexts to strengthen its generalizability.