.png){kind=icon}
.png){kind=icon}
Return
A study on sediment forecasting using machine learning with structure-aware mechanisms
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jhydrol.2026.136203.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
•
Proposes a novel structural-aware ML model (ML-P-EF) for sediment transport.
•
Significantly improves sediment peak forecasting and captures nonlinear behaviors.
•
Outperforms traditional models in long-term forecasting of extreme sediment events.
•
Provides a robust tool for sediment management and disaster prevention in river basins.
Abstract
In recent years, machine learning has been widely applied in sediment forecasting; however, existing research has largely focused on the model algorithms themselves, with insufficient integration of the mechanistic characteristics of sediment transport processes (such as the response mechanism of sediment peaks and time-lag effects). This has resulted in limited forecasting capabilities for sudden sediment peaks in complex river basins. To address the challenge of highly abrupt and spatio-temporally variable sediment transport processes in the typical high-sediment- concentration basin of the middle Yellow River, this study proposes a machine learning model for sediment forecasting (ML-P-EF) incorporating a structural enhancement mechanism. Building upon LSTM (Long Short-Term Memory), ANN (Artificial Neural Network) and Transformer models, systematically incorporates Sediment Process Vectorization (SPV), Dynamic Lag Encoding (DLE) and Event-Driven Features (EDF) to enhance the model’s ability to perceive and simulate sediment physical processes. Through validation in typical catchment areas of the Jing River and Beiluo River in the middle reaches of the Yellow River, taking Hongde Station as an example, for a 1-hour forecast horizon, the ANN-P-EF model achieved a Nash efficiency coefficient (NSE) of 0.97 and a Root Mean Square Error (RMSE) of 52.31 kg∙m⁻3, and the peak sediment error was –3.81 %. Under a 6-hour forecast horizon, the NSE of the LSTM-P-EF model improved by more than 40 % compared to the LSTM-P and LSTM models, the RMSE decreased by over 50 %, and the peak error was reduced by 6 %–12 %. This study innovatively integrates the mechanisms of sediment transport processes with deep learning models, enhancing the model’s ability to capture the non-linear response of sediment peaks. It provides a new technical approach to accurate sediment forecasting in high-sediment river basins, with significant implications for regional water and sediment management as well as disaster prevention and mitigation.
Keywords:
Sediment forecasting
Machine learning
Structural awareness mechanisms
Middle Yellow River
Journal
IF:
6.3
Papers:
2.3W
Citations:
9.8W