Toward multi-day-ahead forecasting of suspended sediment concentration using ensemble models

Area of Science:

• Environmental Science
• Hydrology
• Data Science

Background:

• Accurate suspended sediment concentration (SSC) forecasting is crucial for water resource management and environmental protection.
• Traditional artificial neural network (ANN) models often face limitations in multi-step-ahead forecasting accuracy.

Purpose of the Study:

• To enhance ANN-based models for improved multi-step-ahead suspended sediment forecasting.
• To investigate the impact of incorporating both observed and forecasted time series as input variables.
• To develop and evaluate ensemble models for superior forecasting performance.

Main Methods:

• Developed improved ANN models by integrating observed and forecasted time series data.
• Implemented least-square ensemble models combining multiple wavelet-ANN models.
• Evaluated different wavelet families linked with ANN models using error measures.
• Selected the Skagit River as a case study, using daily flow discharge and SSC data.

Main Results:

• Incorporating both observed and predicted variables significantly improved model performance compared to using only observed data.
• Ensemble models consistently outperformed the best single wavelet-ANN model across all lead times.
• The proposed methodology achieved acceptable daily SSC forecasts up to three days in advance.

Conclusions:

• The integration of observed and forecasted data enhances ANN model efficacy for SSC prediction.
• Ensemble modeling provides a superior approach to single-model forecasting for suspended sediment.
• The developed methodology offers a reliable tool for short-term suspended sediment forecasting.