[A-HW20] Integrated Watershed Management under the Future Extreme Disaster

[AHW20-P03] Development of a multi-prediction model to investigate sediment flux in the Reservoir

Cheng-Chia Huang², *YU-JIA SU¹ (1.Postgraduate,Department of Water Resources Engineering and Conservation,Feng Chia University,Taiwan (R.O.C), 2.Assistant Professor,Department of Water Resources Engineering and Conservation,Feng Chia University,Taiwan(R.O.C))

Keywords:multi-prediction model, numerical model, machine learning, empirical formula , Shihmen Reservoir, sediment discharge

The massive sediment produced by extreme flooding often causes deposition problems in reservoirs and disturbs water resources management. To grasp the reservoir sedimentation issue, this study selects the numerical model, machine learning, and empirical formula to establish a multi-prediction model. A representative reservoir, Shihmen Reservoir was adopted to be the field site case due to the serious deposition issue. This study collects watershed precipitation, inflow material, measuring station along the reservoir, and outflow material of each sluice gate to be the training and testing data. Firstly, we collect historical typhoon cases and employ numerical models to fill in missing data, providing sufficient data for machine learning input. Subsequently, we utilize XGBoost to predict sediment concentration at different lead times, ensuring acceptable accuracy by using different evaluation indexes. Lastly, an empirical formula is developed for the reservoir and is applied to forecast the sediment discharge through different sluice gates. Through the integration of machine learning and traditional empirical formulas, we effectively assess the spatiotemporal variations in sediment discharge, serving as a reference for reservoir operations.