Interdisciplinary application of numerical and machine-learning-based models to predict half-hourly suspended sediment concentrations during typhoons

Abstract

Accurate forecasting of hourly suspended sediment concentration is a critical issue for reservoir management, especially during typhoon periods. This research proposes a two-stage forecasting approach integrating numerical and machine-learning-based models to provide accurate real-time forecasts of half-hourly suspended sediment concentration. In the first stage, the density current model, SRH2D, was verified and applied by the historical typhoon events due to the lack of measured suspended sediment concentration. Next, in the second stage, the calculated results from SRH2D based on the spatial-temporal relation were employed as the input of the MGSVM-based model. Finally, an application in the Shi-Men reservoir was conducted to demonstrate the forecasting performance of the proposed approach. The results indicated that the calculated data from the SRH2D model corresponded to the trend of measured data in spatial-temporal variation. Moreover, the proposed forecasting approach enables reasonably acceptable 0.5- to 3-h ahead forecasts of suspended sediment concentration by using the simulated data of half-hourly suspended sediment concentration. The accurate forecast conducted by the proposed approach is expected to be properly applied to reservoir management.