Modeling the daily suspended sediment concentration in a hyperconcentrated river on the Loess Plateau, China, using the Wavelet–ANN approach

Abstract

The complexity of the relationship between suspended sediment concentration (SSC) and river discharge (Q) remains a challenge for SSC prediction in hyperconcentrated rivers. In this study, the wavelet-artificial neural network model (WANN) was built to predict SSC in the Kuye River, a representative hyperconcentrated river in the middle Yellow River catchments of China. In the WANN model, the observed daily time series for Q and SSC of 2193days (from 1967 to 1972) were decomposed into subseries at different scales using discrete wavelet analysis. Then, the effective subseries were selected to construct Q/SSC inputs to the feed-forward back-propagation artificial neural network (BP ANN) to predict SSC 1day in advance (the time resolution of the observed data). The coefficient of determination (R2) and root-mean square error (RMSE) were adopted to evaluate the model's performance. The WANN model showed higher prediction accuracy (R2=0.846 and RMSE=29.82) than the sediment rating curve (SRC) model (R2=0.537 and RMSE=55.40) or the ANN model (R2=0.664 and RMSE=43.13). The WANN model exhibited more robust performance than the SRC and ANN models, indicated by the appropriate values of error autocorrelation and input-error correlation. Negative values of predicted SSC occurred in ANN and in WANN models. By adjusting the negative values to zero, the WANN R2 was improved by 4.3% from 0.846 to 0.882. In general, the results illustrate that the WANN model better predicts SSC in a hyperconcentrated river setting, with highly nonlinear and nonstationary time series.