Predication of Sediment Yield Using Wavelet-Neural Networks

Accurate prediction of watershed sediment yield is important for ecological environment and water resources engineering. Based on the advantages of wavelet analysis and neural networks, a new hybrid model of wavelet transform and BP neural network (wavelet-neural network model, WANN) for predicting the sediment yield, has been suggested in this paper. The established WANN model is applied to make quantitative prediction of annual sediment yield for Hepinggou´s small watershed, located in the southwest of China´s Henan Province. First, the annual sediment yield time series is decomposed and reconstructed into the low-frequency component and the high-frequency components at one-scale level by db2 wavelet, and then both are forecasted respectively with BP neural networks. Finally, the sum of two parts is the predicting result of the annual sediment yield. Results show that the suggested model can improve the forecasting accuracy; it can also be successfully applied to prediction of hydrological time series.