Piecewise prediction model for watershed-scale erosion and sediment yield of individual rainfall events on the Loess Plateau, China

Establishing a universal watershed-scale erosion and sediment yield prediction model represents a frontier field in erosion and soil/water conservation. The research presented here was conducted on the Chabagou watershed, which is located in the first sub-region of the hill-gully area of the Loess Plateau, China. A back-propagation artificial neural model for watershed-scale erosion and sediment yield was established, with the accuracy of the model, then compared with that of multiple linear regression. The sensitivity degree of various factors to erosion and sediment yield was quantitatively analysed using the default factor test. On the basis of the sensitive factors and the fractal information dimension, the piecewise prediction model for erosion and sediment yield of individual rainfall events was established and further verified. The results revealed the back-propagation artificial neural network model to perform better than the multiple linear regression model in terms of predicting the erosion modulus, with the former able to effectively characterize dynamic changes in sediment yield under comprehensive factor conditions. The sensitivity of runoff erosion power and runoff depth to the erosion and sediment yield associated with individual rainfall events was found to be related to the complexity of surface topography. The characteristics of such a hydrological response are thus closely related to topography. When the fractal information dimension is greater than the topographic threshold, the accuracy of prediction using runoff erosion power is higher than that of using runoff depth. In contrast, when the fractal information dimension is smaller than the topographic threshold, the accuracy of prediction using runoff depth is higher than that of using runoff erosion power. The developed piecewise prediction model for watershed-scale erosion and sediment yield of individual rainfall events, which introduces runoff erosion power and runoff depth using the fractal information dimension as a boundary, can be considered feasible and reliable and has a high prediction accuracy. Copyright (c) 2013 John Wiley & Sons, Ltd.