LAUSR

This study focused on a model for predicting debris flow mean velocity. A total of 50 debris flow events were investigated in the Jiangjia gully. A modified radial basis function (MRBF) neural network was developed for predicting the debris flow mean velocity in the Jiangjia gully. A three-dimensional total error surface was used for establishing the predicting model. A back propagation (BP) neural network and the modified Manning formula (MMF) were used as benchmarks. Finally, the sensitivity degrees of five variables that influence debris flow velocity were analyzed. The results show that the mean error and the relative mean error of the 10 testing samples were only 0.31 m/s and 5.92%, respectively. This proves that the MRBF method performed very well in predicting debris flow mean velocity. Gradient of channel and unstable layer thickness have a greater impact on debris flow mean velocity than the other three influencing variables. This proves that the proposed MRBF neural network is reliable in predicting debris flow mean velocity.