A Semi-virtual Watershed Model by Neural Networks

James C.Y. Guo, PhD, P.E. Professor, Department of Civil Engineering, University of Colorado at Denver, Denver, Colorado 80204. E-mail: [email protected] ___________________________________________________________________________________________ Abstract: A semi-virtual watershed model is presented in this study. This model places the design rainfall distribution on the input layer and the predicted runoff hydrograph on the output layer. The optimization scheme developed in this study can train the model to establish a set of weights under the guidance of the kinematic wave theory. The weights are time dependent variables by which rainfall signals can be converted to runoff distributions by weighting procedures only. With the consideration of time dependence, the computational efficiency of virtual watershed models is greatly enhanced by eliminating unnecessary visitations between layers. The weighting procedure used in the semi-virtual watershed model expands the Rational method from peak runoff predictions to complete hydrograph predictions under continuous and non-uniform rainfall events.