Semidistributed Form of the Tank Model Coupled with Artificial Neural Networks

Abstract

Traditional conceptual rainfall–runoff models in the lumped form are usually developed without consideration of the spatial variation of rainfall and the heterogeneity of the watershed geomorphological nature. As an improvement to traditional conceptual models of the lumped form, a semidistributed form of the Tank model coupled with artificial neural networks (ANNs) is proposed herein. As a result, the effect of spatial variations of rainfall and model parameters can be investigated by dividing the entire catchment into a number of subcatchments and applying the spatially varied rainfall inputs and parameters to each subcatchment. Furthermore, in contrast to the linear summation commonly used in watershed routing that usually regards the total simulated runoff at the entire catchment outlet as a linear superposition of the routed runoff from all individual subcatchments, artificial neural networks are employed to explore nonlinear transformations of the runoff generated from the individual subcatchments into the total runoff at the entire watershed outlet. As illustrated in this study, coupling ANNs with traditional conceptual models reveals a promising new approach to catchment rainfall-runoff modeling.