Comparison of ANNs and empirical approaches for predicting watershed runoff

Prediction of watershed runoff resulting from precipitation events is of gr eat interest to hydrologists. The nonlinear response of a watershed tin ter ms of runoff) to rainfall events makes the problem very complicated. In add ition, spatial heterogeneity of various physical and geomorphological prope rties of a watershed cannot be easily represented in physical models. In th is study, artificial neural networks (ANNs) were utilized for predicting ru noff over three medium-sized watersheds in Kansas. The performances of ANNs possessing different architectures and recurrent neural networks were eval uated by comparisons with other empirical approaches, Monthly precipitation and temperature formed the inputs, and monthly average runoff was chosen a s the output. The issues of overtraining and influence of derived inputs we re addressed. It appears that a direct use of feedforward neural networks w ithout time-delayed input may not provide a significant improvement over ot her regression techniques. However, inclusion of feedback with recurrent ne ural networks generally resulted in better performance.