Improving TOPMODEL performance in rainfall-runoff simulating based on ANN

TOPMODEL is a simple physically based rainfall-runoff model and has become increasingly popular and widely used in various applications in recent years. However, it performs worse than the Artificial Neural Network (ANN)-based rainfall-runoff models in stream flow prediction. In order to overcome this weakness inherent in TOPMODEL, a new approach based on ANN and TOPMODEL is proposed in the present study. The present approach uses the output of an ANN-based rainfall-runoff model in validation period as the 'observed discharge' to calibrate the parameters of TOPMODEL. The calibrated TOPMODEL is then directly employed for stream flow prediction, rather than experienced traditional two stages: calibration period and validation period. To test the new method, Baohe River basin (2413 km(2)), located at the upper stream of the Hanjiang Catchment in Yangtze River Basin, China, is selected as the study area. The results show that the daily stream flows simulated by the new approach are in general agreement with the observed ones, while the daily stream flows simulated by the traditional one, i.e. only using TOPMODEL for stream flow predictions, greatly overestimates some peak flows. And the new method resulted in a Nash and Sutcliffe efficiency coefficient value of 0.764, which is significantly larger than that of the traditional one, which suggests that the new approach combining the advantages of ANN and TOPMODEL is more suitable for daily stream flow forecasting.