A novel integrated rainfall-runoff model based on TOPMODEL and artificial neural network

摘要

Conventional process-based rainfall-runoff models are difficult to catch the non-linear factors and to take full advantages of previous information of rainfall and runoff. However, these factors are closely related to the initial watershed average saturation deficit at each time step. Therefore, in order to address the issue, this study selected the parameter about initial underground flow in TOPMODEL (TOPOgraphic driven Model) as the breakthrough point. Then we used the previous two-day observed runoff and rainfall data as the inputs of an artificial neural network(ANN) and initial subsurface flow of present day as an output, then integrated ANN into runoff generation module in TOPMODEL and finally applied the integrated model for daily runoff modeling in Yingluoxia watershed with 10009 km~2, China. In addition, this work also utilized particle swarm optimization technique (PSO) to avoid the local optimization, especially for the integration of black-box and physical models. The result shows that during the validation period the Nash-Sutcliffe efficiency coefficient (NE) and root mean square error (RMSE) of TOPMODEL are 0.45 and 3.88×10~(-4) m respectively while the NE of 0.70 and RMSE of 2.85×10~(-4) m for the integrated model. Significantly, the integrated model performs much better than the traditional model. Hence, this new method of integrating ANN with the runoff generation module of TOPMODEL is promising and easily extended to other process-based rainfall-runoff models as well.