Stream Flow Forecasting by Artificial Neural Network and TOPMODEL in Baohe River Basin

Abstract

Black box-based ANN (Artificial Neural Network) models and the process-based model TOPMODEL have been increasingly applied to various water resources system problems in recent years. One of main focuses in this work is to develop ANN models for daily stream flow forecasting and determine a suitable combination of input variables and a more accurate architecture in the design phase. Another focus is to compare the performance of ANN models and TOPMODEL in one day ahead stream flow forecasting. Baohe River basin, with a humid climate, is selected as the study area. The results show that ANN models with flow data plus precipitation data as the input variables perform much better than that with only precipitation data or only flow data as the input variables. The performance of ANN models will be slightly reduced if evaporation data are added into the input vector. ANN has a very good performance against the TOPMODEL in terms of Nash-Sutcliffe efficiency. Nevertheless, they both can not capture the main peak flow: ANN underestimates the main peak flows while TOPMODEL overestimates two or three peak flows in validation years.