Flow forecasting models using hydrologic and hydrometric data

Abstract

River flow forecasting is important for successful water resources planning and management. The current study investigated the applicability of the artificial neural network (Ann), adaptive neuro-fuzzy inference system (Anfis), wavelet-Ann (Wann) and wavelet-Anfis (Wanfis) for daily river flow forecasting in Karaj River. Three scenarios were used. In the first scenario, meteorological data were used as input variables for model development. In the second, flow discharge data were used, while the third scenario used a combination of scenarios 1 and 2 as input variables for model development. For each scenario, different combinations of time series were considered. In the Wann and Wanfis models, the effective sub-time series components obtained by discrete wavelet transform were used as the new inputs to Ann and Anfis models to predict daily river flow time series. Wann was found to be the best technique for daily flow prediction throughout this study. The results indicate that the best Wann model (R2test = 0·993, RMSEtest = 0·004) outperformed the best Wanfis model (R2test = 0·976, RMSEtest = 0·008). The results show that only the input structures of scenarios 2 and 3 are efficient input structures for flow prediction in the study area. It was also shown that Wann models are capable of delivering an accurate flow prediction up to the next 4 d.