Flood hydrograph modeling using artificial neural network and adaptive neuro-fuzzy inference system based on rainfall components

Abstract

Different limitations such as the lack of enough hydrometric stations, difficulty in collecting hydrometric data with costly data collection are caused to create hydrologic models for estimating the flood hydrograph. Based on the easy and more access to rainfall statistics, preparing the hydrologic model based on rainfall characteristics and data seems to be the very applicable and logical method. Data-driven models have increasingly been used to describe the behavior of hydrological systems, which can be used to complement or even replace physical-based models. In this study, the efficiency of two data mining models including artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) was evaluated in order to model flood hydrograph characteristics based on rainfall components in Kasilian Watershed, northern Iran. For this purpose, fifteen characteristics of rainfall (hyetograph) and eight characteristics of flood hydrograph were respectively considered independent and dependent variables for 60 rainfall-runoff events from 1975 to 2009. ANN with two functions of hyperbolic tangent and sigmoid and ANFIS with grid partitioning and subtractive clustering were used to estimate flood hydrograph. Variance inflation factor (VIF) (for selecting variables that are minimal multicollinearity) were used to select the input variables. ANFIS model with the grid partitioning method performs better than the ANFIS model with the subtractive clustering method. ANFIS with Nash-Sutcliff efficiency (NSE) of 0.87, root mean squared error (RMSE) of 0.38 m3/s, and deviation of peak time of observed and estimated hydrographs (DPOT) of 4.33 h was found to be superior to ANN with NSE of 0.40, RMSE of 0.88 m3/s, and DPOT of 1.14 h accurately and efficiently for modeling flood hydrograph. Therefore, ANFIS model is proposed for modeling the flood hydrograph based on rainfall characteristics.