Runoff and Sediment Estimation Using ANN and ANFIS: Case Study of Godavari Basin, India

Knowledge of stage-discharge and runoff-sediment relationships are extremely important for the planning and management of water resources. This study focuses on the issues of water resources management through assessment of the runoff and sediment vulnerability in the Godavari basin. To estimate the runoff and suspended sediment daily data of stage, runoff, and sediment for monsoon period (from 1^st June to 30^th September) of 1995 to 2010 were explored using artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) techniques for Pathagudem and Polavaram sites located in Chhattisgarh and Andhra Pradesh, India. Gamma test (GT) was used to select the best combination of input variables for both runoff and sediment prediction. Selected input combinations were then used as input vectors of ANN and ANFIS models for both runoff and sediment estimation. In the case of ANN models, back-propagation algorithm, and log sigmoid activation while in ANFIS models, triangular, trapezoidal, generalized bell, and Gaussian membership functions (MF) were used to train and test the models. The correlation coefficient (CC), root mean square error (RMSE), coefficient of efficiency (CE), and pooled average relative error (PARE) were utilized to validate the developed models. ANFIS model (Gauss, 3) with CC, RMSE, CE, and PARE values of 0.987, 299.75 m³/sec, 0.98 and –0.005 for runoff prediction and ANFIS model (Triangular, 3) with CC, RMSE, CE, and PARE values of 0.836, 0.159 g/l, 0.918 and –0.0074 for sediment prediction were selected as the best performing models at the Pathagudem site. While for the Polavaram, ANFIS model (Triangular, 2623) with CC, RMSE, CE, and PARE values of 0.994, 859.93 m³/sec, 0.995 and –0.0011 and ANFIS model (Gauss, 3) with CC, RMSE, CE, and PARE values of 0.939, 0.113 g/l, 0.966, –0.0036 were better than other ANN and ANFIS-based models for runoff and sediment prediction, respectively. The sensitivity analysis indicated that the current day stage is the most sensitive parameter for runoff prediction and current-day runoff is the most sensitive parameter for sediment prediction at both sites.