ANFIS-ARIMA modelling for scheming re-aeration of hydrologically altered rivers

Abstract

The discharge of a significant amount of wastewater from the urban agglomerations in the rivers is the primary cause for the degradation of the river’s assimilative capacity. The re-aeration process represents an interaction among the air–water interface to absorb oxygen from the atmosphere and indicates the capacity of the water to hold the oxygen that can be used in the degradation of pollutants without affecting the river’s health. Therefore, accurate estimation of re-aeration capacity is the prime requirement to maintain the riverine ecosystem. In this paper, an adaptive neuro-fuzzy inference system (ANFIS) is used to simulate the non-linear behaviour of various water quality parameters by estimating the re-aeration coefficient of Yamuna River, Delhi. The hydraulic and water quality parameters are used to identify the most optimal combination of input parameters for simulation through the Takagi-Sugeno (TS) ANFIS model. The results of ANFIS are ensembled with the time series autoregressive integrated moving average (ARIMA) model to minimize bagging, non-stationarity and biasing. The performance of the ANFIS models was measured using the coefficient of determination (R2), root mean square error (RMSE), and the coefficient of correlation (R). The ARIMA models are evaluated based on standard error, p-value, and t-test. Out of the five different ANFIS models, ANFIS-5 provides the optimal values of R during training and testing of the model as 0.966 and 0.989, respectively, and integrated with the ARIMA model to obtain the re-aeration coefficient for each sampling location. The results indicate that the ensembled ANFIS-ARIMA model reduces the variance significantly with the minimum error of 0.0075 and produces an exceptional improvement in predicting the re-aeration coefficient.