Spatiotemporal Predictions of BOD Levels in River Ganga: Use of NARX Model

Abstract Rivers finding their path through different hamlets, industries and residential etc. are the end receivers of liquid and solid waste. This plight of water bodies has been deteriorating the life of the river stretches and greatly impacting the quality of water. The identification of potentially polluted locations and timely action can help in restoring the water quality of the river stretches and improving the life of water body as well. The dynamics of river Ganga stretches has been modeled using Non-linear Autoregressive network with exogenous inputs (NARX) model spatially and temporally simultaneously. This method helps in developing one single model to understand all the stretches simultaneously which is not possible otherwise. The change in behavior of NARX model under consecutive and non-consecutive time period has also been studied using statistical methods and multi-step ahead predictions. The model developed in this study captures all the seasons together in one single model and has been used to predict BOD values at seven different locations for nine months ahead. The results highlight the performance ability of NARX model to understand future water quality changes of river Ganga stretch owing to continuous pollution adding to the river. The results predicted for 9 months ahead for year 2015 using two different models, a lower root mean square error (RMSE) as 0.026 and higher correlation coefficient ‘r’ as 0.992 was obtained for model with consecutive days while for other model with non-consecutive days the RMSE was 0.031 and r was 0.989. This information may provide some guidance to the policy makers and water managers to prepare and suggest the pollution mitigation measures for the life line of millions; River Ganga.