A Novel Approach to Pollution Source Identification in Groundwater Using Hybrid Ensemble of Wavelet Neural Network (WNN) and ARIMA

Abstract

The groundwater is a significant source of readily available freshwater. However, these groundwater aquifers are susceptible to contamination by the presence of many potential pollution sources. These potential sources may be active at any location for any duration with an unknown magnitude of source fluxes. The identification of these potential sources is necessary for remediation and management. In this work, the groundwater contamination sources are identified for two probable situations, i.e., Error-free concentration measurements and erroneous concentration measurements. The proposed methodology exploits the capability of the multistage approach of recent hybrid soft computing models (ensemble of Wavelet Neural Network [WNN]; Autoregressive Integrated Moving Average [ARIMA]) to identify potential pollution sources effectively. The models are trained by the simulated concentrations for randomly generated source fluxes by Groundwater Modelling System (GMS). The multistage approach has been adopted here, which includes three-stage models. To exploit the effectiveness of WNN and ARIMA, they are used at different stages in the multistage models, thereby developing various hybrid ensembles. Total of four hybrid models are developed utilising the multi-staging approach of WNN and ARIMA models, Namely Hybrid-II, III, IVA, and IVB. Predicted results from these models were compared on the same data sets. They are compared using various performance indices, such as RMSE, NE, correlation coefficient, etc. Error values of Hybrid II ensemble (employing WNN at first stage and ARIMA at second stage) are lowest as compared to only WNN and other Hybrid models. Hence, the Hybrid-II model can be utilised to accurately predict potential contamination sources in any sample area in groundwater.