Hydrochemical Signatures, Common Pollutants and Modified Water Quality Index Using Machine Learning Model in Central Ganga Plain, India

The study highlights a fresh, moderately hard, and slightly alkaline groundwater regime with marked variation in time and space. Hydrochemical analyses indicate a unique chemical environment masked with silicate weathering, dissolution of HCO3− and SO42−, fertilizers, and sewage pollution controlling the groundwater chemistry. Concomitant occurrences of SO42− and NO3− are controlled by multiple common pollution sources. Elevated NO3− concentrations are found to be positively correlated with trace elements like Mn, As, Zn, and U, facilitated by oxidation of organic matter and reductive dissolution of metal oxides. Natural Background Levels (NBLs) for NO3− were 7.81 mg/L (dry season) and 21.87 mg/L (wet season), while those for Mn were 15.17 µg/L (dry season) and 10.88 µg/L (wet season), respectively. Seasonal NBL changes (+ 2.8 times for NO3−, -1.4 times for Mn) highlight their distinct mobility patterns influenced by various factors, including rainfall recharge, irrigation return flows, fertiliser application, aquifer properties, etc. A new method, Water Quality Index- Guideline Ratio (WQIGR), is employed that removes the bias of the existing WQI models in weights’ calculation. The WQIGR indicates a contaminant load increase during the wet season, with poor quality clusters linked to excess NO3 and Mn. Subsequent machine learning based modelling of WQI was performed on 174 samples in an 80:20 ratio for training and validation, respectively. The model proved efficient in WQI prediction with an R2 score of 0.93, MAE RMS of 3.3991, and MLE RMS of 5.3826. The study recommends controlled fertilizer, adequate waste management, and improved sewerage systems for safe and sustainable groundwater management. Based on the graphical abstract, the study involved sampling, data acquisition, and analyses for pre- and post-monsoon seasons, considering both spatial and chemical controls, to better understand the underlying mechanisms that influence the occurrence and variation of these key elements, taking into account the hydrogeochemical environment. Hydrochemical parameters, specifically pH and Electrical Conductivity (EC), were measured in situ immediately, and Subsequent groundwater analyses were conducted in the laboratory following standardized procedures outlined by the American Public Health Association (APHA 1995). The results of the chemical analysis were processed and interpreted using numerous standard plots, like the Piper-Trilinear plot, which reveals a dominant Ca-HCO3 facies, transitioning to Ca-Mg-HCO3-SO4 and Na-Cl along groundwater flow paths. Bivariate plots indicate silicate weathering, anthropogenic input, and the dissolution of HCO3− and SO42−, primarily control groundwater chemistry. The study highlights the control of hydrochemical facies on the groundwater salinity, trace elemental concentration, and its variance with changing hydrodynamics. Natural background level (NBLs) of groundwater contaminants is a novel approach that utilizes the inflection points in cumulative probability functions of hydrochemical species to distinguish between different groups of population and ascertain their genesis to natural and or anthropogenic processes The baseline pollution was established to be 7.81 mg/L & 21.87 mg/L for NO3− and 15.17 µg/L & 10.88 µg/L for Mn in dry and wet season respectively. A new method, Water Quality Index- Guideline Ratio (WQIGR), is proposed that removes the bias of the existing WQI models in weights’ calculation. The WQI map shows that contamination due to trace elements is low in the area; however, even in low quantities, the trace elements are capable of damaging the groundwater by working as catalytic pathways for the dissolution of harmful major ions such as NO3− and Mn. A Sequential ANN model was developed with six fully connected hidden layers and trained for 100 epochs using a batch size of 16. This study demonstrates exceptional performance in predicting the Water Quality Index, showcasing high generalisation ability and reliability. The low error metrics and the robust visual correlation between actual and predicted values render it highly suitable for real-time environmental monitoring systems. Ca-HCO3− is the dominant hydrochemical facies, altering into a mix-type facies. Facies alteration reports an increase in groundwater salinity and trace elemental load. NO3−, SO42−, Mn, and Zn emerge as the dominant anthropogenic pollutants. Modified WQI incorporating revised weights classifies 25%to 16% samples in the marginal to poor category. ANN model predicts water quality index reliability with an R2 score of 0.934.