Machine learning approaches for predicting water quality towards climate-resilient groundwater management in southern India

Climate studies identified many regions in India as water-stressed, effective prediction of the quality and quantity of groundwater is, therefore, necessary for climate-resilient, sustainable groundwater management. In this study, machine learning (ML) models are applied to predict the drinking water quality index (DWQI) of groundwater from site measurements of hydrochemical data. The Central Telangana site, located in Southern India, exhibited increasing levels of contamination and steeply falling water levels due to increased irrigation needs and urbanisation. ML models, such as random forest (RF) and multi-linear regression (MLR), were employed for modelling DWQI. In addition, the accuracy of the models was evaluated for the seasonal variations in chemical data. The results suggest that MLR provides better predictions and is applicable across seasons, while RF models are constrained by seasonal variations. The accuracy achieved by MLR was: correlation close to unity and errors of the order of 10−5, and by RF was as follows: correlation more than 0.99, and median error around 5.5. The outcome of this study demonstrates the ability of ML models in predicting the DWQI of groundwater in hard-rock aquifers, which often exhibit complex dependency of the DWQI on the variables and lack apparent correlation.