Predicting water quality and water treatment operations under future climate scenarios using machine learning models

Climate change is altering the water quality through changes in temperature, rainfall patterns, and weather events. Predicting future water quality becomes crucial as the impacts of climate change intensify, altering surface water quality and subsequent treatment operations. This study developed predictive models that were trained on climate parameters to predict inflow water quality and treatment plant operations using Multivariate Linear Regression, XGBoost, and Nonlinear Autoregressive with Exogenous Inputs (NARX). The NARX model consistently demonstrated superior performance across most water quality and treatment-operation parameters, achieving R² values of 0.55 (Electricity Usage) to 0.99 (Reservoir Level) and low error metrics; however, it shows limited capability to capture electricity usage during treatment processes, suggesting the influence of additional operational factors. Further, the model is used to predict future water quality and operations based on bias-corrected CORDEX climate data for near (2026-2050), mid (2051-2075), and far (2076-2100) future under three representative pathways: RCP2.6, RCP4.5, and RCP8.5. The prediction results reveal 3% to 5% reduction in turbidity in the RCP2.6 scenario and 4% to 25% increase in the RCP4.5 and RCP8.5 scenarios, with a higher probability of exceeding 50 NTU during monsoons, resulting in a 110% increase in coagulant use. The moderate emissions scenario exhibits non-monotonic patterns, characterised by 3%-7% reductions in alkalinity. The high-emissions scenario exhibits severe deterioration, with alkalinity dropping to 10%-12% and frequently falling below 50 mg/L. Overall, water quality is expected to worsen under higher emissions, though the low- and mid-emission scenarios show mixed results, with some parameters improving while others deteriorate.