Probabilistic analysis of water quality deterioration and health risks assessment incorporating machine learning techniques in Mhlathuze catchment, South Africa

Amidst declining water quality issues in South African coastal aquifers, a study was proposed and assessed 40 water samples from the Mhlathuze catchment to evaluate water quality, identify pollution sources, and assess health risks. A machine learning and integrated Positive Matrix Factorization–Monte Carlo Simulation approach were used for probabilistic risk assessment. The average water quality index is good, poor, and unsuitable category ranging from 44 to 406, where 65% of the samples fell under poor to unsuitable for drinking mainly situated in central and eastern regions. The dominance of contamination sources was noted as: seawater intrusion > N-fertilizer application > Leaching from coal stockpile > industrial landfill leachate > smelting industry > mining activities. The probabilistic health risk assessment revealed negligible risk toward children and adults. Mining activities (> 70%) and high N-fertilizer applications (>14%) were found to contribute highest toward human health risks. Metals such as lead, cobalt, and nickel from such sources were found to be dominant and require paid special attention by authorities for future mitigation studies and strategies. This study contributes to the scientific community by providing insights into contamination trends in industrialized coastal aquifers, supporting advancements in water treatment methods while benefiting the global community with informed policies to regulate industrial and agricultural discharge.