Integrated Hydrochemical Assessment and Machine Learning-Based Spatial Prediction of Groundwater Quality in a Coastal Aquifer System of Bangladesh

Abstract This study investigates groundwater quality in the coastal aquifers of Patenga, Chattogram, using an integrated approach combining hydrochemical analysis, spatial mapping, and machine learning (ML) techniques. To compare groundwater characteristics, a comparative assessment between shallow tube wells (STWs) and deep tube wells (DTWs) was conducted. The results reveal distinct hydrochemical characteristics: deep aquifers are predominantly affected by salinity intrusion, while shallow aquifers have higher concentrations of total dissolved solids (TDS), hardness, and arsenic due to surface influences and geochemical interactions. Subsequently, the Water Quality Index (WQI) was used to evaluate the suitability of drinking water. The findings indicate that, while a significant portion of the study area falls within excellent to good categories, localized zones exhibit poor to unsuitable conditions, primarily due to elevated Arsenic and Iron concentrations. Further, correlation analysis highlights strong relationships among chloride, salinity, and TDS, confirming seawater intrusion as a dominant process. In contrast, arsenic and iron show strong positive correlations with WQI, indicating they are critical contaminants. Next, machine learning models were developed to predict groundwater quality, with Linear Regression demonstrating superior performance across multiple evaluation metrics (R² ≈ 0.99). Sensitivity analyses using Sobol, Morris, and feature importance methods consistently identify arsenic and iron as the most influential parameters, collectively explaining over 95% of WQI variability. Finally, spatial analysis reveals a deterioration of groundwater quality toward coastal and riverine zones, emphasizing the combined effects of natural hydrogeological processes and anthropogenic activities.