249-12 Machine Learning Models for Mapping Groundwater Pollution Risk: Advancing Water Security and Sustainable Development Goals in Georgia, USA

Session: Emerging Contaminants: Geochemical Insights and Impacts on Human and Environmental Health

Presenting Author:

Authors:

Abstract:

Machine learning assessment of groundwater atrazine and malathion contamination: advancing water security and SDG 6.1 in Georgia, USA

Shivank Pandey¹, Srimanti Duttagupta², Avishek Dutta^2,3

¹Department of Computer Science, University of Georgia, Athens, Georgia, USA

²Department of Geology, University of Georgia, Athens, Georgia, USA

³Savannah River Ecological Laboratory, University of Georgia, Aikens, South Carolina, USA

Approximately 92% of rural residents in Georgia depend on groundwater for consumption. This study underscores the importance of understanding groundwater contamination and its implications for safe water consumption in Georgia, USA. Random Forest classification models were developed and analyzed to enhance water security and support Sustainable Development Goal 6.1 by predicting county-level contamination of atrazine and malathion. The models utilized 2019 datasets from national agencies and aquifer lithology maps. Among the predictors used were rainfall, depth to the water table, population density, pesticide usage, and aquifer lithologies. To address problems of skewness and class imbalance, pesticide concentrations, presented as ranges, were divided into low, moderate, and high classes. For each pesticide, a Random Forest classifier was trained and evaluated via five-fold cross-validation, achieving mean test accuracies of 55% for atrazine and 60% for malathion. A second classifier for each pesticide was then applied to predict contamination in counties where groundwater contamination data was unavailable. As the sample was small and the data was restricted by range, models performed well on the training data (100% accuracy) but showed moderately good test results (50–75%). Feature-importance analysis indicates that the predominant factor controlling atrazine presence in aquifers is the average pesticide application rate. In contrast, malathion occurrence is predominantly influenced by underlying lithology, with precipitation also significantly impacting its distribution. Comparison of predicted and actual contamination areas identified regions needing targeted monitoring and cleanup. Although model performance improves when incorporating multiple environmental factors and human activities, further enhancements would result from integrating more detailed and extensive datasets. Further research will use regression-based models to project absolute concentrations. Moreover, pesticide concentration measurement in groundwater samples can be used to verify the model predictions in different hydrogeological regions of Georgia. By using machine learning models and geospatial analytic techniques, this study delivers actionable insights into groundwater pollution risk, guiding resource allocation and policy decisions to safeguard water quality and public health in Georgia and beyond. 

Geological Society of America Abstracts with Program. Vol. 57, No. 6, 2025

doi: 10.1130/abs/2025AM-5068

© Copyright 2025 The Geological Society of America (GSA), all rights reserved.