77-4 Predicting Trace Element Concentrations in Major Texas Aquifer systems and associated Groundwater Sustainability using a Multimodal Machine Learning approach

Session: Groundwater and Sustainability: Integrating Science, Technology, and Policy

Presenting Author:

Authors:

Abstract:

Trace elements even at low concentrations can be toxic due to the risk of accumulation in tissue and disruption of cellular and genetic functions thus leading to severe health effects. We developed a bi-layered tool combining a multimodal machine learning (ML) model with an ingestion-based health risk assessment to predict the ambient concentration of trace elements including lead (Pb), strontium (Sr), chromium (Cr), arsenic (As) and nonmetals like nitrate (NO₃^-) and fluoride (F^-) in the major aquifer system in Texas. Our model, an ensemble of artificial neural network, gradient boosted trees and kriging, incorporated hydrogeological (aquifer type etc.), geochemical (concentration of ions like Ca²⁺, HCO₃^-) and physical parameters to capture nonlinear interactions influencing ambient trace element concentration in groundwater. We used these insights to estimate the ingestion-related health risks and map their potential public health impacts. Our initial ensemble model achieved an aggregate r² and a mean squared error (MSE) value of 0.66 and 0.55 respectively across 5-fold cross-validation. We estimated that the peak concentration of Pb in the Edwards-Trinity (Plateau) aquifer was 19.957 ± 0.049 µg/L (x̄ = 0.377 ± 0.083), followed by As at 17.662 ± 0.052 µg/L (x̄ = 0.768 ± 0.088), Cr at 16.068 ± 0.019 µg/L (x̄ = 0.508 ± 0.030), Sr at 7.010 ± 0.019 µg/L (x̄ = 0.448 ± 0.041). Among nonmetals, our model predicted a maximum of 13.566 ± 0.045mg/L (x̄ = 0.475 ± 0.063) for NO₃^- and 4.876 ± 0.021 mg/L (x̄ = 1.159 ± 0.038) for F^-. Over a 30-year ingestion exposure period, we estimated an upper bound cancer risk from As at 3.43 x 10^-4, corresponding to some 350 cancer cases per million population for direct untreated water ingestion and a hazard quotient (HQ) of 3.78 for F^-. The findings highlight the vulnerability of groundwater to contamination under environmental stress. By integrating ML with hydrogeochemical and health risk, we demonstrated how inter-disciplinary science, and technology can guide effective oversight and policy making, thus making it a powerful tool to safeguard and preserve our groundwater resources.

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

doi: 10.1130/abs/2025AM-8851

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