Graph Neural Networks Predict Groundwater Arsenic Levels

William Xing, Stephanie Yang, Aarush Bandemegal, Anushree Misra, Ananya Kalapatapu, Brennan Lagasse, Kevin Zhu· July 23, 2026 View original

Summary

This research uses Graph Neural Networks (GNNs) to predict groundwater arsenic concentrations across the US, integrating over 74,000 samples from various sources into a spatially informed dataset. GNNs demonstrated the ability to match or outperform gradient-boosted trees by accounting for spatial dependence, enhancing environmental prediction and risk mapping.

Arsenic contamination in groundwater poses a significant public health threat, particularly for households relying on private wells in the United States. Accurate, spatially aware prediction models are crucial for identifying high-risk areas and directing mitigation efforts. This study addresses the lack of generalizable models for continuous arsenic concentration variation by framing it as a regression task. The researchers constructed a comprehensive, spatially integrated dataset by aggregating over 74,000 arsenic samples from the Water Quality Portal, Mineral Resources Data System, and Gridded National Soil Survey Geographic Database, using k-Nearest Neighbors and GIS for location-based joining. They evaluated various machine learning models, including tree-based ensembles, multilayer perceptrons, and spatially aware Graph Neural Networks (GNNs). The findings indicate that while gradient-boosted trees remain strong, GNNs effectively leverage spatial dependence to achieve comparable or superior performance, laying a foundation for improved groundwater risk mapping and monitoring.

Why it matters

Environmental and public health professionals can utilize more accurate, spatially informed models to identify and prioritize areas at high risk of arsenic contamination, leading to more effective public health interventions and resource allocation.

How to implement this in your domain

  1. 1Explore integrating Graph Neural Networks into environmental monitoring and prediction systems for spatially dependent data.
  2. 2Leverage GIS and k-NN techniques to create spatially integrated datasets from disparate environmental data sources.
  3. 3Benchmark GNN performance against traditional machine learning models for environmental prediction tasks to identify optimal approaches.
  4. 4Collaborate with data scientists to develop interactive risk maps based on GNN predictions for public health communication and policy making.

Who benefits

Environmental ProtectionPublic HealthUtilitiesGovernmentAgriculture

Key takeaways

  • GNNs can accurately predict groundwater arsenic concentrations by leveraging spatial data.
  • Spatially integrated datasets are crucial for effective environmental modeling.
  • GNNs match or outperform traditional models by accounting for spatial dependence.
  • This approach enhances groundwater risk mapping and mitigation efforts.

Original post by William Xing, Stephanie Yang, Aarush Bandemegal, Anushree Misra, Ananya Kalapatapu, Brennan Lagasse, Kevin Zhu

"arXiv:2607.19392v1 Announce Type: new Abstract: Arsenic contamination in groundwater presents a longstanding public health crisis in the United States, especially for households depending on private wells. Accurate and spatially informed prediction of arsenic concentration is vit…"

View on X

Originally posted by William Xing, Stephanie Yang, Aarush Bandemegal, Anushree Misra, Ananya Kalapatapu, Brennan Lagasse, Kevin Zhu on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses