Related Experiment Video
Updated: May 31, 2026
09:23
Removal of Trace Elements by Cupric Oxide Nanoparticles from Uranium In Situ Recovery Bleed Water and Its Effect on Cell Viability
Published on: June 21, 2015
Cross-Regional Transfer Learning for Interpretable Prediction of Groundwater Uranium Contamination
Hailong Cao1,2, Xianjun Xie3,4, Ziyi Xiao1,2
1Hubei Key Laboratory of Petroleum Geochemistry and Environment, Yangtze University, Wuhan 430100, China.
Environmental Science & Technology
|May 29, 2026
Summary
Transfer learning accurately predicts groundwater uranium (U) levels in data-scarce regions. A model trained on California data successfully predicted U risks in China, identifying hotspots and guiding management.
Area of Science:
- Environmental Science
- Hydrogeology
- Machine Learning
Background:
- High uranium (U) concentrations in groundwater present significant health risks, including kidney damage and cancer.
- Predicting U occurrence in data-limited regions is a major challenge for public health and environmental management.
Purpose of the Study:
- To develop and validate a cross-regional transfer learning framework for predicting groundwater U contamination.
- To assess the feasibility of using a pretraining-fine-tuning strategy with the TabNet neural network for U prediction in under-monitored areas.
Main Methods:
- A TabNet neural network model was pretrained using groundwater U data from California.
- The pretrained model was fine-tuned with limited U data from the Datong Basin, China.
- The fine-tuned model was used to predict basin-wide groundwater U exceedance probabilities (>30 μg/L).
Main Results:
- The fine-tuned model achieved balanced sensitivity and specificity for U exceedance prediction.
- High-risk zones (probability >0.6) with low epistemic uncertainty were identified, covering 11.4% of the Datong Basin.
- Five percent of the population was potentially exposed to U, with an uncertainty range of 2-12%, and previously unrecognized U hotspots were pinpointed.
Conclusions:
- Cross-regional knowledge transfer is feasible for predicting groundwater U contamination in data-scarce regions.
- The developed framework effectively identifies high-risk areas and potential human exposure, guiding targeted monitoring and management strategies.
- Model interpretation revealed precipitation, slope, and cropping intensity as key predictors, though their influence varied by region, highlighting local hydrogeochemical context.
