Machine learning models to predict 1,2,3-trichloropropane occurrence in groundwater of the Central Valley, CA, and implications for domestic well water quality
Abstract
Groundwater contamination by 1,2,3-trichloropropane (1,2,3-TCP), a probable human carcinogen, is widespread in California's Central Valley. Recent regulatory changes have prompted statewide monitoring of 1,2,3-TCP in public supply wells, but the risk to domestic well users is not well known.
We used machine learning models, including random forest classification and gradient boosted machine models, to evaluate the predicted probability of 1,2,3-TCP occurrence above regulatory limits in domestic wells of California's Central Valley. The model predictors include hydrologic features (including aquifer particle size distribution and flow paths), near-surface properties (including land use and soil geochemistry), and groundwater chemistry. To validate the results of these models, we are conducing community-based domestic well testing in the San Joaquin Valley for 1,2,3-TCP and a range of other contaminants. Well owners are surveyed about water use and consumption to characterize routes of exposure to contaminants. Our findings suggest the importance of land use and aquifer particle size distribution (texture) in predicting 1,2,3-TCP occurrence. Both modeling and preliminary field sampling suggest that domestic well owners often face a mixture of contaminants in drinking water along with 1,2,3-TCP. Exposure to these contaminants through drinking, cooking, and bathing is a pressing and environmental health and justice concern.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMSY039..08S
- Keywords:
-
- 0498 General or miscellaneous;
- BIOGEOSCIENCES;
- 0299 General or miscellaneous;
- GEOHEALTH;
- 1699 General or miscellaneous;
- GLOBAL CHANGE;
- 1880 Water management;
- HYDROLOGY