Fluoride in Basin-Fill Aquifers of Western U.S. Aquifers: Predicting Concentrations at Domestic and Public Supply Depths Using Machine Learning Methods
Abstract
Fluoride in groundwater has both geogenic and anthropogenic sources and is considered an important nutrient for healthy structure of teeth and bones. However, elevated concentrations above the U.S. Environmental Protection Agency water quality human-health benchmark in groundwater may result in skeletal disorders or tooth discoloration. From 2013 to 2017, the U.S. Geological Survey collected untreated groundwater samples at public supply depths from principal aquifers of the Western U.S. and compared them to available federal regulatory water-quality benchmarks for protecting human health. Fluoride was among several trace elements that commonly exceeded the federal regulatory health-based and aesthetic water-quality benchmarks in basin-fill aquifers of the Western U.S—California Coastal aquifers, California Central Valley aquifer, Basin and Range aquifers and the Rio Grande aquifer system. Elevated fluoride concentrations were associated with geogenic sources: groundwater with long residence times resulting in increased interaction with aquifer materials, higher total dissolved solids concentrations, and increases in pH that result from mineral reactions contributing to calcite precipitation.
Compromised water-quality conditions can limit water availability for potable, ecological, recreational and other uses. To assess regional concentrations of fluoride in basin-fill aquifers, we developed a boosted regression trees machine learning model and mapped predicted regional-scale fluoride concentrations at domestic and public supply depths. A database of predictor variables representing well characteristics, geochemical conditions, land use, climate, soil characteristics, and soil chemistry was compiled; values were assigned to over 13,000 wells with measured fluoride concentrations from federal and state databases. Models were trained on 80 % of wells and tested on the remaining 20 % (hold-out dataset). Prediction uncertainty was evaluated using USGS developed methods that compute percentile confidence intervals for each hold-out observation. The resulting fluoride prediction maps can aid local, state, and federal water resource managers by identifying domestic and public supplies vulnerable to exceedances of the health-based and aesthetic water-quality benchmarks.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFMGH22A..06R
- Keywords:
-
- 0240 Public health;
- GEOHEALTH;
- 0240 Public health;
- GEOHEALTH;
- 1831 Groundwater quality;
- HYDROLOGY;
- 1831 Groundwater quality;
- HYDROLOGY;
- 1847 Modeling;
- HYDROLOGY;
- 1847 Modeling;
- HYDROLOGY;
- 1880 Water management;
- HYDROLOGY;
- 1880 Water management;
- HYDROLOGY