Machine-Learning Predictions of High Arsenic and High Manganese at Drinking Water Depths of the Glacial Aquifer System, Northern Continental United States.

Globally, over 200 million people are chronically exposed to arsenic (As) and/or manganese (Mn) from drinking water. We used machine-learning (ML) boosted regression tree (BRT) models to predict high As (>10 µg/L) and Mn (>300 µg/L) in groundwater from the glacial aquifer system (GLAC), which spans 25 states in the northern United States and provides drinking water to 30 million people. Our BRT models' predictor variables (PVs) included recently developed three-dimensional estimates of a suite of groundwater age metrics, redox condition, and pH. We also demonstrated a successful approach to significantly improve ML prediction sensitivity for imbalanced data sets (small percentage of high values). We present predictions of the probability of high As and high Mn concentrations in groundwater, and uncertainty, at two nonuniform depth surfaces that represent moving median depths of GLAC domestic and public supply wells within the three-dimensional model domain. Predicted high likelihood of anoxic condition (high iron or low dissolved oxygen), predicted pH, relative well depth, several modeled groundwater age metrics, and hydrologic position were all PVs retained in both models; however, PV importance and influence differed between the models. High-As and high-Mn groundwater was predicted with high likelihood over large portions of the central part of the GLAC.

Occurrence and Geochemistry of Lead-210 and Polonium-210 Radionuclides in Public-Drinking-Water Supplies from Principal Aquifers of the United States.

On the basis of lifetime cancer risks, lead-210 (210Pb) and polonium-210 (210Po) ≥ 1.0 and 0.7 pCi/L (picocuries per liter), respectively, in drinking-water supplies may pose human-health concerns. 210Pb and 210Po were detected at concentrations greater than these thresholds at 3.7 and 1.5%, respectively, of filtered untreated groundwater samples from 1263 public-supply wells in 19 principal aquifers across the United States. Nationally, 72% of samples with radon-222 (222Rn) concentrations > 4000 pCi/L had 210Pb ≥ 1.0 pCi/L. 210Pb is mobilized by alpha recoil associated with the decay of 222Rn and short-lived progeny. 210Pb concentrations ≥ 1.0 pCi/L occurred most frequently where acidic groundwaters inhibited 210Pb readsorption (felsic-crystalline rocks) and where reducing alkaline conditions favored dissolution of iron-manganese- (Fe-Mn-) oxyhydroxides (which adsorb 210Pb) and formation of lead-carbonate complexes (enhancing lead (Pb) mobility). 210Po concentrations ≥ 0.7 pCi/L occurred almost exclusively in confined Coastal Plain aquifers where old (low percent-modern carbon-14) groundwaters were reducing, with high pH (>7.5) and high sodium/chloride (Na/Cl) ratios resulting from cation exchange. In high-pH environments, aqueous polonium (Po) is poorly sorbed, occurring as dihydrogen polonate (H2PoO3(aq)) or, under strongly reducing conditions, as a hydrogen-polonide anion (HPo-). Fe-Mn- and sulfate-reduction and cation-exchange processes may mobilize polonium from mineral surfaces. Po2+ occurrence in low-to-neutral-pH waters is attenuated by adsorption.

Regression models for estimating concentrations of atrazine plus deethylatrazine in shallow groundwater in agricultural areas of the United States.

Tobit regression models were developed to predict the summed concentration of atrazine [6-chloro--ethyl--(1-methylethyl)-1,3,5-triazine-2,4-diamine] and its degradate deethylatrazine [6-chloro--(1-methylethyl)-1,3,5,-triazine-2,4-diamine] (DEA) in shallow groundwater underlying agricultural settings across the conterminous United States. The models were developed from atrazine and DEA concentrations in samples from 1298 wells and explanatory variables that represent the source of atrazine and various aspects of the transport and fate of atrazine and DEA in the subsurface. One advantage of these newly developed models over previous national regression models is that they predict concentrations (rather than detection frequency), which can be compared with water quality benchmarks. Model results indicate that variability in the concentration of atrazine residues (atrazine plus DEA) in groundwater underlying agricultural areas is more strongly controlled by the history of atrazine use in relation to the timing of recharge (groundwater age) than by processes that control the dispersion, adsorption, or degradation of these compounds in the saturated zone. Current (1990s) atrazine use was found to be a weak explanatory variable, perhaps because it does not represent the use of atrazine at the time of recharge of the sampled groundwater and because the likelihood that these compounds will reach the water table is affected by other factors operating within the unsaturated zone, such as soil characteristics, artificial drainage, and water movement. Results show that only about 5% of agricultural areas have greater than a 10% probability of exceeding the USEPA maximum contaminant level of 3.0 µg L. These models are not developed for regulatory purposes but rather can be used to (i) identify areas of potential concern, (ii) provide conservative estimates of the concentrations of atrazine residues in deeper potential drinking water supplies, and (iii) set priorities among areas for future groundwater monitoring.

Predicting regional fluoride concentrations at public and domestic supply depths in basin-fill aquifers of the western United States using a random forest model.

A random forest regression (RFR) model was applied to over 12,000 wells with measured fluoride (F) concentrations in untreated groundwater to predict F concentrations at depths used for domestic and public supply in basin-fill aquifers of the western United States. The model relied on twenty-two regional-scale environmental and surficial predictor variables selected to represent factors known to control F concentrations in groundwater. The testing model fit R2 and RMSE were 0.52 and 0.78 mg/L. Comparisons of measured to predicted proportions of four F-concentrations categories (<0.7 mg/L, 0.7-2 mg/L, >2 mg/L - 4 mg/L, and > 4 mg/L) indicate that the model performed well at making regional-scale predictions. Differences between measured and predicted proportions indicate underprediction of measured F at values by between 4 and 20 mg/L, representing less than 1% of the regional scale predicted values. These residuals most often map to geographic regions where local-scale processes including evaporative discharge in closed basins or intermittent streams concentrate fluoride in shallow groundwater. Despite this, the RFR model provides spatially continuous F predictions across the basin-fill aquifers where discrete samples are missing. Further, the predictions capture documented areas that exceed the F maximum contaminant level for drinking water of 4 mg/L and areas that are below the oral-health benchmark of 0.7 mg/L. These predictions can be used to estimate fluoride concentrations in unmonitored areas and to aid in identifying geographic areas that may require further investigation at localized scales.

Mapped Predictions of Manganese and Arsenic in an Alluvial Aquifer Using Boosted Regression Trees.

Manganese (Mn) concentrations and the probability of arsenic (As) exceeding the drinking-water standard of 10 µg/L were predicted in the Mississippi River Valley alluvial aquifer (MRVA) using boosted regression trees (BRT). BRT, a type of ensemble-tree machine-learning model, were created using predictor variables that affect Mn and As distribution in groundwater. These variables included iron (Fe) concentrations and specific conductance predicted from previously developed BRT models, groundwater flux and age estimates from MODFLOW, and hydrologic characteristics. The models also included results from the first airborne geophysical survey conducted in the United States to target an entire aquifer system. Predictions of high Mn and As occurred where Fe was high. Predicted high Mn concentrations were correlated with fraction of young groundwater (less than 65 years) computed from MODFLOW results. High probabilities of As exceedance were predicted where groundwater was relatively old and airborne electromagnetic resistivity was high, typically proximal to streams. Two-variable partial-dependence plots and sensitivity analysis were used to provide insight into the factors controlling Mn and As distribution in groundwater. The maps of predicted Mn concentrations and As exceedance probabilities can be used to identify areas where these constituents may be high, and that could be targeted for further study. This paper shows that incorporation of a selected set of process-informed data, such as MODFLOW results and airborne geophysics, into a machine-learning model improves model interpretability. Incorporation of process-rich information into machine-learning models will likely be useful for addressing a wide range of problems of interest to groundwater hydrologists.

Machine Learning Predictions of pH in the Glacial Aquifer System, Northern USA.

A boosted regression tree model was developed to predict pH conditions in three dimensions throughout the glacial aquifer system of the contiguous United States using pH measurements in samples from 18,386 wells and predictor variables that represent aspects of the hydrogeologic setting. Model results indicate that the carbonate content of soils and aquifer materials strongly controls pH and, when coupled with long flowpaths, results in the most alkaline conditions. Conversely, in areas where glacial sediments are thin and carbonate-poor, pH conditions remain acidic. At depths typical of drinking-water supplies, predicted pH >7.5-which is associated with arsenic mobilization-occurs more frequently than predicted pH <6-which is associated with water corrosivity and the mobilization of other trace elements. A novel aspect of this model was the inclusion of numerically based estimates of groundwater flow characteristics (age and flowpath length) as predictor variables. The sensitivity of pH predictions to these variables was consistent with hydrologic understanding of groundwater flow systems and the geochemical evolution of groundwater quality. The model was not developed to provide precise estimates of pH at any given location. Rather, it can be used to more generally identify areas where contaminants may be mobilized into groundwater and where corrosivity issues may be of concern to prioritize areas for future groundwater monitoring.

Efficiency of conventional drinking-water-treatment processes in removal of pharmaceuticals and other organic compounds.

Samples of water and sediment from a conventional drinking-water-treatment (DWT) plant were analyzed for 113 organic compounds (OCs) that included pharmaceuticals, detergent degradates, flame retardants and plasticizers, polycyclic aromatic hydrocarbons (PAHs), fragrances and flavorants, pesticides and an insect repellent, and plant and animal steroids. 45 of these compounds were detected in samples of source water and 34 were detected in samples of settled sludge and (or) filter-backwash sediments. The average percent removal of these compounds was calculated from their average concentration in time-composited water samples collected after clarification, disinfection (chlorination), and granular-activated-carbon (GAC) filtration. In general, GAC filtration accounted for 53% of the removal of these compounds from the aqueous phase; disinfection accounted for 32%, and clarification accounted for 15%. The effectiveness of these treatments varied widely within and among classes of compounds; some hydrophobic compounds were strongly oxidized by free chlorine, and some hydrophilic compounds were partly removed through adsorption processes. The detection of 21 of the compounds in 1 or more samples of finished water, and of 3 to 13 compounds in every finished-water sample, indicates substantial but incomplete degradation or removal of OCs through the conventional DWT process used at this plant.

Persistence of pharmaceuticals and other organic compounds in chlorinated drinking water as a function of time.

Ninety eight pharmaceuticals and other organic compounds (POOCs) that were amended to samples of chlorinated drinking-water were extracted and analyzed 1, 3, 6, 8, and 10 days after amendment to determine whether the total chlorine residual reacted with the amended POOCs in drinking water in a time frame similar to the residence time of drinking water in a water distribution system. Results indicated that if all 98 were present in the finished drinking water from a drinking-water treatment plant using free chlorine at 1.2 mg/L as the distribution system disinfectant residual, 52 POOCs would be present in the drinking water after 10 days at approximately the same concentration as in the newly finished drinking water. Concentrations of 16 POOCs would be reduced by 32% to 92%, and 22 POOCs would react completely with residual chlorine within 24 h. Thus, the presence of free chlorine residual is an effective means for transforming some POOCs during distribution.

Mercury concentrations in water from an unconfined aquifer system, New Jersey coastal plain.

Concentrations of total mercury (Hg) from 2 microg/L (the USEPA maximum contaminant level) to 72 microg/L in water from about 600 domestic wells in residential parts of eight counties in southern New Jersey have been reported by State and county agencies. The wells draw water from the areally extensive (7770 km(2)) unconfined Kirkwood-Cohansey aquifer system, in which background concentrations of Hg are about 0.01 microg/L or less. Hg is present in most aquifer materials at concentrations <50 microg/kg, but is at 100--150 microg/kg in undisturbed surficial soils. No point sources of contamination to the affected areas have been conclusively identified. To determine whether high levels of Hg in ground water are related to a particular land use and (or) water chemistry, water samples from 105 wells that tap the aquifer system were collected by the United States Geological Survey. These included randomly selected domestic wells, domestic and observation wells in selected land uses, and sets of clustered observation wells--including two sets that are downgradient from residential areas with Hg-contaminated ground water. Hg concentrations in filtered samples (Hg(f)) were at or near background levels in water from most wells, but ranged from 0.1 to 3.8 microg/L in water from nearly 20% of wells. Hg(f) concentrations from 0.0001 to 0.1 microg/L correlated significantly and positively with concentrations of other constituents associated with anthropogenic inputs (Ca, Cl, Na, and NO(3)) and with dissolved organic carbon. Hg(f) concentrations >0.1 microg/L did not correlate significantly with concentrations of the inorganic constituents. Hg(f) concentrations near or exceeding 2 microg/L were found only in water from wells in areas with residential land use, but concentrations were at background levels in most water samples from undeveloped land. The spatial distribution of Hg-contaminated ground water appears to be locally and regionally heterogeneous; no extensive plumes of Hg contamination have yet been identified.

Persistence of pharmaceutical compounds and other organic wastewater contaminants in a conventional drinking-water-treatment plant.

In a study conducted by the US Geological Survey and the Centers for Disease Control and Prevention, 24 water samples were collected at selected locations within a drinking-water-treatment (DWT) facility and from the two streams that serve the facility to evaluate the potential for wastewater-related organic contaminants to survive a conventional treatment process and persist in potable-water supplies. Stream-water samples as well as samples of raw, settled, filtered, and finished water were collected during low-flow conditions, when the discharge of effluent from upstream municipal sewage-treatment plants accounted for 37-67% of flow in stream 1 and 10-20% of flow in stream 2. Each sample was analyzed for 106 organic wastewater-related contaminants (OWCs) that represent a diverse group of extensively used chemicals. Forty OWCs were detected in one or more samples of stream water or raw-water supplies in the treatment plant; 34 were detected in more than 10% of these samples. Several of these compounds also were frequently detected in samples of finished water; these compounds include selected prescription and non-prescription drugs and their metabolites, fragrance compounds, flame retardants and plasticizers, cosmetic compounds, and a solvent. The detection of these compounds suggests that they resist removal through conventional water-treatment processes. Other compounds that also were frequently detected in samples of stream water and raw-water supplies were not detected in samples of finished water; these include selected prescription and non-prescription drugs and their metabolites, disinfectants, detergent metabolites, and plant and animal steroids. The non-detection of these compounds indicates that their concentrations are reduced to levels less than analytical detection limits or that they are transformed to degradates through conventional DWT processes. Concentrations of OWCs detected in finished water generally were low and did not exceed Federal drinking-water standards or lifetime health advisories, although such standards or advisories have not been established for most of these compounds. Also, at least 11 and as many as 17 OWCs were detected in samples of finished water. Drinking-water criteria currently are based on the toxicity of individual compounds and not combinations of compounds. Little is known about potential human-health effects associated with chronic exposure to trace levels of multiple OWCs through routes such as drinking water. The occurrence in drinking-water supplies of many of the OWCs analyzed for during this study is unregulated and most of these compounds have not been routinely monitored for in the Nation's source- or potable-water supplies. This study provides the first documentation that many of these compounds can survive conventional water-treatment processes and occur in potable-water supplies. It thereby provides information that can be used in setting research and regulatory priorities and in designing future monitoring programs. The results of this study also indicate that improvements in water-treatment processes may benefit from consideration of the response of OWCs and other trace organic contaminants to specific physical and chemical treatments.