Consideration of grid cell size to represent stream network geometry for the conterminous United States
No abstract available.
conterminous United States↗
Geology topics
Kenneth Belitz
Publications and source records attributed to Kenneth Belitz.
A spatial machine learning model developed from noisy data requires multiscale performance evaluation: Predicting depth to bedrock in the Delaware River Basin, USA
Spatial machine learning models can be developed from observations with substantial unexplainable variability, sometimes called ‘noise’. Traditional point-scale metrics (e.g., R 2 ) alone can be misleading when evaluating these models. We present a multi-scale performance evaluation (MPE) using two additional scales (distributional and geostatistical). We apply the MPE framework to predictions of depth to bedrock (DTB) in the Delaware River Basin. Geostatistical analysis shows that approximately one third of the DTB variance is at spatial scale smaller than 2 km. Hence, we interpret our point-scale R 2 of 0.3 (testing data) to be sufficient for regional-scale modelling. Bias-correction methods improve performance at two of the three MPE scales: point-scale change is negligible, while distributional and geostatistical performance improves. In contrast, bias correction applied to a global DTB model does not improve MPE performance. This work encourages scale-appropriate performance evaluations to enable effective model intercomparison.
Delaware, New Jersey, New York, Pennsylvania↗
Gross alpha-particle activity and high 226Ra concentrations do not correspond with high 210Po in the Atlantic and Gulf Coastal Plain aquifers of the United States
210 Po, which is of human-health concern based on lifetime ingestion cancer risk, is indirectly regulated in drinking water through the U.S. Environmental Protection Agency’s gross alpha-particle activity (GAPA) maximum contaminant level of 15 pCi/L (picocuries per liter). This regulation requires independent measurement of 226 Ra for samples exceeding the GAPA screening level of 5 pCi/L. There is no such requirement for 210 Po. Co-occurrence of 226 Ra and 210 Po, alpha-emitting 238 U-decay-series progeny, might be helpful in locating high- 210 Po waters but is unverified. Relations among 210 Po, 226 Ra, and GAPA evaluated for samples from 257 public-supply wells from Coastal Plain aquifers showed that concentrations of 226 Ra correlated with GAPA but neither correlated with 210 Po concentrations. The highest concentrations of 226 Ra and 210 Po were found under differing geochemical conditions. The highest 226 Ra occurred in low-pH oxidizing waters and in neutral-pH reducing waters, where geochemical conditions render Fe–Mn-hydroxide sorbents inefficient. 210 Po was highest (10.1 pCi/L) in reducing waters with high pH (>7.5, which results from progressive cation exchange), where 226 Ra was lowest─exchanged to clay minerals. Because 226 Ra and 210 Po did not co-occur, the GAPA screening might not be protective for 210 Po. Independent 210 Po analysis is prudent, especially where groundwater is reducing with high pH and low 226 Ra concentrations.
Environmental Science & Technology Water↗
Predicted uranium and radon concentrations in New Hampshire (USA) groundwater—Using Multi Order Hydrologic Position as predictors
Two radioactive elements, uranium (U) and radon (Rn), which are of potential concern in New Hampshire (NH) groundwater, are investigated. Exceedance probability maps are tools to highlight locations where the concentrations of undesirable substances in the groundwater may be elevated. Two forms of statistical analysis are used to create exceedance probability maps for U and Rn in NH groundwater. The first, Boosted Regression Tree (BRT), was selected for estimating U exceedance values. It computes exceedance values directly using the Bernoulli distribution function. The second method of statistical analysis used for Rn to determine exceedance probabilities is ordinary least squares (OLS) regression. In the process of determining exceedance probabilities for U and Rn, the utility of a new dataset is investigated. That new predictor dataset is the Multi-Order Hydrologic Position (MOHP) dataset. MOHP raster datasets have been produced nationally for the conterminous United States at a 30-m resolution. The concept behind MOHP is that, for any given point on the earth's surface, there is the potential for a longer groundwater flow path as one goes deeper beneath the land surface. MOHP predictors were tested in both models. Three MOHP predictors were found useful in the BRT model and two in the OLS model. MOHP data were found useful as predictors along with other site characteristics in predicting U and Rn exceedance probabilities in New Hampshire groundwater.
New Hampshire↗
Quality of groundwater used for public supply in the continental United States: A comprehensive assessment
The presence of contaminants in a source water can constrain its suitability for drinking. The quality of groundwater used for public supply was assessed in 25 principal aquifers (PAs) that account for 84% of groundwater pumped for public supply in the U.S. (89.6 million people on a proportional basis). Each PA was sampled across its lateral extent using an equal-area grid, typically with 60 wells per PA. Samples were analyzed for 502 constituents, of which 374 had either a regulatory or nonregulatory human health benchmark (HHB). Nationally, elevated concentrations (relative to HHBs) of geogenic constituents have a larger effect than anthropogenic constituents, as indicated by three metrics: detection frequency, 35% versus 8.1%; prevalence (based on area), 41% versus 6.4%; and population potentially affected, 31.2 million versus 7.1 million. Prevalence of any constituent at elevated concentrations was high─40 to 75%─in PAs comprising unconsolidated sediment (eight PAs) and sandstone or interbedded sandstones and carbonates (four PAs) in the West and Central Interior. Prevalence was lower─15 to 35%─in PAs comprising sediment and sedimentary rocks along the Gulf and Atlantic Coasts (four PAs), carbonates distributed across the continental U.S. (seven PAs), and hard rock (two PAs).
continental United States↗
Over a third of groundwater in USA public-supply aquifers is Anthropocene-age and susceptible to surface contamination
The distribution of groundwater age is useful for evaluating the susceptibility and sustainability of groundwater resources. Here, we compute the aquifer-scale cumulative distribution function to characterize the age distribution for 21 Principal Aquifers that account for ~80% of public-supply pumping in the United States. The aquifer-scale cumulative distribution function for each Principal Aquifer was derived from an ensemble of modeled age distributions (~60 samples per aquifer) based on multiple tracers: tritium, tritiogenic helium-3, sulfur hexafluoride, chlorofluorocarbons, carbon-14, and radiogenic helium-4. Nationally, the groundwater is 38% Anthropocene (since 1953), 34% Holocene (75 – 11,800 years ago), and 28% Pleistocene (>11,800 years ago). The Anthropocene fraction ranges from <5 to 100%, indicating a wide range in susceptibility to land-surface contamination. The Pleistocene fraction of groundwater exceeds 50% in 7 eastern aquifers that are predominately confined. The Holocene fraction of groundwater exceeds 50% in 5 western aquifers that are predominately unconfined. The sustainability of pumping from these Principal Aquifers depends on rates of recharge and release of groundwater stored in fine-grained layers.
Nature Communications Earth & Environment↗
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.
Arkansas, Louisiana, Mississippi, Tennessee↗
Populations using public-supply groundwater in the conterminous U.S. 2010; Identifying the wells, hydrogeologic regions, and hydrogeologic mapping units
Most Americans receive their drinking water from publicly supplied sources, a large portion of it from groundwater. Mapping these populations consistently and at a high resolution is important for understanding where the resource is used and needs to be protected. The results show that 269 million people are supplied by public supply, 107 million are supplied by groundwater and 162 million are supplied by surface water. The population using public supply drinking water was mapped in two ways: the census enhanced method (CEM) evenly distributes the population across populated census blocks, and the urban land-use enhanced method (ULUEM) distributes the population only to certain urban land use designations. In addition, a two-dimensional polygon dataset was created for the conterminous U.S. that identifies 177 unique Hydrogeologic Mapping Units (HMUs) with similar hydrogeologic characteristics. The HMUs do not overlap, but they can delineate areas where stacked hydrogeologic regions (HRs) contribute drinking water from below the surface. HRs are waterbearing geologic regions identified as either a principal aquifers (PA) or secondary hydrogeologic regions (SHR). Within each HMU, the wells were used to determine the proportion of each HR that is providing groundwater to the HMU. In 63% of the HMUs, a single HR is providing water to the public supply wells located within it, while the rest of the HMUs show that the wells are tapping up to a maximum of four stacked HRs. In total, groundwater from 108 HRs provide drinking water for public supply, six of which provide more than 50% of the groundwater used for public supply drinking water. The aquifer serving the largest number of equivalent people (>17 million) is the glacial aquifer. The HR providing the greatest number of people per km 2 is the Biscayne aquifer in Florida at nearly 453 people per km 2 .
Science of the Total Environment↗
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 R 2 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.
Arizona, California, Colorado, New Mexico, Nevada,↗
Volatile organic compounds in groundwater used for public supply across the United States: Occurrence, explanatory factors, and human-health context
This systematic assessment of occurrence for 85 volatile organic compounds (VOCs) in raw (untreated) groundwater used for public supply across the United States (U.S.), which includes 43 compounds not previously monitored by national studies, relates VOC occurrence to explanatory factors and assesses VOC detections in a human-health context. Samples were collected in 2013 through 2019 from 1537 public-supply wells in aquifers representing 78% of the volume pumped for public drinking-water supply. Laboratory detection limits for VOCs generally were less than 0.1 μg/L. Detections were reported for 36% of the sampled principal-aquifer area (38% of sampled wells) and were most common in wells in shallow, unconfined aquifers in urban areas that produce high proportions of modern-age and oxic groundwater. The disinfection by-product trichloromethane (chloroform) was the most commonly detected VOC associated primarily with anthropogenic sources (24% of the sampled area, 25% of sampled wells), followed by the gasoline oxygenate methyl tert-butyl ether (8.4% of area, 11% of wells). Carbon disulfide (12% of area, 14% of wells) was examined separately because of likely substantial contributions from natural sources. Newly monitored VOCs were each detected in <1% of the sampled area. Although detections of 1,4-dioxane in this first national study of its occurrence in raw groundwater were rare, measured concentrations exceeded the most stringent (non-enforceable) human-health benchmark in 0.5% of the sampled area (9 wells). Two wells had exceedances of enforceable benchmarks for tetrachloroethylene and trichloroethylene, and 50 wells total (representing 2.0% of the sampled area, 3.3% of sampled wells) had combined VOC concentrations exceeding 10% of benchmarks of any type. Compared with previous national findings, this study reports lower rates of VOC detection, but confirms widespread anthropogenic influence on groundwater used for public supply, with relatively few concentrations of individual VOCs or mixtures that approach or exceed human-health benchmarks.
Science of the Total Environment↗
Relation between road-salt application and increasing radium concentrations in a low-pH aquifer, southern New Jersey
The Kirkwood–Cohansey aquifer in southern New Jersey is an important source of drinking-water supplies, but the availability of the resource is limited in some areas by high concentrations of radium, a potential carcinogen at elevated concentrations. Radium ( 226 Ra plus 228 Ra) concentrations from a network of 25 drinking-water wells showed a statistically significant increase over a decadal time scale ( p < 0.05), with a median increase of 0.35 picocuries per liter. Increases in Ra are correlated with road-salt application rates, and we hypothesize that the correlation is causal. Geochemical processes associated with road-salt applications that can mobilize Ra into solution include competition by excess sodium for sorption sites and formation of chloride complexes (RaCl + and RaCl 2 ). The largest increases in Ra were in groundwater with low pH (≤5), which is an indirect surrogate for low cation-sorption capacity. Correlations with other potential anthropogenic causes for the increase in Ra were not observed, further suggesting a road-salt effect. Given the significant increase in Ra concentrations in this drinking-water source, the known carcinogenic risks from Ra, the direct link to road-salt application, and the likelihood for continued increases, additional monitoring is necessary in areas with similar hydrogeologic and geochemical settings.
New Jersey↗
Depth of groundwater used for drinking-water supplies in the United States
Groundwater supplies 35 percent of drinking water in the United States. Mapping the quantity and quality of groundwater at the depths used for potable supplies requires an understanding of locational variation in the characteristics of drinking-water wells (depth and open interval). Typical depths of domestic- and public-drinking-water supply wells vary by and within aquifer across the United States. The depths to the top and bottom of the zones from which drinking water is withdrawn are important predictor variables in regional- and national-scale statistical water models, but spatially extensive maps of the depths to drinking-water-supply sources are not consistently available in modeled regions. Therefore, it was necessary to generate a set of grids representing surfaces of the approximate common depth and length of open intervals in the wells from which water is withdrawn for domestic- and public-drinking-water supply (withdrawal zones) within the conterminous United States. Well data (about 7.6 million records) were compiled from several sources, including the U.S. Geological Survey’s National Water Information System (600,922 records), the U.S. Environmental Protection Agency’s Safe Drinking Water Information System dataset (66,540 records, primarily public-supply wells), a groundwater ambient monitoring dataset (31,448 records, primarily domestic-supply wells), individual State data (6,096,503 records), a national brackish aquifer study (96,885 records), and a glacial aquifer study (729,564 records). Fifty-seven principal aquifers and 65 secondary hydrogeologic regions have been designated in the conterminous United States. The principal aquifers and secondary hydrogeologic regions vary in depth, thickness, lithology, and transmissivity characteristics. Some principal aquifers underlie secondary hydrogeologic regions, and may in turn be overlain by glacial sediment or basin and valley fill aquifers, which may also be used as drinking-water sources. The principal aquifer and secondary hydrogeologic region polygons were merged with overlying sediment polygons, where present, including glacial sediment, coarse glacial sediment, and stream valley alluvium (alluvium) polygons, to generate unique hydrogeologic settings across the conterminous United States. A total of 288 distinct hydrogeologic settings resulted from the merging of principal aquifer, secondary hydrogeologic region, glacial sediment, coarse glacial sediment, and alluvium polygons. Each well was assigned to a hydrogeologic setting on the basis of location. Hydrogeologic setting well groupings were used to guide calculations of the median value for well depth and depth to and length of open intervals across the hydrogeologic setting. Where well data were sparse or missing, wells from hydrogeologic settings with similar well construction properties, geology, physiography, and topography were grouped and used to calculate the moving median depth (if less than five wells in a 100-kilometer [62.1-mile] radius) and to estimate open interval length (if not available within hydrogeologic setting). Grids were generated to represent what might be considered as the “typical” or “median” domestic- and public-supply well in an area. The well properties are defined with moving median grids of top depth, bottom depth, and length of open interval at a 1-square-kilometer (0.38-square-mile) grid cell scale. Median depths and open intervals of domestic- and public-supply wells varied by lithology of the hydrogeologic setting and overlying sediment. Overall, the median depths were 142 feet (43.3 meters) for all domestic-supply wells and 202 feet (61.6 meters) for all public-supply wells. The median open intervals were 21 feet (6.4 meters) for domestic-supply wells and 49 feet (14.9 meters) for public-supply wells. The shallowest median bottom open interval depths for domestic-supply wells were in the secondary hydrogeologic regions with coarse glacial sediment, which suggests that the wells are most commonly completed in the permeable coarse glacial sediment and not in the underlying secondary hydrogeologic region. Public-supply wells were completed at relatively shallow median depths when drilled in permeable sediment that overlie secondary hydrogeologic regions. When public-supply wells were completed in principal aquifers, the median depths were typically greater than wells completed in secondary hydrogeologic regions. Well data used in this study were limited to those available from national or State digital databases. Several quality-assurance checks were performed during data compilation, but a comprehensive quality assurance inspection for each of the data sources was outside the scope of this study. Grids defining typical open intervals in domestic- and public-supply wells are presented. Although there are many places where multiple aquifers are stacked, these results correspond primarily to the aquifer with the highest documented number of wells for each use.
Scientific Investigations Report↗
Factors affecting uncertainty of public supply, self-supplied domestic, irrigation, and thermoelectric water-use data, 1985–2015—Evaluation of information sources, estimation methods, and data variability
The U.S. Geological Survey (USGS) Water-Use Program is responsible for compiling and disseminating the Nation's water-use data. Working in cooperation with local, State, and Federal agencies, the USGS has collected and published national water-use estimates every 5 years, beginning in 1950. These water-use data may vary because of actual changes in water use, because of changes in estimation methods, or because of errors. Comparison and interpretation of these data is difficult without first determining the factors that contribute to data variability. This report describes factors that may affect data quality and documents ways to investigate the variability of public supply, self-supplied domestic, irrigation, and thermoelectric water-use data for the 1985–2015 compilations. The USGS produces national water-use estimates for various categories of water use for every county in the United States. Knowledge about the sources of data for county estimates is important because factors such as estimation methodology and reporting affect data uncertainty Determination of meaningful patterns and trends in the data are contingent on the use of consistent methodology throughout the period of interest. With the many ways that water-use data have been collected, assembled, and estimated, multiple factors likely contribute to data uncertainty, Data used to produce these estimates may be furnished from agencies that collect information from entities who report water use; gaps in reported data are typically estimated to achieve a comprehensive county estimate. For example, public supply and thermoelectric category data are based primarily on furnished site-specific data; whereas crop irrigation is often furnished or estimated at the county scale. Public supply deliveries for domestic use and self-supplied domestic withdrawals are most often estimated by USGS personnel using per capita use rate coefficients. Irrigation may be estimated using crop water requirements, application rates, or other soil water balance methods when furnished reported data are not available. Rates, percentages, medians, and interquartile ranges were used to investigate variability in the water-use data among States, regions, and years. The purposes of these evaluations were to (1) identify extreme values that may reflect changes in information sources, estimation methods, or errors; (2) indicate areas of variable or consistent values that are unexpected; and (3) indicate areas where values change because of local climate or other factors. Where factors are identified that contribute to data variability, such as a change in methodology, additional work could determine uncertainty because of these factors. These evaluations identified the availability of information that is needed to address data limitations. Factors such as estimation methodology affect data quality. Some updates to method codes assigned in 2015 and assignment of method codes to earlier compilation datasets for all categories would provide much needed metadata for users of the data. Improvements in data documentation describing sources of information and estimation methods and additional metadata information from agencies and entities that furnish water-use data, would enable a more complete understanding and depiction of water-use patterns and trends. Additional metadata are needed for users of the data to better understand the water-use data and interpret changes in water use across the United States and with time.
Scientific Investigations Report↗
Groundwater-quality and select quality-control data from the National Water-Quality Assessment Project, January 2017 through December 2019
Groundwater-quality environmental data were collected from 983 wells as part of the National Water-Quality Assessment Project of the U.S. Geological Survey National Water Quality Program and are included in this report. The data were collected from six types of well networks: principal aquifer study networks, which are used to assess the quality of groundwater used for public water supply; land-use study networks, which are used to assess land-use effects on shallow groundwater quality; major aquifer study networks, which are used to assess the quality of groundwater used for domestic supply; enhanced trends networks, which are used to evaluate the time scales during which groundwater quality changes; vertical flow-path study networks, which are used to evaluate changes in groundwater quality from shallow to deeper depths; and modeling support studies, which are used to provide data to support groundwater modeling. Groundwater samples were analyzed for many water-quality indicators and constituents, including major ions, nutrients, trace elements, volatile organic compounds, pesticides, radionuclides, microbiological indicators, and some constituents of special interest (arsenic speciation, hexavalent chromium [chromium (VI)], and perchlorate). These groundwater-quality data, along with data from quality-control samples, are tabulated in this report and in an associated data release. Data for microbiological indicators for samples collected in 2016 are included in the companion data release.
Data Series↗
Evaluation of six methods for correcting bias in estimates from ensemble tree machine learning regression model
Ensemble-tree machine learning (ML) regression models can be prone to systematic bias: small values are overestimated and large values are underestimated. Additional bias can be introduced if the dependent variable is a transform of the original data. Six methods were evaluated for their ability to correct systematic and introduced bias. Method performance was evaluated using four case studies of groundwater quality: the units of the dependent variable were pH in two and log-concentration in the others. When performance metrics (bias and RMSE for both points and the CDF) were computed using the same units as those in the ML model, empirical distribution matching (EDM) provided the best results. When the metrics were computed using retransformed concentration, EDM and a method incorporating Duan's smearing estimate were both effective. A method based on the Z-score transform approximates EDM if the correlation coefficient between rank-ordered ML estimates and rank-ordered observations approaches one.
Environmental Modeling and Software↗
Lithium in groundwater used for drinking-water supply in the United States
Lithium concentrations in untreated groundwater from 1464 public-supply wells and 1676 domestic-supply wells distributed across 33 principal aquifers in the United States were evaluated for spatial variations and possible explanatory factors. Concentrations nationwide ranged from <1 to 396 μg/L (median of 8.1) for public supply wells and <1 to 1700 μg/L (median of 6 μg/L) for domestic supply wells. For context, lithium concentrations were compared to a Health Based Screening Level (HBSL, 10 μg/L) and a drinking-water only threshold (60 μg/L). These thresholds were exceeded in 45% and 9% of samples from public-supply wells and in 37% and 6% from domestic-supply wells, respectively. However, exceedances and median concentrations ranged broadly across geographic regions and principal aquifers. Concentrations were highest in arid regions and older groundwater, particularly in unconsolidated clastic aquifers and sandstones, and lowest in carbonate-rock aquifers, consistent with differences in lithium abundance among major lithologies and rock weathering extent. The median concentration for public-supply wells in the unconsolidated clastic High Plains aquifer (central United States) was 24.6 μg/L; 24% of the wells exceeded the drinking-water only threshold and 86% exceeded the HBSL. Other unconsolidated clastic aquifers in the arid West had exceedance rates comparable to the High Plains aquifer, whereas no public supply wells in the Biscayne aquifer (southern Florida) exceeded either threshold, and the highest concentration in that aquifer was 2.6 μg/L. Multiple lines of evidence indicate natural sources for the lithium concentrations; however, anthropogenic sources may be important in the future because of the rapid increase of lithium battery use and subsequent disposal. Geochemical models demonstrate that extensive evaporation, mineral dissolution, cation exchange, and mixing with geothermal waters or brines may account for the observed lithium and associated constituent concentrations, with the latter two processes as major contributing factors.
Science of the Total Environment↗
Machine learning predictions of nitrate in groundwater used for drinking supply in the conterminous United States
Groundwater is an important source of drinking water supplies in the conterminous United State (CONUS), and presence of high nitrate concentrations may limit usability of groundwater in some areas because of the potential negative health effects. Prediction of locations of high nitrate groundwater is needed to focus mitigation and relief efforts. A three-dimensional extreme gradient boosting (XGB) machine learning model was developed to predict the distribution of nitrate. Nitrate was predicted at a 1 km resolution for two drinking water zones, each of variable depth, one for domestic supply and one for public supply. The model used measured nitrate concentrations from 12,082 wells and included predictor variables representing well characteristics, hydrologic conditions, soil type, geology, land use, climate, and nitrogen inputs. Predictor variables derived from empirical or numerical process-based models were also included to integrate information on controlling processes and conditions. The model provided accurate estimates at national and regional scales: the training (R 2 of 0.83) and hold-out (R 2 of 0.49) data fits compared favorably to previous studies. Predicted nitrate concentrations were less than 1 mg/L across most of the CONUS. Nationally, well depth, soil and climate characteristics, and the absence of developed land use were among the most influential explanatory factors. Only 1% of the area in either water supply zone had predicted nitrate concentrations greater than 10 mg/L; however, about 1.4 M people depend on groundwater for their drinking supplies in those areas. Predicted high concentrations of nitrate were most prevalent in the central CONUS. In areas of predicted high nitrate concentration, applied manure, farm fertilizer , and agricultural land use were influential predictor variables. This work represents the first application of XGB to a three-dimensional national-scale groundwater quality model and provides a significant milestone in the efforts to document nitrate in groundwater across the CONUS.
Science of the Total Environment↗
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.
Groundwater↗