Geologic Setting

Geologic names in this report are consistent with prior work done by WVGES (McColloch and others, 2012). The southern West Virginia coal field (located in the central Appalachian coal region) is differentiated from the northern West Virginia coalfield (located in the northern Appalachian coal region) by depositional, stratigraphic, and lithologic controls (Ruppert and others, 2010). These two coalfields have distinct lithologies because of their differing paleoenvironmental settings and are separated by the hinge line defined by Arkle (1974), which differentiates the high sulfur northern coalfield (in the northern Appalachian coal region) from the low sulfur southern coalfield (in the central Appalachian coal region; Ruppert and others, 2010).

The boundary between the northern and southern West Virginia coalfields generally corresponds to the geologic contact between the older Kanawha and Allegheny Formations and the younger Conemaugh Formation (Ruppert and others, 2010). The depositional environments for the two coal regions were different. As a result, the lithologic composition of the older rocks comprising the southern West Virginia coalfield displays a higher proportion of sandstone and thin carbonate units than the lithologic composition of the northern West Virginia coalfield, which displays a higher proportion of shale, siltstone, and mudstone.

Data Limitations

The data used for this analysis are from several studies that utilized varied field collection and laboratory methods. As such, many samples from these studies have a specific number of analytes chosen for the purpose of each study, and representative sampling distributions are lacking when data are scrutinized at smaller scales. Although these data can be used for many purposes, multi-sourced data restrict the level of interpretation that can be applied. Multi-source data are also hard to validate, which then makes the qualification of the data necessary, casting uncertainty upon any resulting analysis.

Samples were not evenly distributed across all coal beds and within coal beds in this study—some coal beds had over 100 samples, whereas others had only 1 sample. Also, data interpretation and dataset utility are confounded by many sites having multiple samples. For this study, which is a regional analysis, all samples were considered to be independent observations to maximize the number of samples that could be used for univariate and multivariate statistics. Repeated samples from the same site or same mine can potentially skew the summary statistics, particularly if there is little to no variability among the samples or where only a few sites are used to represent entire coal beds. The use of repeated samples may indicate that some locations have water quality that varies over time or within the same mine complex. Using consistent field sampling and laboratory analysis methods and compiling addition site metadata would strengthen the dataset and subsequent analysis.

Determination of Net Alkalinity

Alkalinity and acidity balance is often used to determine treatment strategies for acidic coal mine discharges. For this study, calculated net alkalinity was used to understand the suitability of the water in abandoned underground coal mine aquifers for different uses because (1) data required for calculating net alkalinity are routinely collected for regulatory and research purposes, and (2) net alkalinity accounts for possible metal hydrolysis reactions that may precipitate undesirable minerals that adversely affect water quality (Rose and Cravotta, 1998). The acidity of a solution is represented by the release of hydrogen by oxidation, hydrolysis, and precipitation reactions that may result in the subsequent consumption of alkalinity. Acidity in acid mine drainage comes from elevated hydrogen ion, dissolved iron, aluminum, and manganese concentrations (Rose and Cravotta, 1998).

Net alkalinity or net acidity (in cases where the net alkalinity is negative), is reported in milligrams per liter (mg/L) as calcium carbonate and may be calculated by subtracting acidity (calculated by eq. 1; Kirby and Cravotta, 2005a) from field alkalinity (measured in the field by titration). Kirby and Cravotta (2005a) state that the preferred method of measuring net alkalinity is by standard hot acidity titration (American Public Health Association, 1995) but also show that net alkalinity may be calculated by equation 2 (Kirby and Cravotta, 2005a) when standard hot acidity titration results are unavailable. The relation between calculated net acidity and measured acidity has been confirmed by several authors (Rose and Cravotta, 1998; Kirby and Cravotta, 2005b). For this investigation, the calculated values were used to maximize the number of data points available for calculating net alkalinity.

To calculate acidity, Hedin and others (1994) proposed an equation like equation 1; however, they included the common redox states of iron—ferrous (Fe²⁺) and ferric (Fe³⁺). Kirby and Cravotta (2005b) determined that the effects of excluding ferric iron when calculating acidity were minimal for water with a pH range of 2.5 to 8.3 (most waters). For this report, the contribution of iron was assumed to be in the ferrous redox state because the available dataset did not distinguish between redox states and the pH was predominantly greater than (>)3. A more precise calculation of acidity can be determined by full speciation using geochemical modeling software (Kirby and Cravotta, 2005b), but this dataset had sparse data for which calculation of acidity from full speciation could be performed; therefore, acidity was calculated by equation 1.

Statistical Analyses

Univariate and multivariate statistical methods were used to evaluate constituents affecting abandoned underground coal mine aquifer water quality, the types of water in abandoned underground coal mine aquifers, and how abandoned underground coal mine aquifer water quality varies throughout the State. Statistical analyses were completed, and graphics were created, using the R statistical computing environment (ver. 3.5.3) (R Core Team, 2021). Nonparametric techniques were used to compute descriptive and multivariate statistics from water-quality data; data that were sometimes censored at multiple levels. Censored values are water-quality results that are reported at less than a specified laboratory value and for which an exact value was not determined.

Prior to descriptive and multivariate statistical analyses, censored data were coded as u-scores with the codeU function in the U.S. Geological Survey smwrQW package (Lorenz, 2018). The u-score is the sum of the algebraic sign of the differences between each value. Using u-scores allows for the computation of multivariate relations without censoring at the highest reporting limit and retains information at multiple reporting limits. In cases where a column of data has only one censoring level, the u-scores are the same as ordinal methods of ranking for one reporting limit (Helsel, 2012).

Summary Statistics for Censored Data

The nonparametric models—Kaplan-Meier (KM) and robust regression on order statistics (ROS)— were used to estimate the summary statistics for censored data following the methods described by Ryberg (2006) and Helsel (2012). When more than 80 percent of a constituent was censored, the maximum concentration was reported. The minimum concentration was reported as less than the lowest reporting level. When a sample had 50 to 80 percent of its values censored, the robust ROS model was used. The robust ROS model was used for this range of censoring because of its ability to make accurate estimates and handle multiple censoring levels with a small number of observations (less than [<]50 observations). When the data were censored at >50 percent, the KM model was used to estimate the summary statistics. The KM estimate was used to account for multiple censoring levels, because it does not depend on the assumption of a distributional shape with data that are censored at >50 percent. When no values were censored, nonparametric methods were not necessary and summary statistics were computed using standard methods of statistical analysis.

Truncated boxplots are used throughout this report to display statistical distributions of data (fig. 4). Values below the 10th and above the 90th percentiles are not shown in figures throughout this report, but minimum and maximum values are reported along with summary statistics in table 1. The interquartile range is represented by the box in the boxplot with the whiskers extending to the 10th and 90th percentiles. Samples may have many reporting levels that were used in the calculation of the summary and multivariate statistics, and the dotted lines on boxplots represent the highest censoring level. Where the robust ROS method is used to estimate the statistical distribution, a red line represents imputed values above the highest censoring level.

Water Quality in Relation to Drinking Water Standards

Assessing the quality of water within underground mines in relation to drinking water standards is important because many private households and public utilities in the southern part of the State use abandoned underground coal mine aquifers as the primary source of water (Kozar and others, 2020). Constituents that exist in groundwater because of natural processes or are introduced by human activities may pose a risk to health where concentrations of those constituents exceed U.S. Environmental Protection Agency (EPA) maximum contaminant levels (MCLs; also known as primary drinking water standards). The MCLs that regulate public supplies can also be used as guidance for private households (U.S. Environmental Protection Agency, 2018). Other EPA water-quality guidelines include secondary maximum contaminant levels (SMCLs) and health advisories (HAs). SMCLs are given for select constituents that have no known health risks when in drinking water but may still produce bad taste, odor, or staining. The HAs can be listed for constituents that have no MCL or in addition to an MCL (U.S. Environmental Protection Agency, 2019). It is important to note that samples in this analysis were collected and analyzed using many different methods, at different times of the year, and under varying conditions. Chemistry at these sites may vary over time or on a seasonal basis, and these results may not fully characterize the water quality or risk to human health. Additional analyses can be completed at individual sites under consideration as potential water supply sources to understand the full range of analyte concentrations.

The number of samples and minimum, maximum, and median concentrations are provided for several analytes in the dataset (table 1). These analytes were compared to MCLs, HAs, and SMCLs and are displayed as the number and percentage of samples exceeding those standards. Several analytes in table 1 do not have drinking water standards or were not detected at levels exceeding drinking water standards, so they are not explicitly discussed relative to drinking water standards. The only field parameter in the dataset that had a drinking water standard is pH (table 1), and it is represented by a SMCL range of 6.5–8.5 pH units. Five hundred and fifty-six samples had field pH values in the dataset, and 356 of those samples had values that were outside of the SMCL range, comprising 64 percent of the total. Laboratory pH was reported in 270 samples, of which only 30 samples (11 percent) were outside of the SMCL range. Relatively high values, exceeding the 500 mg/L SMCL, were reported for total dissolved solids concentrations in 278 samples (77 percent of the samples)

Measurements of major ions, nutrients, and trace constituents were included in this study’s dataset. Five major ion analytes exceeded drinking water standards, and sulfate is the analyte that most frequently exceeds its drinking water standard. Elevated sulfate concentrations are common in mine discharges and can contribute to the non-carbonate hardness and anionic balance in a sample. Water types are characterized by major ions, and sulfate makes up the largest percentage of anion mass (fig. 5). Sulfate exceeded the 250 mg/L SMCL in 457 (69 percent) of 656 samples (table 1), and 652 samples (99 percent) were above the laboratory reporting limit for sulfate, which indicates that high concentrations of sulfate are prevalent throughout the study area.

Commonly monitored and identified trace constituents in mine discharges include iron, aluminum, and manganese. Out of the 43 dissolved and total trace constituents in table 1, these three analytes have the most reported values in the dataset. The dissolved (that is, filtered) component of the trace constituents was more commonly analyzed than the total (that is, unfiltered) component, but there was usually less than a 10-percent difference between dissolved and total (table 1). This report focuses on the dissolved component, but understanding which component is reported is important because that information may have an effect on the results. Iron had 619 dissolved samples with 490 (79 percent) above the 300 micrograms per liter (μg/L) SMCL, aluminum had 547 dissolved samples with 349 (63 percent) above the lower SMCL (50 μg/L), and manganese had 368 dissolved samples with 333 (90 percent) above the 50 μg/L SMCL (table 1). Manganese also has a HA (300 μg/L), and 284 manganese samples (77 percent of the total) exceeded this HA. Other constituents related to the geochemistry of acid mine drainage also exceeded drinking water standards; 24 percent of arsenic samples exceeded the MCL of 10 μg/L.

Net Alkalinity in Abandoned Underground Coal Mine Aquifers

In addition to drinking water standards, water discharging from mines is often characterized by net alkalinity or net acidity. Net alkalinity (eq. 2) is the measurement primarily used to determine the requirements for the site-specific treatment of acid mine drainage but may also be used to evaluate general measures of water quality and water’s suitability for a particular application. Calculation of net alkalinity (determined by using eqs. 1 and 2), instead of measured acidity, affords a better spatial representation because the constituents required for computation of net alkalinity are more likely to be available than complete water chemistry data needed to calculate acidity from aqueous speciation and mineral equilibria (Kirby and Cravotta, 2005b).

Before acidity was calculated to determine net alkalinity, differences in filtered and unfiltered iron and aluminum and manganese values were evaluated to assess the effect of filtration on the measured analyte concentrations. Several authors have shown the differences between filtered and unfiltered sample values, as well as the differences between laboratory pH and field pH values (Gammons and others, 2000; Kirby and Cravotta, 2005b). These differences in reported values may be because of changes in equilibrium caused by oxidizing conditions and may or may not represent in situ aquifer conditions. Many samples in this dataset had filtered and unfiltered results, as well as field and laboratory pH measurements, which permitted examination of any potential effects filtering may impose on elemental concentrations and pH (figs. 6 and 7). Iron and aluminum clearly show deviation from the 1:1 line and substantial increases (>100 percent) in the concentrations of iron and aluminum observed in some unfiltered samples. The processes controlling iron and aluminum concentrations in mine water have been explained by several investigations (Rose and Cravotta, 1998; Bigham and Nordstrom, 2000; Kirby and Cravotta, 2005a). Those investigations found that differences in filtered and unfiltered samples are commonly the result of metal oxidation and hydrolysis reactions, which may precipitate iron and aluminum hydroxide or hydroxysulfate minerals. In unfiltered samples, these precipitates are redissolved by nitric acid, a preservative used for samples in this study, and the resulting measured concentration represents the total of the dissolved and mineral component. In addition to inorganic precipitation of iron-bearing mineral species, additional contributions from iron-precipitating bacteria may exacerbate aquifer degradation (Smith and Tuovinen, 1985). Filtering the sample before preservation removes these mineral particulates and reduces microbially mediated transformations, resulting in concentrations that represent only the dissolved fraction of the sample that can pass through a filter. For this study, only filtered values of iron and aluminum were used for the calculation of acidity (eq. 1) to reduce the possibility of overestimating acidity caused by the difference in filtered and unfiltered samples.

When possible, filtered manganese samples were used for the calculation of acidity, but unfiltered samples were used when filtered samples were not available. Unlike iron and aluminum, a strong relation was observed between filtered and unfiltered manganese concentrations (fig. 6), indicating that the dissolved fraction generally contained the majority of the manganese. Because of this linear relation observed in figure 6, manganese concentrations in filtered and unfiltered samples were assumed to not differ greatly. The use of filtered and unfiltered manganese values increased the number of values that could be used to calculate acidity by 179.

Although unfiltered manganese concentrations were also used in this study to increase the number of data points, manganese can precipitate under certain geochemical conditions, and this may not be suitable for every study. Kinetics of manganese oxidation are strongly affected by pH and the abiotic oxidation of manganese is very slow at pH<8, which includes most of the samples in this study. Hydrolysis of iron and manganese in aqueous solutions is a sequential process; most of the iron must be removed before manganese (Hedin and others, 1994), making manganese precipitation less likely in these samples.

Field-measured pH data and laboratory-measured pH data had bimodal distribution groups of near neutral (around 7) and acidic (below 4) when compared (fig. 7). The near neutral group had higher values for laboratory-measured pH, and the acidic group had slightly higher values for field-measured pH. Similar observations were made by Kirby and Cravotta (2005b), who attributed this difference of values to the disequilibria of field pH measured in mine water exposed to atmospheric conditions at the time of sample collection. pH becomes stable over time with equilibrium among gaseous, aqueous, and mineral components of the sample, which is likely the reason for discrepancies in pH from field values measured at the time of sample collection and laboratory values measured after equilibration. For this study, field pH was used to calculate acidity when possible, and laboratory pH was only used when field pH was missing. Equilibrium reactions that affect pH may include metal hydrolysis reactions that precipitate minerals and decrease pH or dissolution of carbonates that may increase pH and alkalinity, but these reactions are accounted for when calculating acidity and net alkalinity using equations 1 and 2.

Net acidity and laboratory-measured acidity were compared in 150 samples (fig. 8). A linear relation demonstrating good agreement among values >10 mg/L as calcium carbonate was observed between these two measures of acidity in this study, which is consistent with other studies (Rose and Cravotta, 1998; Kirby and Cravotta, 2005b) and validates the use of net acidity values calculated by way of equation 2. Sites with negative net acidity were not included in the comparison because acidity reported by the laboratory was censored at values of 5 mg/L as calcium carbonate, allowing no negative values to be available. At low values of net acidity (<10 mg/L as calcium carbonate), acidity reported by the laboratory is consistently higher than the net acidity calculated by equations 1 and 2. This is consistent with the findings of Kirby and Cravotta (2005b), but their samples were net alkaline, and the divergence from linearity may have been because of calcite precipitation. Calcite precipitation is unlikely to happen with samples in this study because they are net acidic and unlikely to precipitate calcite at low pH. Ultimately, the number of values at this range of acidity are sparse, so the divergence from linearity cannot be definitely determined from these data.

Sufficient data were available to compute the net alkalinity of 387 samples from 180 sites, which represented 134 individual underground mines (table 2). Twenty-three coal beds are represented by at least 1 sample. Data were collected at these sites from April 5, 1978, to September 7, 2016; however, 89 percent of samples were collected after January 1, 2000 (McAdoo and Kozar, 2017b). The coal beds in table 2 are in stratigraphic order (McColloch and others, 2012); the table demonstrates that the younger coal beds of northern West Virginia have most of the available data used to calculate acidity.

The northern region of West Virginia contains mostly younger coal beds, including the Sewickley, Pittsburgh, Bakerstown, Upper Freeport, Lower Kittanning (also mined to a lesser degree in the southern region), and Number 5 Block (fig. 2), and the southern region contains mostly older coal beds spanning the Stockton to Pocahontas 3 (table 2). The coal beds in the northern part of West Virginia show a wide distribution of net alkalinity: the Pittsburgh coal bed has the lowest alkalinity values (−5,292 mg/L as calcium carbonate), and the Upper Freeport bed has the highest alkalinity value (1,761 mg/L as calcium carbonate). Most samples from the northern part of the State had negative net alkalinity (fig. 9), which is indicative of high metal concentrations, low pH, and low alkalinity. Net alkalinity values calculated for the southern part of West Virginia (n equal to [=] 133) had a tighter distribution of values (fig. 9); the lowest value came from the Pocahontas 3 coal bed (−181 mg/L as calcium carbonate), and the highest value came from the Eagle coal bed (620 mg/L as calcium carbonate).

Sulfur content in coal has been directly linked to the subsequent water quality of mine discharges; poor water quality is directly related to high sulfur content in coalbeds (Rose and Cravotta, 1998). Moving up the stratigraphic column (from older to younger formations), water quality clearly changes from being primarily positive (net alkaline) to being primarily negative (net acidic). This change occurs in the Middle Pennsylvanian age as sulfur content increases between the Number 2 Gas and Coalburg coal beds (fig. 10). Cecil and others (1985) concluded that the sulfur content of older coal beds in the southern part of the State is lower than the sulfur content of the younger coal beds in northern West Virginia. They also proposed that this difference in sulfur content is the result of a shift in climate from an ever-wet marine tropical setting in the Early Pennsylvanian age to a subtropical fluvial depositional setting in the Middle and Late Pennsylvanian age. This change in climate fundamentally altered peat bog morphology (for example, domed or planar), leading to contrasting geochemical conditions that increased the sulfur content of coal deposits in the Middle Pennsylvanian age at the contact of the Kanawha and Allegheny Formations. This stratigraphic and spatial hinge line in the Middle Pennsylvanian coal beds was observed by Cecil and others (1985) and coincides with the transition of net alkaline to net acidic water quality between the Number 2 Gas and Coalburg coal beds (fig. 10). Further evidence that climate has affected sulfur content is corroborated by the general overlap of coal beds with high sulfur content and the subsequent deposition of marine sediments (Williams and Keith, 1963); all this evidence suggests that a sustained temporal shift in paleoenvironmental conditions took place from the early to late Pennsylvanian age.

Sites Grouped by Water-Quality Characteristics

The primary output from HACA is a dendrogram depicting the clustering structure of the input data (fig. 11). Individual samples are represented by vertical lines at the bottom of the figure, and the merging of similar clusters is represented by horizontal lines connecting clusters. Clusters that merge near the bottom of the dendrogram have similar water chemistry. Clusters that do not merge until the top of the dendrogram represent groups of samples that have contrasting water chemistries. Specific analytes and parameters used for the HACA in this study included pH, SC, alkalinity, aluminum, calcium, iron, magnesium, manganese, potassium, sodium, chloride, and sulfate which resulted in a HACA with 102 samples. These samples were collected at 81 different sites which represented 71 individual mines and 16 different coal beds.

The major cluster groups in this study’s HACA were represented by capital letters (A, B, and C), and cluster subgroups were represented by numbers (1 through 6; fig. 11). The agglomerative coefficient—a value between 0 and 1 used to assess the HACA (Kaufman and Rousseeuw, 1990)—was 0.935, a number that provides confidence in the HACA structure. The similarity profile test identified 11 significant clusters (p<0.001), 6 of which were used for further analysis. Most of the samples in the HACA are from the Pittsburgh coal bed (42), whereas several coal beds only had 1 sample in the HACA (table 3). The Pittsburgh, Sewell, and Upper Freeport coal beds have the highest number of samples in the HACA and show variability among major cluster groups. Variability among coal bed clusters could be a product of greater sample density within a cluster or other factors in samples, such as mine age, depth of the coal mine, degree of flooding, and subsequent mine recovery. The specific reason why clusters are classified differently within a specific coal bed is not apparent with these data. Group A has 36 samples that are generally net alkaline but has a nearly neutral pH; these samples also have lower SC relative to groups B and C (fig. 12). Group B has 38 samples that range from net alkaline to net acidic (pH ranging from <5 to >7). Group B has the highest values for SC. Group C’s 28 samples had the lowest values for net alkalinity and pH.

Distinct differences in major ion chemistry were observed between HACA cluster groups (fig. 13). Cluster group A’s samples have the widest distribution of water types ranging from samples with higher proportions of sodium and bicarbonate to samples with higher proportions of calcium and sulfate. Most samples in cluster group B showed higher proportions of calcium and sulfate but five samples in this cluster showed higher proportions of sodium. Samples from cluster group C had the smallest distribution of water types; all samples had higher proportions of calcium and sulfate.

Cluster subgroups are within larger cluster groups (A, B, and C) and represented by numbers 1 through 6 (fig. 11; table 3). Subgroup 1 has the lowest number of samples (6), and subgroup 2 has the highest number of samples (30; fig. 14). These two subgroups have the lowest values for acidity, aluminum, iron, and manganese, which is expected because cluster group A has relatively high net alkalinity and pH (fig. 12). Cluster subgroups 3 through 6 all had high acidity relative to subgroups 1 and 2. However, the metals associated with acidity in these subgroups varied. Cluster subgroup 6 had the highest median values for aluminum and manganese, whereas cluster subgroup 3 had the highest median values for iron (fig.14).

Cluster group B had the highest values for SC (fig. 12), which is explained by the high values for major ion chemistry (fig. 15). Subgroups 3 and 4 have the highest values of the chemical constituents that contribute to SC, values such as calcium, magnesium, sodium, chloride, and sulfate. Higher concentrations of calcium and magnesium indicate higher rates of carbonate dissolution which buffers acidity by adding alkalinity. Sodium and chloride may be higher in the samples of subgroups 3 and 4 because they may be representative of chemistry located deeper in the subsurface or because of cation exchange of calcium and magnesium with sodium.

The spatial distribution of cluster groups across West Virginia shows a distinct difference between the northern and southern coal fields in the State (fig. 16). The southern part of the State has more samples from cluster group A than from B and C. The northern part of the State has more samples from cluster groups B and C than A. Abandoned underground coal mine aquifers in the southern part of the State have more samples that are characterized by relatively high alkalinity (net alkaline throughout) and low concentrations of aluminum, iron, and manganese (table 4). In contrast, the northern part of the State is characterized by low to negative net alkalinity, higher SC, and higher concentrations of aluminum, iron, and manganese.