Purpose and Scope

This report describes methods and results of a two-part study designed to (1) assess water use and temporal trends and changes in streamflow, and to (2) characterize baseline water quality (2021–23) in eastern Ohio, as they relate to hydraulic fracturing and (or) other oil and gas extraction-related activities. Streamflow time-series data from eight gages were used to assess temporal trends and changes in streamflow. Water-quality data were collected at about 3-month intervals between 2021 and 2023 at eight sampling sites on streams near the eastern edge of Ohio where shale-gas drilling is particularly focused (fig. 2). Six of the sampling sites were colocated with gages whose streamflow data were used to assess temporal trends and changes in streamflow, and the other two sampling sites were colocated with other gages.

Coal Mining

The study area lies within the Ohio coal region where more than 4 billion tons of coal have been extracted since 1800 (Wright, 2022). Lands associated with coal mines (active, remediated, and abandoned) are present in all eight drainage basins used to establish baseline water quality—although only about 0.2 percent of the Little Muskingum River (station 03115400) drainage basin contains lands associated with coal mines (fig. 4). In 2022, Belmont County was the second highest coal-producing county in Ohio (Wright, 2022). Mine drainage can increase specific conductance and concentrations of total dissolved solids, sulfate, iron, manganese, arsenic, aluminum, total carbon, and dissolved organic carbon in receiving streams and cause water quality in drainage basins to be impaired so that they do not support a normal assemblage of aquatic life (Helsel, 1983; Haefner and Simonson, 2010; Ohio Department of Natural Resources Division of Mineral Resources Management [ODNR–DMRM], 2016).

Oil and Gas Drilling

Oil and gas drilling has been active in Ohio for more than 150 years (Ohio Oil & Gas Association, 2013), most recently in the eastern one-third of the State. As of March 2023, there were approximately 3,794 vertical oil and gas wells and 438 horizontal oil and gas wells in the drainage basins used to establish baseline water quality (fig. 3). Most of the vertical wells were completed in the Ohio Shale (23.6 percent; Late Devonian), Gordon Sandstone (13.0 percent; Pennsylvanian), Berea Sandstone (10.0 percent; Late Devonian), Clinton Sandstone (3.7 percent; Early Silurian), and Clinton Group (3.2 percent; Early Silurian). Most of the horizontal wells were completed in the Point Pleasant Limestone (90.4 percent), Utica Shale (8.0 percent), and Marcellus Shale (1.6 percent; ODNR–OGRM, 2023b).

Waste injection wells, depending on their type, can be used to dispose of hazardous and (or) nonhazardous fluids in deep underground porous geologic formations. Seven active and nine plugged waste injection wells are within the study basins used to establish baseline water quality—one active injection well each in the Stillwater Creek (station 03125338) and Leatherwood Creek (station 03141870) drainage basins and three plugged injection wells in the Little Muskingum River (station 03115400) drainage basin. The Duck Creek (station 03115786) drainage basin contained the most injection wells—five active and six plugged (fig. 3; ODNR–OGRM, 2023b). In addition to oil-field fluids from Ohio, fluids disposed of in Ohio injection wells may originate from out of state.

The ODNR–DMRM regulates the spreading of oil-field fluids on township roads for dust control in the State of Ohio. Within the eight drainage basins of the baseline water-quality sites, permits to spread oil-field fluids were issued during the study sampling period (2021–23) to two townships in Carroll County, two townships in Harrison County, two townships in Monroe County, three townships in Noble County, and three townships in Washington County (Wendy Booth, ODNR–OGRM, written commun., 2023).

Streamflow Trend and Change Methods of Analysis

The “Exploration and Graphics for RivEr Trends” (EGRET) package (Hirsch and De Cicco, 2015) provides a variety of tabular and graphical outputs that can be used to evaluate whether selected streamflow characteristics are changing over time. EGRET analyses of long-term changes in streamflow characteristics are based on time-series smoothing methods pioneered by Cleveland (1979) and Cleveland and Devlin (1988). EGRET performs Locally WEighted Scatterplot Smoothing (lowess) on annual streamflow statistics (relevant to low, high, and mean flows) to produce plots that show patterns of change over time spans of about a decade or more. Lowess is a nonparametric method that can determine the coordinates of a smooth line (hereafter referred to as a “lowess smooth”) representing the central tendency of the relation between dependent and independent variables. Plotting the lowess smooth along with the annual data makes it easier to visualize how the dependent and independent variables covary. When the independent variable is time, the lowess smooth shows the central tendency of how the dependent variable has changed over time. A detailed description of the smoothing method can be found in Hirsch and others (2010).

Mann-Kendall tests (Mann, 1945) were used in some cases to assess temporal trend. The Mann-Kendall test makes no assumption about the shape of the trend (for example, whether it is linear or nonlinear) but, instead, tests whether values tend to increase or decrease overall. The null hypothesis for the Mann-Kendall test is that there is no trend in the series. Unless stated otherwise, all tests of statistical significance described in this report are based on a significance level (alpha value) of 0.05. Sen’s slope (the median of the slopes of all lines through all pairs of points) was computed to determine the median linear rate of change in the natural logarithms of annual streamflow statistics (Sen, 1968).

Quantile-Kendall plots were prepared for each study gage using the EGRET framework and methods described by Hirsch (2018). In a quantile-Kendall plot, daily streamflow nonexceedance probability is plotted versus the trend slope for streamflows associated with those probabilities. The trend slope is determined with Sen's slope (Sen, 1968) and expressed in percent change per year. Larger nonexceedance probabilities are associated with larger streamflows and smaller nonexceedance probabilities are associated with smaller streamflows. Each point in the Quantile-Kendall plot is color coded according to the p-value for a test of the null hypothesis that the trend slope is zero. Points plotted in red indicate a statistically significant likelihood that streamflows at those nonexceedance probabilities have nonzero trend slopes.

The Indicators of Hydrologic Alteration (IHA) software, version 7.1 (Nature Conservancy, 2009), was used to compute 67 potentially ecologically relevant statistical parameters from the USGS daily streamflow time series for study and reference gages. The IHA statistical parameters provide information on high, low, and median flow characteristics. This study focused on the potential impact of water use for development of hydraulic fracturing wells; therefore, only statistics related to low flows were examined because that is where the impact of water withdrawal would be most likely to be observed. The statistics examined include the following:

The classification of daily streamflow data into different flow categories was based on nonparametric criteria. The IHA software makes three passes through the daily streamflow data for each gage to assign each day to a flow class. During the first pass, each day was initially classified as a low or high flow. During the second pass, all days that were initially classified as high flows were reclassified as either a small flood, large flood, or high-flow pulse. During the third pass, some of the initial low-flow days were reclassified as extreme low flows. Flows that exceeded the 75th percentile of the daily flows for the period were classified as high flows. All flows below that level initially were classified as low flows; however, low flows were subsequently reclassified as extreme low flows if they were less than the 10th percentile of daily flows. High-flow pulses were defined as contiguous periods when the daily flow was greater than or equal to 75 percent of the flow values and low-flow pulses were defined as contiguous periods when the daily flow was less than or equal to 25 percent of the flow values. A screen capture of the Environmental Flow Component selections used in the IHA analyses is shown in figure 7. For more detail on the classification process, see the IHA user’s manual (Nature Conservancy, 2009).

The gages whose streamflow records were analyzed (including Captina Creek [station 03113990] where records from two gages were merged) had between 21 and 85 equivalent years of observed flows (table 3, determined by dividing the number of reported daily flows divided by 365.25 days/year) and averaged about 68.7 years. The IHA user’s manual (Nature Conservancy, 2009) suggests 20 years of record as a baseline requirement for the amount of data needed for an IHA analysis; however, it states that more years of record may be needed depending on the (1) degree of climate variability, (2) the frequency of occurrence or variability of the particular parameter, (3) the severity of the hydrologic alteration that you are trying to detect, (4) and whether the goal is to characterize the central tendency or range of interannual variability.

R language (R Core Team, 2017) scripts were developed to process outputs from the IHA analyses and create time-series plots with lowess smooths overlaid on the annual IHA statistics. The lowess smooths were used to assess temporal trends in the IHA statistical parameters. In particular, the lowess smooths were examined to see whether there was an observable change in slope of the smoothed line around 2012 when annual hydraulic fracturing water use in Ohio first exceeded 100 million gallons. If withdrawals for hydraulic fracturing began to appreciably affect flows around 2012, one might expect the annual 1- and 7-day minimum flows and the median of extreme low-flow period minimums to show indications of decline beginning around 2012. Also, metrics describing the frequency and duration of low-flow pulses and extreme low-flow periods might increase (indicating more frequent and (or) longer low-flow pulses and extreme low-flow periods) because of the withdrawals.

The daily streamflow records for each gage were also analyzed to determine daily flows associated with selected daily nonexceedance probabilities (that is, the probabilities that selected flows are not exceeded) in each water year. In the absence of persistent streamflow-altering factors such as climate change or changes in regulation, if withdrawals for hydraulic fracturing started to appreciably reduce streamflows beginning around water year 2012, then streamflows associated with at least some nonexceedance percentiles should decrease. In that case, points in a scatterplot of streamflows versus nonexceedance percentiles computed for the entire period of record should plot on or below points computed using only data from the period before water year 2012.

Streamflow Trend and Change Results

The sections “Trends in Annual Minimum and Median Daily Streamflows and Annual Daily Streamflow Nonexceedance Probabilities,” “Temporal Trends in Indicators of Hydrologic Alteration (IHA) Statistics” and “Changes in Nonexceedance Percentiles of Daily Streamflows” describe results from analyses designed to assess temporal trends in a variety of streamflow characteristics. EGRET analyses provided information on trends in climatic-year (April 1–March 31, designated by the beginning year) minimum daily streamflows, water-year (October 1–September 30, designated by the ending year) median daily streamflows, and flows over a range of nonexceedance probabilities. Analyses of statistics determined with the IHA software program provided information on temporal trends in the magnitude, frequency, and duration of a variety of low-flow characteristics. Finally, analyses of nonexceedance percentiles of daily streamflows provided information about the relative magnitude of flows in the lower quartile of daily streamflows before and after annual hydraulic fracturing water use in Ohio first exceeded 100 million gallons.

Trends in Annual Minimum and Median Daily Streamflows and Annual Daily Streamflow Nonexceedance Probabilities

Sen’s slopes of the period of record climatic-year annual minimum daily streamflows and water year median daily streamflows were positive for all reference and study gages (table 3); however, the slopes were not statistically different than zero for some gages. Figures 8 and 9 show examples of trend plots for a study gage and a reference gage, respectively. The Sen’s slope value was statistically significant for the data shown in figure 8 but not in figure 9. Irrespective of statistical significance, the uniformly positive Sen’s slopes for the study and reference gages indicate a consistent tendency for low and median flows to have increased over the periods of record. Quantile-Kendall plots showed positive trend slopes for flows with nonexceedance probabilities of 0.90 and less for all reference and study gages (see figures 10 and 11 for example). While trend slopes were not statistically significant for all nonexceedance probabilities, these results indicate a consistent pattern for all but high flows to have increased over the periods of record at both the study and reference gages.

Changes in Nonexceedance Percentiles of Daily Streamflows

The percentage of daily streamflows that did not exceed a given streamflow magnitude is referred to as its nonexceedance percentile. Nonexceedance percentiles were computed for the reference and study gages for two time periods. The first time period was computed based on the record through water year 2011 (referred to as the “pre-2012 period”), before annual withdrawals for hydraulic fracturing exceed 100 million gallons in Ohio. The second time period was based on the entire period of record through water year 2021 (referred to as the “full period”). Scatterplots of nonexceedance percentiles less than or equal to 25 percent (the lower quartile of daily streamflows) versus streamflow were prepared for each gage record for both the pre-2012 and full periods. The streamflows associated with each nonexceedance percentile determined for the full period of record plotted at or, more commonly, above those determined for pre-2012 period for all study and reference gages (for examples, see figures 12 and 13). If low flows decreased (relative to the pre-2012 period) in the post-2011 period when withdrawals were being made for hydraulic fracturing, the percentage of observations that did not exceed a given low-flow value would increase when the two periods were combined. In fact, the opposite was observed, indicating that low flows did not decrease during the post-2011 period.

Previous Studies

The study area has an extensive history of coal, oil, and gas production and subsequently, several water-quality studies have been done in eastern Ohio (Pfaff and others, 1981; Engelke and others, 1981; Helsel, 1983; Jones, 1988; Sedam, 1991; Sedam and Francy, 1993; ODNR–DMRM, 2016; OEPA, 2017). More recently, the U.S. Geological Survey, in cooperation with the Muskingum Watershed Conservancy District, did a study to assess baseline (2015–16) surface-water quality at 30 surface-water sites within the Muskingum River watershed that were in the early years of shale-gas development (Covert and others, 2018).

Sampling Methods and Laboratory Analysis

Water-quality data were collected eight times at each of eight sites from June 2021 through March 2023. Sampling was done during a variety of flow conditions and included field parameters and major ions, trace elements, and nutrients (table 4). When sites were fully wadable, multiple vertical, depth-integrated, isokinetic samples were collected with a DH–81 sampler using an equal-width-increment method described in the USGS National Field Manual (USGS, variously dated). When sites were not fully wadable, multiple vertical, depth-integrated, isokinetic samples were collected from one or more points in the wadable portion of the stream or, if available, from a nearby bridge. Samples from the multiple verticals were composited in a churn splitter and then processed using standard USGS methods (USGS, variously dated; Wilde and others, variously dated). All samples were analyzed for turbidity, total dissolved solids, major ions, trace elements, nutrients, and dissolved organic carbon (table 4). Field measurements of water temperature, specific conductance, dissolved oxygen concentration, pH, and alkalinity were made onsite according to standard protocols (USGS, variously dated). Water temperature, specific conductance, pH, and dissolved oxygen were measured using YSI data sondes, and the reported value of each parameter was the median of at least three readings taken across the stream cross section. Sulfide concentration was measured in the field with a Hach DR2000 spectrophotometer by use of the methylene blue method (Hach Company, 1989) for grab samples collected in the thalweg. Turbidity was measured in the field using a portable Hach 2100P turbidimeter on a subsample of the unfiltered, composited sample (obtained from churn splitter). Alkalinity was measured in the field using a Beckman 250 pH meter on a subsample of the unfiltered, composited sample (obtained from churn splitter). Analyses for total dissolved solids, major ions, trace elements, nutrients, and dissolved organic carbon were done at the USGS National Water Quality Laboratory (NWQL) in Denver, Colorado. Analytical methods are listed in table 4.

The terms “detection limits,” “reporting levels,” and “censored values” are often used in discussions of chemical concentrations. The detection limit is traditionally defined as the lowest concentration (or amount) of an analyte that can be measured and reported with 99 percent confidence that the analyte concentration is higher than zero (Electronic Code of Federal Regulations, 2018). The reporting level is the lowest measured concentration that may be reliably measured and is statistically determined (USGS–NWQL, 2015). Censored values are values less than the reporting level and are reported using a less-than sign (<) in front of the reporting level value. The analytical results are available online from the USGS National Water Information System (NWIS) database (USGS, 2023a). USGS station numbers (table 1) and parameter codes (table 4) are needed to retrieve these data.

Geospatial Analyses

Geospatial data were obtained and compiled using a geographic information system (GIS; ArcMap [Esri, 2023]) to assess potential relations between geospatial characteristics and water quality. Drainage basin boundaries for the sampling sites were delineated using StreamStats—a USGS, GIS-based web application (USGS, 2019). The drainage basins were used to aggregate and summarize land cover data and other characteristics within each sampling site’s upstream drainage basin. Summary geospatial data were divided by the drainage area of the site—producing a number per square mile—to better facilitate comparisons among sites with different sized drainage areas.

The following geospatial data types, sources, and transformations were used in analysis of:

Statistical Analyses

Descriptive and summary statistics for censored data (those with values less than the reporting level) and uncensored data were computed in Statistical Analysis System (SAS) JMP version 14.3.0 statistical and graphing software (SAS Institute, Inc., 2023). Censored-style boxplots were created using a USGS-developed function (boxPlot.lcens/smwrQW) in RStudio software package (RStudio, Inc., 2023) to compare the range, median, and outliers of the water-chemistry data. The Wilcoxon rank-sum test (Wilcoxon, 1945) was used to test differences among median constituent concentrations in samples with a brine component versus samples containing no brine. Spearman’s rank correlation coefficients (Spearman, 1904) were used to identify correlations between water chemistry and drainage-basin characteristics that could affect water quality. The Spearman’s rank correlation coefficient is a nonparametric test that measures the strength and direction of monotonic association between two variables. The Spearman’s rank correlation coefficient is computed on the ranks of the data values instead of on the values themselves. All concentrations less than the reporting level for a constituent were assigned a single value that was less than all reported concentrations, so that ranks of values less than the reporting level were tied, yet less than the rank for the lowest measured concentration. The strength of the relation is denoted by the Greek letter rho—the closer rho is to +1 or −1, the stronger the association. Positive values of rho signify a positive correlation and negative values of rho signifies a negative correlation. For purposes of this report, the strength of the correlation between two properties or constituents is defined as follows, based on the absolute value of rho: very strong correlation 0.80–1.0; strong correlation 0.60–0.79; moderate correlation 0.40–0.59; and weak correlation 0.20–0.39.

Data Treatment Based on Quality-Control Results

Sampling and analytical variability were evaluated using replicate analyses. When analysis of water-quality data identified statistically significant differences between samples containing a component of brine versus samples without a brine component, the authors assessed whether the observed difference in median concentrations between groups was higher than the maximum difference in concentration between QC replicate pair samples. There were no instances of differences being the result of sampling and analytical variability rather than actual environmental differences between groups.

Hydrologic Conditions Captured by Sampling Events

To characterize water chemistry over a range of seasons and hydrologic conditions, a total of eight samples were collected at about 3-month intervals at each of eight water-quality sampling sites. The variation in hydrologic conditions that was captured by the sampling was assessed by creating boxplots showing the distribution of the 15-minute instantaneous streamflows (or gage heights) at the sampling sites for the entire study period (6/23/2021–3/16/2023) and overlaying them with points representing the streamflows (or gage heights) at the times of sampling (fig. 14). Continuous streamflow was not determined at Stillwater Creek (station 03125338) during the study period, so its sampled variation is shown using gage heights in figure 14. Streamflows (and gage heights for Stillwater Creek [station 03125338]) at the times of sampling were distributed throughout the box representing the interquartile range (the range containing 50 percent of the observed instantaneous values), and frequently extended a little above and below the interquartile range. That indicates that although sampling did not capture the extremes in flow (both high and low), it did capture a wide range of flows, typically including one or more streamflows in the first and fourth quartile of streamflows that occurred during the sampling period.

Water Types Based on Major Ions

Patterns in major-ion composition at eight sampling sites were examined by constructing trilinear diagrams (fig. 15). This type of diagram uses the relative percentages of major cations and anions to classify water types (Hem, 1985). The study samples were plotted with two other sources of water (oil and gas brines and samples affected by road salt) to provide context. Data for brines originating from vertical (58 brine samples from Blondes and others, 2018) and horizontal (2 Marcellus Shale brines [AppMSB16–17] and 20 Utica Shale brines [AppMSB36–55] from Blondes and others, 2020) oil- and gas-drilling activities in Ohio and Pennsylvania (hereafter referred to as “oil and gas brines”) were plotted on the trilinear diagram as points of reference, as were results for water samples affected by road salt. Horizontal well brines did not have bicarbonate values, but, typically, because of the high concentrations of chloride in brines, the relative percentage makeup of chloride compared to bicarbonate for total anions is enough to mask most bicarbonate value differences where brine samples are plotted. To put this in perspective, all but one of the vertical brines had bicarbonate percentages of 0.001 to 0.3 (not including the maximum value of 4.4 percent). For the purposes of this study, it was acceptable to use the horizontal brine data (without bicarbonate values) that plot in the middle of the linear pattern of oil and gas brines (indicated as grey squares) in the upper diamond graph of figure 15. The water samples affected by road salt included a sample of road salt runoff from beneath a bridge in Western Springs, Illinois (sample RS–2 in table 2 of Panno and others [2005]) and samples of groundwater collected from monitoring wells at a salt-storage facility in Valparaiso, Indiana (wells TW–2, TW–94D, TW–94E, MW–3, MW–4, MW–5, and MW–7 in table 1 of Risch and Robinson [2000]). The samples that plot close to the oil and gas brines or road salt-affected samples have a higher proportion of sodium and chloride ions than other stream samples and might contain a component of road salt or oil-field brine.

The trilinear diagrams showed that samples from Short Creek (station 03111500), Stillwater Creek (station 03125338), and Wheeling Creek (station 03111548) were calcium-magnesium-sulfate-type waters (hereafter referred to as “sulfate-type waters”; fig. 15). The highest specific conductance (higher than 756 microseimens per cm [µS/cm]), elevated concentrations of total dissolved solids (higher than 541 milligrams per liter [mg/L]), sulfate (higher than 254 mg/L), calcium (higher than 93 mg/L), and magnesium (higher than 40 mg/L) relative to other samples were observed in the sulfate-type waters. A relation between elevated sulfate concentrations and coal-mine drainage was determined in past studies (Helsel, 1983; Haefner and Simonson, 2010). Surface and underground coal mines are prevalent in the region (fig. 4). The Ohio Department of Natural Resources, Abandoned Mine Land Program assessed drainage basins in Ohio that were impacted by abandoned mine lands (ODNR–DMRM, 2016). The program developed criteria (table 6 of ODNR–DMRM, 2016) to identify and assess the severity of mining impacts based on concentrations of eight water-quality constituents and properties. For the purposes of this study, describing severity of impact was not an objective, and therefore, if any constituent concentration for a site met one of the ODNR–DMRM criteria for mild, moderate, or severe mine drainage, then the site could be classified as “mine impacted.” Sulfate values for seven of the eight sites in the baseline water-quality analysis fell into one of the mild, moderate, or severe mine drainage impact categories of the ODNR–DMRM criteria. The only exception was Little Muskingum River (station 03115400) with sulfate concentrations (12.1 to 19.4 mg/L) less than 76 mg/L (table 6 of ODNR–DMRM, 2016), which is considered to have “no detectable mine drainage impact.”

The trilinear diagrams also showed that samples from Little Muskingum River (station 03115400) were the only calcium-bicarbonate-type waters (fig. 15). Most of the remaining samples were mixed bicarbonate-chloride-type waters and included Captina Creek (station 03113990), Yellow Creek (station 03110000), and a few samples each from various other drainage basins (fig. 15).

The oil and gas brines and road salt-affected waters are sodium-chloride-type waters. Ion compositions enriched in chloride and sodium plus potassium relative to the other surface-water samples were observed in a few samples at Captina Creek (station 03113990; depicted as light blue squares in figure 15), indicating a chemistry intermediate between that of calcium-bicarbonate waters and oil and gas brines or road salt—best seen on the cations triangle (lower left) of figure 15. The highest chloride concentration (44.2 mg/L) among all sites was measured in a sample collected at Captina Creek (station 03113990). These samples at Captina Creek (station 03113990) might contain a component of oil-field brine or road salt.

Sources of Salinity

Binary mixing curves (fig. 16) were prepared following methods described in Whittemore (1988) to help identify samples with components of oil-field brine or saturated halite (road salt) solution (should they exist). The binary mixing curves show how the chloride:bromide (Cl:Br) ratio of dilute stream water would change with the addition of increasing amounts of two concentrated solutions—oil-field brine and saturated halite (road salt) solution. The brine and halite solutions and the dilute, unaffected stream water are referred to as “endmembers” because they represent the starting and stopping points of the possible mixing process. Data from the eight sites used in the analysis of baseline water quality were plotted with the binary mixing curves to evaluate whether any samples were representative of road salt or brine endmembers (fig. 16). The Cl:Br ratios of the brine endmembers were determined using chemical data from multiple samples of oil and gas brines produced from the Berea Sandstone and Clinton Formation, which are two of the primary target formations for vertical wells drilled in the study area (Blondes and others, 2018), as well as from the Utica Shale (Blondes and others, 2020). Two road salt endpoints were used to show a range of possible concentrations. The uppermost Cl:Br value for the road salt endpoint (depicted as a light blue square in [fig. 16]) is road salt runoff collected beneath a bridge in Western Springs, Ill. (Panno and others, 2005). The lesser road salt endpoint (depicted as a yellow square in [fig. 16]) was computed as the mean of three samples of road salt applied in northeastern Ohio (collected by Knuth and others, 1990). The light blue square shown between these two values represents a concentrated saline solution made by Panno and others (2005) from road salt collected at an Illinois Department of Transportation salt-storage facility. The area designated as dilute water in (fig. 16) is based on the maximum chloride concentration (12 mg/L) and the maximum Cl:Br ratio (521) reported by Panno and others (2005) for samples of Midwestern precipitation, soil water, and groundwater deemed to be unaffected by sodium chloride contamination.

Water samples collected as part of this baseline study in which uncensored bromide concentrations were observed are plotted on figure 16. Twenty-four samples (representing all sites except Little Muskingum River [station 03115400]) plotted outside of the dilute water region and between the bounds of the lower road salt (depicted as dotted black line) and upper Berea Sandstone brine mixing curves (depicted by dashed black line; figure 16). Panno and others (2005) determined that septic effluent, wastewater treatment plant effluent, animal wastes, and water from field tiles (agricultural subsurface drainage systems) plots between the brine and road salt mixing curves.

Eleven samples from four sites showed a component of brine in the binary mixing curves. These samples plotted outside of the dilute water region and within the bounds of the upper and lower mixing curves, between dilute water and brines from the Clinton Formation (depicted by solid black lines) and Berea Sandstone (depicted by dashed black lines). No samples plotted inside of the Utica Shale mixing curves (depicted by solid green lines). Captina Creek (station 03113990) had six samples (depicted as red diamonds) with a component of brine. Three samples (depicted as light blue dashes) from Little Muskingum River (station 03115400) plotted within the bounds of the Berea Sandstone and Clinton Formation brine mixing curves (fig. 16). Single samples from two other sites plotted within the bounds of the Berea Sandstone and Clinton Formation brine mixing curves—Leatherwood Creek (station 03141870; denoted by plus sign) and Duck Creek (station 03115786; denoted by triangle; fig. 16). Based on binary mixing calculations, the maximum percentage of brine input denoted by these samples is less than one percent—0.08 percent—Berea Sandstone brine, and 99.92 percent dilute water.

The Wilcoxon rank-sum test was used to test differences between median constituent concentrations for samples with a component of brine (11 samples) and samples containing no brine (42 samples; figs. 17A, B, C, and D). Results indicated statistically significant differences between median concentrations of bromide, chloride, arsenic, and barium for sites with and without a component of brine in their drainage areas (table 7). Boxplots for these constituents showed higher median concentrations at sites with a component of brine in their drainage areas compared to sites without a component of brine (figs. 17A, B, C, and D). It should be noted that the increased concentrations of constituents in samples containing a component of brine does not necessarily mean that the elevated concentrations were caused by the brine.

OEPA Aquatic Life Use Criteria

Measured pH and constituent concentrations at the eight sampling sites were compared to OEPA aquatic life use criteria based on beneficial use designations to assess the potential for adversely affecting stream biota (table 8). Stillwater Creek (station 03125338), Leatherwood Creek (station 03141870), Wheeling Creek (station 03111548), Short Creek (station 03111500), and Duck Creek (station 03115786) are designated as warmwater habitat (WWH), while Captina Creek (station 03113990) and Yellow Creek (station 03110000), as well as Little Muskingum River (station 03115400), are designated as exceptional warmwater habitat (EWH; Ohio Administrative code 3745–1–24 [LAWriter Ohio Laws and Rules, 2023a]). If a constituent or property is not included in table 8, the OEPA has not established an aquatic life use criterion for that constituent or property. The results of the baseline assessment of surface-water quality in the study area showed no exceedances of aquatic life use criteria as contained in Ohio Administrative Code 3745–1–35 (LAWriter Ohio Laws and Rules, 2023b).

Relation Between Water Chemistry and Drainage-Basin Characteristics

Spearman’s rank correlation coefficients were used to identify correlations between water chemistry and selected drainage-basin characteristics that could affect water quality (tables 9 and 10). Spearman’s rank correlation coefficients were calculated using median water-chemistry values for each site and drainage-basin characteristics that included (1) the density of producing vertical oil and gas wells, (2) the density of producing horizontal gas wells that were tapped into the Utica Shale, Point Pleasant Limestone, and Marcellus Shale, (3) the density of active and plugged injection wells, (4) the density of wastewater treatment plants, (5) the percentage of lands associated with underground coal mines, (6) the percentage of lands associated with surface coal mines, (7) the density of road miles, (8) the percentage of township lands with permits to spread brine for dust control, and (9) the percentage of land cover for agriculture, developed, and forest. In the “Summary and Discussion” section, only significant positive correlations are mentioned. A positive correlation means that as the drainage basin characteristic increases (either in density or percentage), so does the median concentration or level of the constituent or property. Caution must be used when interpreting the correlations, because correlations were among the drainage-basin characteristics themselves (table 11).

Table 10 shows correlations between water chemistry and selected drainage-basin characteristics for the eight sites used in the analysis of baseline water quality. There were no statistically significant correlations between the density of vertical or horizontal wells and median characteristic values. Median parameter values were predominantly positively correlated with the density of wastewater treatment plants and the percentage of land associated with surface mines and predominantly negatively correlated with township lands with permits to spread brines.

Water Use and Changes in Streamflow

Between calendar years 2010 and 2019, hydraulic fracturing withdrawals were reported in 643 locations, with streams and reservoirs being the most common sources for withdrawals (Bradley Lodge, ODNR–Division of Water, written commun., 2022). Annual withdrawals for hydraulic fracturing in Ohio totaled about 27,168 million gallons between 2010 and 2019. Annual withdrawals peaked at about 5,755 million gallons in 2017 and then decreased to about 4,371 million gallons in 2019. Average withdrawals by calendar month ranged from about 226 to 349 million gallons, with April and May typically being the highest withdrawal months—January and December being the lowest.

Water withdrawn within a drainage basin for hydraulic fracturing may or may not be detectable in a streamgage’s record depending on factors such as the proximity of the gage to the withdrawal location(s), the magnitude of the withdrawal(s) relative to the magnitude of the streamflow, and whether the withdrawal(s) were made from streams instead of ponds or reservoirs. Withdrawals made from ponds or reservoirs may have no detectable impact on downstream flows depending on the outlet characteristics of the pond or reservoir and (or) if outflow from the pond or reservoir is regulated. Other factors, such as changes in other sources of water use, discharges within the drainage basins, or changing climate also can confound the ability to detect streamflow changes caused by hydraulic fracturing withdrawals.

The amount of water used for hydraulic fracturing in Ohio was substantial but represented a small fraction of the water discharged in streams. The environmental impact of those withdrawals was not evaluated in this study. The reader is referred to Dieter and others (2018) for a brief discussion of some potential environmental impacts associated with hydraulic fracturing, including issues surrounding water availability, water quality, and the disposal of wastewater and associated potential for seismicity.

Daily streamflow data from nine USGS gages in eastern and east-central Ohio were used to assess temporal streamflow trends and changes. The records for two gages on Captina Creek (stations 03113990 and 03114000) were merged into a single record intended to be representative of flows at the Captina Creek (station 03113990) gage in current (2023) operation. Two of the gages that had no hydraulic fracturing withdrawals within their drainages were used as “reference gages.” The remaining streamgages (referred to as “study gages”) had hydraulic fracturing withdrawals within their drainages.

Sen’s slopes of the period of record annual minimum and median daily streamflows were positive for all reference and study gages, indicating a consistent tendency for low and median flows to have increased over the periods of record. In addition, the trend slopes were positive for all flows with nonexceedance probabilities of 0.90 and less for all reference and study gages. These results indicate a consistent pattern for all but high flows to have increased over the periods of record at both the study and reference gages.

The Indicators of Hydrologic Alteration (IHA) software (Nature Conservancy, 2009) was used to compute annual values of 67 potentially ecologically relevant statistics from daily streamflow time series for the study and reference gages. Analyses of annual IHA statistics showed no general indication that low flows or extreme low flows have gotten lower, become more frequent, or increased in duration since 2010, when records for hydraulic fracturing water withdrawals began. In fact, in almost all cases, the opposite was indicated.

Analyses of nonexceedance percentiles of daily streamflows from reference and study gages indicated flows in the lower quartile of nonexceedance percentiles were typically larger for the period of record than for the pre-2012 record (before annual withdrawals for hydraulic fracturing exceeded 100 million gallons in Ohio). This indicates that low flows did not decrease during the post-2011 period when hydraulic fracturing water withdrawals were occurring. It is possible that the higher discharges for the low flows at both study and reference gages reflect the results of shorter-term climate fluctuations or possibly more permanent changes in climate. In either case, reductions in streamflow caused by hydraulic fracturing withdrawals (if measurable at the gages) appear to have been more than offset by wetter climatic conditions.

Baseline Water Quality

Lands associated with coal mines are prevalent in the study area region. Sulfate values for seven of the eight sites in the baseline water-quality analysis indicated some degree of mine drainage impact based on ODNR–DMRM criteria. The only exception was Little Muskingum River (station 03115400). Because of the potential impacts from mine drainage at these seven streams, the use of constituents such as specific conductance, total dissolved solids, sulfate, iron, manganese, arsenic, total carbon, and dissolved organic carbon as indicators of oil and gas influence was complicated.

Wheeling Creek (station 03111548), Short Creek (station 03111500), and Stillwater Creek (station 03125338) had the highest maximum specific conductance values and the highest median concentrations of total dissolved solids and sulfate. Stillwater Creek (station 03125338) also had the highest maximum (223 μg/L) and median (183 μg/L) concentration of manganese. These three sites were classified as sulfate-type waters based on major-ion chemistry and have some of the highest percentages of lands associated with surface and underground coal mines in their drainage basins—a potential source for these elevated values.

Trilinear diagrams also revealed that a few samples at Captina Creek (station 03113990) had a chemistry intermediate between that of calcium-bicarbonate-type waters and oil and gas brines or road salt and might contain a component of oil-field brine or road salt. Sources of chloride in water samples were also evaluated using binary mixing curves and ratios of chloride-to-bromide. These ratios indicated that 11 samples from 4 sites (Captina Creek [station 03113990], Little Muskingum River [station 03115400], Leatherwood Creek [station 03141870], and Duck Creek [station 03115786]) likely contained a component of brine from the Clinton Formation or Berea Sandstone. Captina Creek (station 03113990) had the most samples (six) with a component of brine. The other three sites with samples having a component of brine, potentially indicative of oil and gas contamination, all had brine injection wells in their upstream drainages as well as having the top three highest densities of vertical oil and gas wells. Concentrations of arsenic, barium, bromide, and chloride were statistically higher in samples with a component of brine than in samples without a component of brine.

The results of the baseline assessment of surface-water quality in the study area showed no exceedances of OEPA aquatic life use criteria. Spearman’s rank correlation coefficients indicated no significant positive correlations with the density of vertical or horizontal oil and gas wells.