Science of the Total Environment 545–546 (2016) 114–126

Contents lists available at ScienceDirect

Science of the Total Environment

j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenv

Elucidating hydraulic fracturing impacts on groundwater quality using aregional geospatial statistical modeling approach

Taylour G. Burton a, Hanadi S. Rifai b,⁎, Zacariah L. Hildenbrand c,d, Doug D. Carlton Jr d,e,Brian E. Fontenot d, Kevin A. Schug d,e

a Civil and Environmental Engineering, University of Houston, W455 Engineering Bldg. 2, Houston, TX 77204-4003, United Statesb Civil and Environmental Engineering, University of Houston, N138 Engineering Bldg. 1, Houston, TX 77204-4003, United Statesc Inform Environmental, LLC, Dallas, TX 75206, United Statesd Collaborative Laboratories for Environmental Analysis and Remediation, University of Texas at Arlington, Arlington, TX 76019, United Statese Department of Chemistry & Biochemistry, The University of Texas at Arlington, Arlington, TX, United States

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• Migration pathways from fracturedwells to groundwater are poorly under-stood

• Geospatial modeling correlated ground-water chemicals to Barnett fracturedwells

• Increased Beryllium strongly associatedwith hydraulically fractured gas wells

• Indirect evidence of pollutant migrationvia microannular fissures in well casing

• Large-scale and spatial approach neededto detect groundwater quality changes

⁎ Corresponding author.E-mail addresses: tgburton@uh.edu (T.G. Burton), rifai

uta.edu (B.E. Fontenot), kschug@uta.edu (K.A. Schug).

http://dx.doi.org/10.1016/j.scitotenv.2015.12.0840048-9697/© 2015 Elsevier B.V. All rights reserved.

A relative increase in beryllium concentrations in groundwater for the Barnett Shale region from 2001 to 2011

was visually correlated with the locations of gas wells in the region that have been hydraulically fractured overthe same time period.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 25 October 2015Received in revised form 17 December 2015Accepted 18 December 2015Available online xxxx

Editor: D. Barcelo

Hydraulic fracturing operations have been viewed as the cause of certain environmental issues including ground-water contamination. The potential for hydraulic fracturing to induce contaminant pathways in groundwater isnot well understood since gas wells are completed while isolating the water table and the gas-bearing reservoirslay thousands of feet below the water table. Recent studies have attributed ground water contamination to poorwell construction and leaks in the wellbore annulus due to ruptured wellbore casings. In this paper, a geospatialmodel of the Barnett Shale region was created using ArcGIS. The model was used for spatial analysis of ground-water quality data in order to determine if regional variations in groundwater quality, as indicated by variousgroundwater constituent concentrations, may be associated with the presence of hydraulically fractured gaswells in the region. The Barnett Shale reservoir pressure, completions data, and fracture treatment data wereevaluated as predictors of groundwater quality change. Results indicated that elevated concentrations of certain

Keywords:Barnett ShaleNatural gas

@uh.edu (H.S. Rifai), zac@informenv.com (Z.L. Hildenbrand), doug.carlton@mavs.uta.edu (D.D. Carlton), brian.fonteno@mavs.

115T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

groundwater constituents are likely related to natural gas production in the study area and that beryllium, in thisformation, could be used as an indicator variable for evaluating fracturing impacts on regional groundwater qual-ity. Results also indicated that gas well density and formation pressures correlate to change in regional waterquality whereas proximity to gas wells, by itself, does not. The results also provided indirect evidence supportingthe possibility that micro annular fissures serve as a pathway transporting fluids and chemicals from the frac-tured wellbore to the overlying groundwater aquifers.

© 2015 Elsevier B.V. All rights reserved.

Bottom-hole pressureGeographic Information Systems (GIS)Cluster analysisMicro annular defects

1. Introduction

Hydraulic fracturing is a technology used in oil and gas production toincrease hydrocarbon recovery from low-permeability formations. Dur-ing a hydraulic fracturing operation, fluid is injected into an oil and gaswell at high pressures—a process that fractures the rock of thehydrocar-bon bearing formation thereby increasing its hydraulic conductivity andthe rate of flow of oil and gas from the formation to the wellbore. Hy-draulic fracturing techniques, developed as early as 1949, have signifi-cantly improved since the 1980s such that they currently allowproduction from low-permeability shale formations that have histori-cally been considered a non-producible resource (Murray, 2013). Thefirst hydraulic fracturing treatment in a horizontal wellbore was per-formed in 1992 in the Barnett Shale. However, the combined advancesin horizontal drilling and hydraulic fracturing (in addition to othernovel technologies such as use of high-volume of fracturing fluids,clustered-multi-well pads, and long laterals) have propelled fracturingof horizontal wells to become an industry standard practice in the de-velopment of low-permeability shale formations (Smith and Hannah,1996).

Hydraulic fracturing use has increased significantly since the mid-2000s and has become a subject of controversy concerning potentialrisks to human health and the environment (Finkel and Hays, 2013;Rahm, 2011; Walton and Woocay, 2013; McKenzie et al., 2012;Preston et al., 2014; Ziemkiewicz et al., 2014; Eaton, 2013; Meng,2015; Révész et al., 2012; and Werner et al., 2015). Research is neededto address health and safety issues in the development of oil and gas re-sources including the cumulative impacts of tightly spaced wells thatare more difficult to quantify (Vidic et al., 2013).

A heightened interest in the impact of hydraulic fracturing ongroundwater exists since this subject is not as well understood despitethe fact that several studies have been undertaken. Well casing failures,contaminant migration through fractures, surface spills, and/or waste-water disposal are all potential pathways that could lead to groundwa-ter contamination. A risk-model by Rozell and Reaven (2012) proposedthat disposal of wastewater had the highest risk for contaminatingground water while other studies demonstrated that contaminationmay be from the subsurface. Methane concentrations in groundwater(primarily in the Marcellus Shale), for example, were evaluated insome studies as an indicator of potential communication betweenwater aquifers and gas wells; the distance to gas well operations wasshown to be a statistically significant variable for methane concentra-tions in ground water samples by Osborn et al. (2011) and Jacksonet al. (2013).

The study of methane concentrations alone, however, may not be astraightforward indication that groundwater contamination has oc-curred, particularly since other research studies have demonstratedthat methane concentrations and chemical properties were correlatedto the geophysical environment and topography (Molofsky et al.,2011; Warner et al., 2012; Molofsky et al., 2013), and to the distanceto natural faults (Moritz et al., 2015). A study by Fontenot et al. (2013)evaluated heavy metal concentrations in groundwater as indicators ofgroundwater contamination and presented statistically significanthigher median concentrations of heavy metals in water qualitysamples taken in proximity to natural gas extraction activities in theBarnett Shale region in Texas (the region studied in this work). The

aforementioned studies suggested that some impact to groundwaterfrom hydraulic fracturing operations could be observed; however, it isunclear whether the migration of methane gas coincided with the mi-gration of other groundwater contaminants.

The toxic elements found in hydraulic fracturing wastewaterstreams should be considered in the study of hydraulic fracturing im-pacts on groundwater. These elements have the potential to contactthe groundwater system andmay serve as indicator variables for exam-ining changes in groundwater quality related to hydraulic fracturing op-erations. Wastewater from hydraulic fracturing operations containstoxic elements originating in the shale and from chemicals used in thefracturing treatment including total dissolved solids, volatile sub-stances, bromide, naturally occurring radioactive materials, and heavymetals such as arsenic, barium, beryllium, uranium, and zinc (Gordallaet al., 2013; Ternes, 2012; Rahm et al., 2013; Lester et al., 2015; andChermak and Schreiber, 2014). Harkness et al. (2015) attributed thehigh bromide and chloride content in the wastewater to the brinefrom the shale reservoir; Rowan et al. (2011) demonstrated that highsalinity mobilizes radionuclides, increasing exposure to radioactivewaste such as radium 226.

Even less well understood than the impacts of fracturing on groundwater quality are the potential pathways for pollutant migration togroundwater from the shale formations. The study presented in thispaper addresses this knowledge gap and investigates gas migration asa transportation mechanism of contaminants into groundwater. Recentstudies have concluded that groundwater contamination is due to poorwell construction (Jackson et al., 2013) and that leaks in the wellboreannulus are due to ruptured wellbore casings (Darrah et al., 2014). Astudy presented by Ingraffea et al. (2014) developed a risk assessmentmodel for casing and cement impairment for oil and gas wells in Penn-sylvania concluding that unconventional wellbores are at a greater riskfor impairment than conventional wellbores, and periods of intensedrilling have resulted in loweredwellbore integrity. Another study indi-cated that wellbore integrity failure rates vary significantly based ongeographical region and noted that more wellbore monitoring wouldbe required to better understand failure rates (Davies et al., 2014).

In this paper, the research presented differs from the aforemen-tioned studies where groundwater contamination was attributed to anoticeable failure in the wellbore systems. The research presented inthis work investigates how minor defects in the wellbore system,which are farmore common than amajor defect, may still be significantto cause widespread impacts of fracturing operations on groundwater.Gas can permeate through small cracks in the annular cement sheath(see Section 4). Theworking hypothesis is that the expansion of naturalgas, released from the producing formation during the hydraulic frac-turing process, is the mobilizing mechanism that allows chemicals in agas–fluid mixture to make contact with the water table above theshale formation (in the case ofwellboreswith a defect in the annular ce-ment sheath). Because the formation pressure in a well will drive thegas velocity and volume of gas generated, the reservoir pressure is ex-amined as a predictor variable for elevated concentrations of indicatorconstituents in groundwater. It is presumed that a larger volume ofgas flow will contribute to a greater accumulation of contaminants inthe aquifer system.

Additionally, the expansion of gas hypothesis presented above dic-tates that groundwater quality changes, when present, will only be

116 T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

evidenced via a regional analysis that takes into account the relativelysignificant number of hydraulically fractured wells in a given produc-ing region, and the natural variability in ground water constituents(complicated by the paucity of groundwater quality data). The re-gional analysis has to also take into account the differing geologicstrata, the separation in depth between the groundwater formations,and the varying formation pressure over the region. This approachis novel and has not been applied to a shale producing formation/groundwater aquifer system as of yet to the best of the knowledgeof the authors.

The present study focuses on the Barnett Shale region in Texas from2001 (pre-fracturing) to 2011 (post- peak fracturing period)mainly dueto the availability of historical ground water quality data and more re-cent, albeit relatively limited, detailed ground water quality samplingstudies (Fontenot et al., 2013; Thacker et al., 2015 and Hildenbrandet al., 2015). The study relies on geospatial modeling using GeographicInformation Systems (GIS) tools to correlate the changes in groundwa-ter quality to fracturedwell characteristics such as bottomhole pressureand distance from the fractured well. The study illustrates regionaltrends and correlations and demonstrates the possibility of using con-stituents uniquely associated with fracturing activities as indicator var-iables of regional ground water quality changes. The results from thestudy provide indirect evidence supporting the hypothesis thatwellbore construction is a key potential pathway for contaminant trans-port to groundwater.

2. Study area — the Barnett Shale

The Barnett Shale, located in the Fort Worth Basin of northeastTexas, is predominately a gas-bearing hydrocarbon formation with anabove-normal pressure gradient of 0.52 psi/ft (Bowker, 2007) that re-quires fracturing to extract the gas. The Barnett Shale was chosen forfour reasons: (1) there were relatively extensive water quality datasets for the region available from the Texas Water Development Board(TWDB) (TWDB, 2014), and the University of Texas Arlington (UTA)(Fontenot et al., 2013; Hildenbrand et al., 2015); (2) the production his-tory in the Barnett presented an opportunity for comparingwater qual-ity samples taken in the year 2000 and prior, to samples taken between2011 and 2014; (3) the Barnett is predominately a gas bearing forma-tion; and (4) wells in the Barnett Shale require hydraulic fracturing forproduction; in this sense gas migration could be attributed to the hy-draulic fracturing process where gas flow would not occur without thewell having been fractured. The study presented here evaluated theBarnett Shale wells, both vertical and horizontal since nearly all havebeen fractured.

Horizontal drilling has been very active in the Barnett Shale since2002 with lateral lengths varying between 500 and 3500 ft(Montgomery et al., 2006). Gas well data for the region were obtainedfrom the Texas Railroad Commission (RRC, 2014), fracfocus.org, anddrillinginfo.com.Water and gas well data were collected for the Bosque,Clay, Collin, Cooke, Dallas, Denton, Ellis, Erath, Grayson, Hamilton, Hill,Hood, Hunt, Jack, Johnson, Kaufman, Montague, Palo Pinto, Parker,Rockwall, Somervell, Tarrant, and Wise counties in north Texas (seeFig. S1 in Supporting information). There were over 30,000 gas wellsin the study area with more than 18,000 of them concentrated in Den-ton, Tarrant, Parker, and Wise counties as can be seen in Fig. S1. Fig. S1also shows that the deepest part of the Barnett is to the northeast inDenton County.

3. The Trinity aquifer

The major water aquifer for the region is the Trinity, a group of foursandstone layers with varying lateral extents (Harden, 2004). The Trin-ity wasmodeled as a single continuous sandstone layer that mainly em-bodies the Paluxy sandstone formation, the uppermost layer in theregion. Vertical water well depth references from the TWDB were

used to create a contour of this sandstone layer in ArcGIS (see Fig. 1,black dots indicate water well sample locations). As shown in Fig. 1,the depth of the aquifer ranges from approximately 30 ft (~10 m) tomore than 4000 ft (~1219m) below the surface, thus placing the Paluxysand between 2500 and 7500 ft (~762 to 2286 m) above the BarnettShale, with a greater thickness between the formations in the northeast.The hydraulic conductivity for the Paluxy is 5.8 ft/day (1.77 m/day)(Harden, 2004) with an average gradient of 0.009 ft/ft (m/m) over thecontoured layer shown in Fig. 1. The travel distance over the 10-year pe-riod of the study was estimated to be on the order of 1000 ft (~305 m)using Darcy's Law and the aforementioned estimates of gradient andhydraulic conductivity for the Paluxy. This distance was taken into con-sideration throughout the study as will be seen subsequently in thepaper.

4. Contamination via micro-annular defects in a wellbore

In order to demonstrate the potential for the wellbore to serve as acontaminant migration pathway, an analysis was undertaken to quan-tify the flow velocity in a micro-annular crack or fissure, based on awidth of defect between 10 and 100 μm. The velocity was calculatedusing an equation previously presented in a study of CO2 storage wells(Deremble et al., 2010):

vf ¼ − w2

12c f=

dPdS

þ ρg cos αð Þ� �

ð1Þ

where vf is the mean fluid velocity,w is the width of the defect, dPdS is thechange in reservoir pressure over a vertical well distance, µf is the fluiddensity, g is the gravitational constant, ρ is the density of the groundwa-ter, and α is the angle of the wellbore.

Using a dPdS value of −0.54 psi/ft, (based on the reservoir pressure of

the study area), and a µf value of 0.0113 cP for a gas mixture of 85%methane and 15% CO2, themeanfluid velocitywas determined to be be-tween 81.4 and 8137 ft/day (~24.8 and 2480 m/day) for the 10 and100 μm defect widths, respectively. Clearly, this flow velocity is signifi-cant, indicating that even in a cemented wellbore, natural gas couldmake contact with the water table on a scale of 1–10 days in an8000 ft (~2438 m) deep vertical well.

In a hydraulically fracturedwell, a fluidmixture of water and naturalgas would flow through such micro-annular pathways. The fluid mix-ture has the potential to entrain contaminants existing naturally in themethane gas and formation brine, as well as chemicals from the fracturetreatment. While the extent to which these contaminants are soluble inthe fluid mixture is not taken into consideration in the study, it is rea-sonable to assume that some change in groundwater quality over theregion may occur given the significantly large number of gas wellsthat would be present in the region. It also follows that such a changein groundwater constituent concentrations may be possible to observeusing reservoir pressure gradients and the locations of gas wells in thestudy region as predictor variables. It should be noted that the flow ve-locity calculated above would be even greater in a scenario of greaterchange in reservoir pressure or when more wells are drilled in a givenregion.

5. Research approach and methodology

The research approach used qualitative and quantitativemethods to study ground water quality changes over the relativelylarge Barnett Shale region. The research methodology relied on sta-tistical testing and GIS geospatial and correlation analyses as will beseen in the remainder of the section. Two types of analyses wereconducted: in the first set of analyses, visual and statistical analyseswere undertaken to determine if there were correlations betweenthe ground water quality data, distance from hydraulic fracturing

Fig. 1. Location of water wells in the Paluxy overlain with Contoured Aquifer Depth.

117T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

wells, fracturing well density, and reservoir pressure gradients. Inthe second set of analyses, the gas wells were clustered based on10 specific properties related to gas well construction. The clusterswere then correlated to constituent concentrations in order to de-termine if there were correlations between specific well construc-tion variables and observed constituent concentrations posthydraulic fracturing.

5.1. Reservoir pressure gradient in the Barnett

Amodel of the reservoir pressure gradient (RPG) in the Barnett wasdeveloped usingflowing bottomhole pressure (BHP) data for over 2046gas wells from G-10 completion records stored at the RRC in Austin, TX(only pressure values taken prior to production were used in this anal-ysis since the purpose of the analysis was to evaluate the magnitude ofthe RPG prior to production). The BHP values estimate the excess pres-sure in the rock (in excess of normal hydrostatic) that is due to the con-version of oil to gas over time (gas is compressed when trapped in thereservoir rock; as gas is generated, it will expand, Barker, 1990). Corre-sponding True Vertical Depths (TVD) of well locations with a recorded

BHP were used to calculate the reservoir pressure gradient at eachwell location using the equation:

Reservoir Pressure Gradientpsift

� �

¼Flowing BHP psið Þ þ 0:433

psift

� Vertical Shale Depth ftð Þ� �

Vertical Shale Depth ftð Þ : ð2Þ

The RPG values at their corresponding locations were contoured inArcGIS resulting in the plot shown in Fig. 2. As can be seen in Fig. 2, res-ervoir pressure gradients ranged from 0.45 to 0.90 psi/ft (trending up-wards in a northeasterly direction towards Denton County). Theaverage gradient in the dataset was 0.53 psi/ft, which correspondedwell with literature values (see for example, Bowker, 2007).

5.2. Water quality data

The water quality data used in the study, combining samples fromthe TWDB and UTA and containing data for 31 groundwater constitu-ents, are listed in Table 1. The samples used in the study had a depth

118 T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

(of well sample) reference that corresponded to the aquifer depth fromthe contour plot shown in Fig. 1 at that water sample location.

Scatter plots of concentration data for all sampled ground waterwells as a function of time and spatial plots of the concentration datawere prepared and visually inspected for each constituent. Based on avisual inspection of the resulting graphs/plots (cannot all be showndue to space limitations), it became evident that: (i) the data were rel-atively very sparse for many of the constituents (e.g., dissolved radium226 concentrations shown in Fig. S2); (ii) the data for all constituentsexhibited significant variability over time and spatially (e.g., dissolvediron concentrations shown in Fig. S3 (top shows over time and bottomshows spatial distribution); and (iii) patterns or trends were not dis-cernible for many constituents from the data (e.g., iron data shown inFig. S3 show no discernible trends whereas those in Fig. S4 for bariumand beryllium show amarked increase in the range of observed concen-tration ranges in recent years). These findings confirmed that it wouldbe difficult to determine the associations, if any, between change ingroundwater quality over time and hydraulic fracturing activities with-out incorporating a spatial analysis of thewater samples and their prox-imity to hydraulic fracturing operations. Changes in the context of

Fig. 2. Contour plot of Barnett Shale p

specific variables from hydraulic fracturing activities, became the guid-ing principle for the methodology and approaches used in the study asdescribed below.

5.3. Visual analysis of regional groundwater quality constituent change

Qualitative visual analyses were undertaken to evaluate trends orchanges, if any, in the ground water quality data over time and space.Contour plots were created in ArcGIS for 20 of the constituents thathad both historical (pre-2001) and current (2011–2014) water qualitydata. For each constituent, two contour plots were created, one usingwater samples taken before the year 2001, and one using samplestaken during 2011–2014 (non-detect values were replaced with azero for contouring purposes). In ArcGIS, the Radial Basis Function(RBF), generally used in groundwater modeling (Kresic, 2006, p. 78)when water quality data sets are small, was used. Using this contourfunction, someof the plots created a negative value contour. The InverseDistanceWeighting (IDW) method was used in such cases to avoid thenegative values generated by the RBF.

ressure gradient (RPG) in psi/ft.

Table 1Constituents evaluated in the Barnett region.

Groundwater constituent Source Sample years Number of samples

Dissolved alpha (pc/L) TWDB 1976–2011 380Dissolved aluminum (ppb) TWDB 1939–2011 786Total arsenic (ppb) TWDB/UTA 1949–2014 570Total barium (ppb) TWDB/UTA 1985–2014 542Total benzene (mg/L) UTA 2012–2014 379Total beryllium (ppb) TWDB/UTA 1994–2014 466Total boron (ppb) TWDB 1948–2011 756Total bromide (mg/L) UTA 2012–2014 379Dissolved bromide (mg/L) TWDB 1988–2011 738Total chloride (mg/L) UTA 2012–2014 379Total copper (ppb) TWDB/UTA 1980–2014 503Dissolved oxygen (mg/L) TWDB/UTA 1983–2014 464Total ethanol (mg/L) UTA 2011–2014 454Total ethyl benzene (mg/L) UTA 2012–2014 379Total iron (ppb) TWDB/UTA 1923–2014 1899Total methanol (mg/L) UTA 2011–2014 454Total molybdenum (ppb) UTA 2012–2014 379Dissolved molybdenum (ppb) TWDB 1989–2011 756Total nickel (ppb) TWDB/UTA 1994–2014 461Total nitrate (mg/L) TWDB/UTA 1975–2014 540pH UTA 2011–2014 452Dissolved phosphorus (mg/L) TWDB 1952–2011 316Dissolved radium 226 (pc/L) TWDB 1977–2011 150Dissolved radium 228 (pc/L) TWDB 1988–2011 150Redox potential (mV) TWDB/UTA 1990–2014 721Total selenium (ppb) TWDB/UTA 1977–2014 559Total sulfate (mg/L) UTA 2012–2014 379Water temperature ( C) TWDB/UTA 1963–2014 1167Total dissolved solids (mg/L) UTA 2011–2014 452Dissolved vanadium (ppb) TWDB 1989–2011 734Total zinc (ppb) TWDB/UTA 1980–2014 514

Notes:

1. The constituents were based upon data availability and relatedness to hydraulicfracturing.

2. A total of 20 of the 31 constituents had historical (pre-2001) and current (2011–2014)sample data.

119T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

A spatial extraction and visualization methodology was developedto evaluate the spatial changes in groundwater quality between thetwo contour plots and correlate the observed changes to fracturing ac-tivities. A grid of 19,594 data points, covering an area of 15,048 mi2

was draped over the contour plots described above. The contour valuesat the grid data points were extracted from the pre-2001 and post-2011plots using the Extract Multi-Values to Points function in ArcGIS. In theinstances where both a historical and current value was extracted, thedifference was taken between the two values and then reimportedinto the ArcGISmodel. The IDWcontour functionwas applied to the dif-ference values to create a continuous contour plot of constituent con-centration change. While this method involves a certain amount ofsmoothing and spatial interpolation, it is a reasonably valid approachwhen comparing spatially variable datasets at two points in time, as isthe case here. It should be noted that the spatial extent of the contourplots that were generated does not cover the entire study area of inter-est. Additionally, the contour plots for each constituent and year werenot the same spatial extent (since this depended on sample availability).The resulting plots of groundwater change varied in extent and shapeand while they were used to qualitatively model large scale trends ingroundwater quality changes over the region, it must be kept in mindthat inaccuracies may exist within the contour due to data availabilitylimitations.

The locations of the gas wells in the region were plotted on the con-stituent contour plots to determine if there was a visual correlation thatcan be observed between the change in concentration plot and the loca-tion of the gaswells. In a second visual analysis, the RPG contour (Fig. 2)was plotted over the constituent concentration change to evaluate thehypothesis that a higher RPG would correlate to a more observablechange in groundwater constituent concentrations.

5.4. Proximity to gas wells nonparametric statistical Mann–Whitney U-Test

The Mann–Whitney U-Test, a nonparametric comparison test, wasused to determine if there was a significant statistical difference ingroundwater quality measurements between water samples takennear gas wells, and those that were not. The Mann–Whitney U-Test istypically used to compare the distribution of two data sets that are ran-domly sampled, independent, and that could be ranked, where samplesets are not necessarily equal in size/and or not normally distributed(Paulson, 2003). In this study, the P-value of significance was for atwo-tailed test with a 0.95 confidence interval. Results yielding a P-value less than 0.05 were considered statistically significant.

The water samples were divided into two groups: (1) those with nogas wells existing within 1 mile (control group), and (2) water sampleshaving at least one gas well within a 1000 ft distance (test group, recallthat the average travel distance within the Paluxy was estimated to be1000 ft in the 10 years of fracturing activities) (see Fig. 3). The Matlabsoftware was used to calculate distances between water and gas wells,where the locations of water wells were calculated relative to the loca-tions of gas wells using the haversine function (the haversine functiongives the shortest distance over the earth's surface between two pointson a sphere based on their longitude and latitude).

Many of the samples in the groundwater datawere non-detect (ND)values. The nonparametric Mann–Whitney test was thus performed foreach constituent three times: (1) in the first test, the ND valueswere setto zero, (2) in the second test, the ND values were set to 1/2 the detec-tion limit (as defined by themeasuring instrument), and (3) in the thirdtest, the ND valueswere excluded. Non-detect valueswere not differen-tiated within the TWDB samples, yielding similar results for dissolvedalpha (also known as gross alpha particle activity — a measure of thetotal amount of radioactivity in awater sample attributable to the radio-active decay of alpha-emitting elements), dissolved aluminum, dis-solved boron, dissolved molybdenum, dissolved radium 226, dissolvedradium 228, and dissolved vanadium for the three methods dealingwith the ND values described above.

5.5. Gas well density as a predictor variable of regional GW constituentchange

The purpose of this analysis was to analyze the impact of the densityof gas wells in a given area on groundwater quality. Contour plots werecreated for the groundwater samples taken between 2011 and 2014 for31 constituents. The data extracted from the contour plots were used inthe analysis as were the data extracted from plots of reservoir pressuregradients (RPG) shown in Fig. 2. The first step in the analysis was to cre-ate a rasterfile denoting the relative gaswell density in the study region.This was done by converting the shapefile of gas well locations to a ras-ter dataset using the Point to Raster tool in the Arc Toolbox (see Fig. 4).The raster displays a grid of rectangular cells, each cell being 0.01squared degrees representing an area of 0.04 mile2 (note that the cellboundaries 0.2 miles are greater in length than the maximum 1000 ftdistance of groundwater transport calculated in Section 3 above). Thecount of gas wells within each pixel was color coded in the raster(shown in Fig. 4), thus red indicates a higher count of gas wells withinthe pixel than the yellow does, for example.

The second step was to generate a grid of evenly spaced data pointsfor further analysis. The grid of data points was created such that thedata point locations were in the centroid of each raster cell created instep 1 above. The values of the local reservoir pressure gradient, wellcount, and the water constituent concentrations were extracted ateach data point using the Extract Multi-Value Points function. Thedata points were categorized by well density using two categories: ahigh well density category and a zero density category. Cells consideredto have a high density had a well count of 18–54 per cell, and zero den-sity had no wells. The high well density category was further separatedby pressure gradient into categories of 0.4–0.49 psi/ft, 0.5–0.59 psi/ft,

Fig. 3. Locations of control group and test group groundwater samples (the black dots indicate the location of gas wells).

120 T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

0.6–0.69 psi/ft, 0.7–0.79 psi/ft, and 0.80–0.89 psi/ft. The aforementionedcategorization procedures resulted in a total of six subgroups of data asshown in Table 2. As can be seen in Table 2, subgroup zero with 4161cells had no wells in any of those cells; whereas subgroups 1 through6 had 18–54 wells per cell with an increasing RPG as the subgroupnumber increased from 1 to 6. As might be expected, the numberof cells exhibiting a large number of wells was much smaller than4161 and ranged from a minimum of 9 cells for subgroup 5 (RPG be-tween 0.7–0.79 psi/ft) to 68 cells for Subgroup 3 (RPG between 0.5–0.59 psi/ft). Subgroup 6 with the largest RPG range of 0.8–0.89 psi/fthad 11 cells.

5.5.1. Comparison of control group (Subgroup 1) to test group (Subgroup 6)In this test, a difference in data distribution between Subgroup 1 and

Subgroup 6 was evaluated using theMann–Whitney U test. Subgroup 6was considered to be the group with the highest risk areas in themodeldue to a high density of gaswells and a high reservoir pressure gradient,whereas Subgroup 1 represents the lowest risk areaswhere no gaswellswere present. The data in Subgroup 1 and Subgroup 6 for each of the 31constituents were plotted as Boxplots in Minitab for distribution com-parison (plots not shown).

5.5.2. Inter-subgroup correlationsIn this analysis, correlations between constituent concentrations

within the data points in each subgroupwere found considering the hy-pothesis that strong correlations between constituent concentrationswould be related to elevated constituent levels andwould indicate a re-lationship between gas well fracturing and contaminant migration.Matlab was used to find correlations between the constituent levels inSubgroups 1–6. The R-squared value was obtained from a linear regres-sion model for each constituent pair within each data Subgroup. TheMatlab code was run 6 times; for each run, a 31 × 31 output matrixwas created, a row and column for each constituent — the cellintersected by each row and column contains the R-squared value forthe two constituent variables. For each output matrix, the results weredivided into R-squared values less than 0.5 (weaker correlations) andR-squared values greater than 0.5 (stronger correlations) since R-squared varies between 0 and 1.

5.6. Cluster analysis

A spatial clustering analysis was undertaken in ArcGIS for 2049 gaswells in order to explore the relationships between variables associated

Fig. 4. Gas well density raster map. (For interpretation of the references to color in this figure, the reader is referred to the web version of this article.)

121T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

with well completions/hydraulic fracturing and change in groundwaterconstituent concentrations. The cluster analysismethodology presentedhere developed spatial clusters of wellbore completions/fracturing datawith values that were similar in magnitude. The analysis used the LocalMoran's I method built-in within the ArcGIS Cluster Analysis tool thatreturns the Local Moran's I index, z-score, p-value, and cluster/outliertype. I is the spatial statistic of spatial association (the function identifieswhere high or low values cluster spatially, and features with values thatare very different from surrounding feature values). The z-scores and p-values are measures of statistical significance that indicate whether theapparent similarity or dissimilarity is more pronounced than onewouldexpect in a random distribution. The analysis returns an output of clus-ter locations denoted byHH, LL, andNot Significant rankings of the clus-tered variable (where HH denotes statistically significant (0.05 level)cluster of high values and LL for a statistically significant (0.05 level)cluster of low values).

A total of ten well completions/hydraulic fracturing properties wereanalyzed as shown below. The studied variables were all related to thepotential for a well to rupture during hydraulic fracturing:

Table 2Data subgroup descriptions.

Subgroupnumber

Wells per 0.4mi2

Reservoir pressuregradient(psi/ft)

Number of datapoints

1 0 – 41612 18–54 0.43–0.49 353 18–54 0.50–0.59 684 18–54 0.60–0.69 225 18–54 0.70–0.79 9

1–2) Surface casing/bottomhole casing: The size of the casing is impor-tant to thewellbore system, where pressure ratings decreasewithsmaller pipe diameter, increasing the risk for wellbore failure;

3) Injected fluid volume: A greater volume of injected fluid wouldexpose the wellbore system to high pressures for a longer dura-tion, potentially weakening the integrity of the well. Additionally,larger volumes of pumped fluid indicate that the hydraulic frac-tures may be larger with a potential to release a greater amountof gas;

4) Injected weight of sand: Injecting more sand could cause erosionof the perforations and form a microannular pathway. A largersand volume indicates a larger-scale fracture treatment andgreater potential for erosion;

5) Total vertical depth: A deeper gas well has increased distance be-tween the shale and water aquifer thereby potentially decreasingthe risk of groundwater contamination;

6) Volume of N2: Nitrogen gas is usually injected into shallowerwells and could act as a mobilizer of contaminants as it flowsfreely back to the surface;

7) Length of lateral: A longer lateral section in the wellbore could in-crease the risk of an insufficient cement barrier due to the poten-tial settling of cement in the horizontal wellbore. The laterallength was found by subtracting the Total Vertical Depth of theWellbore from the Measured Depth of the entire wellbore. Notethat some wells were vertical, resulting in a lateral length ofzero. This cluster analysis effectively made a comparison betweenvertical and horizontalwells,where LL clusterswere vertical wellswith a lateral length of zero, and HH clusters designated the lon-gest horizontal section;

8) Aquifer-perforation thickness: Perforatedwellbore sections abovethe Total VerticalWell Depth increase the risk ofwellbore rupture.

122 T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

Thus, a lower value of the thickness between the perforations andthe water aquifer indicates an increased risk of groundwatercontamination;

9) Wells existing in 1 mile: The number of gas wells within a 1 miledistance from the gas well of interest was evaluated to assess thecumulative effects of tightly spaced hydraulic fracturingwells; thehypothesis being that a greater number of gas wells may increasethe possibility of a contaminant pathway to groundwater from agas well even if the pathway has not been specified; and

10) Bottom hole pressure/reservoir pressure gradient: A high reser-voir pressure would increase the flow rate and volume of naturalgas from the reservoir, possibly increasing the amount of contam-inants mobilized.The aforementioned wellbore completion parameters were pri-marily sourced from the Texas Railroad Commission G-1 comple-tion forms which were stored on the Texas Railroad Commissiononline servers; formsfiled prior to 2010were stored on a separateserver (http://www.rrc.state.tx.us/about-us/resource-center/research/online-research-queries/imaged-records-menu/) thanthose filed after 2010 (http://webapps.rrc.state.tx.us/CMPL/publicHomeAction.do). Some wells were missing data related tofluid and sand volumes; the data were searched for in Fracfocus.org. At each well location, the depth of the Paluxy and the BottomHole Pressure (BHP)were extracted from theArcGISmodel. In theanalysis, BHP was analyzed instead of the RPG since there is notenough variability in the data values of RPG for ameaningful clus-ter analysis in ArcGIS.Beryllium concentrations were extracted from ArcGIS at theresulting cluster locations and were evaluated to determine if el-evated concentrations were associated with specific clusters ofthe wellbore variables 1 through 10 discussed above. Berylliumwas used as an indicator variable in this analysis based upon theresults from the gas well density correlations (as will be seen inthe Results and discussion section of the paper), where this con-stituentwas found to exhibit a relationship to hydraulic fracturingin all analyses. TheHHand LL clusters for each of thewellbore var-iables 1 through 10 were compared using the Mann–Whitney UTest to the corresponding beryllium concentrations at the clusterlocations. Thus, if a statistically significant difference in berylliumconcentration was detected between locations of the HH and LLclusters, then this would indicate that the specificwellbore designparameter is important and indicative of a potential contaminantpathway in the wellbore system.

6. Results and discussion

6.1. Visual analyses of groundwater quality constituent change

A total of 40 plots for the 20 groundwater constituents considered inthe analysis were evaluated and the change in each constituent wascompared to the spatial distribution of gas wells and variations in thereservoir pressure gradient. A qualitative assessment of the plots dem-onstrated that a correlation between the changes in groundwater con-stituent concentrations, gas well locations, and the reservoir pressuregradient could not be clearly deduced for most of the plots, with the ex-ception of total beryllium. Fig. 5 illustrates a visual correlation betweenincreased total beryllium concentrations and the location of gas wells.As can be seen in Fig. 5, the areas with the greatest positive change intotal beryllium, denoted by red, are associated with the presence ofgas wells while the area of the plot showing a decrease in total beryl-lium, denoted in green, is not. While qualitative in nature, andrepresenting trends in groundwater quality over a regionwith inherentinaccuracies within the plot due to data availability limitations, thetrends in Fig. 5 were considered significant particularly since similartrends were not apparent in the majority of the other plots.

The visual correlation between the reservoir pressure gradient andthe change in total beryllium shown in Fig. 6 is not strong, however, itshould be noted that the reddish pressure contours overlay areas ofred shading only indicating that the highest changes in total berylliumconcentration correlate well with the areas of highest RPGs. Whilesome of the green RPG contours overlap areas of red total berylliumshading, the green contours tend to emanate from the yellow-greenareas of the plot.

The aforementioned findingwhen taken in conjunctionwith the factthat the majority of the other constituents were not well correlated towell density and/or RPGs led to the conclusion that beryllium deservesconsideration as an indicator in gas well production, and the potentialimpact from fracturing on groundwater quality. Berylliumalso deservesconsideration as a potential indicator variable for wellbore integrity is-sues in hydraulic fracturing operations. Since beryllium is almostnever found at detectable concentrations in ground water aquifers, itspresence at relatively elevated levels can be construed to indicate mi-gration through microannular defects in the wellbore. The averagetransport distance of 1000 ft (~305 m) in 10 years does not providean alternate explanation in this case because of the level of observed be-ryllium concentrations as will be seen later in the paper.

6.2. Proximity to gas well nonparametric statistical Mann–Whitney U-Test

The results of Mann–Whitney U-Tests do not strongly indicate thatproximity to gas wells was associated with degraded water quality(see Table S1 in the Supporting information). The results from Test 1(non-detects as zeros) and 2 (non-detects as 1/2 the detection limit) in-dicated a statistically significant difference between the control groupand test group for arsenic (P= 0.04), chloride (P= 0.01), dissolved ox-ygen (P= 0.03), selenium (P= 0.0), water temperature (P= 0.0), andtotal dissolved solids (P = 0.0). The median of the test group samples(the ones expected to be affected by proximity to gas wells), however,was lower than the control group, with the exception of selenium(P = 0.0) and dissolved oxygen (P = 0.03) that were higher. These re-sults emphasize the need to address groundwater quality change at aregional scale taking into account the density of gas wells, their depthand the pressure gradient.

In Test 3, where the non-detect values were omitted, the resultsdemonstrated a statistically significant difference in median concentra-tion between arsenic (P = 0.04), beryllium (P = 0.03), chloride (P =0.01), dissolved oxygen (P = 0.02), water temperature (P = 0.0), andtotal dissolved solids (P=0.0). Outside of dissolved oxygen, themedianof the test group was less for all constituents except for beryllium,where themedian concentration was higher. This provided further evi-dence that beryllium concentrations are related to gas well productionin the Barnett.

6.3. Gas well density nonparametric Mann–Whitney statistical test

The maximum, minimum, median, and mean values for each con-stituent for the six data subgroups described in Table 2 are shown inTable S2 in Supporting information (recall that Subgroup 1 representsa zero-well density and Subgroup 6 represents a high well densitywith high pressure gradient). The values in the table exceeding theEPA Primary Water Quality drinking standard are highlighted in yellow(note that the Primary drinking water quality standards are not avail-able for all constituents studied).

As can be seen in Table S2, the mean, median, and maximum valuesof beryllium exceed the EPA threshold in Subgroup 6, which is not thecase for any of the other constituents or Subgroups. The maximum con-centration values detected for arsenic, benzene, and beryllium exceedthe enforceable standard and are present in Subgroup 1. A boxplot ofthe beryllium concentrations for Subgroup 1 and Subgroup 6 was cre-ated in Minitab and is shown in Fig. 7. As can be seen in Fig. 7, the me-dian beryllium concentration is elevated in Subgroup 6. This finding

Fig. 5. Change in total beryllium and location of gas wells in the Barnett. (For interpretation of the references to color in this figure, the reader is referred to theweb version of this article.)

123T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

supports the use of beryllium as an indicator variable for evaluating hy-draulic fracturing impacts on ground water quality within a region.

6.3.1. Comparison of control group (Subgroup 1) and test group (Subgroup6)

The Mann–Whitney U-Test demonstrated a statistically significantdifference between themedian concentrations for most of the constitu-ents. Total arsenic (P = 0.04), total beryllium (P = 0.0), dissolved bro-mide (P = 0.02), total copper (P = 0.0), total ethanol (P = 0.04), totalmethanol (P=0.0), dissolved radium228 (P=0.0), andwater temper-ature (P= 0.0) have a significant greater median concentration in Sub-group 6 than in Subgroup 1. Dissolved aluminum (P = 0.04), totalbromide (P = 0.01), total chloride (P = 0.0), dissolved oxygen (P =0.02), total iron (P = 0.0), total selenium (P = 0.01), total sulfate(P = 0.0), total dissolved solids (P = 0.0), and dissolved vanadium(P = 0.0) had a significant lower median concentration in Subgroup 1.The median concentration of nitrate (P = 0.03) in both groups was 0.The results of the Mann–Whitney test for all constituents are shownin Table S3 in the Supporting information.

As can be seen in Table S3, nine of the constituents tested(highlighted in gray) have a relationship to hydraulic fracturing activitywhere the mean of Subgroup 6 is greater than Subgroup 1, except for

dissolved oxygen which showed a lower concentration as would be ex-pected (due to higher groundwater temperatures). The increased arse-nic in the datamay be expected since arsenic is present in natural gas astrimethylarsine and processing plants are equipped to remove it(Kidnay et al., 2006). Bromide from the shale reservoirmay be dissolvedin water particles produced with the natural gas where it is contactingthe water table as it travels through the micro-annulus, explaining theincrease in the presence of dissolved bromide. Interestingly, there isno statistically significant difference in total bromide concentrations be-tween the Subgroups. The increased copper may be associated with thehydraulic fracturing chemicals or shale rock properties and increasedethanol is likely related to the hydraulic fracturing chemicals. The in-crease in radium 228, and beryllium (a radionuclide) may be attributedto the produced gas as shale formations have naturally occurring radio-active materials.

6.3.2. Inter-subgroup correlationsThe analysis demonstrated a strong correlation between constituent

concentrations in the data subgroups associated with high reservoirpressure gradient and high density of gas wells. The number of constit-uent correlations with a value greater than 0.5 was significantly higherin the Subgroups associated with a high reservoir pressure and high

Fig. 6. Change in total beryllium and contoured RPG in the Barnett.

124 T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

well density than in other subgroups. Subgroup 5 and Subgroup 6 hadthe greatest correlation between constituents, whereas the controlgroup, Subgroup 1, had almost no correlations. A summary of the resultsis shown in Table 3.

A general trend was seen whereby an increase in reservoir pressurewas accompanied by an increase in the correlation between constituentvariables. Despite the low number of data points in Subgroups 5–6,there were high correlations between various constituents yielding astrong indication that high well density and reservoir pressure are pre-dictor variables for groundwater quality changes in the Barnett. In Sub-groups 5–6, the concentration of Beryllium was strongly correlated tothe concentration of the other constituents. Since beryllium concentra-tionwas demonstrated to be related to gaswell operations, a correlationbetween beryllium and another constituent would indicate that ele-vated levels of other constituents in the groundwater may have somerelationship to gas well operations as well.

6.4. Cluster analysis

The Cluster analysis performed in ArcGIS showed significant cluster-ing for 8 of the 10 clustering variables—Injected Fluid (Cluster 3) and ni-trogen (Cluster 5) did not have significant clustering and were omittedfrom the evaluation. Thus, HH and LL cluster locations were found for

the remaining 8 clustering variables. The concentration of beryllium(from 2011 to 2014 sample data) extracted at the HH and LL cluster lo-cations for 2 of the variables: Surface Casing and Bottom Hole Casingfound no statistical significance for the two variables. The results ofthe Mann–Whitney U-Test for the remaining clusters, however, weresignificant and of interest.

As expected, a higher density ofwells (variable 9) andhigher bottomhole pressure (variable 10)were associatedwith a highermedian beryl-lium concentration. Clusters with a greater vertical depth (variable5) had a lower median concentration of beryllium. Likewise, clusterswith a decreased thickness between the Trinity and the uppermost per-foration (variable 8) were associated with a higher median concentra-tion. The results of the Lateral Length clusters (variable 7) showedthat the LL clusters (vertical wells having a lateral length of zero)were associated with a greater median beryllium concentration thanthe HH cluster values. Additionally, clusters of high injected weight ofsand (variable 4) were not associated with a higher median berylliumconcentration, further indicating that the contamination pathway isnot related to the horizontal wellbore. These results are logical andshould be expected based upon an understanding of fracture extension,where fractures tend to extend upwards. In vertical wellbores, thismeans that the fractures are parallel to the annulus, possibly creatinga breach in the wellbore system.

Fig. 7. Beryllium concentration boxplot for Subgroup 1 (right) and Subgroup 6 (left).

125T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

7. Conclusions

The results from this research emphasize the need to study ground-water quality and hydraulic fracturing relationships in a spatial contextat the regional scale, andwith respect to the geophysical characteristicsof thewellbore environment. This is particularly noted in comparing theresults in Sections 6.2 and 6.3, where it was demonstrated that the den-sity of wells per area establishes a relationship between groundwaterquality changes and hydraulic fracturing. Modeling water quality withrespect to specific characterization of the wellbore environment(Section 5.5) resulted in statistically significant differences in medianconstituent concentrations that indicated that degraded groundwaterquality has some relationship to hydraulic fracturing operations. Addi-tionally, this research demonstrated that while a number of constitu-ents can serve as indicators of groundwater quality, total berylliumwas found to be associatedwith gaswell production and to be the stron-gest indicator variable for detecting a pathway between gas wells andgroundwater.

By identifying an appropriate indicator variable such as beryllium inthis case, the results of a cluster analysis of well design and hydraulicfracturing parameters allowed identification of a possible origin of thecontaminant pathways in the wellbore environment. The results fromthe study indicated that contaminant pathways are formed in the verti-cal section of a wellbore, where the fracture extends parallel to thewellbore potentially creating a microannular pathway in the cementsheath. Thus, improving hydraulic fracturing treatment and wellbore

Table 3Correlated constituent levels per data subgroup.

Subgroupnumber

Number of Pearsoncoefficients above0.5

Number of berylliumPearsoncoefficients above 0.5

1 18 02 73 63 23 24 228 11

designs may reduce the potential impact of natural gas production onfresh water resources.

While most of the constituents tested did not have sample concen-trations exceeding the EPAMCL threshold, strong correlations betweenvarious constituents with beryllium (which appears to be highly relatedto hydraulic fracturing and exceeds the MCL) may indicate that theother constituents do indeed have elevated concentration levels thatmay also be associated with the presence of hydraulic fracturingoperations.

This study demonstrated that while the quality of groundwater maynot be directly associated with proximity to gas wells; it is impacted bythe high density of gas wells in an area. In the Barnett Shale region, thehighest density of gas wells is located in the highest pressure gradientregion. A high density of gas wells treated in a small area may causean intersection of pressure cones in the subsurface, possibly increasingthe reservoir pressure and/or fracture treatment pressure and affectingthe integrity of the vertical wellbore. To what extent that may be occur-ring is unknown within the context of data available for this study.However, future work on this subject should further investigate thisissue by incorporatingmore specific knowledge of the hydraulic fractur-ing treatment pressures (particularly the cumulative effects of multiplefracturing events within a mile of a groundwater water well), wellborepressure limitations, and reservoir rock properties into the model.

Acknowledgments

TheTexasWater Commission on Environmental Quality (TCEQ)pro-vided support for this research; their support is gratefully acknowl-edged. However, it is noted that the work presented in the paper isthe sole product of the authors. Mr. TomHolley, Interim Chair of the Pe-troleum Engineering Department at the University of Houston is ac-knowledged for providing access to DrillingInfo.com and for hisconstructive comments and support during the development of theresearch.

References

Barker, C., 1990. Calculated volume and pressure changes during the thermal cracking ofoil to gas in reservoirs (1). AAPG Bull. 74 (8), 1254–1261.

126 T.G. Burton et al. / Science of the Total Environment 545–546 (2016) 114–126

Bowker, K.A., 2007. Barnett shale gas production, FortWorth Basin: issues and discussion.AAPG Bull. 91 (4), 523–533. http://dx.doi.org/10.1306/06190606018.

Chermak, J.A., Schreiber, M.E., 2014. Mineralogy and trace element geochemistry of gasshales in the United States: environmental implications. Int. J. Coal Geol. 126, 32–44.

Darrah, T.H., Vengosh, A., Jackson, R.B., Warner, N.R., Poreda, R.J., 2014. Noble gases iden-tify the mechanisms of fugitive gas contamination in drinking-water wells overlyingtheMarcellus and Barnett Shales. Proc. Natl. Acad. Sci. U. S. A. 111 (39), 14076–14081.http://dx.doi.org/10.1073/pnas.1322107111.

Davies, R.J., Almond, S., Ward, R.S., Jackson, R.B., Adams, C., Worrall, F., Herringshaw, L.G.,Gluyaa, J.G., Whitehead, M.A., 2014. Oil and gas wells and their integrity: implicationsfor shale and unconventional resource exploitation. Mar. Pet. Geol. 56, 239–254.

Deremble, L., Loizzo, M., Huet, B., Lecampion, B., Quesada, D., 2010. “Assessment of leak-age pathways along a cemented annulus”. Paper presented at the Society of Petro-leum Engineers. SPE International Conference on CO2 Capture, Storage, andUtilization 2010.

Eaton, T.T., 2013. Science-based decision-making on complex issues: Marcellus shale gashydrofracking and NewYork City water supply. Sci. Total Environ. 461–462, 158–169.http://dx.doi.org/10.1016/j.scitotenv.2013.04.093.

Finkel, M.L., Hays, J., 2013. The implications of unconventional drilling for natural gas: aglobal public health concern. Public Health 127 (10), 889–893.

Fontenot, B.E., Hunt, L.R., Hildenbrand, Z.L., Carlton, D.D., Oka, H., Walton, J.L., ... Schug,K.A., 2013. An evaluation of water quality in private drinking water wells near naturalgas extraction sites in the Barnett shale formation. Environ. Sci. Technol. 47 (17),10032–10040.

Gordalla, B.C., Ewers, U., Frimmel, F.H., 2013. Hydraulic fracturing: a toxicological threatfor groundwater and drinking-water? Environmental Earth Sciences 70 (8),3875–3893.

Harden, R.W., 2004. Northern Trinity,Woodbine Aquifer Groundwater Availability Model.RW Harden & Assoc.

Harkness, J.S., Dwyer, G.S., Warner, N.R., Parker, K.M., Mitch, W.A., Vengosh, A., 2015. Io-dide, bromide, and ammonium in hydraulic fracturing and oil and gas wastewaters:environmental implications. Environ. Sci. Technol. 49 (3), 1955–1963.

Hildenbrand, Z.L., Carlton, D.D., Fontenot, B.E., Meik, J.M., Walton, J.L., Taylor, J.T., Thacker,J.B., Korlie, S., Shelor, C.P., Henderson, D.J., Kadjo, A.F., Roelke, C., Dasgupta, P.K.,Hudak, P., Burton, T., Rifai, H., Schug, K.A., 2015. A comprehensive analysis of ground-water quality in the Barnett Shale Region. Environ. Sci. Technol. http://dx.doi.org/10.1021/acs.est.5b01526.

Ingraffea, A.R., Wells, M.T., Santoro, R.L., Shonkoff, S.B., 2014. Assessment and risk analysisof casing and cement impairment in oil and gas wells in Pennsylvania, 2000–2012.PNAS 111 (30), 10955–10960.

Jackson, R.B., Vengosh, A., Darrah, T.H., Warner, N.R., Down, A., Poreda, R.J., Osborn, S.,Zhao, K., Karr, J.D., 2013. Increased stray gas abundance in a subset of drinkingwater wells near Marcellus shale gas extraction. Proc. Natl. Acad. Sci. U. S. A. 110(28), 11250–11255. http://dx.doi.org/10.1073/pnas.1221635110.

Kidnay, A.J., Parrish,W.R., McCartney, D.G., 2006. Fundamentals of Natural Gas Processing.218. CRC Press.

Kresic, N., 2006. Hydrogeology and Groundwater Modeling. CRC press, Boca Raton, FL,p. 78.

Lester, Y., Ferrer, I., Thurman, E.M., Sitterley, K.A., Korak, J.A., Aiken, G., Linden, K.G., 2015.Characterization of hydraulic fracturing flowback water in Colorado: implications forwater treatment. Sci. Total Environ. 512–513, 637–644.

McKenzie, L.M., Witter, R.Z., Newman, L.S., Adgate, J.L., 2012. Human health risk assess-ment of air emissions from development of unconventional natural gas resources.Sci. Total Environ. 424, 79–87.

Meng, Q., 2015. Spatial analysis of environment and population at risk of natural gasfracking in the State of Pennsylvania, USA. Sci. Total Environ. 515–516, 198–206.

Molofsky, L.J., Connor, J.A., Farhat, S.K., Wylie Jr., A.S., Wagner, T., 2011. Methane in Penn-sylvania water wells unrelated to Marcellus shale fracturing. Oil Gas J. 109 (19),54–67+93.

Molofsky, L.J., Connor, J.A., Wylie, A.S., Wagner, T., Farhat, S.K., 2013. Evaluation of meth-ane sources in groundwater in northeastern Pennsylvania. Ground Water 51 (3),333–349. http://dx.doi.org/10.1111/gwat.12056.

Montgomery, S.L., Jarvie, D.M., Bowker, K.A., Pollastro, R.M., 2006. Mississippian BarnettShale, Fort Worth basin, north-central Texas: gas-shale play with multi-trillioncubic foot potential: reply. AAPG Bull. 90 (6), 967–969. http://dx.doi.org/10.1306/02090605186.

Moritz, A., Hélie, J.-F., Pinti, D.L., Laroque, M., Barnetche, D., Retaileau, S., Lefebvre, R.,Gélinas, Y., 2015. Methane baseline concentrations and sources in shallow aquifersfrom the shale gas-prone region of the St. Lawrence Lowlands (Quebec, Canada). En-viron. Sci. Technol. 49 (7), 4765–4771.

Murray, K.E., 2013. State-scale perspective on water use and production associated withoil and gas operations, Oklahoma, U.S. Environ. Sci. Technol. 47 (9), 4918–4925.http://dx.doi.org/10.1021/es4000593.

Osborn, S.G., Vengosh, A., Warner, N.R., Jackson, R.B., 2011. Methane contamination ofdrinking water accompanying gas-well drilling and hydraulic fracturing. Proc. Natl.Acad. Sci. U. S. A. 108 (20), 8172–8176.

Paulson, D.S., 2003. Applied Statistical Designs for the Researcher. CRC Press, NY, NewYork, p. 525.

Preston, T.M., Chesley-Preston, T.L., Thamke, J.N., 2014. A GIS-based vulnerability assess-ment of brine contamination to aquatic resources from oil and gas development ineastern Sheridan County, Montana. Sci. Total Environ. 472, 1152–1162.

Rahm, D., 2011. Regulating hydraulic fracturing in shale gas plays: the case of Texas. EnergPolicy 39 (5), 2974–2981. http://dx.doi.org/10.1016/j.enpol.2011.03.009.

Rahm, B.G., Bates, J.T., Bertoia, L.R., Galford, A.E., Yoxtheimer, D.A., Riha, S.J., 2013. Waste-water management and Marcellus Shale gas development: trends, drivers, and plan-ning implications. J. Environ. Manag. 120, 105–113.

Railroad Commission (RRC), 2014. http://www.rrc.state.tx.us/aboutus/resource-center/research/online-research-queries/#OilandGas.

Révész, K.M., Breen, K.J., Baldassare, A.J., Burruss, R.C., 2012. Carbon and hydrogen isotopicevidence for the origin of combustible gases in water-supply wells in north-centralPennsylvania. Appl. Geochem. 27 (1), 361–375.

Rowan, F., Engle, M., Kirby, C., Kraemer, T., 2011. Radium content of oil and gas field pro-duced waters in the northern Appalachian Basin (USA)—Summary and discussion ofdata. US Geological Survey Scientific Investigation Report, pp. 2011–5135 31 pp.

Rozell, D.J., Reaven, S.J., 2012. water pollution risk associated with natural gas extractionfrom the Marcellus Shale. Risk Anal. 32 (8), 1382–1393.

Smith, M.B., Hannah, R.R., 1996. High-permeability fracturing: the evolution of a technol-ogy. J. Pet. Technol. 48, 628–633.

Ternes, M.E., 2012. Regulatory programs governing shale gas development. Chem. Eng.Prog. 108 (8), 60–64.

Thacker, J.B., Carlton Jr., D.D., Hildenbrand, Z.L., Kadjo, A.F., Schug, K.A., 2015. Chemicalanalysis of flowback and produced water from unconventional drilling operations.Water 7, 1568–1579.

Texas Water Development Board (TWDB), 2014. https://www.twdb.texas.gov/groundwater/data/gwdbrt.sp.

Vidic, R.D., Brantley, S.L., Vandenbossche, J.M., Yoxtheimer, D., Abad, J.D., 2013. Impact ofshale gas development on regional water quality. Science 340. http://dx.doi.org/10.1126/science.1235009.

Walton, J., Woocay, A., 2013. Environmental issues related to enhanced production of nat-ural gas by hydraulic fracturing. Journal of Green Building 8 (1), 62–71.

Warner, N.R., Jackson, R.B., Darrah, T.H., Osborn, S.G., Down, A., Zhao, K., White, A.,Vengosh, A., 2012. Geochemical evidence for possible natural migration of Marcellusformation brine to shallow aquifers in Pennsylvania. Proc. Natl. Acad. Sci. U. S. A. 109(30), 11961–11966.

Werner, A., Vink, S., Watt, K., Jagals, P., 2015. Environmental health impacts of unconven-tional natural gas development: a review of the current strength of evidence. Sci.Total Environ. 505, 1127–1141.

Ziemkiewicz, P.F., Quaranta, J.D., Darnell, A., Wise, R., 2014. Exposure pathways related toshale gas development and procedures for reducing environmental and public risk.Journal of Natural Gas Science and Engineering 16, 77–84.