Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Paper: Water Pollution Risk Associated with Natural Gas Extraction from the Marcellus Shale


Published on

A research paper written by doctoral student Daniel Rozell and professor Dr. Sheldon Reaven, published in the August 2012 issue of the journal Risk Analysis. The paper creates mathematical risk models for five possible ways fresh water may become contaminated by fracking fluids during the process of drilling for natural gas. The authors conclude the most likely way it might happen is through the disposal of fracking wastewater.

  • Be the first to like this

Paper: Water Pollution Risk Associated with Natural Gas Extraction from the Marcellus Shale

  1. 1. Risk Analysis DOI: 10.1111/j.1539-6924.2011.01757.xWater Pollution Risk Associated with Natural GasExtraction from the Marcellus ShaleDaniel J. Rozell∗ and Sheldon J. Reaven1 In recent years, shale gas formations have become economically viable through the use of hor- izontal drilling and hydraulic fracturing. These techniques carry potential environmental risk due to their high water use and substantial risk for water pollution. Using probability bounds analysis, we assessed the likelihood of water contamination from natural gas extraction in the Marcellus Shale. Probability bounds analysis is well suited when data are sparse and pa- rameters highly uncertain. The study model identified five pathways of water contamination: transportation spills, well casing leaks, leaks through fractured rock, drilling site discharge, and wastewater disposal. Probability boxes were generated for each pathway. The potential contamination risk and epistemic uncertainty associated with hydraulic fracturing wastewater disposal was several orders of magnitude larger than the other pathways. Even in a best-case scenario, it was very likely that an individual well would release at least 200 m3 of contam- inated fluids. Because the total number of wells in the Marcellus Shale region could range into the tens of thousands, this substantial potential risk suggested that additional steps be taken to reduce the potential for contaminated fluid leaks. To reduce the considerable epis- temic uncertainty, more data should be collected on the ability of industrial and municipal wastewater treatment facilities to remove contaminants from used hydraulic fracturing fluid. KEY WORDS: Marcellus; probability bounds analysis; water1. INTRODUCTION Shale formations are a very promising source of nat- ural gas.(4) Shale is a sedimentary rock formed from Natural gas has become a preferred fossil fuel clay-rich mud in slow moving waters. The mud is afrom an environmental and political perspective.(1) precursor to natural gas and oil deposits owing to itsCompared to coal or oil, natural gas generates less high organic material content. By 2030, it is expectedair pollution and greenhouse gases (although natu- that half of all natural gas produced in the Unitedral gas production potentially can generate excessive States will come from unconventional sources, pri-methane emissions that could offset the beneficial ef- marily shale formations.(5) The Marcellus Shale is thefects of reduced CO2 emissions(2,3) ). In the United largest of the newly developing shale gas deposits inStates, natural gas is a primarily domestically pro- the United States.duced fuel that creates jobs and does not increase The Marcellus Shale is a thin, black forma-international trade deficits. Finding new supplies of tion that covers approximately 124,000 km2(6) fromnatural gas to keep up with demand is a challenge. New York to West Virginia at depths ranging from ground level to over 2,500 m (Fig. 1). As recently1 Department of Technology and Society, State University of as 2002, the entire formation was estimated to hold New York at Stony Brook, Stony Brook, NY, USA.∗ Address correspondence to Daniel Rozell, Department of Tech- 53 billion m3 of natural gas.(7) However, more recent nology and Society, 347A Harriman Hall, Stony Brook Univer- estimates of recoverable natural gas are as large as sity, Stony Brook, NY 11794-3760, USA; 13.8 trillion m3 .(8,9) By using the advanced techniques 1 0272-4332/11/0100-0001$22.00/1 C 2011 Society for Risk Analysis
  2. 2. 2 Rozell and Reaven probability ranges of water contamination risk for a typical natural gas well in the Marcellus Shale re- gion that is being developed using high-volume hy- draulic fracturing and horizontal drilling. The prob- ability bounds constitute the best case (smallest possible contamination) and worst case (largest pos- sible contamination) for a single well. The distance between these bounds represents the amount of epis- temic uncertainty (lack of knowledge) regarding the process. The analysis is intended to inform the con- tentious public debate over shale gas extraction in the region. Stakeholders and the public generally can decide if they are willing to accept potential water contamination risks within the probability bounds and policymakers can decide if additional research is needed to decrease the epistemic uncertainty before a policy decision can be made.Fig. 1. The Marcellus Shale formation in northeastern UnitedStates (modified from Milici and Swezey(71) ). 2. DATA AND METHODS 2.1. Analysis Methodof horizontal drilling and hydraulic fracturing, it now The risk analysis was performed using probabil-seems to be economically feasible to extract natural ity bounds analysis (PBA) as developed by Yager,(17)gas from the Marcellus Shale. Although these tech- Frank et al.,(18) Williamson and Downs,(19) Fersonniques are well established, they are not without po- and Ginzburg,(20) and Ferson.(21) PBA is used to cre-tential risk. ate probability boxes (p-boxes) that combine proba- Hydraulic fracturing uses high-pressure solutions bility distributions to represent aleatory uncertaintyto create and prop open fractures in rock to enhance (natural variation) and interval arithmetic to rep-the flow of oil, gas, or water. More than 750 dis- resent epistemic uncertainty (lack of knowledge).tinct chemicals, ranging from benign to toxic, have Karanki et al.(22) provides a more complete overviewbeen used in hydraulic fracturing solutions.(10,11) Al- of interval analysis and PBA calculations with exam-though these additives are less than 2% by vol- ples. Because p-boxes emphasize the bounded rangeume of the total fracturing fluid, hydraulic fractur- of a class of possible distributions that might be gen-ing is a water-intensive process and at least 50 m3 erated by techniques such as second-order Monteof chemicals would be used for a typical 10,000 m3 Carlo or Bayesian sensitivity analysis, p-boxes arehydraulic fracturing project.(12) Given the extensive particularly well suited to analyses where distribu-use of hydraulic fracturing in recovering gas from the tion parameters are highly uncertain and correla-Marcellus Shale, large quantities of wastewater are tions are unknown. The computationally simple andexpected to be generated. In 2010, the U.S. Environ- mathematically rigorous bounds generated by PBAmental Protection Agency started an investigation, are useful for analyses where tail risks and best-scheduled for completion in 2012, of the potential case/worst-case scenarios are of special interest.(23)impact of hydraulic fracturing on drinking water.(13) For these reasons, PBA has been recently used inCurrently, hydraulic fracturing is exempt from reg- a variety of environmental risk assessments.(24−27)ulation under the Safe Drinking Water Act due to The method has been critiqued as a simple worst-an exemption written in the Energy Policy Act of case technique,(28) but risk managers can use PBA2005.(14,15,16) to determine if a desirable or undesirable outcome The public policy decision to pursue natural gas resulting from a decision is even possible, whetherextraction from the Marcellus Shale involves several the current state of knowledge is appropriate forpotential risks and benefits. The crucial unknown making a decision, or as a complement to other riskis the potential risk of water contamination from analysis methods. For this study, the software Riskhydraulic fracturing. This study generates bounded Calc 4.0(21) was used for all calculations.
  3. 3. Marcellus Shale Water Pollution Risk 3 where CV T is the contaminant volume spilled from transportation in m3 per well, F is the drilling and fracturing fluid on site in m3 , PR is the portion of drilling and fracturing fluid returned from the well, N C is the number of hazmat truck crashes in the United States each year, N S is the total number of hazmat shipments in the United States each year, PS is the portion of hazmat tankers that spill in crashes, and PL is the portion of a tanker truck load thatFig. 2. Model of water contamination pathways. would spill in a crash. 2.2.2. Well Casing Failure2.2. Model A failure in a well casing that would cause a leak There are many types of water contamination of fluids to the surrounding groundwater was mod-that can result from the shale gas extraction process, eled as:including: gases (e.g., methane and radon), liquids(e.g., hydraulic fracturing fluids), and solids (e.g., drill CVW = PWFail × PWLeak × F, (2)cuttings). Because the hydraulic fracturing process where CV W is the contaminant volume leaked fromgenerates primarily liquid waste products, this risk the well casing in m3 per well, PWFail is the probabilityassessment only considers water contamination from that a Marcellus Shale gas well fails, PWLeak is portiondrilling and hydraulic fracturing fluids. For the pur- of the injected fluids that leak from the well, and F ispose of the assessment, the model defines contami- the drilling and fracturing fluid on site in m3 .nation as anything that could potentially exceed thelimits of the U.S. Clean Water Act or Safe Drink-ing Water Act. Given recent public attention to the 2.2.3. Contaminant Migration Through Fracturespotential environmental risks of hydraulic fracturing, The potential for hydraulic fracturing fluid todrillers have been making the transition to hydraulic travel through fractures into overlying aquifers wasfracturing components that are considered largely modeled as:benign.(29) However, even a benign hydraulic frac-turing fluid is contaminated once it comes in contact CVF = PFL × PFluid × F(1 − PR ), (3)with the Marcellus Shale. Recovered hydraulic frac- where CV F is the contaminant volume leakedturing fluid contains numerous materials from the through fractures in m3 per well, PFL is the prob-Marcellus Shale formation in excess of drinking wa- ability that well fractures will leak to an overlyingter standards, including: sodium, chloride, bromide, aquifer, PFluid is the portion of fluids leaked througharsenic, barium, and naturally occurring radioactive fractures, F is the drilling and fracturing fluid on sitematerials such as uranium, radium, and radon.(30) in m3 , and PR is the portion of the fracturing fluidThus, any drilling or fracturing fluid is suspect for the returned from the well.purposes of this study. The proposed potential water contaminationpathways are shown in Fig. 2 for a hypothetical shale 2.2.4. Drilling Site Surface Contaminationgas well drilled in the Marcellus Shale using high- The potential for water contamination fromvolume hydraulic fracturing. The pathways followed drilling site spills due to improper handling or leaksthe life-cycle of the water used and were modeled as from storage tanks and retention ponds was modeleddescribed below. as: CVDS = PD × PFD × F × PR , (4)2.2.1. Transportation where CV DS is the contaminant volume discharged at Potential water contamination due to tanker the drilling site in m3 per well, PD is the probabilitytruck spills to and from the well site was modeled as: that the drilling site will experience some discharge, F(1 + PR ) × NC × PS × PL PFD is the portion of drilling site fluids discharged, CVT = , (1) F is the drilling and fracturing fluid on site in m3 , and NS
  4. 4. 4 Rozell and Reaven Table I. Data Used for the Model VariablesVariable Description P-Box Used SourceF Drilling and fracturing fluid on site MinMaxMean (9e3, 3e4, 1.7e4) 10, 33PR Portion of drilling and fracturing fluid returned from the well MinMaxMean (0.1, 1, 0.3) 10, 33, 35NC Number of hazmat truck crashes in the U.S. each year MinMaxMean (5,000, 7,800, 5,200) 32NS Total number of hazmat shipments in U.S. each year MinMax(2.9e8, 3.3e8) 32PS Portion of hazmat tankers that spill in crashes MinMaxMean (0.1, 0.4, 0.36) 32PL Portion of a tanker truck load that would spill in a crash MinMax (0.01, 0.99) EstimatePWFail Probability that gas well fails MinMaxMean (2e-8, 2e-2, 1.5e-3) 37, 39–42PWLeak Portion of the injected fluids that leak from the well MinMax (1e-6, 0.1) EstimatePFL Probability that well fractures will leak to aquifer MinMax (1e-6, 0.1) EstimatePFluid Portion of fluids leaked through fractures MinMax (1e-6, 0.1) EstimatePD Probability that site has some discharge MinMaxMean (0.1, 0.5, 0.3) 53PFD Portion of on-site fluids discharged MinMaxMean (1e-6, 1, 1e-4) 42, 53PT Portion of wastewater treated and released MinMax (0.75, 0.85) 42, 53PCR Portion of contaminants released after treatment MinMax (0.3, 1) 10, 58PNT Portion of wastewater not treated MinMax (0, 0.05) 42, 53PR is the portion of the fracturing fluid returned from 2.2.6. Correlationsthe well. For each of the pathways described above, the individual variables may or may not be independent.2.2.5. Contamination from Disposal of Used Some relations are most likely independent. For ex- Hydraulic Fracturing Fluids ample, there is no reasonable mechanistic relation between the number of hazmat truck crashes, N C , The potential for water contamination from dis- and the portion of the fracturing fluids returned fromposal of the used drilling and hydraulic fracturing flu- a gas well, PR . Other variables may be theoreticallyids depends on the method of disposal. If the fluid is correlated. For example, the portion of the fracturingreused, it was assumed there are no disposal losses fluids returned from a gas well, PR , the probabilityfor the well and any further losses are accrued by the that well fractures will leak to an overlying aquifer,next well. Deep well injection, to be discussed later, PFL , and the portion of fluids leaked through frac-was not considered. The disposal contamination was tures, PFluid , are all at least partially dependent on themodeled as: hydrogeological conditions in the local shale. Simi- larly, there could be some positive correlation be- CVWD = (F × PR − CVDS )(PT × PCR + PNT ). tween the total fluid used, F, and various spill and (5) treatment rates due to the difference in difficulty of managing the operations of a small versus a large hy- Substituting in Equation (4) for CV DS yields: draulic fracturing project. There is also a likely cor- relation among the possible pathways. An assump- CVWD = (F × PR (1 − PD × PFD )) tion of independence suggests that each step of the × (PT × PCR + PNT ), (6) drilling process is separately managed and operated. A more likely scenario is that all of the contami-where CV WD is the contaminant volume from nant pathways are positively correlated because theywastewater disposal in m3 per well, F is the drilling are managed by a single company that is consistentlyand fracturing fluid on site in m3 , PR is the portion of conscientious or careless during each step of the pro-the fracturing fluid returned from the well, PD is the cess. To find the most conservative bounds, all calcu-probability that the drilling site will experience some lations used the Frechet(31) method, which does not ´discharge, PFD is the portion of drilling site fluids dis- assume a specific dependence.charged, PT is the portion of the wastewater treatedand released to surface waters, PCR is the portion of 2.3. Datacontaminants that are released to surface waters af-ter treatment, and PNT is the portion of wastewater Table I lists the model variables and the datanot treated and released to surface waters. used to create the p-boxes. Because the published
  5. 5. Marcellus Shale Water Pollution Risk 5data are sparse for Marcellus Shale gas extraction tankers that spill in crashes, PS , was set to a meanand the drilling process is inherently uncertain, there of 0.36, a minimum of 0.1 based on the assumptionis considerable uncertainty regarding appropriate that nonfatal and unreported accidents have lowervariable values and distributions. This analysis used spill rates, and a maximum of 0.4. The portion of anonparametric p-boxes, a capability of PBA with tanker truck load that would spill in a crash was setminimal data distribution assumptions that gener- as a wide interval between 0.01 and 0.99. That is, aates conservative bounds. It can be useful in PBA tanker truck spill can range between very little andto select boundaries that are more conservative than almost the entire load.available data to account for uncertainty and to en-sure that the final p-box encloses the true proba- 2.3.2 Drilling and Fracturingbility. However, the selected level of conservatismcan be contentious. This study used publicly avail- Although the fluids used in the drilling mud doable data and estimates where available without in- not have the same characteristics or recovery rateflating the bounds. Generally, the best-case boundary as the hydraulic fracturing fluids, the two are not(left side of p-box) showed a contamination probabil- treated separately in this analysis because the drillingity resulting from the estimates of hydraulic fractur- fluid is only about 1% of the total combined vol-ing proponents (e.g., the natural gas drilling indus- ume.(33) Average estimated water usage for drillingtry). Similarly, the worst-case boundary (right side and hydraulic fracturing a well in the Marcellusof p-box) showed a contamination probability result- Shale ranges from 13,000 m3(10) to 21,000 m3(33) withing from the estimates of hydraulic fracturing oppo- limits of 9,000 m3 to 30,000 m3 for a typical 1,200 mnents (e.g., environmental organizations). Details of horizontal well.(10) For this analysis, the drilling andthe variables are described later. fracturing fluid used, F, was set as a nonparametric distribution with a mean of 17,000 m3 , a minimum of 9,000 m3 , and a maximum of 30,000 m3 . The rate2.3.1. Transportation of return of hydraulic fracturing fluid for shale gas Because there are no specific statistics avail- wells has been stated to range from: 9% to 35%,(10)able for hydraulic fracturing fluid transportation 30% to 60%,(34) 15% to 60%,(35) 15% to 80%,(36) orspills, the spill rates for liquid hazardous materials even 10% to 100%.(33) For this analysis, the portion(hazmat) transport in the United States are used as of drilling and fracturing fluid returned from the well,a proxy. Hazmat spill rates were chosen because they PR , was set as a mean of 0.3 with minimum and max-are tracked more reliably than non-hazmat. Accord- imum bounds of 0.1 and 1, to the U.S. Department of Transportation, there Regarding the likelihood of a well casing fail-are over 800,000 shipments of hazmat by truck each ure, a commonly cited statistic(10) originated with anday and about 5,200 hazmat truck accidents annually American Petroleum Institute (API) report(37) thatin the United States. However, the accident statistics estimates the absolute risk of contaminating an un-are underreported because an estimated one-quarter derground source of drinking water from a Class IIof hazmat truck accidents involve intrastate carriers (oil & gas) injection well as between 2 × 10−5 (1 inand the federal statistics are only collected on inter- 50,000) and 2 × 10−8 (1 in 50 million) well-years. Thestate carriers; similarly, the data are on the basis of report uses historical well failure rate data and isself-reporting by carriers.(32) When accidents involve based on the simultaneous failure of multiple wellfatalities, the spill rate for hazmat tankers is approxi- casings and fluids moving between the deep injec-mately 36%.(32) tion reservoir and surface aquifer.(38) It has been ar- For this assessment, the total number of hazmat gued(10) that the API study serves as an upper boundsshipments in the United States each year, N S , was for well failure risk because wastewater injectionset as an interval between 2.9 × 108 and 3.3 × 108 , wells continuously operate at higher than the geo-which equates to between 800,000 and 900,000 daily logic formation pressure whereas hydraulically frac-shipments. The number of hazmat truck crashes in tured wells only operate above the formation pres-the United States each year, N C , was set as a non- sure for a few days during construction. However,parametric distribution with a mean of 5,200, a mini- this assumption does not consider the much highermum of 5,000, and a maximum of 7,800 based on the pressures involved in hydraulic fracturing and theassumption that hazmat truck accidents could be as specific intent to generate additional fracturing inmuch as 50% underreported. The portion of hazmat the formation. Other studies have not supported the
  6. 6. 6 Rozell and ReavenAPI low failure rates. For example, Browning and erable incertitude on the subject, the assessmentSmith(39) find an average 10% failure rate for me- assumes an interval probability of fracture contam-chanical integrity tests of oil and gas injection wells. ination, PFL , as somewhere between extremely rarePresumably, the actual leakage rate is much lower. (1 in 1 million) and relatively common (1 in 10).Although the sample size was only 43 wells, a subse-quent report by the Underground Injection Practices 2.3.3. Drilling Site Leaks and SpillsCouncil (UIPC)(40) finds a 2% leak rate into under-ground sources of drinking water for Class I wastew- According to PA-DEP records(53) from July 2009ater injection wells. The Pennsylvania Department of to June 2010, there were about 4,000 permittedEnvironmental Protection (PA-DEP) found 52 sepa- Marcellus wells in PA. Of these wells, about 850 wellsrate cases of methane migration in a five-year period were producing gas, 400 wells were not producing,ending in 2009.(41) There are approximately 71,000 and 2,800 wells were planned or in process. Duringactive gas wells in Pennsylvania.(42) This corresponds this same time, 630 environmental, health, and safetyto a 1.5 × 10−4 (1 in 7,000) chance of a well leak- violations were issued for Marcellus wells in PA, ofing each year. Assuming a short 10-year well lifespan, which approximately half were for discharges up tothe lifetime well leak risk is 1 in 700. For this analy- 60 m3 or for potential to cause discharge. These datasis, the probability that a gas well fails, PWFail , was set are acknowledged by public officials to be underre-as a mean of 1.5 × 10−3 (1 in 700) with a minimum ported.(42) For this analysis, the probability that abound of 2 × 10−8 , based on the API study,(37) and site has some discharge, PD , was set as a mean ofa maximum bound of 2 × 10−2 , based on the UIPC 0.3, based on the number of discharge violations di-study.(40) Lacking reliable data, the portion of the in- vided by the number of active wells, a minimum ofjected fluids that leak from the well, PWLeak , was con- 0.1, based on the number of discharge violations di-servatively set as an interval ranging from 1 × 10−6 vided by the number of total permitted wells, and ato 0.1. maximum of 0.5 under the assumption of substantial The risk of hydraulic fracturing fluid migrating underreporting. The portion of drilling site fluids dis-through fractures to an overlying aquifer is a point charged, PFD , was set as a mean of 1 × 10−4 , basedof considerable debate.(43) Proponents of hydraulic on the mean violation discharge divided by the meanfracturing posit that there is no known mechanism total fluid used, a minimum of 1 × 10−6 , which repre-by which fractures and fluid can propagate through sents the smallest spill volume of interest, and a max-over 1,000 m of sedimentary strata; some recent data imum of 1, which assumes a catastrophic retentioneven suggest that Marcellus Shale wells should be pond failure where the entire contents are spilled.more closely spaced due to shorter effective frac-ture lengths than originally estimated.(44) In reply, 2.3.4. Wastewater Treatmentcritics say that several potential mechanisms can-not be ruled out:(10,45) unexpected vertical fractur- Although some well operators recycle and reuseing through overlying strata,(46,47) and fracturing fluid hydraulic fracturing fluids for multiple wells, mostpreferentially traveling through naturally occurring operators do not due to the cost of separation andfractures and faults.(44,48) Besides, critics argue, cur- filtration.(54) Instead, the used hydraulic fracturingrent conventional fracturing models are not appro- fluid is transported to a wastewater treatment facil-priate for shale gas reservoirs,(49) and the interpreta- ity and discharged to streams. From July 2009 totion of microseismic data used to monitor hydraulic June 2010, 729,000 m3 of Marcellus Shale hydraulicfracturing is still controversial.(50) Bredehoeft(51) dis- fracturing wastewater was reported in PA.(53) Of thecusses the substantial portion of conceptual ground- total, 77.5% was sent to approved industrial wastew-water models that are found to be invalid once fur- ater treatment facilities, 16% was reused in otherther data are available. As an initial assessment, wells, 5% was sent to municipal wastewater treat-Myers(52) simulated a Marcellus Shale well that had ment facilities, 1% had unknown disposal, 0.5% was19,000 m3 of fracturing fluid injected over five days. injected into deep wells, and 0.007% was spread onGiven the uncertainty in hydrogeological parameters roads. Although deep well injection is a common dis-(particularly conductivity, local flow gradients, and posal method in other shale gas areas of the Uniteddepth of shale), it was determined that fracturing flu- States, the Marcellus Shale region has relatively fewids could flow into overlying aquifers in timescales suitable deep injection sites(42) and the permittingranging from years to millennia. Given the consid- process for these wells is protracted.(55) Therefore,
  7. 7. Marcellus Shale Water Pollution Risk 7 Table II. Comparison of Contaminant Concentrations Before and After Industrial Wastewater Treatment Typical Untreated Mean Effluent Portion of Contaminant Wastewater Concentrations Concentration Not RemovedContaminant (Min, Median, Max) in ppm(10) in ppm(58) as Interval RangeaBarium (Ba) (0.5, 1450, 15700) 27 [0.001, 1]Bromide (Br) (11.3, 607, 3070) 1,069 [0.35, 1]Strontium (Sr) (0.5, 1115, 5841) 2,983 [0.51, 1]Chloride (Cl) (287, 56900, 228000) 117,625 [0.51, 1]Magnesium (Mg) (9, 177, 3190) 1,248 [0.39, 1]Total Dissolved Solids (TDS) (1530, 63800, 337000) 186,625 [0.55, 1]a Assumes that the effluent had not been substantially diluted when measured in the stream and that upstream sources were not substantiallyadding to the effluent concentrations.deep well disposal is considered negligible in thisanalysis. Municipal wastewater treatment facilities are notdesigned to handle hydraulic fracturing wastewatercontaining high concentrations of salts or radioactiv-ity two or three orders of magnitude in excess of fed-eral drinking water standards.(42,56) As a result, highsalinity and dissolved solids in Appalachian rivershave been associated with the disposal of MarcellusShale hydraulic fracturing wastewater after standardwastewater treatment.(57) The amount of wastewatertreated in public sewage facilities seems to be under-reported and actual levels may be as high as 50%.(42)Volz has presented data(58) that industrial wastewa-ter treatment facilities may also release effluent inexcess of drinking water standards (Table II). Forthis analysis, the portion of wastewater treated andreleased to surface waters, PT , was set as an inter-val with a minimum of 0.75 and a maximum of 0.85 Fig. 3. P-box generated for transportation spill risk.based on data reported to the PA-DEP and whethertreatment at a municipal wastewater treatment fa- (CDF) of the smallest potential water contaminationcility counted as treatment. The portion of contam- (best-case scenario). Similarly, the right side of theinants released after treatment, PCR , was set as an p-box represents the CDF of the largest potentialinterval with a minimum of 0.3 and a maximum of 1 water contamination (worst-case scenario). The hor-based on data in Table II. The portion of wastewater izontal distance between the right and left side of thenot treated, PNT , was set as an interval with a mini- p-box represents the epistemic uncertainty or lackmum of 0 and a maximum of 0.05 using the assump- of knowledge of the risk. Comparing the p-boxestion that treatment at a municipal wastewater treat- in Table III, the risks of water contamination arement facility could be categorized as nontreatment listed in increasing order with transportation (Fig. 3)because it is not designed to treat Marcellus Shale risks and epistemic uncertainty being negligible com-brines. pared to the other pathways. The contamination risks and epistemic uncertainties associated with well cas- ing failure (Fig. 4) and migration of fluids through3. RESULTS fractures (Fig. 5) were potentially substantial, but The p-boxes generated for each fluid loss path- minor compared to the contamination risk and epis-way given the assumptions stated above are shown temic uncertainty associated with disposal of used hy-in Figs. 3–7. In each p-box, the left side of the p- draulic fracturing fluids (Fig. 7). Although both thebox represents the cumulative distribution function best- and worst-case 50th percentile risk of drilling
  8. 8. 8 Rozell and Reaven Fig. 6. P-box generated for drilling site discharge risk.Fig. 4. P-box generated for well casing failure surface contamination (Fig. 6) were modest, at at least 200 m3 of contaminated fluids (by compari-the 99.9th percentile, a rare, but serious retention son, an Olympic-size swimming pool has a volume ofpond failure could generate a very large contami- 2500 m3 ).nated water discharge to local waters. An importantfeature of Fig. 7, the left side, or best-case boundary, 4. DISCUSSIONis expanded in Fig. 8 which more clearly shows that itwas very likely that an individual well would generate Estimating the risks of contamination scenarios is subject to more general underlying methodologi- cal and conceptual dilemmas than can be addressedFig. 5. P-box generated for fracture leaks to aquifer risk. Fig. 7. P-box generated for wastewater discharge risk.
  9. 9. Marcellus Shale Water Pollution Risk 9 Table III. Comparison of Water Contamination Pathway Risks from Hydraulic Fracturing in a Typical Macellus Shale Gas Well Best-Case 50th Worst-Case 50th Maximum Epistemic Percentile Contamination Percentile Contamination Uncertainty BetweenPathway Volume (m3 ) Volume (m3 ) Best and Worst Case (m3 )Transportation < 0.01 0.3 0.6Well casing failure < 0.01 9 60Fracture migration < 0.01 225 270Dilling site spills < 0.01 3 15,000Wastewater disposal 202 13,500 26,900 or shortly after the well construction phase. Second, any well casing failure would impact the surrounding areas during or after construction depending on dis- tance to the well. Finally, the time for fluid to migrate through fractures is poorly understood, but would likely affect drinking water sources years or decades after the well was constructed. Similarly, the rate at which hydraulic fracturing fluids would leak from a failed well casing is not con- sidered. Methane has a low molecular weight and vis- cosity and is expected to leak at a higher rate and travel farther than a heavier gas like radon and much more than many of the hydraulic fracturing compo- nents. However, it is likely that any gas leak large enough to be publicly noticeable is also leaking hy- draulic fracturing fluids. The estimate for transportation losses may be overestimated because some sites mix hydraulic frac- turing fluids on site. For these sites, most fluid goingFig. 8. Left bound (best-case) for wastewater disposal per wellp-box. to the site will consist of plain water. This overestima- tion decreases with the increasing reuse of hydraulichere—in assigning probabilities or probability dis- fracturing fluid. Regardless, the overall contributiontributions to the pertinent scientific theories and of transportation to the total potential water contam-mathematical models of groundwater contamination ination turns out to be negligible.themselves, in ensuring that all major contamination Although PBA generates mathematically rigor-scenarios have been considered, in characterizing the ous bounds(62) (i.e., the bounds are mathematicallyvarieties of uncertainty within Bayesian and non- guaranteed to enclose the true value), the bounds canBayesian frameworks, and in designing “how clean be too narrow or broad based on the validity of theis clean enough?” policy. These dilemmas are diag- data and assumptions inherent in the selection of thenosed by Reaven.(59−61) input variables. In this analysis, input bounds were No time variable was included in the present selected from currently available data and estimates.analysis to minimize the number of variables. As Any change in hydraulic fracturing practices in thesuch, only the total contamination potential at some Marcellus Shale would change the PBA. Similarly,future steady state was calculated. However, it is as the Marcellus Shale is developed, additional infor-not expected that the contamination from shale gas mation will become available and the epistemic un-would be experienced all at once or evenly over certainty should decrease. From Table III, it is clearsome predefined time period. Instead, each contam- that the lack of knowledge associated with the dis-ination pathway has a unique timeframe. First, the posal of used hydraulic fracturing fluid is the most im-spills from transportation, drilling site handling and portant area for further research. Within the disposalstorage, and disposal would be experienced during pathway, the critical variables are: the drilling and
  10. 10. 10 Rozell and Reavenfracturing fluid used, F (the primary variable that in- temic uncertainty was largest for wastewater disposalfluences all pathways) and the portion of contami- and for the rare, but serious, retention pond breachnants that are released after wastewater treatment, that could cause a large drilling site discharge. The p-PCR . box for the contamination risk from fluid migrating Regulators have recently moved to control the through fractures to an overlying aquifer (Fig. 5) hadsurface disposal of hydraulic fracturing wastewater substantial epistemic uncertainty. That is, the p-boxmore stringently.(63) The expected increased costs of was almost box-shaped—a representation of simpledisposal may cause drillers to increase fluid reuse interval uncertainty. This was a result of the veryor select alternative fracturing techniques. If not, large interval estimates used to represent the prob-regulators should explore the option of mandating ability that well fractures would leak and the por-alternative fracturing methods to reduce the wa- tion of fluid that would leak through the fractures.ter usage and contamination from shale gas extrac- Normally, this would suggest that future research ef-tion in the Marcellus Shale. Some extensively tested, forts be focused on the fluid fracture migration path-widely used alternatives are: N2 gas,(64) N2 -based way. However, the total uncertainty of fracture leaksfoams,(65,66) CO2 ,(67) and even liquefied petroleum was very small compared to the wastewater disposalgas (LPG).(68) Although water-based fracturing is potential risk and epistemic uncertainty. Hence, fu-the most commonly used for reasons of cost, famil- ture research efforts should be focused primarily oniarity, and effectiveness on low permeability, high wastewater disposal and specifically on the efficacypressure formations like the Marcellus,(69) nitrogen- of contaminant removal by industrial and municipalbased fracturing is the least expensive and most of- wastewater treatment facilities. Even in a best-caseten used alternative due to its ability to improve scenario, an individual well would potentially releaselow-pressure well production and reduce waste costs. at least 200 m3 of contaminated fluids.Carbon dioxide is a more expensive and corrosive Given typical well spacing in the Marcellusfracturing fluid, but it can preferentially displace Shale,(10) if only 10% of the region is developed, thismethane, thereby sequestering carbon dioxide(70,71) would equate to 40,000 wells. Using the best-case me-and lowering the carbon footprint of shale gas ex- dian risk determined above, this volume of contam-traction. Although it is counterintuitive, even LPG inated water would equate to several hours flow offracturing potentially could decrease the total wa- the Hudson River or a few thousand Olympic-sizedter pollution generated from Marcellus Shale gas swimming pools. This potential substantial risk sug-extraction. The PBA for Marcellus Shale hydraulic gests that additional steps be taken to reduce the po-fracturing found that the main sources of potential tential for contaminated fluid release from hydraulicwater pollution were from drilling site spills and fracturing of shale gas.wastewater disposal. LPG fracturing would reducedrilling site spills because there would be no reten- ACKNOWLEDGMENTStion ponds, only storage tanks. Similarly, LPG va-porizes after fracturing so it has a very high recovery We would like to thank Scott Ferson for his as-rate; moreover LPG is a saleable product that cre- sistance and helpful suggestions regarding probabil-ates a strong incentive to minimize loss. Given the ity bounds wastewater disposal options in the Marcel-lus Shale region, any of these fracturing fluid alter- REFERENCESnatives that substantially reduce wastewater gener- 1. Afgan NH, Pilavachi PA, Carvalho MG. Multi-criteria evalu-ation could make Marcellus Shale gas development ation of natural gas resources. Energy Policy, 2007; 35(1):704–more environmentally benign and a less controver- 713.sial source of future energy for the United States. 2. Hayhoe K, Kheshgi HS, Jain AK, Wuebbles DJ. Substitution of natural gas for coal: Climatic effects of utility sector emis- sions. Climatic Change, 2002; 54:107–139. 3. Howarth RW, Santoro R, Ingraffea A. Methane and the5. CONCLUSION greenhouse-gas footprint of natural gas from shale forma- tions. Climatic Change, 2011; 106. The study model identified five pathways of wa- 4. Ambrose WA, Potter EC, Briceno R. An “Unconventional”ter contamination: transportation spills, well casing Future for Natural Gas in the United States. Alexandria, VA:leaks, leaks through fractured rock, drilling site sur- Geotimes, AGI, 2008. 5. EIA. Annual Energy Outlook 2007 with Projections to 2030.face discharge, and wastewater disposal. Probability Washington, DC: Energy Information Administration, Dept.boxes were generated for each pathway. The epis- of Energy, 2007.
  11. 11. Marcellus Shale Water Pollution Risk 11 6. Moss K, McCoy C, O’Dell P, Garcia G, Simon S, 24. Ferson S, Burgman MA. Correlations, dependency bounds Rosenlieb G, Penoyer P, Salazer H. Development of the and extinction risks. Biological Conservation, 1995; 73:101– Natural Gas Resources in the Marcellus Shale: New York, 105. Pennsylvania, Virginia, West Virginia, Ohio, Tennessee, and 25. Regan HM, Hope BY, Ferson S. Analysis and portrayal of Maryland. Denver, CO: National Park Service, U.S. Dept. In- uncertainty in a food-web exposure model. Human and Eco- terior, 2009. logical Risk Assessment, 2002; 8:1757–1777. 7. Milici RC, Ryder RT, Swezey CS, Charpentier RR, Cook TA, 26. Regan HM, Sample BE, Ferson S. Comparison of determin- Crovelli RA, Klett TR, Pollastro RM, Schenk CJ. Assessment istic and probabilistic calculation of ecological soil screen- of Undiscovered Oil and Gas Resources of the Appalachian ing levels. Environmental Toxicology and Chemistry, 2002; Basin Province, 2002. USGS Fact Sheet FS-009–03, February 21:882–890. 2003. 27. Sander P, Bergback B, Oberg T. Uncertain numbers 8. Engelder T, Lash G. Marcellus Shale play’s vast resource po- and uncertainty in the selection of input distributions— tential creating stir in Appalachia. American Oil & Gas Re- Consequences for a probabilistic risk assessment of contami- porter, 2008; 52:76–87. nated land. Risk Analysis, 2006; 26(5):1363–1375. 9. Engelder T. Marcellus 2008: Report card on the breakout 28. Van der Voet H, Slob W. Integration of probabilistic ex- year for gas production in the Appalachian Basin. Fort Worth posure assessment and probabilistic hazard characterization. Basin Oil and Gas Magazine, 2009; 2(20):18–22. Risk Analysis, 2007; 27(2):351–371.10. New York State Dept. of Environmental Conservation. Pre- 29. Dezember R. Edible ingredients used to drill for gas. Wall liminary Revised Draft Supplemental Generic Environmen- Street Journal, December 15, 2010. tal Impact Statement of the Oil, Gas and Solution Mining 30. Harper JA. The Marcellus Shale—An old “new’ gas reser- Regulatory Program: Well Permit Issuance for Horizontal voir in Pennsylvania. Pennsylvania Bulletin, Bureau of To- Drilling and High-Volume Hydraulic Fracturing to Develop pographic and Geologic Survey, Pennsylvania Department of the Marcellus Shale and Other Low-Permeability Gas Reser- Conservation and Natural Resources, Middletown, PA, 2008; voirs. October 5, 2009. Revised July 2011. 38(1):2–13.11. U.S. House Committee on Energy and Commerce. Chemi- ´ ´ ´ ´ ` 31. Frechet M. Generalisations du theoreme des probabilites to- ´ cals used in hydraulic fracturing: Minority Staff, April 2011. tales. Fundamenta Mathematica, 1935; 25:379–387. Available at: 32. Craft R. Crashes involving trucks carrying hazardous mate- sites/default/files/documents/Hydraulic%20Fracturing% rials. Federal Motor Carrier Safety Administration. Wash- 20Report%204.18.11.pdf, Accessed April 18, 2011. ington, DC: U.S. Dept. of Transportation. Publication #:12. Soeder DJ, Kappel WM. Water resources and natural gas pro- FMCSA-RI-04–024, 2004 duction from the Marcellus Shale. USGS, 2009. Fact Sheet 33. Ingraffea AR. Letter to Gary Abraham, May 17, 2010. 2009–3032. Available at: drilling/13. EPA. Draft Hydraulic Fracturing Study Plan. 2010. Avail- NEWSNY in Chemung/Ingraffea ltr. and cv 5–17-10.pdf, able at: Accessed March 2, 2011. hydraulicfracturing/index.cfm, Accessed on February 24, 34. Houston N, Blauch M, Weaver D, Miller DS, O’Hara 2011. D. Fracture-stimulation in the Marcellus Shale—Lessons14. The Halliburton Loophole. New York Times, 2009 November learned in fluid selection and execution, in SPE Eastern Re- 2. gional Meeting Proceedings. Paper 125987. Charleston, WV:15. Zeller T. EPA to study chemical used to tap natural gas. New Society of Petroleum Engineers, 2009 York Times, September 9, 2010. 35. Chase B, Chmilowski W, Marcinew R, Mitchell C, Dang Y,16. Arthur JD, Hochheiser HW, Coughlin BJ. State and federal Krauss K, Nelson E, Lantz T, Parham C, Plummer J. Clear regulation of hydraulic fracturing: A comparative analysis, fracturing fluids for increased well productivity. Oilfield Re- in SPE hydraulic fracturing technology Conference Proceed- view, 1997; 3:20–33. ings, Paper 140482. Woodlands, TX: Society of Petroleum En- 36. EPA. Hydraulic Fracturing Research Study. EPA/600/F- gineers, January 2011. 10/002. June 2010. Available at: Yager RR. Arithmetic and other operations on Dempster- safewater/uic/pdfs/hfresearchstudyfs.pdf, Accessed May Shafer structures. International Journal of Man-Machine 12, 2011. Studies, 1986; 25:357–366. 37. Michie & Associates. Oil and Gas Industry Water Injection18. Frank MJ, Nelsen RB, Schweizer B. Best-possible bounds for Well Corrosion, report prepared for the American Petroleum the distribution of a sum—A problem of Kolmogorov. Prob- Institute, February 1988. Washington, DC: API, 1988. ability Theory and Related Fields, 1987; 74:199–211. 38. Michie TW, Koch CA. Evaluation of injection-well risk man-19. Williamson RC, Downs T. Probabilistic arithmetic I: Numer- agement in the Williston Basin. Journal of Petroleum Tech- ical methods for calculating convolutions and dependency nology, 1991; 43(6):737–741 bounds. International Journal of Approximate Reasoning, 39. Browning LA, Smith JB. Analysis of the rate of and rea- 1990; 4:89–158. sons for injection well mechanical integrity test failure, pre-20. Ferson S, Ginzburg LR. Different methods are needed to sented at SPE/EPA Exploration and Production Environ- propagate ignorance and variability. Reliability Engineering mental Conference, March 7–10, 1993, San Antonio, TX. and System Safety. 1996; 54:33–144. 40. Underground Injection Practices Council (UIPC). A Class I21. Ferson S. RAMAS Risk Calc 4.0 Software: Risk Assessment Injection Well Survey (Phase II Report): Survey of Opera- with Uncertain Numbers. Boca Raton, FL: Lewis Publishers, tions, December 1987. Oklahoma, OK: UIPC, 1987. 2002. 41. Lustgarden A. Water problems from drilling are more22. Karanki DR, Kushwaha HS, Verma AK, Ajit S. Uncer- frequent than PA officials said. Available at: http://www. tainty analysis based on probability bounds (p-box) ap- proach in probabilistic safety assessment. Risk Analysis, 2009; more-frequent-than-officials-said-731 2009, Accessed on 29(5):662–675. March 3, 2011.23. Tucker WT, Ferson S. Probability bounds analysis in envi- 42. Urbina I. Regulation lax as gas wells’ tainted water hits rivers. ronmental risk assessments. Applied Biomathematics Report, New York Times, February 26, 2011. 2003, 63. Available at:, 43. Urbina I. Politics seen to limit E.P.A. as it sets rules for natu- Accessed December 18, 2010. ral gas. New York Times, March 3, 2011.
  12. 12. 12 Rozell and Reaven44. Edwards KL, Weissert S, Jackson J, Marcotte D. 58. Volz CD. Written Testimony Before the Senate Committee Marcellus Shale hydraulic fracturing and optimal well on Environment and Public Works and its Subcommit- spacing to maximize recovery and control, in SPE Hydraulic tee on Water and Wildlife, Joint Hearing “Natural Gas Fracturing Technology Conference Proceedings. Paper Drilling, Public Health and Environmental Impacts.” April 140463. Woodlands, TX Society of Petroleum Engineers, 12, 2011. Available at: 2011 cfm?FuseAction=Files.View&FileStore id=57d1bfd4–9237-45. Shakanov T, Gumarov S, Ronderos J, Fragachan FE. 488e-999f-4e1e71f72e52, Accessed on July 27, 2011. Multiple-fractures propagation, orientation and complexities 59. Reaven S. The methodology of probabilistic risk assessment: in drilling waste injection, in SPE Hydraulic Fracturing Tech- Completeness, subjective probabilities, and the “Lewis Re- nology Conference Proceedings. Paper 139504. Woodlands, port.” Explorations in Knowledge, 1988; 5(1):11–32. TX. Society of Petroleum Engineers, 2011 60. Reaven S. Choosing among risk management alternatives for46. O’Brien DG, Larson, Jr. RT, Parham RC, Thingelstad BL, mitigating groundwater pollution. Pp. 215–230 in McTernan Aud WW, Burns RA, Weijers L. Using real-time downhole W, Kaplan E (eds). Risk Assessment for Groundwater Pol- microseismic to evaluate fracture geometry for horizontal lution Control. New York: American Society of Civil Engi- packer-sleeve completions in the Bakken Formation, Elm neers, 1990. Coulee Field, Montana, in SPE Hydraulic Fracturing Tech- 61. Reaven S. The reliability of individual and pooled expert nology Conference Proceedings. Paper 139774. Woodlands, judgments. Pp. 189–206 in Covello V (ed). Uncertainty in TX. Society of Petroleum Engineers, 2011. Risk Assessment and Decision Making. London: Plenum47. Walker K, Sink J, Le Calvez J, Frantz J, Martin J. Using mi- Press, 1987., croseismic monitoring to better understand completion of the 62. Ferson S, Hajagos JG. Arithmetic with uncertain numbers: Medina–Whirlpool formations: A NYSERDA–Sponsored Rigorous and (often) best possible answers. Reliability Engi- Project, in SPE Eastern Regional Meeting Proceedings. Pa- neering and System Safety, 2004; 85:135–152. per 125979. Charleston, WV: Society of Petroleum Engineers, 63. Pennsylvania Dept. of Environmental Protection. DEP Calls 2009. on Natural Gas Drillers to Stop Giving Treatment Facili-48. Gu H, Weng X, Lund J, Mack M, Ganguly U, Suarez- ties Wastewater: Efforts in Motion to Quickly Stop Drilling Rivera R. Hydraulic fracture crossing natural fracture at non- Wastewater from Going to Treatment Works “Grandfa- orthogonal angles, a criterion, its validation an applications, in thered” by the Previous Administration. News Release April SPE Hydraulic Fracturing Technology Conference Proceed- 19, 2011. ings. Paper 139984. Woodlands, TX. Society of Petroleum En- 64. Hamilton R, Williams S, Martin J, Stalnaker C, Davis T. Re- gineers, 2011. liable method of achieving annular isolation for a multi-ste49. Weng X, Kresse O, Cohen C, Wu R, Gu H. Modeling of hy- fracture treatment in an air drilled hole, in SPE Eastern Re- draulic fracture network propagation in a naturally fractured gional Meeting Proceedings. Paper 125866. Charleston, WV: formation, in SPE Hydraulic Fracturing Technology Confer- Society of Petroleum Engineers, 2009 ence Proceedings. Paper 140253. Woodlands, TX. Society of 65. Ribeiro LH, Sharma MM. Multi-phase fluid-loss properties Petroleum Engineers, 2011. and return permeability of energized fracturing fluids, in SPE50. Taleghani AD, Lorenzo JM. An alternative interpretation of Hydraulic Fracturing Technology Conference Proceedings. microseismic events during hydraulic fracturing, in SPE Hy- Paper 139622. The Woodlands, TX: Society of Petroleum En- draulic Fracturing Technology Conference Proceedings. Pa- gineers, 2011. per 140468. 24–26 January 2011. Woodlands, TX. Society of 66. Brannon HD, Kendrick DE, Luckey E, Stipetich A. Multi- Petroleum Engineers. stage fracturing of horizontal wells using ninety-five quality51. Bredehoeft J. The conceptualization model problem— foam provides improved shale gas production, in SPE Eastern Surprise. Hydrogeology Journal, 2005; 13:37–47. Regional Meeting Proceedings. Paper 124767. Charleston,52. Myers T. Technical Memorandum: Review and Analysis of WV: Society of Petroleum Engineers, 2009. DSGEIS for NRDC, 2009. Available at: 67. Yost II AB, Mazza RL, Gehr JB. CO2/Sand fracturing in de- energy/files/ene 10092901d.pdf, Accessed March 23, vonian shales, in Society of Petroleum Engineers 1993 East- 2010. ern Regional Conference, Pittsburgh, 1993, 14p.53. PA DEP Oil & Gas Reporting Website. Available at: 68. Tudor EH, Nevison GW, Allen S, Pike B. 100% gelled LPG fracturing process: An alternative to conventional water- Modules/Welcome/Welcome.aspx, Accessed on 12–8-2010. based fracturing techniques, in SPE Eastern Regional Meet-54. Arthur JD, Bohm B, Coughlin BJ, Layne M. Evaluating the ing Proceedings. Paper 124495. Charleston, WV: Society of environmental implications of hydraulic fracturing in shale Petroleum Engineers, 2009. gas reservoirs, in International Petroleum & Biofuels Envi- 69. Britt LK, Schoeffler J. The geomechanics of a shale play: ronmental Conference. Albuquerque, NM. November 11–13, What makesa shale prospective, in SPE Eastern Regional 2008. Meeting Proceedings. Paper 125525. Charleston, WV: Society55. Arthur JD, Dutnell SL, Cornue DB. Siting and permitting of Petroleum Engineers, 2009 of class II brine disposal wells associated with development 70. Nuttall BC, Eble CF, Drahovzal JA, Bustin RM. Anal- of the Marcellus Shale, in SPE Eastern Regional Meet- ysis of Devonian black shales in Kentucky for potential ing Proceedings. Paper 125286. Charleston, WV: Society of carbon dioxide sequestration and enhanced natural gas Petroleum Engineers, 2009 production. Lexington Kentucky Geological Survey, 2005.56. Blauch ME, Myers RR, Moore TR, Lipinski BA, Available at: Houston NA. Marcellus Shale post-frac flowback waters— 2L7ng5/native/842849.pdf, Accessed April 19, 2011. Where is all the salt coming from and what are the implica- 71. Pruess K. Enhanced Geothermal Systems using CO2 as a tions? In SPE Eastern Regional Meeting Proceedings. Paper working fluid—A novel approach for generating renewable 125740. Charleston, WV: Society of Petroleum Engineers, energy with simultaneous sequestration of carbon. Berkeley: 2009. Lawerence Berkeley National Laboratory, 2008.57. Pennsylvania DEP investigates elevated TDS in 72. Milici RC, Swezey CS. Assessment of Appalachian Basin Monongahela River. Water and Wastes Digest. October 27, oil and gas resources: Devonian shale—Middle and Up- 2008. Available at: per Paleozoic total petroleum system: U.S. Geological DEP-Investigates-Elevated-TDS-in-Monongahela-River- Survey Open-File Report 2006–1237, 2006. Available at: newsPiece16950, Accessed on May 9, 2010., Accessed May 9, 2010.