lable at ScienceDirect

Atmospheric Environment 155 (2017) 11e20

Contents lists avai

Atmospheric Environment

journal homepage: www.elsevier .com/locate/atmosenv

Impact of natural gas development in the Marcellus and Utica shaleson regional ozone and fine particulate matter levels

Yusuf H. Roohani a, Anirban A. Roy b, Jinhyok Heo c, Allen L. Robinson a, Peter J. Adams a, *

a Center for Atmospheric Particle Studies, Carnegie Mellon University, Pittsburgh, PA 15217, USAb Department of Earth and Atmospheric Sciences, University of Houston, Houston, TX 77004, USAc Department of Civil and Environmental Engineering, Cornell University, Ithaca, NY 14853, USA

h i g h l i g h t s

� Effects of Marcellus gas development on regional air quality were investigated.� The Medium Emissions scenario increased ozone design values by as much as 2.5 ppbv.� The same scenario increased PM2.5 concentrations by 0.27 mg/m3 in some locations.� Premature mortality for this scenario is predicted to be 200 to 460 annually.� Ozone and PM2.5 impacts were driven primarily by NOx emissions.

a r t i c l e i n f o

Article history:Received 4 April 2016Received in revised form8 December 2016Accepted 2 January 2017Available online 3 January 2017

Keywords:PM2.5

OzoneShale gasEnergyAir qualityChemical transport modelingHealth effects

* Corresponding author.E-mail address: petera@andrew.cmu.edu (P.J. Adam

http://dx.doi.org/10.1016/j.atmosenv.2017.01.0011352-2310/© 2017 Published by Elsevier Ltd.

a b s t r a c t

The Marcellus and Utica shale formations have recently been the focus of intense natural gas develop-ment and production, increasing regional air pollutant emissions. Here we examine the effects of theseemissions on regional ozone and fine particulate matter (PM2.5) levels using the chemical transportmodel, CAMx, and estimate the public health costs with BenMAP. Simulations were performed for threeemissions scenarios for the year 2020 that span a range potential development storylines. In areas withthe most gas development, the ‘Medium Emissions’ scenario, which corresponds to an intermediate levelof development and widespread adoption of new equipment with lower emissions, is predicted to in-crease 8-hourly ozone design values by up to 2.5 ppbv and average annual PM2.5 concentrations by asmuch as 0.27 mg/m3. These impacts could range from as much as a factor of two higher to a factor of threelower depending on the level of development and the adoption of emission controls. Smaller impacts(e.g. 0.1e0.5 ppbv of ozone, depending on the emissions scenario) are predicted for non-attainment areaslocated downwind of the Marcellus region such as New York City, Philadelphia and Washington, DC.Premature deaths for the ‘Medium Emissions’ scenario are predicted to increase by 200e460 annually.The health impacts as well as the changes in ozone and PM2.5 were all driven primarily by NOx emissions.

© 2017 Published by Elsevier Ltd.

1. Introduction

The Marcellus and Utica shales are rock formations located inthe Appalachian Basin; they contain large natural gas reserves andrecently have been the focus of intense drilling and leasing activity(Annual Energy Outlook, 2015; Drilling Productivity Report, 2015).Natural gas development, production, and processing activities canbe a significant source of air pollution (Roy et al., 2014). For

s).

example, equipment involved in setting up a well such as trucks,drill rigs, and hydraulic fracturing pumps are powered by heavyduty diesel engines that emit oxides of nitrogen (NOx), volatileorganic compounds (VOCs) and fine particulate matter (PM2.5)(Field et al., 2014; Vinciguerra et al., 2015; Moore et al., 2014; Grantet al., 2009). Natural gas fired compressors emit NOx and VOCs(USEPA, 2000). Venting and fugitive emissions can also be a sig-nificant source of VOCs (Vinciguerra et al., 2015; Kemball-Cooket al., 2010).

In the atmosphere, VOCs and NOx react in the presence ofsunlight to produce ozone, which causes several health problemssuch as asthma and decreased lung function (Levy et al., 2001). NOx

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e2012

emissions also lead to the formation of PM2.5 nitrate that, alongwith primary emissions of PM2.5, leads to an increase in fine par-ticulate concentration in the atmosphere that can ultimately resultin decreased life expectancy (Pope et al., 2009).

Field and modeling studies have also shown that the aggregateemissions from natural gas activities, including shale gas, can haveimportant impacts on local and regional air quality. Schnell et al.(2009) measured peak 1-hr ozone levels as high as 100 ppbv inWyoming. Elevated VOC levels were also found in and around gas-producing regions of Colorado, Utah, Pennsylvania and West Vir-ginia (Vinciguerra et al., 2015; Southwestern Pennsylvania, 2010;Helmig et al., 2014; Brantley et al., 2015). Kemball-Cook et al.(2010) used a chemical transport model to predict that gas devel-opment in the Haynesville Shale could increase the maximum daily8-hr average (MDA8) ozone levels by as much as 17 ppbv over partsof Louisiana and Texas. Many counties in and around the Marcellusregion currently violate the National Ambient Air Quality Standards(NAAQS) for ozone and PM2.5 (The Green Book), and gas develop-ment in the Appalachian basin may complicate these existingproblems (Drilling Productivity Report, 2015; Adgate et al., 2014).Litovitz et al. (2013) estimated that the air-quality-related damagesin 2011 from Marcellus-related activity in Pennsylvania, includingemissions of NOx, SOx, VOCs, and PM, ranged from $7.2 million to$32 million.

This work investigates the impacts of Marcellus and Utica(hereafter referred to as Marcellus-Utica or MU) development onregional ozone and PM2.5 levels by implementing an updatedversion of the Marcellus emissions inventory developed previouslyin a 3-D chemical transport model (Roy et al., 2014). Specifically, weaccount for impacts resulting from emissions of NOx, VOCs, andprimary PM2.5 from natural gas production, including condensateandwet gas, but not from oil production, which has been small thusfar in the Utica formation. Different emission scenarios are evalu-ated corresponding to different levels of development and/or airpollution control. The health impacts of the predicted changes in airquality, namely additional mortality stemming from increases inPM2.5 and ozone, are evaluated using the BenMAP model. Non-mortality health end points were not considered. Emissions of airtoxics and their potential health effects were not considered due toinsufficient data.

2. Methods

A series of simulations were performed with different emissioninventories to assess the potential impacts of Marcellus and Utica(MU) shale gas activities on regional air quality (Table 1). All sim-ulations were performed using 2005 meteorology for a full year tocapture seasonally varying sensitivities to MU emissions. The MUemissions scenarios are based on projected MU activities for 2020.The year 2020was chosen because it is far enough in the future thata significant amount of development activity is expected to have

Table 1List of simulations and emissions.

Simulation name Non-Marcellus-Utica emissio

2005 2005 emissions2020 No-MU (No Marcellus-Utica) 2005 emissions scaled to 202High Emissions 2005 emissions scaled to 202Medium Emissions 2005 emissions scaled to 202Low Emissions 2005 emissions scaled to 202Marcellus-Utica NOx only 2005 emissions scaled to 202Marcellus-Utica VOC only 2005 emissions scaled to 202

a Elemental carbon.

occurred by then and also due to the availability of scaling factorsfor non-MU emissions for that year. While existing regulations onnon-MU emissions dramatically lower predicted ozone and PM2.5concentrations in 2020 compared to 2005, this study focuses onregional air pollution caused by MU activity.

2.1. Chemical transport modeling and non-Marcellus emissions

A 3-D chemical transport model, CAMx (version 5.41), was usedto simulate ozone and PM2.5 concentrations over the continentalUnited States. The modeling domain was discretized into a 36 kmby 36 km grid with 14 vertical layers up to a height of 16 km. Detailsof the model are described in the CAMx User's guide (CAMx, 2011).Briefly, CAMx simulates advection, dispersion, gas- and particle-phase chemistry, emission, wet and dry deposition, and aerosolmicrophysical processes. Chemistry was simulated using theCarbon Bond Chemical Mechanism, 2005 (CB05) (Carbon BondChemical Mechanism, 2005). The simulations used meteorologyfields generated by the Weather Research and Forecasting (WRF)model for the year 2005 (Skamarock et al., 2008). The chemicalboundary conditions were taken from the output of a globalchemical transport model, GEOS-Chem (GEOS-Chem, 2015).

Emissions from all non-Marcellus and Utica (non-MU) sourcesare from an inventory developed by Ramboll Environ Inc. for aregulatory impact assessment conducted by the US EPA (RegulatoryImpact Analysis, 2010). The Ramboll Environ inventory is for 2005and takes the US National Emissions Inventory (NEI) version 2 as itsstarting point. CAMx has previously been evaluated for both ozoneand PM2.5 using this inventory at a finer grid resolution(12 km � 12 km) than used here (Air Quality Modeling, 2011). Wesupplement those results with evaluations using our36 km� 36 km grid; details are in the online supplemental material(Section S.1). In summary, the model meets the performancestandards for modeling ozone recommended by the EPA andcompares favorably with other models (UAM, 1991; Solazzo et al.,2012). For PM2.5, model performance was within the recom-mended thresholds of Boylan et al (Boylan and Russell, 2006). Biasand error metrics for PM2.5 and ozone at our 36 km resolutionwerevery similar to those for the finer 12 km grid.

We investigated the effects of MU emissions in the year 2020.Therefore, we re-scaled the 2005 emissions inventory for all non-MU sources to 2020 using EPA NEI data and projections for themid-Atlantic/northeastern United States compiled by the Mid-Atlantic Regional Air Management Association (MARAMA)(MARAMA, 2012). The scaling factors account for effects of controlstrategies promulgated by the US EPA, such as the phasing in of Tier4 for the fleet of diesel engines and the Transport Rule. They aresector-specific and uniformly applied across space and time.Further details are discussed in Section S.3.

The non-MU emission scaling factors are listed in Table S.1. Be-tween 2005 and 2020, overall non-MU area source (stationary and

ns Marcellus-Utica (tons day�1)

NOx VOC ECa

0 0 00 0 0 00 212 282 4.20 112 76 2.50 38 32 1.50 112 0 2.50 0 76 2.5

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e20 13

mobile) NOx emissions are reduced by 56%, SO2 emissions by 72%,and VOC emissions by 28%. Over the same period, point sourceemissions, NOx emissions are reduced by 37% and SO2 by 82%.Primary PM2.5 emissions are reduced by 20%. No scaling factorswere applied to the biogenic emissions.

2.2. Marcellus-Utica (MU) emissions and simulation scenarios

Table 1 summarizes the simulations performed to evaluate theimpact of MU activities on regional air quality. This section de-scribes how those emissions estimates were generated. The first ofthe future simulations was a 2020 ‘No-Marcellus-Utica’ (no-MU)simulation to calculate a baseline against which to compare thesimulations with MU emissions. MU emissions are then added ontothe No-MU 2020 inventory and the impacts of theMU emissions onozone and PM2.5 concentrations were determined by differencebetween CAMx simulations with and without MU emissions.

The emission inventories for MU activities are based on per-unit-activity emission factors developed by Roy et al. (2014) forwell development, gas production, and midstream processing inthe Marcellus region. The interested reader is referred to that paperfor additional details regarding relative contributions of specificcontributing activities. Briefly, Roy et al. (2014) uses a bottom-upapproach to estimate emissions of NOx, VOCs, and primary PM2.5for the Marcellus region from well development, gas production,and midstream activities. Per-unit-activity emission factors includecontributions from several sub-categories of equipment and ac-tivities including compressor stations, wellhead compressors,trucking, fracing, drilling, completion, pneumatic devices,condensate tanks, etc. Roy et al. (2014) uses EPA AP-42 data forsome activities but goes beyond this by including measured emis-sion factors from in-use equipment available from published liter-ature, deriving results specific to the Marcellus region, including anexplicit uncertainty analysis, accounting for the effects of futureregulations and fleet turnover on future emissions, and derivingresults per unit gas production or well development. They foundthat drilling, hydraulic fracturing, compressor stations, andcompletion venting were the major sources of air pollutants fromMarcellus activities.

We combined the Roy et al. (2014) process-level emission fac-tors with activity data to develop three different emissions sce-narios for MU activities: Low, Medium, and High Emissions(Table 1). Storylines for each scenario are described below todemonstrate their plausibility and to provide a context for theirinterpretation. These scenarios designed to represent a range ofpotential gas development activity and emissions controls for 2020.In each case, the same emissions could be obtained with othercombinations of activity levels and emissions factors. As shownlater, predicted ozone and PM2.5 concentrations vary almost line-arly with changing NOx emissions between these three scenarios.Therefore, estimates for additional emissions scenarios can be ob-tained by simply interpolating among the three defined here.

The Medium Emissions scenario corresponds to the median2020 process-level emission factors from Roy et al. (2014) and EIAprojections of natural gas production in 2020. The Roy et al. (2014)median emission factors account for implementation of the EPA'sOil and Gas Rule (Oil and natural Gas), the Tier 4 (Clean Air NonroadDiesel Rule, 2004) standard for off-road diesel engines, and the2007/2010 standard for on-road heavy-duty diesel vehiclesassuming high fleet turnover – 75% for drill rigs, frac pumps, andtrucks and 98% for compressor stations and wellhead compressors.Therefore, the majority of the sources in the Medium Emissionsscenario comply with the latest emission standards. The MediumEmissions scenario assumes 17.4 billion cubic foot per day (bcf) ofgas production in the Marcellus shale and 4 bcf in the Utica shale in

2020, which are based on the EIA predicted (Annual EnergyOutlook, 2015) national gas production in 2020 and the fractionof current production in Marcellus and Utica shales (22% and 5% oftotal dry gas production in the United States) (Drilling ProductivityReport, 2015). The Medium Emissions scenario assumes that thenumber of wells drilled in 2020 in each state is the same as theaverage number drilled in 2012e2014. Over this period welldevelopment was relatively steady with an average of 1400 wellsdrilled per year in Pennsylvania, 225 in West Virginia, and 500 inOhio (Pennsylvania Department; West Virginia; Ohio Department).

The High and Low Emissions scenarios were designed to span arange of potential combinations of future well development, gasproduction, and emission controls. Both of these scenarios corre-spond to plausible development scenarios. For example, the HighEmissions scenario would occur if there were a combination ofhigh-level of MU activity (3000 wells drilled per year and 23.8 bcfgas produced versus 2125 wells and 21.4 bcf in the Medium Sce-nario) and slower fleet turnover. This level of well developmentcorresponds to the peak drilling observed in each state since theend of the US economic recession in 2009; this level of gas pro-duction corresponds to the “High Oil and Gas Resource” scenariodefined by the EIA (Annual Energy Outlook, 2015). The HighEmissions scenario has slower fleet turnover than the MediumEmissions scenario – 20% turnover for drill rigs, frac pumps, andtrucks and 30% for compressors from 2009-compliant equipment.Therefore the High-Emissions scenario has somewhat higherprocess-level emission factors and a higher level of MU activitythan the Medium Emission scenario.

The Low Emissions scenario represents a lower level of activity(1350 wells drilled and 20.5 bcf gas produced) with a fleet turnoverrate of 100% for all 2009-compliant equipment. Additionally, itassumes that 75% of the compressors and 40% of the drill rigs, fracpumps and trucks have adopted state-of-the-art control technol-ogy, which exceeds what is required by existing regulations. Thisnumber of wells drilled corresponds to the lowest period of welldevelopment in each state since 2009 (Pennsylvania Department;West Virginia; Ohio Department). This level of production corre-sponds to the “Low Oil Price” scenario defined by the EIA (DrillingProductivity Report, 2015). Therefore the Low-Emissions scenariowould if occur if there was both less MU development and muchstricter emission controls than the Medium Emission scenario.

Two additional simulations were performed to investigate thesensitivity of the model predictions to MU VOC and NOx emissions.This was done by removing the VOC emissions entirely and onlyapplying the Medium Emissions NOx emissions or vice-versa.

Lacking detailed projections for the spatial distribution of futuredevelopment, the MU emissions were spatially allocated assumingthat future development would follow a similar spatial pattern tothe development to date. First, the MU region was defined withinthe states of Pennsylvania, Ohio, and West Virginia. The projected2020 MU emissions for each state were then distributed withineach state based on the number of presently active wells withineach grid cell with the results shown in Figure S1 (PennsylvaniaDepartment; West Virginia; Ohio Department). Areas of highestemissions are evident in Figure S1 in eastern Ohio, southwesternPennsylvania, and northeastern Pennsylvania, corresponding toareas that have seen significant development already. With theseassumptions, 65% of the 2020 MU NOx emissions occur in Penn-sylvania and 12% and 23% occur in West Virginia and Ohio,respectively. We also performed a sensitivity simulation in whichMU emissions had a uniform spatial distribution throughout theMarcellus fairway. In that simulation (not shown), PM2.5 and ozoneimpacts were broadly similar, but themaximum impacts were ~25%lower in high activity areas of southwestern Pennsylvania, adjacentareas of West Virginia and Ohio, and northeastern Pennsylvania. In

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e2014

contrast, impacts in central Pennsylvania were similarly higher.Emissions are assumed to occur uniformly throughout the year.Speciation profiles for VOC emissions from condensate tanks are anaverage of profiles taken from Hendler et al. (2009). Otherwise,speciation profiles for VOC emissions, which are primarily frompneumatics, compressor stations, completion venting, gas plantand transmission, are based on EPA's SPECIATE database.

2.3. Air quality metrics

Ozone impacts were quantified using the Maximum Daily 8-hraverage (MDA8) mixing ratio of ground-level ozone, which iscalculated by taking a running 8-hr average of CAMx ozone outputand then picking the maximum amongst these values for eachsimulation day. The current National Ambient Air Quality Standard(NAAQS) for ozone was recently revised to an MDA8 of 70 ppb(National Ambient) after a review by the Clean Air ScientificAdvisory Committee (CASAC) that recommended that the ozonestandards be tightened to somewhere between 60 and 70 ppb(Review of the EPA, 2014).

Another metric considered was the ozone design value.Observed design values correspond to the 3-year average of the 4th-highestmeasuredMDA8 value for a givenmonitor. Base case (2005)design values are published online by the US EPA (Air Trends) andwere used as a starting point for estimating future design valuesusing the relative response factor method prescribed by the US EPA(Guidance on the Use of Models (2007). The EPA protocol is espe-cially useful since 3-year simulations would be computationallyintensive. For each of the 2020 scenarios (No-MU, Low, Medium,and High), estimated future design values are calculated as theproduct of the base case (2005) observed design value and a rela-tive response factor computed by the model. For each future sce-nario and monitor site, the relative response factor is the modeledratio of average MDA8 ozone levels in that scenario to the same forthe 2005 base case. Those modeled MDA8 levels are time averagesover a set of ‘high ozone’ days defined for each monitor. ‘Highozone’ days were selected for all monitored locations in New York,Pennsylvania, Ohio, West Virginia, New Jersey, Maryland and Dis-trict of Columbia using appropriate thresholds as described in theUS EPA guidance document. This was possible since every sitesatisfied theminimum threshold requirement in 2005, i.e. having atleast five days with MDA8 values greater than 70 ppb. A recentupdated draft of the same guidance report (that is still under re-view) recommended selecting the ten highest days at each siteregardless of a threshold (Modeling Guidance, 2014). Results fromthe proposed method (not shown here) predicted slightly largerchanges for counties with a higher base design value, furtheraggravating the impacts especially in non-attainment counties.

Impacts on PM2.5 concentrations were quantified using meanannual concentrations of total PM2.5 mass. The current annualprimary and secondary NAAQS for PM2.5 are 12 mg/m3 and 15 mg/m3

respectively (National Ambient ). Species-wise contributions werealso studied to assist in the development of regulatory solutions.

2.4. Estimating public health effects

The premature mortality associated with PM2.5 and ozoneformed from MU emissions of NOx, primary PM2.5, and VOCs wasestimated using BenMAP-CE software (Final Ozone, 2008; RTI,2015). Emissions of air toxics and their effects were not consid-ered. The method involves two steps: the estimation of prematuredeaths based on concentration-response relations from epidemio-logical studies and their valuation using the Value of a StatisticalLife (VSL). Other health endpoints were not analyzed becausemortality costs generally dominate other costs (Final Ozone, 2008).

Monte-Carlo simulations are done to explore the uncertaintiessurrounding the concentration-response relations and VSL.

The changes in premature deaths were estimated for changes inPM2.5 (annual average) and O3 (MDA8) concentrations betweenscenarios (No-MU, High, Medium, and Low Emissions). For PM2.5,concentration-response relations from Krewski et al. (2009) andLepeule et al. (2012) were used, which are the latest follow-upstudies from two landmark cohort studies on PM2.5 health effects.For ozone, two studies were used: Bell et al. (2004), an analysis ofthe National Morbidity, Mortality, and Air Pollution Study, and Levyet al. (2005), a meta-analysis. Other ozone concentration-responsefunctions fall between the two used here (Bell et al., 2005; Ito et al.,2005; Schwartz, 2005; Huang et al., 2005). BenMAP populationprojections for 2020 were used. Baseline mortality rates for 2020were based on a series of Census Bureau mortality projections (RTI,2015). The valuation of the estimated changes in mortality relied ona Weibull distribution of VSL based on 26 VSL studies, which isrecommended by U.S. EPA (Economic Analyses, 2010). Adjusted forincome level in 2020 using U.S. EPA's official adjustment factorsprovided in BenMAP, the distribution has the mean value of $9.6million (in 2010 USD).

3. Results

3.1. Ozone impacts

3.1.1. Spatial impacts on MDA8 ozoneFig. 1 shows maps of predicted MDA8 ozone mixing ratios

averaged across the entire ozone season (MarcheOctober).Fig. 1(b)e1(d) illustrate the impacts of various MU emission sce-narios on ozone levels in 2020, each map showing the difference inozone-season (MarcheOctober) average MDA8 between each MUscenario and the 2020 No-MU simulations. Impacts of at least0.2 ppbv are predicted in all three scenarios in most of Pennsyl-vania, southern New York, eastern Ohio and northern West Vir-ginia. The largest increases in ozone levels are predicted to occur innortheastern and southwestern Pennsylvania and eastern Ohiowhere most of the MU-related activity is taking place, but someimpacts occur downwind in major East Coast cities. Within themodeling domain, the greater sensitivity of rural areas to emissionsfrom MU activity is likely due to their NOx-limited nature.

As seen in Fig. 1(c), the Medium Emissions scenario increasesthe ozone season average MDA8 ozone by 0.5e0.8 ppbv acrossmuch of the Marcellus Fairway. Fig. 1(b) shows that the maximumozone impacts for the Low Emissions scenario over the same areaare ~0.3 ppbv, thus highlighting the potential air quality benefits oflower levels of MU development in the region and tighter emissioncontrols. Similarly, the ozone impacts for the High Emissions sce-nario are nearly twice as severe as those in the Medium Emissions.

The ozone impacts shown in Fig. 1 are considerably less severethan those predicted by Kemball-Cook et al. (2010) for the Hay-nesville region. This difference is likely due to a number of factors.First, there are several notable differences in how they developedtheir emissions inventory versus the one used here. For example,they did not assume any controls on midstream sources likecompressor stations, which contribute a significant fraction of NOx

emissions. Additionally, Kemball-Cook did not account for the ef-fects of the EPA's Oil and Gas Rule (Oil and natural Gas) since it wasnot implemented at that time. The Kemball-Cook low and highemissions scenarios for 2012 have 61 and 140 tons day�1 of NOxfrom gas development activities, which is similar to the 129 tonsday�1 emitted in our Medium Emissions (2020) scenario. However,the Kemball-Cook emissions are more concentrated in the Hay-nesville region, which is more than 10 times smaller than the MU.Moreover, we expect that the impact of NOx emissions depends

Fig. 1. Mean MDA8 ozone mixing ratios (ppbv) during the ozone monitoring season (MarcheOctober) in the MU shale region for: (a) the No-MU case in 2020. Panels (b) - (d) showthe differences between the No-MU base case simulation and the three MU emission scenarios: (b) Low Emissions (c) Medium Emissions (d) High Emissions.

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e20 15

strongly on the pre-existing VOC/NOx ratios resulting from thebaseline (without gas development) emissions.

3.1.2. Impacts on ozone design valuesTo understand the predicted ozone impacts from a regulatory

perspective, future ozone design values were estimated for selectedurban and rural sites (Fig. 2) using the methods described above.The projections were done for the 2020 No-MU case and for thethree MU emission scenarios. By definition, changes in designvalues are larger than ozone-season MDA8 changes because theyrepresent effects during high ozone episodes when ozone levels aremore sensitive to changes in emissions. Overall, the results indicatethat MU development could result in pronounced ozone changes inregions with significant gas development. Long-range transportfrom the MU region could still affect downwind nonattainmentregions including New York, Philadelphia and Washington, DC.

The greatest impacts occur in counties located in and aroundregions with high MU activity, such as northeastern and south-western Pennsylvania and eastern Ohio. For the Medium Emissionsscenario, design values for counties in the Pittsburgh, Cleveland,and Charleston areas are predicted to increase by 1e2 ppbv due toMU activities in 2020 (Fig. 2(a)). For the High Emissions scenario,this increases to 2e4 ppbv in the most-affected regions. In Alle-gheny County, which is located in southwestern Pennsylvania, theimpact of the Medium Emissions case on the ozone design value is1.7 ppbv while the impact of the High and Low Emissions scenariosare approximately a factor two higher and lower. Smaller impacts,0.1e0.5 ppbv, depending on the emissions scenario considered, arepredicted for non-attainment areas such as New York City, Phila-delphia and Washington, DC that are located downwind of the MUregion.

Although the rural regions experiencing the most MU devel-opment presently meet existing ozone standards, the effect of MU-related activities on ozone levels is predicted to be noticeably largerin these regions than in urban areas. For example, rural counties insouthwestern Pennsylvania and eastern Ohio such as GreeneCounty in PA and Washington County in OH show design value.

Impacts between 1.9 and 2.3 ppbv for the Medium Emissionsscenario, while adjacent urban counties such as Allegheny Countyin PA and Washington county in PA have lower impacts of around1.5 ppbv. Marcellus development would contribute to non-attainment status in several of these rural counties if a futureregulatory review lowered the ozone standard to 60 ppbv, a valuethat was considered in the most recent review. Figure S2 andTable S6 in the Supplemental Information show the predicted MUimpacts on ozone design values for all monitored counties in andaround the MU region.

Cumulative distribution functions (CDFs) of the MU ozone im-pacts in select locations are shown in Figure S1 to further illustratethe temporal variability of increases in MDA8 values during highozone days for different MU related emissions scenarios. While thelargest impacts are in northeastern and southwestern Pennsylvaniaand eastern Ohio, similar trends are predicted across the entire MUregion. For the Medium Emissions scenario, the largest increasesare in the range of 2e2.5 ppbv. Another clear trend is the sub-stantial reduction in the severity of impacts for the Low Emissionscenario which corresponds to a future with tighter emissionscontrols and less MU development.

3.2. PM2.5 impacts

Fig. 3 shows the predicted annual-average PM2.5 concentrations

Fig. 2. (a) Impacts of Marcellus emissions on future design values for different MU emissions scenarios. Plotted results show the difference between the estimated future designvalue for the Low, Medium, and High emissions scenarios and the No-MU case in 2020. (b) Projected design values (ppbv) for selected monitored counties shown as the height ofthe black bar for the No-MU case in 2020. The yellow bars correspond to the change in design value caused by the Medium Emissions scenario. The solid horizontal line representsthe current attainment standard of 70 ppbv while the dotted line represents the stricter of the CASAC recommended standards. Both panels are vertically divided into sevenmetropolitan areas (numbered from left to right): Pittsburgh, Cleveland/Youngstown, Charleston/Morgantown, Buffalo/Rochester, Philadelphia, Washington, DC and New York. Rednumbers correspond to regions that are in non-attainment of ozone as of 2015. The eighth panel displays some of the most highly impacted counties. (For interpretation of thereferences to colour in this figure legend, the reader is referred to the web version of this article.)

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e2016

in 2020 for the No-MU emission scenario as well as the PM2.5 in-creases due to Low, Medium, and High Emissions scenarios. Emis-sions fromMU activities are predicted to appreciably increase PM2.5levels. For example, the Medium Emissions scenario predicts in-creases of ~0.25 mg/m3 in parts of northwestern Pennsylvania andeastern Ohio. The maximum impacts range from ~0.1 to ~0.4 mg/m3

for the Low Emissions and High Emissions case, respectively,highlighting the substantial effects of the scale of activity andemission controls. Outside of the MU development region, PM2.5increases tend to be less than 0.1 mg/m3. Nitrate, ammonium, andelemental carbon account for all of the predicted increases in totalmean PM2.5 mass concentrations (Fig. 4). Across most of the MUregion and downwind areas, ammonium nitrate accounts for80e90% of the predicted PM2.5 increase. Around Pittsburgh andnorthern West Virginia, elemental carbon accounts for 40e55% ofthe predicted increase in PM2.5, but the impacts of directly emittedelemental carbon are much less elsewhere. The increases inelemental carbon are due to primary emissions from diesel engines.

The PM2.5 nitrate is due to atmospheric oxidation of NOx emissionsfrom diesel and natural gas fired engines and turbines, the latteroccurring at the well pad and within the gathering network. Themodel predicts substantial increases in summertime PM2.5 nitrateconcentrations (see Section S.2).

4. Discussion and conclusions

4.1. Estimated public health costs

Estimated annual premature deaths associated with PM2.5 andozone increases from MU emissions and their economic values arepresented in Tables S4 and S5, respectively. Estimates based onKrewski et al. (2009) and Bell et al. (2004) are presented as lowerbounds and those based on Lepeule et al. (2012) and Levy et al.(2005) as upper bounds. MU emissions are projected to increasepremature deaths by 200e460 annually for the Medium Emissionsscenario, by 350e800 for the High Emissions scenario, and by

Fig. 3. Mean annual PM2.5 concentration (mg/m3) in the MU shale region for the (a) No-MU case in 2020. Panels (b) - (d) show the differences between the No-MU 2020 simulationand the three MU emission scenarios: (b) Low Emissions (c) Medium Emissions (d) High Emissions.

Fig. 4. Speciation of MU-related PM2.5 impacts in the Medium Emissions scenario. Panels (a)e(c) display the fractional contribution of each species to the overall PM2.5 increase for(a) elemental carbon (PEC) (b) nitrate and (c) ammonium.

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e20 17

73e170 for the Low Emissions scenario. The economic value of theannual premature deaths caused by MU emissions are estimated tobe $1600 million to $3700 million for the Medium Emissions sce-nario and proportionally higher and lower in the High and LowEmissions scenarios. Depending on the concentration-responserelation, ozone health effects account for 2%e21% of the total eco-nomic value.

Our estimates, ranging from $590 million to $6900 million, areapproximately two orders of magnitude larger than those ofLitovitz et al. (2013). The differences are due to multiple factors:emissions scenario, air quality modeling, dose-response, and eco-nomic valuation. First, the Litovitz et al. (2013) analysis was donefor 2011, which had lower emissions than the 2020 emissions usedhere (e.g. Litovitz et al. (2013) estimated 17,000e28,000 tons of NOxin 2011 versus 14,000e77,000 tons here). A second factor is the

methodology used to estimate air quality impacts from MU emis-sions. Litovitz et al. (2013) used a reduced-form model, APEEP(Muller and Mendelsohn, 2007, 2012), to value health damages perunit emissions. APEEP relies on a Gaussian dispersion model with asimplified representation of PM2.5 formation from NOx emissionscompared to CAMx, the air quality model used here. Other com-parisons have shown that, all other factors being equal, CAMxpredicts PM2.5 formation from NOx emissions in Pennsylvania thatis ~5 times higher than APEEP (Heo, 2015). In addition, the overallemissions landscape is rapidly changing, which alters the sensi-tivity of ambient concentrations to changes in emissions (Ansariand Pandis, 1998). For example, Holt et al. (2015) found thatsensitivity to NOx emissions increased by as much as a factor of twodue to emissions changes between 2005 and 2012. This is impor-tant as APEEP calculates impacts using 2002 as a baseline year

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e2018

whereas the values presented here account for recent and pro-jected changes in emissions from non-MU sources. Recent updatesfrom APEEP modeling show NOx social costs increasing by ~50% inPennsylvania between 2002 and 2008 (Nick Muller, personalcommunication). A third factor is the concentration-responsefunction. Our lower bound is based on Krewski et al., which is avery similar concentration-response function to the APEEP default(Pope et al., 2002), but our upper bound is more than two timeshigher (Lepeule et al., 2012). Finally, we use VSL to monetize pre-mature mortality here, which is about three times higher than theage-adjusted VSL values from APEEP used in Litovitz et al. (2013).Other differences in methodology include changes in populationdistributions and adjustments to VSL for income growth or infla-tion, but these factors only play a minor role in the differences.

4.2. VOC/NOx sensitivity and control strategies

Fig. 5 illustrates the sensitivity of ozone concentrations to MUVOC and NOx emissions for the Medium Emissions scenario. NOxemissions account for more than 80% of the impact on regionalozone across most of the region while VOC emissions only play alimited role in some urban regions. These patterns are consistentwith the well-known spatial patterns of ozone formation, namelythat rural ozone levels tend to be NOx-limited (Tsimpidi et al.,2008). The 36 km resolution used in this study probably does notfully resolve the VOC-limited nature of urban core regions; there-fore, geographically small but high-population downtown areasmay be more sensitive to VOC emissions than these results predict.

The major sources of MU NOx emissions are drill rigs, trucktraffic, and compressor stations (Roy et al., 2014). Our simulationssuggest that tighter controls on these sources would reduce the MUozone impacts. The Low Emissions scenario (Fig. 1(b)) shows thatimplementing state-of-the-art controls on these sources (equiva-lent to more than a 90% reduction) strongly reduces the impact ofMU activities on ozone levels. Controls on VOC emissions benefitsmaller regions compared to NOx controls, but these are generallyurban areas with the most severe ozone problems and the highestpopulation.

Control of NOx emissions will also reduce the majority of thePM2.5 impacts as more than 80% of the predicted PM2.5 increases inmost of the domain are associated with NOx emissions with thebalance due to direct primary emissions (Section 3.2).

Fig. 5. Assessment of the relative contributions of MU VOC and NOx emissions to the ovcontribution of VOC emissions assessed as MU ozone increases in the Marcellus-Utica VOCNOx emissions using the Marcellus-Utica NOx only scenario.

4.3. Impacts of emission controls

The three emissions scenarios also provide insight into potentialemission controls options. For example, the Low and High Emis-sions scenarios can also be derived using the same activity level asthe Medium scenario level coupled with varying degrees of emis-sions controls. Therefore, the High and Low Emissions scenariosillustrate the benefits of both existing regulations and possiblefurther reductions.

The Low Emissions scenario is probably most interesting. It canbe derived using the Medium Scenario activity coupled with fullimplementation of state-of-the-art control technology on all sour-ces and equipment (e.g. selective catalytic reduction units for NOxcontrol on all engines and turbines). Therefore the differences be-tween the Medium and Low Emissions scenarios illustrate thepredicted ozone and PM2.5 reductions achievable given fullimplementation of stringent control technologies at an interme-diate level of MU activity.

Similarly, the High Emissions scenario can be constructed if oneassumes that no new controls are added to 2009-compliantequipment in 2020 combined with intermediate activity. Thiswould represent a hypothetical future scenario in which there wasvery little equipment turnover or the Oil and Gas Rule, Tier 4standards, and on-road vehicle standards were not implemented inthe oil and gas sector. Thus comparing the differences betweenHigh and Medium Emissions scenario illustrate the benefits ofthese recently promulgated regulations for an intermediate level ofMU activity.

4.4. Potential benefits of reduced end-use emissions

At point of use, natural gas generally has lower emissions thanother fossil fuels, especially coal (De Gouw et al., 2014). However,the analysis presented here accounts only for the emissions asso-ciated with the development and production of MU gas on regionalozone and PM2.5 levels. A broader assessment would also accountfor changes in emissions associated with the end uses of MU gas.For example, regional NOx emissions may decrease if a portion ofthe MU gas is used to repower or replace older coal-fired powerplants in the region because gas-fired power plants emit approxi-mately one tenth of the NOx per unit MWh compared to coal-firedplants (USEPA, 2013). In this scenario, fuel switching from coal togas could bring significant benefits to regional ozone and PM.However, these hypothetical emission reductions must be

erall MU impacts on MDA8 ozone in the Medium Emissions scenario. (a) Fractionalonly scenario normalized by those in the Medium Emissions scenario. (b) Same but for

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e20 19

evaluated in the larger regulatory context. The Cross-State AirPollution Rule (CSAPR) has set strict caps for NOx emissions frompower plants in the eastern US. It is not clear if fuel switching fromcoal to MU gas will reduce regional emissions below the CSAPRcaps. For example, fuel switching from coal toMU gas is likely beingdone by companies in lieu of other control strategies such as theinstallation emission control devices. Therefore, switching to nat-ural gas likely provides the required emissions reductions at lowercost than adding emission controls to coal-fired power plants, but itdoes not necessarily mean that the overall regional emissions fromthe electricity sector will be lower than the CSAPR cap. Alterna-tively, if abundant gas lowers energy prices such that overall energyuse grows, then an increase in NOx emissions beyond that projectedhere may occur (Ansari and Pandis, 1998). These sorts of scenariosare difficult to forecast and therefore beyond the scope of the workpresented here. Nevertheless, our results indicate that the directimpacts of gas development alone will increase ozone and PM2.5levels significantly and therefore merit regulatory attention in theirown right.

Acknowledgments

This work was funded by the Heinz Endowments (Grant Ref#C1705) and the Department of Energy's National Energy Technol-ogy Laboratory (DOE NETL Project Activity Number4000.4.605.920.012.751). The authors also wish to thank RambollEnviron, Inc. for the base emissions inventory and meteorologicalfields used in this work.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.atmosenv.2017.01.001.

References

Adgate, J.L., Goldstein, B.D., McKenzie, L.M., 2014. Potential public health hazards,exposures and health effects from unconventional natural gas development.Environ. Sci. Technol. 48 (15), 8307e8320.

Air Quality Modeling Final Rule Technical Support Document, 2011. US Environ-mental Protection Agency, Office of Air Quality Planning and Standards,Research Triangle Park, NC.

Air Trends e Design Values; http://www.epa.gov/airtrends/values.html.Annual Energy Outlook 2015 with Projections to 2040, 2015. U.S. Energy Informa-

tion Administration; U.S. Department of Energy, Washington, DC. http://www.eia.gov/forecasts/aeo/pdf/0383(2015).pdf.

Ansari, A.S., Pandis, S.N., 1998. Response of inorganic PM to precursor concentra-tions. Environ. Sci. Technol. 32 (18), 2706e2714.

Bell, M.L., McDermott, A., Zeger, S.L., Samet, J.M., Dominici, F., 2004. Ozone andshort-term mortality in 95 US urban communities, 1987-2000. JAMA J. Am.Med. Assoc. 292 (19), 2372e2378.

Bell, M.L., Dominici, F., Samet, J.M., 2005. A meta-analysis of time-series studies ofozone and mortality with comparison to the national morbidity, mortality, andair pollution study. Epidemiology 16 (4), 436e445.

Boylan, J.W., Russell, A.G., 2006. PM and light extinction model performance met-rics, goals, and criteria for three-dimensional air quality models. Atmos. Envi-ron. 40 (26), 4946e4959.

Brantley, H.L., Thoma, E.D., Eisele, A.P., 2015. Assessment of volatile organic com-pound and hazardous air pollutant emissions from oil and natural gas well padsusing mobile remote and on-site direct measurements. J. Air Waste Manag.Assoc. 65 (9), 1072e1082.

CAMx User ’s Guide, Version 5.40, 2011. ENVIRON International Corporation,Novato, CA. http://www.camx.com/files/camxusersguide_v5-40.aspx.

Updates to the Carbon Bond Chemical Mechanism: CB05, 2005. Yocke & Company,Novato, CA. http://www.camx.com/publ/pdfs/cb05_final_report_120805.aspx.

Clean Air Nonroad Diesel Rule, 2004. US Environmental Protection Agency, Office ofAir Quality Planning and Standards, Research Triangle Park, NC. http://www.epa.gov/otaq/documents/nonroad-diesel/420f04032.pdf.

Drilling Productivity Report; U.S., 2015. Energy Information Administration; U.S.Department of Energy, Washington, DC. http://www.eia.gov/petroleum/drilling/pdf/dpr-full.pdf.

Guidelines for Preparing Economic Analyses, 2010. National Center for Environ-mental Economics, Office of Policy, U.S. Environmental Protection Agency.

Field, R.A., Soltis, J., Murphy, S., 2014. Air quality concerns of unconventional oil and

natural gas production. Environ. Sci. Process Impacts. 16, 954e969. http://dx.doi.org/10.1039/c4em00081a.

Final Ozone NAAQS Regulatory Impact Analysis; EPA-452/R-08e003, 2008. U.S.Environmental Protection Agency, Office of Air Quality Planning and Standards,Health and Environmental Impacts Division, Air Benefit and Cost Group (C439-02), Research Triangle Park, NC.

GEOS-chem V10-01 Online User's Guide, 2015. Atmospheric Chemistry ModelingGroup, School of Engineering and Applied Sciences, Harvard University. http://acmg.seas.harvard.edu/geos/doc/man/.

De Gouw, J.A., Parrish, D.D., Frost, G.J., Trainer, M., 2014. Reduced emissions of CO2,NOx, and SO2 from US power plants owing to switch from coal to natural gaswith combined cycle technology. Earth's Future 2 (2), 75e82.

Grant, J., Parker, L., Bar-Ilan, A., Kemball-Cook, S., Yarwood, G., 2009. Draft Report:Development of Emissions Inventories for Natural Gas Exploration and Pro-duction Activity in the Haynesville Shale. Environ International Corporation, CA.http://www.netac.org/UserFiles/File/NETAC/9_29_09/Enclosure_2b.pdf.

Guidance on the Use of Models and Other Analyses for Demonstrating Attainmentof Air Quality Goals for Ozone, PM2.5, and Regional Haze, 2007. US Environ-mental Protection Agency, Office of Air Quality Planning and Standards,Research Triangle Park, NC. http://www.epa.gov/scram001/guidance/guide/final-03-pm-rh-guidance.pdf.

Helmig, D., Thompson, C.R., Evans, J., Boylan, P., Hueber, J., Park, J.H., 2014. Highlyelevated atmospheric levels of volatile organic compounds in the Uintah Basin,Utah. Environ. Sci. Technol. 48 (9), 4707e4715.

Hendler, A., Nunn, J., Lundeen, J., McKaskle, R., 2009. VOC Emissions from Oil andCondensate Storage Tanks. http://files.harc.edu/Projects/AirQuality/Projects/H051C/H051CFinalReport.pdf.

Heo, Jinhyok, 2015. Evaluation of Air Quality Impacts on Society: Methods andApplication (Doctoral dissertation). Carnegie Mellon University.

Holt, J., Selin, N.E., Solomon, S., 2015. Changes in inorganic fine particulate mattersensitivities to precursors due to large-scale US emissions reductions. Environ.Sci. Technol. 49, 4834e4841.

Huang, Y., Dominici, F., Bell, M.L., 2005. Bayesian hierarchical distributed lag modelsfor summer ozone exposure and cardio-respiratory mortality. Environmetrics16 (5), 547e562.

Ito, K., Leon, S.F.D., Lippmann, M., 2005. Associations between ozone and dailymortality: analysis and meta-analysis. Epidemiology 16 (4), 446e457.

Kemball-Cook, S., Bar-Ilan, A., Grant, J., Parker, L., Jung, J., Santamaria, W.,Yarwood, G., 2010. Ozone impacts of natural gas development in the Haynes-ville Shale. Environ. Sci. Technol. 44 (24), 9357e9363.

Krewski, D., Jerrett, M., Burnett, R.T., Ma, R., Hughes, E., Shi, Y., Turner, M.C.,Pope III, C.A., Thurston, G., Calle, E.E., et al., 2009. Extended Follow-up andSpatial Analysis of the American Cancer Society Study Linking Particulate AirPollution and Mortality. Research Report 140. Health Effects Institute, Boston,MA.

Lepeule, J., Laden, F., Dockery, D., Schwartz, J., 2012. Chronic exposure to fine par-ticles and mortality: an extended follow-up of the harvard six cities study from1974 to 2009. Environ. Health Perspect. 120 (7), 965e970.

Levy, J.I., Carrothers, T.J., Tuomisto, J.T., Hammitt, J.K., Evans, J.S., 2001. Assessing thepublic health benefits of reduced ozone concentrations. Environ. Health Per-spect. 109 (12), 1215.

Levy, J.I., Chemerynski, S.M., Sarnat, J.A., 2005. Ozone exposure and mortality: “anempiric bayes metaregression analysis.” Epidemiology 16 (4), 458e468.

Litovitz, A., Curtright, A., Abramzon, S., Burger, N., Samaras, C., 2013. Estimation ofregional air-quality damages from Marcellus Shale natural gas extraction inPennsylvania. Environ. Res. Lett. 8 (1), 014017.

MARAMA Future Year Inventories Version 3, 2012. Mid-Atlantic Regional AirManagement Association. http://www.marama.org/technical-center/emissions-inventory/2007-emissions-and-projections/future-year-inventory-version-3.

Modeling Guidance for Demonstrating Attainment of Air Quality Goals for Ozone,PM2.5, and Regional Haze, 2014. US Environmental Protection Agency, Office ofAir Quality Planning and Standards, Research Triangle Park, NC. http://www.epa.gov/ttn/scram/guidance/guide/Draft_O3-PM-RH_Modeling_Guidance-2014.pdf.

Moore, C.W., Zielinska, B., Petron, G., Jackson, R.B., 2014. Air impacts of increasednatural gas acquisition, processing, and use: a critical review. Environ. Sci.Technol. 48 (15), 8349e8359.

Muller, N.Z., Mendelsohn, R., 2007. Measuring the damages of air pollution in theUnited States. J. Environ. Econ. Manag. 54 (1), 1e14.

Muller, N.Z., Mendelsohn, R., 2012. Efficient pollution regulation: getting the pricesright: corrigendum (mortality rate update). Am. Econ. Rev. 102 (1), 613e616.

National Ambient Air Quality Standards; http://www.epa.gov/ttn/naaqs/standards/ozone/s_o3_history.html.

Ohio Department of Natural Resources e Shale well drilling and permitting; http://oilandgas.ohiodnr.gov/shale.

Oil and Natural Gas Air Pollution Standards; http://www.epa.gov/airquality/oilandgas/.

Pennsylvania Department of Environmental Protection - Oil and gas reportingwebsite; https://www.paoilandgasreporting.state.pa.us/publicreports/Modules/DataExports/DataExports.aspx.

Pope, C.A., Burnett, R.T., Thun, M.J., Calle, E.E., Krewski, D., Ito, K., Thurston, G.D.,2002. Lung cancer, cardiopulmonary mortality, and long-term exposure to fineparticulate air pollution. J. Am. Med. Assoc. 287 (9), 1132e1141.

Pope III, C.A., Ezzati, M., Dockery, D.W., 2009. Fine-particulate air pollution and life

Y.H. Roohani et al. / Atmospheric Environment 155 (2017) 11e2020

expectancy in the United States. N. Engl. J. Med. 360 (4), 376e386.Regulatory Impact Analysis for the Proposed Federal Transport Rule, 2010. US

Environmental Protection Agency, Office of Air Quality Planning and Standards,Research Triangle Park, NC. http://www.epa.gov/ttnecas1/regdata/RIAs/proposaltrria_final.pdf.

Review of the EPA’s Second Draft Policy Assessment for the Review of the OzoneNational Ambient Air Quality Standards; Clean Air Scientific Advisory Com-mittee; 2014; http://yosemite.epa.gov/sab/sabproduct.nsf/5EFA320CCAD326E885257D030071531C/$File/EPA-CASAC-14-004þunsigned.pdf.

Roy, A.A., Adams, P.J., Robinson, A.L., 2014. Air pollutant emissions from thedevelopment, production, and processing of Marcellus Shale natural gas. J. Air &Waste Manag. Assoc. 64 (1), 19e37.

RTI International. Environmental Benefits Mapping and Analysis Program e Com-munity Edition: User's Manual; Data and Tools, 2015. Prepared for Office of AirQuality Planning and Standards, U.S. Environmental Protection Agency.

Schnell, R.C., Oltmans, S.J., Neely, R.R., Endres, M.S., Molenar, J.V., White, A.B., 2009.Rapid photochemical production of ozone at high concentrations in a rural siteduring winter. Nat. Geosci. 2 (2), 120e122.

Schwartz, J., 2005. How sensitive is the association between ozone and daily deathsto control for temperature? Am. J. Respir. Crit. Care Med. 171 (6), 627e631.

Skamarock, W.C., Klemp, J.B., Dudhia, J., Gill, D.O., Barker, D.M., Duda, M.G.,Huang, X.-Y., Wang, W., Powers, J.G., 2008. A Description of the AdvancedResearch WRF Version 3. http://www2.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf (Accessed 24 December 2012).

Solazzo, E., Bianconi, R., Vautard, R., Appel, K.W., Moran, M.D., Hogrefe, C.,Galmarini, S., 2012. Model evaluation and ensemble modelling of surface-level

ozone in Europe and North America in the context of AQMEII. Atmos. Environ.53, 60e74.

Southwestern Pennsylvania Marcellus Shale Short-term Ambient Air SamplingReport, 2010. Pennsylvania Department of Environmental Protection, Bureau ofAir Quality. http://www.dep.state.pa.us/dep/deputate/airwaste/aq/aqm/docs/Marcellus_SW_11-01-10.pdf.

The Green Book Nonattainment Areas for Criteria Pollutants; http://www.epa.gov/oaqps001/greenbk/.

Tsimpidi, A.P., Karydis, V.A., Pandis, S.N., 2008. Response of fine particulate matterto emission changes of oxides of nitrogen and anthropogenic volatile organiccompounds in the Eastern United States. J. Air & Waste Manag. Assoc. 58 (11),1463e1473.

Guidance for Regulatory Application of the Urban Airshed Model (UAM), 1991. USEnvironmental Protection Agency, Office of Air Quality Planning and Standards,Research Triangle Park, NC.

USEPA, 2000. USEPA: Natural Gas-Fired Reciprocating Engines. AP-42 Section 3.2.United States Environmental Protection Agency, Washington, DC. http://www.epa.gov/ttn/chief/ap42/ch03/final/c03s02.pdf.

USEPA, 2013. eGRID2012 Version 1.0 Data Files for Year 2009. http://www.epa.gov/cleanenergy/energy-resources/egrid/index.html (Accessed 26 September 2013).

Vinciguerra, T., Yao, S., Dadzie, J., Chittams, A., Deskins, T., Ehrman, S.,Dickerson, R.R., 2015. Regional air quality impacts of hydraulic fracturing andshale natural gas activity: evidence from ambient VOC observations. Atmos.Environ. 110, 144e150.

West Virginia West Virginia Geological and Economic Survey e Marcellus and othershales; http://www.wvgs.wvnet.edu/www/datastat/devshales.htm.