Modeling the impact of a potential shale gas industry in Germany and the United Kingdom on ozone with WRF-Chem

Introduction Natural gas production in the United States (US) has risen markedly over the past years: in 2018, US dry gas production was 30 trillion cubic feet (861 billion cubic meters; bcm), an increase of 69% over 2005 levels (EIA, 2019a). The majority of the US gas production increase since 2005 was afforded through shale resources (EIA, 2016). In fact, shale gas made up approximately 69% of total US natural gas production in the year 2018 (EIA, 2018). Commercial extraction of shale gas reservoirs has been made possible through relatively recent advancements in horizontal drilling and hydraulic fracturing technologies. While shale gas production has hitherto largely been an American phenomenon (BP, 2016), global shale gas resources1 are vast totaling more than 200 trillion cubic meters (tcm; about one third of world total gas technically recoverable resources), with about 13 tcm in Europe (EIA, 2013). Though there is currently no European commercial shale gas industry, several countries there have considered using domestic supplies to complement their national energy strategy (EC, 2014; JRC, 2012). Shale gas has the potential to reduce dependency on foreign imports and to offset the decline in indigenous production of conventional gas that European countries have been experiencing (BP, 2017). While Poland was initially a promising location for commercial shale gas extraction, exploration efforts there ultimately failed (LaBelle, 2018). On the other hand, Germany and the United Kingdom (UK) have among the top European shale gas-in-place2 volumes (Germany: 7.7 tcm; UK: 37.6 tcm), and have expressed interest in recent years in exploiting their reserves (BGR, 2016; BGS, 2013). Additionally many reports and papers RESEARCH ARTICLE


Introduction
Natural gas production in the United States (US) has risen markedly over the past years: in 2018, US dry gas production was 30 trillion cubic feet (861 billion cubic meters; bcm), an increase of 69% over 2005 levels (EIA, 2019a). The majority of the US gas production increase since 2005 was afforded through shale resources (EIA, 2016). In fact, shale gas made up approximately 69% of total US natural gas production in the year 2018 (EIA, 2018). Commercial extraction of shale gas reservoirs has been made possible through relatively recent advancements in horizontal drilling and hydraulic fracturing technolo-gies. While shale gas production has hitherto largely been an American phenomenon (BP, 2016), global shale gas resources 1 are vast totaling more than 200 trillion cubic meters (tcm; about one third of world total gas technically recoverable resources), with about 13 tcm in Europe (EIA, 2013). Though there is currently no European commercial shale gas industry, several countries there have considered using domestic supplies to complement their national energy strategy (EC, 2014;JRC, 2012). Shale gas has the potential to reduce dependency on foreign imports and to offset the decline in indigenous production of conventional gas that European countries have been experiencing (BP, 2017). While Poland was initially a promising location for commercial shale gas extraction, exploration efforts there ultimately failed (LaBelle, 2018). On the other hand, Germany and the United Kingdom (UK) have among the top European shale gas-in-place 2 volumes (Germany: 7.7 tcm; UK: 37.6 tcm), and have expressed interest in recent years in exploiting their reserves (BGR, 2016;BGS, 2013). Additionally many reports and papers
In the presence of sunlight, VOCs, CH 4 , and CO interact with NO x through a complex series of reactions to form tropospheric ozone (O 3 ) (Sillman, 1999;Sillman, 2003). The relationship between O 3 and its precursors is nonlinear: in a NO x -sensitive regime (high VOC/NO x ratio), O 3 increases with increasing NO x , while increasing VOCs will have little to no impact. In a VOC-sensitive regime (low VOC/NO x ratio), O 3 increases with increasing VOCs, while the addition of NO x decreases O 3 formation. VOC-sensitive regimes are often encountered in urban areas where emissions of NO x from combustion (e.g., in road traffic) are high, while NO x -sensitive cases are often found in rural areas.
Ground-level O 3 is a significant short-lived climate forcer (Myhre et al., 2013), and dangerous to human and environmental health (Amann et al., 2008). O 3 adversely affects the respiratory, cardiovascular and central nervous systems, reproduction and development (EPA, 2013;Horvath and McKee, 1993;Nuvolone et al., 2018), damages ecosystems (EPA, 2013;Horvath and McKee, 1993;Nuvolone et al., 2018), reduces crop yields (Avnery et al., 2011), and impairs infrastructure (Kumar and Imam, 2013;Lee et al., 1996). The European Union's (EU) air quality directive asserts that the maximum daily 8-hour average O 3 (MDA8) should not surpass 120 μg m -3 (60 ppb 4 ) as a long-term objective; as a target value it should not be exceeded on more than 25 days per year averaged over 3 years (EP, 2002). However, the EU's target value for O 3 is higher than the World Health Organization's (WHO) air quality guideline of 100 μg m -3 , which they recommend to adequately protect public health (WHO, 2005). In spite of extensive regulation of precursor emissions, O 3 pollution is still a problem in Europe: according to the European Environmental Agency (EEA), over the 2000-2015 period between 94-99% of the European urban population (EU-28) was exposed to O 3 levels exceeding the WHO guideline (EEA, 2017). Exposure to O 3 pollution is believed to cause thousands of premature deaths of Europeans annually, with 13,600 deaths estimated for 2014 (EU-28) (EEA, 2017). Unhealthy levels of ambient O 3 in Europe can be ascribed to rising hemispheric background O 3 (Monks et al., 2015), the transboundary nature of O 3 and its precursors i.e. they can be transported across national boundaries and regions (Bach et al., 2014), reduced NO x titration in some urban areas whereby O 3 is scavenged through reaction with NO (Amann et al., 2008), and the nonlinear relationship between O 3 and its precursors (EEA, 2014).
Atmospheric chemistry transport modeling is a valuable tool for assessing air quality on the local and regional levels and assisting in air quality management. A widely used model is the regional scale air quality WRF-Chem model (Grell et al., 2004). WRF-Chem has frequently been used in European-based air quality studies in recent years, e.g., Brunner et al. (2015), Fallmann et al. (2016), Forkel et al. (2012, Im et al. (2015), Kuik et al. (2016Kuik et al. ( , 2018, Mar et al. (2016), Solazzo et al. (2012aSolazzo et al. ( , 2012b, Tuccella et al. (2012), Zhang et al. (2013aZhang et al. ( , 2013b, among others. Intensive natural gas production has motivated several modeling studies to assess the role of precursor emissions from this source in regional O 3 formation. Rodriguez et al. (2009) found that increased growth in oil and gas activities in the US Intermountain West could increase MDA8 by a maximum of 9.6 ppb in southwestern Colorado and northwestern New Mexico in the summertime. Kemball-Cook et al. (2010) modeled the impact of shale gas development in the Haynesville Shale play for 2012 and found increases in MDA8 up to 5 ppb within Northeast Texas and Northwest Louisiana. Carter and Seinfeld (2012) looked at the relative contributions of NO x , individual VOCs and HONO (nitrous acid) in O 3 formation during episodes in the wintertime in the gas producing Upper Green River Basin region in Wyoming; they found that the locations varied in their sensitivity to VOC and NO x emissions: one site for the 2008 episode was highly NO x sensitive, while the other 2008 site and both sites for 2011 were highly VOC sensitive. Ahmadov et al. (2015) simulated high O 3 episodes during the winter of 2013 over the intensive oil and gas producing Uinta Basin in the Western US, finding that a simulation based on top-down oil and gas sector emissions was able to capture observations during high O 3 episodes, while a simulation using bottom-up emissions from the US Environmental Protection Agency's (EPA) National Emission Inventory did not. Fann et al. (2018) modeled the impact of oil and gas activities over the US for 2025, predicting increases in summertime MDA8 up to 8.12 ppb in Western Texas. Archibald et al. (2018) modeled over the UK the potential impacts of VOC and NO x emissions related to future shale gas fracking activities there for 2013 and generally found increases in MDA8 across the country throughout the year with a maximum increase of 2.3 ppb in June. In addition to these studies, much of the literature on US shale gas has focused on O 3 , e.g., AACOG Natural Resources Department (2013, Chang et al. (2016), Edwards et al. (2014), Field et al. (2015), Helmig et al. (2014), Koss et al. (2015), NC Division of Air Quality (2015), Olaguer (2012), Schade (2017), Schnell et al. (2009). These works underline the value and need for studies that analyze how O 3 may be affected by oil and gas development.
Shale gas is a contentious topic among the public and political spheres in both Germany and the UK, with critics citing various concerns, e.g., adverse air quality impacts (Althaus, 2014;Cremonese et al., 2015;Vetter, 2016;Yeo, 2019). The German government currently has a moratorium on unconventional fracking activities, e.g. for shale, which can be reevaluated in 2021 (German Federal Government, 2017). On the other hand, the British government recently granted consent for hydraulic fracturing testing to the energy company Cuadrilla in late 2018 (Perry, 2018). However, the UK's safety regulations strictly limit the magnitude of seismic activity allowed by fracking activities, currently set to 0.5 on the Richter scale, which severely constrains the amount of gas that can be extracted (Thomas and Pickard, 2019). Indeed, tremors exceeding this limit have forced Cuadrilla to suspend work on numerous occasions since testing started. In any case, what role shale gas production will have in Europe's future, if any, is uncertain. In order to inform decisions and the greater debate on shale gas, scientifically-based knowledge on potentially dangerous impacts is critical. To this end, air quality modeling can be used to run scenarios and study how varying levels of precursor emissions from a future shale gas sector may impact regional European O 3 pollution. This paper is a companion paper to Cremonese et al. (2019), who developed scenarios on a future shale gas industry in these two countries to quantify the potential impacts on greenhouse gas and pollutant emissions. The present study builds on Cremonese et al. (2019) by using their scenario work as the basis for emissions input to the WRF-Chem model to investigate the range of potential impacts from shale gas on local and regional O 3 , with a major focus on O 3 health-related metrics MDA8 and SOMO35 (annual Sum of Ozone Means Over 35 ppb, daily maximum 8-hour). In addition to the CH 4 leakage rate of Cremonese et al. (2019), we also investigate higher CH 4 leakage rates from other studies up to 6%. Due to our interest in the European region, we base our WRF-Chem setup on Mar et al. (2016), whose evaluated setup over Europe performed well for both meteorology and chemistry. Our objective is to improve understanding on potential air quality impacts from a future shale gas industry in Europe for public and political discourse, and for supporting policymakers and other decision-makers. Further our work can inform regulation of a potentially emerging industry to enact emission control strategies and prevent potentially harmful levels of air pollution. To our knowledge, there have been no published studies of regional air quality impacts from future shale gas activities in Europe, apart from Archibald et al. (2018), which focused on the UK. In this paper, we first describe the methodology, detailing the model setup, initial and boundary conditions and emissions input, the shale gas scenarios, and the preprocessing of shale gas emissions. Subsequently we discuss the results, and finish with a summary and conclusions.

Model description and emissions
We used the Weather Research and Forecasting model (WRF) version 3.8.1 (Skamarock et al., 2008), coupled with chemistry (WRF-Chem) (Fast et al., 2006;Grell et al., 2005). Additionally we employ the same principal WRF-Chem options as done by Mar et al. (2016), who performed an extensive evaluation over Europe for the entire year of 2007 with WRF-Chem version 3.5.1. Here we describe our setup, and the main options we use are summarized in Table 1. The namelist is provided in SM Text S1, Table S1.
The model domain is set over Western Europe and fully covers the countries of study (i.e., Germany and the  (Lin et al., 1983) UK; Figure 1). The horizontal grid resolution was set to 15 km × 15 km, and the model domain was established with 150 grid points in both the west-east and southnorth directions. We used 35 vertically-stretched levels in the model with the top layer at 50 hPa. Our simulation period spans from May 29, 2011 to September 1, 2011, where the days in May serve as model spin-up. We decided on JJA (June, July, August) as the period for which to run our simulations because meteorological conditions during the summer months, i.e., elevated temperatures, increased sunlight and slow-moving high pressure systems, are more favorable to production of high and potentially harmful levels of tropospheric O 3 , and therefore this season commonly experiences adverse impacts on and air quality in terms of this pollutant, e.g., EEA (2017), Volz- Thomas and Ridley (1994). Meteorological initial and lateral boundary conditions (BCs) were obtained from the ERA-Interim reanalysis dataset provided by the European Centre for Mediumrange Weather Forecast (ECMWF; Dee et al., 2011) with a spatial resolution of ~8 0 km every 6 hours. We used nudging by four dimensional data assimilation (FDDA) to reduce model errors in meteorology that may be associated with deviation of simulated large-scale circulation from the observed synoptic conditions. We nudged temperature at all vertical levels, while horizontal winds are nudged above the planetary boundary layer (PBL). Mar et al. (2016) reported erroneous precipitation suppression over Europe in sensitivity studies compared with observations when water vapor nudging was applied; therefore we likewise did not nudge water vapor in our simulations, also following the approach of, e.g., Miguez-Macho et al. (2004) and Stegehuis et al. (2015). Space-and time-varying (i.e., dynamic) chemical initial and lateral BCs were implemented in this study and were provided by simulations from MOZART-4/GEOS-5 through NCAR at http://www.acom.ucar.edu/wrf-chem/mozart.shtml. Note that dynamic BC data can improve simulated O 3 (Gavidia-Calderón et al., 2018), and MOZART-4 global model data is frequently used as the BCs for simulations (Pfister et al., 2011).
Based on the findings of Mar et al. (2016), we apply the MOZART-4 chemical mechanism in our simulations (Emmons et al., 2010). Of the natural gas-relevant VOC species that are considered in this work, ethane and propane are represented explicitly by the MOZART-4 mechanism, while the higher alkanes (i.e., butane, pentane, hexane and heptane) are lumped into the C > 3 group known as BIGALK. Note that VOC shale gas emissions in this study are very light and do not contain aromatics (Table 4), and NO x emissions are likewise low compared with national inventories (Figure 3). This means that their effect on secondary aerosol formation would likely be insignificant, and as such we do not look at this in our study. Anthropogenic emissions of CO, NO x , SO 2 , VOCs, and NH 3 used in WRF-Chem were obtained from the TNO MACC III inventory provided at a resolution of 7 km × 7 km, for the year 2011 (Kuenen et al., 2014). In addition to the aforementioned anthropogenic emissions, shale gas emissions based on Cremonese et al. (2019) were included in our simulations (described in Scenario background).

Scenario background
The present study carries on the work of , who developed a series of drilling and emission scenarios to study the effects of a potential shale gas industry in Europe on greenhouse gas and pollutant release. Specifically, they explore inter alia how different practices, uncertainty in data -i.e., activity data and emission factors, as well as the extent of well productivity, may impact emissions released annually from such an industry in Europe. The scope of Cremonese et al. (2019)'s scenarios covers upstream emissions, i.e., from well pad preparation up to gas processing, from shale gas development in Germany and the UK. Cremonese et al. (2019) assume in their scenarios that annual shale gas production will be equivalent to recent conventional production in these two countries (Germany: 11.6 bcm; UK: 36.6 bcm). All scenarios explored in the present study are based on the 'REm-U P25' scenario in Cremonese et al. (2019). REm stands for -'Realistic Emissions' case, meaning that business as usual practices for US shale gas production are applied; -U stands for 'Upper' meaning that activity data and emission factors are on the high end of the uncertainty range. P25 means that well productivity is at the 25 th percentile, i.e., the lower end of the spectrum of well productivity as determined by the authors. CH 4 leakage rates from REm-H P25 are up to 1.36%, which are well within the range predicted for average upstream natural gas leakage in the US and worldwide ). While European official CH 4 leakage estimates for natural gas systems are, comparatively, substantially lower, independent studies investigating European natural gas CH 4 leakage rates in a transparent way are unavailable  and references therein). Indeed recent studies suggest that CH 4 emissions from European oil and gas systems are higher than indicated by inventories (Riddick et al., 2019;Yacovitch et al., 2018), while numerous studies carried out in the US have reported emissions from natural gas systems which are higher than national estimates (Introduction). Hence it is reasonable to explore higher leakages as unanticipated, yet plausible rates for large-scale gas production in Europe.

Present work
In total, we examine three scenario sets covering low, medium and high shale gas emissions which we refer to respectively as SG1, SG2, and SG3. Our interest is to see how varying emissions of VOCs with constant NO x from shale gas activities may affect air quality in this region. Note that the shale gas emissions from Cremonese et al. (2019) included in our scenarios are NO x , CO and VOCs. NO x and CO are based on combustion processes, while VOC emissions are based on CH 4 leakage as the two species are co-emitted together in gas leaks. Due to its relevance to the literature on natural gas, we explore scenarios based on varying CH 4 leakage rates. Therefore NO x and CO shale gas emissions remain constant in SG1-3, while VOC emissions (based on extent of CH 4 leakage) are the only ones that vary between the three scenario sets. It is necessary to note that changes in the CH 4 emissions are not used in this study due to model limitations as will be explained in Pre-processing shale gas emissions for WRF-Chem. In SG1 we use the CH 4 leakage rates directly from Cremonese et al. (2019), which are tailored specifically to Europe ( Table 2). For SG2 medium level emissions, we chose a rate of 2% based on Alvarez et al. (2018) (the approximate value for the segment of the gas chain considered in our scenarios), as it is relatively high yet still representative of large-scale gas production in the US (Introduction). For SG3 high level emissions, we chose a conservative extreme value of 6% which is on the higher end of the literature for CH 4 leakage (typically ranging from <1.0 to 10%, e.g., Visschedijk et al., 2018). As reported in Saunois et al. (2016), the GAINS model adopted a CH 4 emission factor of 4.5% for shale gas mining with current technology, which also supports our choice of 6% as a conservative extreme rate.
Within each scenario set we explore three "sub" scenarios resulting in a total of nine simulated emission scenarios ( Table 2), in addition to a baseline scenario for comparison. These three scenarios are wet gas composition, dry gas composition, and concentrated NO x emissions, which we abbreviate to 'wet gas', ' dry gas', and ' conNO x '. The first two are explored due to the critical impact that gas composition has on total VOC volumes released, and hence air quality. For example, unprocessed gas typically ranges from 75-90 vol% CH 4 , with the rest mainly consisting of VOCs (Baker and Lokhandwala, 2008;Faramawy et al., 2016;Gilman et al., 2013). Wet gas is leaner in CH 4 and richer in VOCs, while the opposite case is true for dry gas. Because it is not known in advance what the gas composition will be, we explore scenarios that cover both wet and dry options. Furthermore, because NO x emissions are unaffected by gas composition, wet and dry gas scenarios allow us to study the impact of varying levels of VOC loading to NO x on O 3 production. In our scenarios we applied wet and dry gas compositions of 84.6/15.4 and 96/4 (vol% CH 4 /VOCs), respectively, based on Faramawy et al. (2016) and displayed in Table 3. Finally, the interest of the latter scenario was to see if NO x emissions from shale gas activities concentrated in space and time could lead to a significant impact on local O 3 production. Because the conNO x scenario required further development, it is described in detail in the following section below.

conNO x scenario design
The purpose of the conNO x scenario is to explore the sensitivity of the simulation to concentrated NO x emissions on O 3 production. It needs to be pointed out that absolute NO x scenario emissions from Cremonese et al. (2019) are not increased, but rather concentrated over space and time. However, in order to avoid limiting the analysis to an arbitrary segment of the simulation, we run the concentrated NO x emissions for the entire JJA simulation period. Because the NO x emissions would be too high over the whole period to look at cumulative metric results, the analysis for the conNO x scenario is restricted to peak daily values. In this way the model run shows results of concentrated NO x emissions for any time during the study period, with the advantage being that days with more favorable Sub-scenarios dry gas, wet gas, conNO x dry gas, wet gas, conNO x dry gas, wet gas, conNO x a The UK and DE have different leakage rates in SG1 due to the design of the drilling projections in . In SG2 and SG3, a flat leakage rate is applied to gas production of both countries. b Based on several different studies in the past years reporting very high CH 4 leakage rates from shale gas activities, as described in the text. photochemical conditions for O 3 production will automatically be included in the results.
For conNO x scenario development we first determined which sectors of shale gas exploitation from Cremonese et al. (2019) contribute to NO x emissions in a manner that a high volume of emissions may be concentrated over space and time. Drilling and fracking activities fit these requirements: both activities require machinery with high horsepower and consequently result in high NO x emissions, while operation time thereof is relatively short on the order of several days up to a few weeks. Furthermore it is plausible that a contractor would elect to drill/frack multiple wells within the same area, e.g., due to target sweet spots, regulation permits or restrictions, or favorable conditions of a location such as close proximity to roads, pre-existence of pipeline systems etc., so that total NO x emissions from drilling and fracking activities may be concentrated in space.
According to Cremonese et al. (2019)'s shale gas emission scenarios, drilling and fracking combined constitute approximately 25% of total NO x emissions for Germany and 20% for the UK. Therefore we concentrate this percentage of NO x emissions for each respective country, while the rest of the NO x emissions were treated as in the other scenarios (described in Pre-processing shale gas emissions for WRF-Chem). In our conNO x scenario we assumed that one well is drilled and one well is fracked per well pad. Based on our calculations, NO x emissions from drilling and fracking activities are concentrated over an area of 980 km 2 and 1225 km 2 , or 8.0% and 7.2% of the total shale gas basin area, for Germany and the UK, respectively. With this information and the time over which the emissions occur, we determined the concentrated NO x emissions flux. In order to place an upper bound on the effect of concentrated emissions of NO x , we chose a location for these emissions based on which area exhibited sensitivity to added NO x emissions for O 3 production. We achieved this through a sensitivity study in which we looked for coordinates within the shale gas basin region which displayed an increase in MDA8 when a simulation included both shale gas VOC and NO x emissions over a simulation which included shale gas VOCs only (conNO x regions displayed in Figure 2). Nevertheless it is worth noting that the increase displayed by added NO x emissions was low (<4 μg m -3 ). Additionally, it is important to note that the conNO x scenarios are based on wet gas composition like the wet gas scenario ( Table 3). Further information on the development of and calculations for the conNO x scenario is provided in SM Text S1, Text S1.

Scenario emissions in context
In order to put our scenario emissions into context, we compare them with TNO MACC III inventory values for year 2011 (Figure 3). This offers a useful comparison since the TNO MACC III values are likewise used as anthropo- genic emissions input in WRF-Chem (see Model description and emissions). Scenario emissions are equivalent to a higher proportion of country total emissions for the UK than Germany. This is primarily due to British gas production being greater than that of Germany in the scenario storyline. VOC emissions between the scenarios increase greatly from SG1 up to SG3. The importance of gas composition on total VOC emission volumes is evident, where wet gas emissions are about 3-4 times greater than dry gas emissions (more information on wet and dry gas is provided in Present work). With the leakage rate assumed in SG3 under the wet gas scenario, VOC emissions are equivalent to 109% of British country total emissions, and 25% for Germany. In contrast, VOC emissions under SG1 dry gas are equivalent to 4 and 2% of British and German country total emissions.

Pre-processing shale gas emissions for WRF-Chem
The shale gas activities examined in Cremonese et al. (2019) primarily occur within the shale gas basin regions of each respective country. Shale gas emissions from Cremonese et al. (2019) were adapted to the TNO MACC III grid for emissions pre-processing for WRF-Chem (Figure 4). The shale gas basin maps for Germany and the UK were obtained from reports by national institutions of each respective country, i.e., the Institute for Geosciences and Natural Resources (Bundesanstalt für Geowissenschaften und Rohstoffe) for Germany and the British Geological Survey for the UK (BGR, 2016;BGS, 2013). Because there is currently no commercial shale gas production in these countries, it is not known at which locations shale gas activities will occur and consequently where corresponding emissions will be released; therefore we averaged total scenario emissions over the shale gas basin area for each respective country. The impact of averaging emissions is not expected to be significant since VOCs are relatively long-lived (Atkinson, 2000). Furthermore no considerable diurnal variability in shale gas activities is expected, and accordingly total emissions for each respective country are assumed to occur at a constant rate. The only exception to this is that we concentrated some NO x emissions over a subset of the shale gas basin region and JJA period, as described in conNO x scenario design. Averaging the scenario emissions over space and time provided the emission flux for each shale gas species used in our simulations (VOCs, NO x , and CO from Cremonese et al. (2019). While CH 4 is an important component of shale gas emissions and valuable to the scenario storylines explored here, shale gas CH 4 is not included in our simulations: WRF-Chem treats CH 4 as a BC on account of CH 4 's relatively long atmospheric lifetime of circa 12 years (Myhre et al., 2013).
In the scenarios explored here, >99.9% of the VOC emissions result directly from natural gas loss. For this reason we applied a typical natural gas VOC composition as the speciation for all VOC emissions from the shale gas scenarios that we pre-processed into WRF-Chem. The VOC speciation for natural gas was obtained from Faramawy et al. (2016) (Table 4). Since the MOZART-4 chemical mechanism used in our setup lumps C > 3 alkanes into one group (BIGALK), the actual speciation we applied includes 29% ethane, 35% propane and 36% BIGALK, by weight. We applied the wet gas VOC speciation to all scenarios to avoid confounding the results between wet and dry gas scenarios, i.e., so that wet and dry gas scenarios differ only in their extent of VOC emissions and not in their speciation. It also needs to be pointed out that natural gas is indeed a lightweight mixture of alkanes; nevertheless, in reality gas is often found in the reservoir in association with some oil which contains higher alkanes, alkenes, aromatics, etc. Ahmadov et al. (2015) found that aromatic VOCs have a disproportionately greater contribution to O 3 formation relative to all other VOC emissions. Additionally Carter and Seinfeld (2012) reported that both aromatics and alkenes were the most significant contributors to O 3 formation in spite of their comparatively small  VOC contribution. Therefore if oil is present, the impact of VOCs on O 3 production may be even greater.

Evaluation
Due to the similarity of our setups, we use the same approach and offer a comparison of our evaluation results to those of Mar et al. (2016). Notable differences between our setup and Mar et al. (2016) Mar et al. (2016) we compare our data to rural background stations since our horizontal grid resolution is likewise relatively course and therefore more representative of rural conditions. We evaluate our model-simulated results against observations for the following statistics: mean model, mean observations, mean bias (MB), normalized mean bias (NMB), mean fractional bias (MFB), and the temporal correlation coefficient (r). Definitions of the statistical calculations are provided in Mar et al. (2016).

Meteorology
The meteorological evaluation was carried out for the following variables at a 3-hourly temporal resolution: mean sea-level pressure (MSLP), 2m temperature (T2), and 10m wind speed and direction (WS10 and WD10, respectively). A summary of the domain-wide statistical performance for meteorology of the base run setup with WRF-Chem against the observations at rural background stations is shown below in Table 5.
MSLP was reproduced over the domain with a high degree of accuracy. The bias for MSLP is negligible, where both NMB and MFB are 0, and the r value is 0.99. Likewise T2 was found to be reproduced with a high degree of accuracy by WRF-Chem, with a low MB of -0.08°C averaged over the entire domain. The seasonal average T2 spatial distribution statistics are displayed in Figure 5. In general the absolute values of MB in T2 ( Figure 5C) were <1°C, where larger biases are found in the Alps. This greater bias over mountainous regions was also found in Zhang et al. (2013a) and Mar et al. (2016), which the latter notes is likely due to the complex mountain terrain and related unresolved local dynamics. Moreover the r values for T2 ( Figure 5D) are generally >0.9 and do not display considerable geographical variation, which demonstrates the models ability to reproduce this parameter well. Furthermore it was found that the model represents wind speed well, with a domain-wide average MB of +0.07 m⋅s -1 and r value of 0.67. Averaged wind direction over the domain was found to originate from the south-west, with a MB of about 20 degrees (a wind rose diagram for model results over JJA has been provided in SM Text S1, Figure  S1). In total, it was found that our WRF-Chem set-up is capable of reproducing meteorological conditions and their spatial and temporal variations in Europe reasonably well, and notably our values are consistent with those reported in, e.g., Zhang et al. (2013a) and Mar et al. (2016).

Chemistry
We performed an evaluation of the model chemistry with hourly observations of the following species: O 3 , NO x , NO 2 and NO. The domain-wide statistical performance for chemistry of the base run setup with WRF-Chem against the observations at rural background stations is shown below in Table 6.
In Figure 6 we see that the lowest modeled surface O 3 concentrations in our domain are concentrated over the UK, Belgium, the Netherlands and the North Rhine-Westphalia region of Germany, with averaged values around 50-60 μg m -3 , while highest over the Mediterranean region, where values exceed 110 μg m -3 . O 3 is overpredicted throughout most of Italy; similarly, high O 3 was predicted over the Mediterranean with WRF-Chem in Mar et al. (2016) and with various models in Im et al. (2015). This finding may be due to the relatively coarse model resolution causing an underestimation of NO x and in turn an overestimation of O 3 , or causing an excessive diffusion of O 3 from the sea to the land. Predicted O 3 concentrations reproduce the northsouth gradient shown in the observations. In Table 6, we see that our WRF-Chem setup overpredicts O 3 with a MB    (HTAP, 2010). Overall, however, the O 3 BC compares well with the observational time series trend.
In Figure 8 high NO x concentrations (~30 μg m -3 and greater) are visible over parts of the UK, the Netherlands, Belgium, Germany, Poland, and large urban conglomerates like Paris, Barcelona, and Belgrade, as well as the coastal region where the English channel meets the North Sea, resulting from high emission activities in these areas. In Table 6 we see that our setup slightly underestimates domain-average NO x concentrations (MB: -0.27 μg m -3 ) over JJA due to the balancing of underestimated NO (MB: -0.93 μg m -3 ) and overestimated NO 2 (MB: 1.01 μg m -3 ). Our results for JJA are consistent with NO 2 (over-) and NO/NO x (underprediction) trends for the whole year reported in Mar et al. (2016). These biases are likely a combination of uncertainty in NO x emissions (Kuenen et al., 2014) and in the model representation. Reported causes of the latter include deficiencies in mixing in the planetary boundary layer (Kuik et al., 2018), overestimating  nighttime NO 2 , coarse grid resolution (Kuik et al., 2018), and that the atmospheric lifetime may be too long relative to deposition or chemical loss mechanisms (Stern et al., 2008). In the AQMEII project  found NO 2 to be overestimated by one model by 15%, while underestimated by the rest of the models by 9 to 45% for the European domain, where the WRF-Chem simulations used RADM2, CBMZ (Zaveri and Peters, 1999), RACM (Stockwell et al., 1997) (Karlický et al., 2017). Overall, we find that the model performance for chemistry is consistent with Mar et al. (2016) on which we based our setup, and furthermore in line with biases of other studies for Europe, e.g., Solazzo et al. (2012b), Tuccella et al. (2012) and Im et al. (2015). Based on these aspects we find that our setup is performing at a reasonable level for modeling the impact of shale gas industry emissions on European air quality.

Quantification of O 3 impacts from shale gas activities
The maximum difference in daily MDA8 between scenario and base case, over the entire simulation period for each grid cell (referred to as ΔMDA8), is depicted in Figure 9; statistical data is provided in SM Text S1, Table S2. Note that in the following discussion we generally leave out SG2 to be concise as its impacts fall between the SG1 and SG3 cases.
In the SG1 wet gas scenario, predicted ΔMDA8 generally ranges between 2-4 μg m -3 and is restricted to relatively small areas located within the vicinity of the shale gas basin regions. Concentrating NO x emissions leads to prominent differences compared with averaging NO x emissions under SG1: ΔMDA8 values >2 μg m -3 stretch over a larger area for conNO x , especially over Germany and within the vicinity where NO x emissions are concentrated. The peak values likewise show a stark difference, where peak ΔMDA8 for wet gas is 4.5 μg m -3 and for conNO x is 9.5 μg m -3 . This indicates model O 3 sensitivity to NO x , especially in Germany. In the dry gas scenario, ΔMDA8 values are lower compared with wet gas and are more strictly located over the shale gas regions, showing the importance of VOCs in model O 3 formation.
Under SG3 with wet gas speciation the largest peak ΔMDA8 occurs, reaching 28.3 μg m -3 and located over the North Sea off the British coast. This result is consistent with Archibald et al. 2018, where the largest maximum increase in MDA8 occurs under the scenario with the greatest VOC emissions. Also notable is that peak ΔMDA8 under SG3 is seen in the wet gas scenario; this is in contrast to SG1, where peak ΔMDA8 occurs in the conNO x scenario. Nevertheless, conNO x and wet gas are relatively similar under SG3, which also stands in contrast to SG1. ΔMDA8 values >2 μg m -3 extend over a considerably greater area of the domain for the wet gas and conNO x scenarios under SG3 compared with their SG1 counterparts, reaching Scandinavia, Eastern Europe, and the Mediterranean countries. ΔMDA8 values are especially prominent in marine areas including the Baltic Sea, the Mediterranean Sea, and the Atlantic Ocean west of the UK (in addition to the North Sea). Further under SG3 wet gas and conNO x scenarios, most of the UK experiences ΔMDA8 values >6 μg m -3 near the coast, with the inner portion of the country seeing enhancements from 14 up to about 22 μg m -3 . In Germany, the majority of the country experiences ΔMDA8 values >2 μg m -3 , with Northern Germany experiencing the highest values from 8 to 12 μg m -3 (SG3), in part due to the majority of the shale gas basin area being located in the upper half of the country. The much greater impact on MDA8 in the UK compared with Germany is likely due to VOC shale gas emissions being much higher in the UK (Figure 3); this is a result of the British shale gas industry being about three times larger than Germany in terms of gas production according to the scenarios. Additionally, under SG3 many of the neighboring countries experience high O 3 enhancement maxima likely due to long-range Notably, the portion of Italy in our domain experiences ΔMDA8 values of only ~2 -4 μg m -3 . It is interesting to note here that the weather over Europe during the simulation period was dominated by cyclonic fields (low pressure systems). Low pressure systems are generally associated with clouds, storms, and wind. In this case it means that the shale gas O 3 impacts would be more widespread, as seen in the results here. Finally, ΔMDA8 values for dry gas under SG3 are markedly less in extent and magnitude compared with wet speciation (i.e., wet gas and conNO x scenarios). These results show that the O 3 enhancement resulting from shale gas activities taking place inside the UK and Germany can be significant, with the potential to negatively affect air quality on the local and regional scales in Europe. The general trend of model O 3 here is that as shale gas VOC emissions increase from SG1 up to SG3 and from dry gas to wet gas speciation (wet gas and conNO x ), O 3 maxima increase considerably in magnitude as well as in extent over the domain. Total VOC emissions from shale gas activities (as a result of gas composition and leakage) are critical in increasing local and regional O 3 over the base case. Interestingly, when VOC emissions are relatively low (SG1), concentrating NO x emissions leads to greater O 3 increase in magnitude and extent than when NO x emissions are averaged; however, when VOC emissions are relatively high (SG3) O 3 maxima outcomes are similar when NO x emissions are concentrated or averaged. This variance in model O 3 sensitivity to NO x under different VOC loadings is likely the result of the nonlinear relationship between O 3 precursors and its formation in the atmosphere. Further the fact that concentrating NO x emissions leads to similar or slightly greater O 3 than averaged NO x emissions indicates a mixed regime of sensitivity to both NO x and VOC; though, varying NO x leads to a lower impact overall on O 3 than varying VOC emissions does. In sum, the most critical factor on increasing local and regional O 3 over the base case is the amount of VOC emissions, while concentration of NO x emissions plays a further, albeit minor, role. Next we explore extra exceedances of MDA8 at the locations of regulatory measurement stations, namely those which otherwise would not occur but do so as a result of shale gas activities. Here we explore the EU threshold due to its relevance for regional regulations of the study area, and examine the WHO guidelines due to the more comprehensive scope of health as well as international relevance. On account of inherent model bias in MDA8, which in our case is low yet nevertheless present, we calculate extra exceedances by the following: where an observational value is below the threshold, the difference in MDA8 between the base case and scenario at the measurement location is added to the observational value, and, when these two values together surpass the threshold, this is treated as an extra exceedance. Henceforth, we refer to these as exceedances for ease of discussion. AirBase measurement stations considered within our domain have at least 75% temporal data coverage for the JJA simulation period ("valid stations"). In the interest of clarity, it is worth noting that for countries not fully covered by the domain (e.g., Italy), only the stations of that country that are located within the domain are included in our analysis. Figure 10 depicts the total exceedance counts, summed over the full set of valid stations, under the WHO and EU thresholds. The number of exceedances increases substantially over the lower WHO threshold value compared with the EU threshold value, ranging from 56 to 190 (EU threshold) and from 103 to 427 (WHO threshold); this highlights the impact of the threshold value on the total number of exceedances registered. Further we see that exceedances increase considerably as VOC emissions increase: under both thresholds, the lowest number of exceedances occurs in the lowest VOC scenario (SG1dry gas) while the greatest number of exceedances in the highest VOC scenario (SG3-wet gas). In order to put these exceedance counts into context, we examine the number of exceedances per station and then in Table 7 present data on the percentage of stations in each country which have exceedances. The exceedances caused by the scenario emissions are spread out among the stations, rather than a few stations experiencing the majority of exceedances. The mean number of exceedances per station under all scenarios is 1 (not including stations which exhibit zero exceedances), where the maximum exceedance experienced at any station is 2-3. This demonstrates the widespread, episodic nature of shale gas emissions on station exceedances. Additionally it shows that while shale gas activities, even under extreme emission cases, have the potential to cause considerably more exceedances over large areas, the frequency of this effect on any particular station or area is low in this regard. Due to the similarity in trends when applying the EU and WHO thresholds, we show exceedances from here on applying the WHO threshold for the sake of brevity; data for the EU threshold is provided in SM Text S1, Table S3. The number of stations with valid data over our simulation period varies strongly from country to country within our domain (Table 7). France has the highest number of valid stations (386) followed by Italy (244), of which 8-33% and 9-18% show exceedances, respectively (representing SG1-dry gas at the low end of the range and SG3wet gas at the high end). Germany has the third highest number of valid stations (234), of which 6 to 21% of these stations have exceedances (SG1-dry gas to SG3-wet gas). 80 valid stations are located in the UK, of which 4-35% encounter exceedances (SG1-dry gas to SG3-wet gas). The higher percentage of British stations having exceedances  compared with German stations is the result of the majority of the UK having a relatively high ΔMDA8 compared with Germany from shale gas emissions (Figure 9). Again, as VOC emissions increase from SG1 to SG3 and from dry gas to wet gas, exceedances generally increase. Furthermore stations in distant countries have exceedances as a result of long-range transport, e.g., Macedonia, Slovakia and Latvia. Based on these findings shale gas activities have the potential to cause exceedances at a considerable percentage of country stations (locally and in distant countries), where VOC emissions are critical to the extent of exceedances.
Plots of exceedance magnitude are presented in Figure 11 applying the WHO guideline, and in SM Text S1, Figure S2 applying the EU threshold. We define exceedance magnitude as the difference between the scenario and base case when an exceedance occurs. This provides meaningful insight because it shows whether shale gas activities only effected an exceedance because the background O 3 concentration was already very close to the limit value (i.e., when exceedance magnitude is low), or whether shale gas activities had a robust impact on MDA8 during an exceedance (i.e., when exceedance magnitude is high). The maximum exceedance magnitude spans between 0.8 to 15.1 μg m -3 under the WHO threshold. Exceedance magnitudes are low (mostly <1 μg m -3 ) in SG1, indicating that low shale gas VOCs are able to force an exceedance only when the background O 3 level is already close to the threshold. These results again reflect that the lower VOC scenarios lead to a low impact, whereas the impact is stronger with the high VOC scenarios.
While Italy is one of the top exceedance locations among the scenarios, it exhibits a very low exceedance magnitude, where in all scenarios the majority of values are <1 μg m -3 , with only a few values exceeding this value yet are still below 2 μg m -3 . This implies that background MDA8 levels are already high in Italy and close to the limit value, so that only a slight increase in MDA8 from shale gas through long-range transport is required to push it over. Consequently this results in many exceedances occurring in Italy ( Table 7) in spite of being further away from the shale gas activity area and experiencing a low overall Figure 11: Spatial depiction of exceedances and corresponding exceedance magnitude. Exceedance magnitude is defined as the difference between the shale gas scenarios and base case when an exceedance occurs, and is an indicator of the robustness of shale gas emissions on an exceedance. Exceedances are displayed as filled dots at the station locations where they occur, in μg m -3 , over JJA, applying the WHO guideline for O 3 as the threshold (100 μg m -3 ). For stations which experienced more than one exceedance, the maximum exceedance magnitude is shown. The top left-hand corner of each plot indicates the particular scenario, and the top right-hand corner displays the maximum exceedance magnitude value experienced over the domain and simulation period. DOI: https://doi.org/10.1525/ elementa.387.f11 ΔMDA8 (Figure 9). The UK displays the highest exceedance magnitude of the countries in our domain, where maximum values reach >15 μg m -3 under SG3. This finding is in line with the surface plots displayed in Figure 9, which show that ΔMDA8 is highest in the UK among all countries in the domain; again, this is likely due in part to the shale gas industry and consequently VOC scenario emissions being highest in this country. The exceedance magnitude for France is relatively high under the greater VOC scenarios, which is not surprising because of its close proximity to the shale gas activity areas of both the UK and Germany and high ΔMDA8 (Figure 9). Additionally, France experiences a greater number of high magnitude exceedances than does Germany. This is not surprising considering that relatively high ΔMDA8 (4-6 ppb) covered a greater area in France than Germany (Figure 9). SOMO35 is an indicator of accumulated O 3 exposure recommended by the WHO for use in health impact assessment, and is the sum of O 3 values exceeding an MDA8 level of 35 ppb or 70 μg m -3 (Equation 1) (Amann et al., 2008). The 35 ppb cutoff was chosen because the relationship between O 3 and negative health effects, as well as atmospheric models, is very uncertain below this threshold (WHO, 2013 where C iday is the maximum daily 8-hour average concentration and the summation is from i day = 1 to 92 for the JJA simulation period. Surface plots showing the percent impact of each scenario on SOMO35 levels are presented in Figure 12. A surface plot showing SOMO35 values for the base case is provided in SM Text S1, Figure S3. In the base run, a northsouth gradient of increasing SOMO35 is apparent, where values reach as high as 7500 μg/m 3 days in some parts over the Mediterranean Sea. Under the SG1 scenario set, a low increase in SOMO35 is present over the entire domain (generally less than 1%), while slightly greater increases are localized to the German shale gas basin regions and some of the area north of the British basin. Largely in contrast with the German basins, over the British basin area there is a prominent decrease in SOMO35 as a result of NO x titration. However, because the UK already experiences relatively low values of SOMO35 (SM Text S1, Figure  S3), this is not expected to bring about substantial health benefits there through reduced O 3 exposure. The maximum percent increase in SOMO35 for dry gas under SG1 is only slightly less compared with wet gas, though the percent decrease covers a considerably greater area on account of greater NO x titration due to less VOCs. Under SG2 (wet gas) and SG3, greater increases in SOMO35 stretch over a more extensive area of the domain over the UK and Northern Germany and the surrounding region (with effects most concentrated over the UK and the North Sea). Notably, the southern portion of the domain experiences a very low percent increase meaning that VOC emissions are not expected to cause relatively worse health impacts on this region. Under SG3 wet gas, there is essentially no percent decrease in SOMO35 on account of less NO x titration due to greater VOC emissions. The impact of VOC emissions on SOMO35 values is clear, which reach a maximum percent increase of about 28% for the wet gas scenario under SG3. On the other hand, dry gas under SG3 leads to a maximum increase of about 9%, and under SG2 still displays a percent decrease in SOMO35 that covers most of the British basin region (a maximum percent decrease in SOMO35 of circa -7%). These findings underline the important role of VOC emissions in increasing O 3 production and in turn worsening adverse health effects, and that effects are primarily localized to the countries where shale gas is being produced and the closely surrounding region.

Summary and conclusions
Our study offers the opportunity to understand and quantify potential implications from a future European shale gas industry on O 3 air quality. Here we use the WRF-Chem online-coupled regional chemistry transport model where our setup is based on Mar et al. (2016). We explore a total of nine comprehensive emission scenarios which are based on in-depth European shale gas scenarios from Cremonese et al. (2019), and which examine the effects of gas speciation, concentration of NO x emissions over space and time, and a range of three CH 4 leakage rates on model O 3 . Additionally, it is important to emphasize that our results depend on our scenario assumptions; because European shale gas does not yet exist as an industry, the results are not a predictor of what will happen in the future but rather a range of potential impacts, and further the results highlight what may be important for regulation if this industry were to come into existence.
Our results show that shale gas emissions are capable of significantly increasing O 3 concentrations (maximum of 28.4 μg m -3 over the North Sea and maximum of about 22 μg m -3 over land in the UK). Shale gas activities result in up to one third of all valid measurement stations in France and the UK having additional exceedances above the WHO threshold, up to about a fifth of stations in Germany, and a considerable percentage of measurement stations in neighboring and distant countries. Furthermore we find that values of SOMO35, an indicator of health impacts, can be considerable with a maximum percent increase of ~2 8%, which would further burden O 3 -related health issues in Europe (Bell et al., 2014). This also poses concern for a future European shale gas industry where an ageing population is at greater risk to O 3 -related health effects (Amann et al., 2008).
The overarching trend found in our results is that VOC emissions are critical to O 3 formation. Impacts were greatest for a scenario in which the VOC emissions are based on the assumptions that shale gas is wet and CH 4 leakage is relatively extreme. While in practice these assumptions together may be unlikely, high VOC emissions are also possible through greater shale gas production, wet gas with an even greater VOC component, and even higher gas leakage. Additionally the findings here demonstrate that concentrated NO x emissions increase impacts on O 3 . However these impacts from concentrating NO x emissions are relatively low in comparison to impacts from increasing VOC emissions, and become less important as VOC emissions increase.
This study shows a clear potential for a future shale gas industry in Europe to adversely impact local and regional O 3 concentrations and to exacerbate already existing air quality issues in this region. Even in lower VOC emission scenarios, emissions are sufficiently high to effect a considerable number (minimum of 103 when applying the WHO threshold) of additional exceedances. Altogether these results show that future shale gas industries in Germany and the UK pose a threat to European O 3 air quality, and emission control strategies, especially for VOCs and to a lesser extent NO x , are critical to mitigate impacts. While it is not possible to control the speciation of gas extracted, concerted effort would be required to reduce CH 4 leakage from gas production, and in turn associated VOC emissions. This would offer a three-pronged benefit by reducing detrimental outcomes for air quality, mitigating climate change, and improving the economics since less leakage means that more gas can be brought to market.

Data Accessibility Statement
All model results used in this study are available here: Weger, Lindsey; Lupascu, Aurelia; Cremonese, Lorenzo; Butler, Tim (2019), Modeling the impact of a potential shale gas industry in Germany and the United Kingdom on ozone with WRF-Chem, v11, Dryad, Dataset, https:// doi.org/10.5061/dryad.08kprr4xv Notes 1 (Technically recoverable) resources is the volume of gas that can be produced based on current technology. EIA. 2019. Oil and natural gas resource categories reflect varying degrees of certainty. Available at https://www. eia.gov/todayinenergy/detail.php?id=17151. 2 Gas-in-place refers to the total amount of gas estimated in the reservoir, and not the amount that can be recovered. BGS. 2013. The Carboniferous Bowland Shale gas study: geology and resource estimation.
London, UK: British Geological Survey for Department of Energy and Climate Change. 3 CH 4 leakage is expressed as total CH 4 emissions divided by total CH 4 (natural gas) produced, for the segment of the natural gas chain under study. 4 Note that O 3 values are typically reported in units of parts per billion (ppb) for the US while in μg m -3 for the EU, where 1 ppb O 3 is equal to 2.00 μg m -3 at standard temperature and pressure (20°C and 1013.25 mbar). Due to the European focus of this study we will henceforth present our findings in μg m -3 , and list values from other studies in their original format.

Supplemental file
The supplemental file for this article can be found as follows: • Text S1. Supplemental text. Text S1 includes the namelist used in the WRF-Chem simulations. A detailed description of the conNO x scenario development is provided. Statistical data on MDA8 enhancement from shale gas scenarios are included. Exceedance data and plots resulting from scenarios applying the EU threshold for O 3 are also included. Additionally a wind rose diagram of the domain is included, as well as a surface plot of SOMO35 for the base case simulation. DOI: https://doi.org/10.1525/ elementa.387.s1