Potential impacts of emissions associated with unconventional hydrocarbon extraction on UK air quality and human health

Here, we report the first results of model sensitivity simulations to assess the potential impacts of emissions related to future activities linked to unconventional hydrocarbon extraction (fracking) in the UK on air pollution and human health. These simulations were performed with the Met Office Air Quality in the Unified Model, a new air quality-forecasting model, and included a wide range of extra emissions of volatile organic compounds (VOCs) and nitrogen oxides (NOx) to reflect emissions from the full life cycle of fracking-related activities and simulate the impacts of these compounds on levels of nitrogen dioxide (NO2) and ozone (O3). These model simulations highlight that increases in NOx and VOC emissions associated with unconventional hydrocarbon extraction could lead to large local increases in the monthly means of daily 1-h maximum NO2 of up to + 30 ppb and decreases in the maximum daily 8-h mean O3 up to − 6 ppb in the summertime. Broadly speaking, our simulations indicate increases in both of these compounds across the UK air shed throughout the year. Changes in the 1-h maximum of NO2 and 8-h mean of O3 are particularly important for their human health impacts. These respective changes in NO2 and O3 would contribute to approximately 110 (range 50–530) extra premature-deaths a year across the UK based on the use of recently reported concentration response functions for changes in annual average NO2 and O3 exposure. As such, we conclude that the release of emissions of VOCs and NOx be highly controlled to prevent deleterious health impacts.


Introduction
Hydraulic fracturing or 'fracking' describes the industrial process used to extract hydrocarbons by pervasively fracturing a shale unit (termed shale plays) with high-pressure water, mixed with sand and chemical additives (Bickle 2012).
This source of unconventional hydrocarbons has created a large amount of interest from investors in the petroleum industry as well as environmentalists and the general public concerned with the impacts associated with it. Whilst fracking as a technique has a long history, its wide-scale uptake is much more recent. This increased uptake is driven largely by technological breakthroughs such as horizontal drilling, which will ultimately allow those involved to extract more of the oil and gas from geological reservoirs than previous methods would have permitted. In the USA, natural gas (an industry term for gaseous mixtures of hydrocarbons rich in methane) from shale accounted for only~2% of total production in 2004. Owing to proliferation in the fracking of shale plays, by 2035, shale gas is predicted to account for over 63% of total production (BP 2014). This exploitation of shale oil and gas coincides with the USA becoming the world's largest producer of natural gas. In the USA, shale fracking sites are generally found in large geological basins, for example in the Rocky and Appalachian Mountains and Great Plains. Generally speaking, the most productive sites are in very rural areas well away from large population settlements (US Energy Information Administration 2015).
Whilst methane is one of the most important hydrocarbons extracted from fracking, there are a large number of non-methane volatile organic compounds (NMVOCs) held within the shale plays, including alkanes, cycloalkanes, aromatic compounds and bicyclic hydrocarbons .
The potential environmental and health impacts associated with fracking are varied in terms of their sources and mechanisms of action (Adgate et al. 2014). Of particular interest is the influence of fracking on air quality. The compositions of the emissions associated with fracking are dependent on the particular phase in the life cycle of the fracking well (Moore et al. 2014). However, significant emissions of nitrogen oxides (NO x ) and NMVOCs (also referred to as VOCs) are common throughout this life cycle. Both NO x and NMVOCs represent important species in the production of ozone (O 3 ), a secondary pollutant associated with adverse health effects (Brunekreef and Holgate 2002). Elevated levels of O 3 , especially in urban areas, are of particular concern in the USA, where fracking has been adopted on a large scale. As such, much of the literature concerned with air pollution resulting from fracking focuses on the relationship between NO x and NMVOC emissions and O 3 concentrations (e.g. Edwards et al. 2014 and references therein) and these interactions provide an interesting point of comparison for an assessment of the potential atmospheric impacts of fracking in the UK.
Fracking in the UK remains a controversial issue. However, the House of Commons' rejection of a moratorium in January 2014 means that wide-scale adoption of fracking is a significant possibility. Whilst there has been considerable speculation about the environmental impacts of fracking in the UK (Environmental Audit Committee 2015) mainly based on extrapolation from the USA, there have been no studies investigating the impacts of any potential risks. It is, therefore, important to provide some detail of how the consequences of fracking in the UK may differ from those of the USA. This sort of preparatory investigation may also provide vital information for the development of effective, evidence-based regulation.
The aim of this study is to determine how increased emissions of VOCs and NO x in regions where fracking could take place in the UK may impact air quality and ultimately human health. To achieve this, we make use of a recently developed numerical air quality-forecasting model, Air Quality in the Unified Model (AQUM) (Savage et al. 2013). The 'The AQUM model' section provides a description of the AQUM model set-up used in this study. The 'Emission scenarios' section provides information on our emission scenarios for increases in VOCs and NO x in the potential fracking regions, whilst in the 'Results' section, we present the results obtained from our AQUM simulations, including analyses of comparisons with observations of O 3 and NO 2 from the UK Automatic Urban and Rural Network (AURN) and the results from our emission scenarios. In the sections 'Results' and 'Discussion', we discuss and conclude the study and make recommendations for further work.

The AQUM model
AQUM is a limited area configuration of the Met Office Unified Model (UM), used operationally to provide a 5-day air quality forecast for the UK (Savage et al. 2013). The model is operated with a 12-km × 12-km horizontal resolution grid covering part of Western Europe and 38 vertical levels spanning from the ground up to around 40 km altitude. The representation of chemistry in AQUM is given by the regional air quality (RAQ) chemistry scheme of the United Kingdom Chemistry and Aerosols (UKCA) sub-model of the UM, whilst aerosol processes are parameterised within the Coupled Large-Scale Aerosol Simulator for Studies in Climate (CLASSIC) sub-model.
The RAQ chemistry mechanism includes 40 transported species and 18 non-advected species, 116 gas-phase reactions and 23 photolytic reactions. It takes account of emissions for NO x , carbon monoxide (CO) and the following NMVOC species: ethane (C 2 H 6 ), propane (C 3 H 8 ), butane (C 4 H 10 ), ethene (C 2 H 4 ), propene (C 3 H 6 ), formaldehyde (HCHO), acetaldehyde (CH 3 CHO), acetone (CH 3 COCH 3 ), methanol (CH 3 OH), isoprene (C 5 H 8 ), toluene and o-xylene. Removal by wet and dry deposition is considered for 19 and 16 species, respectively. CLASSIC is a bulk aerosol scheme, where the following aerosol species are treated as an external mixture: ammonium sulphate [(NH 4 ) 2 SO 4 ], ammonium nitrate (NH 4 NO 3 ), black carbon (BC), fossil fuel organic carbon (FFOC), biomass burning aerosol, mineral dust, sea salt aerosol and a climatology of biogenic secondary organic aerosol (BSOA). The model has two-way coupling of oxidants between the aerosol and gas-phase chemistry schemes. Within CLASSIC, sulphur dioxide (SO 2 ) and dimethyl sulphide (DMS) are emitted and oxidised into aerosol sulphate (SO 4 2 − ) by four oxidants whose concentrations are calculated in RAQ: O 3 , hydroxyl radical (OH), hydroperoxyl radical (HO 2 ) and hydrogen peroxide (H 2 O 2 ). The reaction of sulphate with ammonia (NH 3 ) produces ammonium sulphate. Any excess ammonia reacts with nitric acid (HNO 3 ), provided by RAQ, to form ammonium nitrate. Finally, the depleted oxidant concentrations are passed back to the RAQ chemistry scheme of UKCA. (For further details on CLASSIC, see Appendix A of Bellouin et al. (2011).) A full description of the RAQ chemistry scheme and the AQUM modelling system is provided by Savage et al. (2013).
Lateral boundary conditions (LBCs) for meteorological fields are provided by the operational forecast of the Met Office global model. LBCs for chemistry and aerosols are derived from the operational forecast of the Monitoring Atmospheric Composition and Climate (MACC) global model (Flemming et al. 2009).
The main difference between the model configuration presented by Savage et al. (2013) and the one used in this study is the treatment of anthropogenic gas-phase emissions within UKCA. In the present study, monthly emission fields used in the AQUM control run are created for the year 2012 by merging data from three emission inventories: the National Atmospheric Emissions Inventory (NAEI) at 1-km resolution over the UK (Passant et al. 2014), ENTEC shipping data around the UK at 5 km (Whall et al. 2010) and the MACC-II regional emission inventory at 0.125°longitude × 0.0625°l atitude resolution (Kuenen et al. 2014) for the rest of the domain. The ENTEC and MACC emission data used here were compiled for 2007 and 2009, respectively, but they are scaled according to published totals from the European Monitoring and Evaluation Programme (EMEP) for 2012. Data from all three sources are interpolated to the AQUM 12-km grid prior to merging. In the version of AQUM introduced by Savage et al. (2013), anthropogenic monthly emission fields for gas-phase species were spread equally over the first four model layers (20, 80, 180 and 320 m mid-altitude) and hourly emission factors derived from an analysis of traffic cycles were applied. This simplistic approach has been replaced here by a more comprehensive system that enables the separate treatment of emission fields according the socalled selected nomenclature for air pollutants (SNAP) source sectors. In particular, the vertical disaggregation of emissions consists of separate vertical profiles for each SNAP sector based on calculations performed with the SMOKE-EU plume rise model (Bieser et al. 2011). Emissions in our simulations are initially redistributed in the vertical among the EMEP vertical layers (Simpson et al. 2012) following the Bieser et al. (2011) recommendations, but adding a supplementary 0-20-m layer as done by Mailler et al. (2013) to account for the fact that the first EMEP layer is relatively thick (92 m). The resulting matrix attributing emissions to vertical layers is equivalent to that of Mailler et al. (2013), but interpolated to the AQUM vertical grid. Hourly and daily emission factors based on those created by TNO for the MACC project (Denier van der  are also applied to account for the diurnal and weekly cycles of each SNAP source sector. The use of a comprehensive treatment of gas-phase emissions has a moderate impact on the modelled nearsurface concentrations of pollutants. Basically, the new vertical disaggregation of emissions results in somewhat higher effective emission heights compared to a previous configuration which resembles that of Savage et al. (2013). This yields some decreases in the peak NO x concentrations at the surface and consequently a reduction in the number of events with very low O 3 concentrations. The combined effect of the new vertical and temporal profiles of emissions also results in improved diurnal cycles for both NO x and O 3 when compared to near-surface observations from the UK Automatic Urban and Rural Network (AURN). However, this does not remove all biases from the model.

Emission scenarios
A number of model simulations, summarised in Table 1, have been conducted to simulate the year 2013 and perform an annual assessment of the impact of emissions related with fracking on air quality over the UK.
The location/extent of the additional emissions was constrained by the locations of shale prospective areas for fracking in the UK identified by the British Geological Survey (Fig. 1). These include (i) the Jurassic Weald Basin in the South East of England (Andrews 2014), (ii) the Carboniferous Bowland-Hodder Shale in the North of England (Andrews 2013) and (iii) the Carboniferous Midland Valley of Scotland (Monaghan 2014). These areas already contain licenced blocks, granted by the Department for Energy and Climate Change (DECC), for the exploration and extraction of oil and gas. Only emissions in the AQUM grid cells that were within these licenced blocks were modified.
It is critical to note that as there are no wide-scale fracking plants in operation in the UK at present, the scenarios we have explored with AQUM test the sensitivity of air pollution to potential changes in emissions. Fracking wells themselves are not likely to be allowed (by law) to become major sources of VOC emissions in the UK. The DECC and other government agencies have acknowledged that the major risk to air pollution is in the associated surface-based activities that rely on heavy-duty internal combustion engines (particularly diesel fuelled engines which are known to have significant impacts on human health (IARC 2012)). As such, we took the grid square in the UK AQUM emission data sets (see the 'The AQUM model' section for details of the sources of emissions data) with the highest emissions of VOC and NO x from SNAP sector 5 (extraction and distribution of fossil fuels and geothermal energy) or SNAP sector 7 (road transport) for 2012 as a starting point under the premise that additional emissions from fracking-related activity could potentially mirror the current UK emission situation. These emissions were then added to each of the grid squares in the model where shale gas extraction licences have been granted ( Fig. 1) to create scenario 1 ( Table 1). The NO x and VOC emissions were then further increased through scenarios 2-4 to span a wide range of potential increases in VOC and NO x emissions (following Edwards et al. (2014)) to determine more about the role of NO x limitation and VOC limitation on potential O 3 increases (Sillman 1999). These increments to the emissions in AQUM have been added to SNAP sector 5 and have the appropriate temporal and vertical profiling applied (see the 'The AQUM model' section).
An important question to ask is whether or not these emission scenarios are realistic. Given the uncertainty present, it is impossible to answer this with a yes or a no. Our aim is to span plausible to potential worst-case scenarios (outline in Table 1).
Vengosh et al. (2017) analysed over 100 papers published since 2013 on the topic of unconventional hydrocarbon extraction. Despite this large number of papers on the topic, very few published studies exist which quantify NMVOC and NO x emission rates. By far, the majority of emission-based studies have focused on CH 4 which is not so important for air quality. The analysis of these papers highlights large basin-basin variability which causes huge uncertainty in trying to estimate emissions from a new basin. Marrero et al. (2016) provide one of the only alternative studies to those in Uintah cited by Edwards et al. (2014), to evaluate how plausible our emission scenarios could be. Marrero et al. (2016) used whole air samples, collected during October 2013 in the Barnett Shale deposit in northern Texas and found that NMVOC emission rates were greater than 600-1000 kg h −1 . Our extra NMVOC emissions range from 88 to 373 kg h −1 (per grid box) and so are of similar magnitude to those observed in the Barnett Shale deposit.
NO 2 and O 3 are key pollutants included in the calculation of the Daily Air Quality Index (DAQI) for the UK (Connolly et al. 2013;Neal et al. 2014). Since the emission scenarios considered here only include additional sources for NO x and NMVOCs which are implemented in AQUM, the main impact will be on the modelled simulation of surface NO 2 and O 3 . The impact on the surface concentrations of PM 2.5 is expected to be minor because the emissions of aerosol precursors in CLASSIC have not been altered. Note, however, that the change in the atmospheric concentrations of oxidants (O 3 , OH, HO 2 , H 2 O 2 and HNO 3 ) within the RAQ chemistry scheme of AQUM-resulting from the photochemical processing of the additional sources related to fracking-will have an impact on the formation of ammonium sulphate and   ammonium nitrate in CLASSIC. Therefore, only a lower limit of the effect of fracking emissions on PM 2.5 can be assessed with the current experimental set-up.
The main focus of our work here is on NO 2 and O 3 and we omit assessing the impacts on PM 2.5 , but note that it is likely that increases in PM 2.5 emissions (especially from the use of heavy diesel machinery (Hasheminassab et al. 2014)) would have significant health impacts.

Results
Here, we present an analysis of hourly mixing ratios of gasphase species simulated at the lowest model layer in AQUM. The calculation of the DAQI treats each pollutant with its own average period and concentration threshold levels, derived from health impact studies (Connolly et al. 2013;Neal et al. 2014). Following those indications, the following analysis focuses on the maximum daily 8-h running average ozone (MDA8 O 3 ) and the maximum daily 1-h average NO 2 . Given the lack of changes in emissions of PM 2.5 and consequently small changes in their abundance, we neglect any further analysis of these species. Figure 2 shows a comparison of the results from the year 2013 control simulation against observations of NO 2 and O 3 at a selection of rural and suburban locations across the UK (see panels for number of sites included in the analysis). Greater than 80% of all simulated mixing ratios of NO 2 and O 3 in the control simulation are within a factor of two of the observations considering hourly data over the full year. The average correlation coefficients for the comparison of modelled against observed hourly NO 2 and MDA8 O 3 are 0.62 and 0.64, respectively, across all sites over the full year. There is a slight positive bias in modelled MDA8 O 3 (average of the mean normalised bias error (MNBE) over all stations~0.04 ppb) and a negative bias in maximum daily 1-h average NO 2 (average MNBE over all stations~− 0.28 ppb), but in general, the comparison with the AURN observations gives us good confidence that AQUM performs reasonably well over the period investigated. A more detailed evaluation of AQUM has been documented by Savage et al. (2013) and the inclusion of the temporal and vertical emission scaling (discussed in the 'The AQUM model' section) moderately improves the fit to observations over previous versions of the model (not shown). A more comprehensive evaluation of the performance of the control simulation against surface observations across the UK is given in the online supplementary information.
The impacts of including emissions related to the processes associated with fracking, as explored through scenarios 1-4 (Table 1), are shown in Fig. 3 and Table 2 Table 3 for NO 2 . Tables 2 and 3 detail the summary statistics from the analysis for the changes in O 3 and NO 2 , respectively, for every month of our simulation for 2013. Table 2 highlights that in general, the largest changes in the area averages for O 3 occur in the summer months (June and July). In Figs. 3 and 4, panel a displays the monthly mean MDA8 O 3 and monthly mean daily 1-h maximum NO 2 for June 2013 in the control simulation, respectively. Whilst the introduction of the fracking-related emissions leads to changes in O 3 and NO 2 year-round (as shown in Tables 2 and 3), photochemical secondary oxidant production is usually greatest in the summer months, hence the focus of our analysis  In general, there is a degree of negative correlation in the spatial distribution of O 3 and NO 2-where NO 2 is high, O 3 is low and vice versa. This pattern is well explained through the reaction of NO with O 3 , which is a primary mechanism for the production of NO 2 and accounts for a process termed O 3 titration in urban areas. Figure 3a highlights that O 3 titration takes place in the English Channel whereas MDA8 O 3 shows highest levels over suburban areas in Northern France and Southern England. Panels b-e in Fig. 3 show the difference  Table 3 Results for changes in monthly mean 1-h daily mean maximum NO 2 mixing ratio (ppb) between control simulation and scenarios for each month in 2013. The results show the area averaged mean change across the study area (avg) (11°W-6°E, 49°N-61°N) as well as the maximum (max) and minimum (min) changes in any grid box. The implementation of large increases in NO x emissions in scenario 2 leads to large maximum increases in NO 2 and the largest area wide average changes in NO 2    in monthly mean MDA8 (ΔMDA8) for June 2013 between the fracking activity scenarios and the control simulation shown in panel a. Panel b displays ΔMDA8 between scenario 1 and the control simulation. Over the domain, there is an average increase in MDA8 O 3 of 0.1 ppb, but larger local decreases in the regions where fracking activity emissions were included (a maximum decrease of − 1.9 ppb). Panel c highlights that the impacts of scenario 2 are large local decreases in O 3 in the grid squares where emission perturbations are applied (up to − 6 ppb), relatively small decreases in the surrounding cells and a smaller regional increase in O 3 elsewhere (up to 1.5 ppb) which results in effectively no change in the net area averaged ΔMDA8 O 3 . Scenario 2 included much larger perturbations to the NO x and VOC emissions than scenario 1. Panels d and e display ΔMDA8 O 3 for the differences between scenarios 3-4 and the control, respectively. Scenario 3 includes increased VOC emissions over scenario 1 and comparison of panels b and d highlights that the increased VOC emissions minimise the magnitude of the local ΔMDA8 decreases in the grid cells that contain fracking licences. Panel e shows that further increases in VOC emissions yield slightly higher increases in regional O 3 levels in the model (area weighted ΔMDA8 0.3 ppb), particularly in the outflow regions of the emission perturbations and over the Irish and North Seas, with maximum local increases of 2.3 ppb. Very similar results are found for July (not shown). Figure 4 (panel a) illustrates that the maximum NO 2 values simulated by AQUM are present over the dense shipping lanes of the English Channel and large urban centres (cities with population > 400,000 persons). Panel b shows that the difference in monthly mean 1-h mean daily maximum NO 2 (ΔHMDM NO 2 ) between scenario 1 and the control simulation for June 2013 ranges from 0 to 10 ppb with the largest increases in the model grid cells where the emission perturbations were applied. In panel c, the results from scenario 2 for ΔHMDM NO 2 are presented and demonstrate that with much larger increases in NO 2 emissions, the simulated increases in ΔHMDM NO 2 are of the same order of magnitude. In panel c, the maximum ΔHMDM NO 2 is~30 ppb (i.e. a factor of 3 increases in ΔHMDM NO 2 for a factor of 4 increases in the additional NO x emissions from fracking compared to panel b). These increases in ΔHMDM NO 2 are very large compared with the control simulation (panel a). Of particular importance is the fact that the large increases in ΔHMDM NO 2 are spread out over significant distances from the grid cells where the emissions were modified and the emission perturbations therefore impact the levels of HMDM NO 2 in several large urban areas in the UK (particularly in the South East, in places like London; in the North of England, in places like Manchester; and in Scotland, in places like Glasgow and Edinburgh). Panels d and e compare the results from scenarios 3-4 for ΔHMDM NO 2 . These scenarios reflect no difference in NO x emissions relative to scenario 1 but increases in VOC emissions (by a factor of 2 and 4, respectively, compared to scenario 1). The impacts of increasing VOC emissions on ΔHMDM NO 2 are minimal (< 0.2 ppb).

Discussion
In general, the scenarios we have explored with AQUM result in very small impacts on O 3 but significant and large impacts on NO 2 . The small impacts on O 3 are surprising given the increases in emissions applied to the precursors ( Table 1).
The general chemistry responsible for controlling the production of ozone is discussed in detail by Monks et al. (2015), but briefly revolves around the following cycles of reactions: where M refers to a third body and RO 2 represent the class of organic peroxy radicals formed from the oxidation of VOCs.
With sufficient amounts of NO x and HO 2 /RO 2 , the reactions outlined in R1-R5 will tend to produce large amount of O 3 . However, as the amount of NO x increases, the following reactions become important: The first of these reactions (R6) leads to local depletion of O 3 but will lead to subsequent O 3 formation if the products of the reaction (NO 2 ) go on to photolyse and reproduce O 3 (R3 + R4). However, if the fate of NO 2 is reaction with OH (R8), then there is loss of NO x from the system as HONO 2 represents a very soluble species with a long lifetime and low yield towards releasing NO x back to the atmosphere. This sequence of reactions (R1-R8) gives rise to the non-linear dependence of ozone to its precursor emissions. The balance between catalytic ozone production (R1-R5) and ozone loss is controlled by the abundance of VOCs and NO x and the rate constants involved in the elementary reactions. Scenarios 1-4 sample a wide range of VOC and NO x emissions to investigate this chemistry. Edwards et al. (2014) used a similar approach to understand the controls on O 3 in the Uintah Basin in Utah. Under the conditions observed in Uintah, VOC chemistry was found to dominate the radical chemistry that propagates O 3 formation and the greatest sensitivity to modelled O 3 was found by modifying NO x emissions. The results presented in Figs. 3 and 4 demonstrate that in our modelling study, there is a traditional non-linear behaviour in O 3 upon modifying NO x and VOC emission. A fourfold increase to the additional NO x emission from fracking (scenarios 2-1) leads to a threefold increase in ΔHMDM NO 2 concentrations, whilst a fourfold increase in VOC emissions from fracking (scenarios 4-1) reduces the change in minimum MDA8 O 3 by 31%.
The fact that the large changes applied to the model emissions of VOC and NO x (see Table 1) did not give rise to the sorts of increases in O 3 reported in Edwards et al. (2014) is at first confusing but can be rationalised by considering the fact that O 3 chemistry in the Uintah Basin (Edwards et al. 2014) was greatly facilitated by local meteorological conditions in the form of stable shallow boundary layer inversions. The Uintah O 3 pollution events happen often during the winter, when the height of the boundary layer is much shallower than in other seasons and where snow cover can act to increase the short-wave radiation flux required to trigger ozone production (Edwards et al. 2014).
In the UK, these sorts of meteorological conditions do not happen to the same extent. Of particular relevance to our study is the fact that most of the locations we investigated are relatively close to the coast and so would be affected by local scale meteorological mixing in the form of sea/land breezes. It is interesting to consider how changes in meteorology may affect the results from the model. For example, under the conditions favourable for photochemical production that occurred over the UK during several weeks of summer of 2003 (e.g. Lee et al. 2006), we speculate that the impacts on simulated O 3 would be much more severe.
In spite of only small changes to surface O 3 in our model simulations, the potential impacts on air quality from our study are still important to determine. In order to assess the impacts that the changes in NO 2 and O 3 could have on human health, we followed the methodology of Lelieveld et al. (2015) to calculate the impacts on premature mortality as a result of the different scenarios. The change in premature mortality ΔMort can be calculated based on Eq. (1): y 0 is the UK baseline mortality rate (taken from the WHO http://www.who.int/healthinfo/global_burden_disease/ projections/en/). The exponential term reflects the relative risk of a change in pollutant-Δ[X] (calculated from the absolute change in the concentration of pollutant between a given scenario and the control simulation)-and the pollutantspecific concentration response factor (CRF) is given by β.
Pop refers to the gridded population density (data taken from  (WHO 2013) recommended CRF values for changes in all-cause mortality associated with changes in annual mean NO 2 of 1.055 (1.031-1.080 range of the 95% confidence interval (CI95)). In our study, we also used 30% lower CRF values for NO 2 to account for the effects of PM 2.5 highlighted by the HRAPIE study following Walton et al. (2015). We also note that in our calculations, we use CRFs for long-term exposure (as the lifetime of the fracking wells is expected to exceed 5 years). The uncertainty in longterm CRFs is much greater than in short-term CRFs and is an area of vital further work.
To account for the thresholds in exposure to pollutants that health effects have, our calculations of ΔMort include only grid boxes that exceeded annual average NO 2 of 20 μg m −3 and seasonal (April-September) average O 3 of 70 μg m −3 (following the recommendations of WHO (2013)). Equation (1) was applied for changes in NO 2 and O 3 in each grid box in the model and the resulting ΔMort taken as the sum of the results for each grid box. Our calculation of the total ΔMort is then the sum of the contributions from O 3 and NO 2 taken respectively. In all cases, the ΔMort calculated reflected the difference from the control simulation of the concentration of the pollutant in the scenario under consideration applying the above thresholds on the control and scenario fields. For each of the four scenarios, the calculation of ΔMort was performed using the range of the 95th percentile confidence interval (CI95) values for the CRF values of O 3 and NO 2 yielding a total of 12 estimates of the mortality impacts of frackingrelated emissions. Figure 5 shows a histogram of the 12 calculations of ΔMort. The median increase in deaths that was calculated based on the range of scenarios explored is 111 per year across the UK with a range of 51 to 533. Figure 5 shows that the results of our calculations tend to have a very skewed distribution. The changes in mortality as we have calculated them here are all due to changes in NO 2 concentrations with the greatest effects on premature all-cause mortality being attributed to the simulated changes in NO 2 in scenario 2. The lack of any health effects from the changes in O 3 is owing to both the small number of grid boxes in the model simulations that have seasonal average (April-September) O 3 above 70 μg m −3 and the lack of any large increases in O 3 in the scenarios we explored.

Conclusions
In this study, we aimed to assess the potential role that increases in emissions of VOCs and NO x -important precursors to ground-level ozone-associated with the life cycle emissions related to unconventional hydrocarbon extraction could have on the burden and health effects of air pollutants in the UK.
Following the assumption of minimal emissions of VOCs and NO x from fracking wells themselves, we derived an emission scenario to explore the impacts of fracking-related emissions based on current emission data for the UK. Present-day model grid-cell maximum emissions from the extraction and distribution of fossil fuels and geothermal energy or road transport were used as a starting point to generate our emission scenario. These extra emissions were included into the AQUM model in the model grid cells where licences for fracking have been granted in three major geological basins-the Weald, the Bowland-Hodder and the Midland Valley.
Further sets of emission scenarios were then generated to allow us to systematically explore the effects of combined and isolated increases in VOCs and NO x emissions. These emission scenarios were then included into a model simulation for the year 2013 and analysis focused on the impacts of inclusion of the extra emissions relative to a control simulation during June (a peak period for photochemical activity). By far, the biggest impact of the inclusion of the extra fracking emissions was on NO 2-with a monthly mean daily maximum 1-h NO 2 increased by~0.4 ppb on average over all of the UK and Ireland and large-scale local increases of up to 30 ppb. Much smaller impacts were simulated for O 3 , in spite of the large increases in VOC and NO x emissions.
The effect of the increases in NO 2 and changes in O 3 was assessed by calculating changes in premature mortality following Lelieveld et al. (2015). Considering our emission scenarios, our calculations suggest that the median increase in deaths associated with fracking-related emissions is 111 per year across the UK with a range of 51 to 533.
This study has provided the first assessment of the air quality impacts of the adoption of wide-scale fracking in the UK using a numerical modelling framework. There is undoubtedly large uncertainty as to what the magnitude and distribution of emissions will look like in the future, but the results of this study provide useful evidence for the impacts of poor controls on fracking-related emissions. We conclude that in order to protect air quality and human health, fracking activities must adopt as stringent as possible emission control technology.
Funding information We acknowledge the NERC and NCAS for contributing to the funding of the development of the UKCA model which underpins AQUM. ATA also acknowledges NERC grant NE/M00273X/ 1 for the funding.