Interactive comment on “ Understanding high wintertime ozone pollution events in an oil and natural gas producing region of the western US ”

The authors present a thorough analysis of the modeling of ozone in the Uintah Basin. The central outcome is that modeling of ozone in 2013 with the use of a top down emissions estimate (from measurements) is better able to reproduce episodic ozone than with bottom up emissions estimate (from inventory calculations). This is an important point in particular given the apparent lack of difference in 2012. The authors explore the importance of model setup besides emissions and provide useful observations. The paper could be improved by some re-working of sections as noted below: Abstract reads like a summary and could be shorter to ensure main points are not too diluted. We shortened the Abstract a bit. However we tried to keep the major statements in the Abstract, as this paper has several important conclusions.


Introduction
The development of new drilling and stimulation technologies has resulted in a large increase in the number of active unconventional oil and NG wells and accelerated oil and NG production onshore in the US during the last decade (www. eia.gov).The quantification of the potential impacts of oil and NG production emissions on climate and air quality is a topic that has received recent attention (Schnell et al., 2009; Published by Copernicus Publications on behalf of the European Geosciences Union.Alvarez et al., 2012;Carlton et al., 2014).Several studies based on atmospheric measurements suggest that the leakage rate of CH 4 from the NG production, transmission, storage and distribution systems in North America may be significantly higher than is estimated within the regulatory emission inventories (Pétron et al., 2012;Karion et al., 2013;Miller et al., 2013;Brandt et al., 2014).At the same time, increased production of NG in the US stimulates switching from coal to NG by the power plants for electricity generation, which has led to a reduction in emissions of carbon dioxide and other pollutants by power plants in the US (de Gouw et al., 2014).
The oil and NG exploration, production, and processing systems also contribute significant amounts of non-methane hydrocarbons to the atmosphere.Katzenstein et al. (2003) reported elevated alkane concentrations in air samples collected at the surface in the southwestern US and attributed them to the oil and NG operations in the region.The authors also measured alkyl nitrate species, indicating enhanced photochemistry in the presence of VOC emissions and NO x .Pétron et al. (2012) analyzed daily air samples collected at the National Oceanic and Atmospheric Administration (NOAA) Boulder Atmospheric Observatory located in northeastern Colorado, and found highly correlated light alkane enhancements in a regionally distributed mix of sources in the Denver-Julesburg Basin.Additionally, CH 4 and VOC data from a 2008 summer pilot study involving mobile-based surveys found that leaks of raw NG and flashing emissions from condensate storage tanks can explain the alkane ratios that were observed in air masses impacted by oil and NG operations.Pétron et al. (2012) suggested that the emissions of the measured species are most likely underestimated in existing emission inventories, which was supported by a recent study (Pétron et al., 2014).Gilman et al. (2013) analyzed extensive VOC measurements at the same site in Colorado during the winter of 2011, confirming an abundance of alkanes associated with the oil and NG production operations in northeastern Colorado based on differences in chemical signatures (e.g., i-pentane/n-pentane ratios and relative amounts of acetylene to propane) between urban sources and oil/gas well emissions.There have also been studies to estimate the emissions of pollutants in the eastern US from NG production, processing, transmission and distribution (Jackson et al., 2014;Roy et al., 2014).
Ozone is a major surface air pollutant that adversely affects human health and vegetation (Jerrett et al., 2009;Anenberg et al., 2010).The US National Ambient Air Quality Standard (NAAQS) for 8 h average O 3 mixing ratios set by the EPA is 75 parts per billion (ppb) (www.epa.gov/air/criteria.html).Traditionally, high O 3 and photochemical smog have been associated with hot and sunny weather conditions over urban areas with high vehicle, residential and industrial emissions (Lelieveld and Dentener, 2000;Trainer et al., 2000;Ryerson et al., 2003;Banta et al., 2011).Consequently, ground-level elevated O 3 was considered to be primarily a summertime problem in urban and suburban areas.Contrary to the traditional view, in the rural Upper Green River basin (UGRB), Wyoming, in the winters of 2005 and 2008, high levels of O 3 were measured, with hourly levels exceeding 140 ppb in 2008 (Schnell et al., 2009).Very high O 3 mixing ratios, with hourly values reaching as high as 166 ppb, were observed in the same region during the winter of 2011, well above estimated surface background levels of 40 ppb (Rappenglück et al., 2014).As of July 2012, EPA declared the UGRB a non-attainment area for the 2008 ground-level 8 h ozone standard (Rappenglück et al., 2014).
Similar phenomena were observed in the UB, Utah, which has experienced rapid development of the oil and NG industry over the last several years (Oltmans et al., 2014a).The high O 3 episodes in both the UGRB and UB are associated with oil and NG production and processing, below freezing temperatures, and when persistent snow cover is present under prevailing light wind conditions and associated clear skies.This combination produces strong, shallow temperature inversions that trap fossil fuel effluents in a stable boundary layer.Sunlight passing through the trapped pollutants and reflecting back off the high-albedo snow drives diurnal photochemical production of O 3 that peaks soon after solar noon (Schnell et al., 2009;Rappenglück et al., 2014).In the absence of the above meteorological conditions, O 3 mixing ratios were near the background levels, in the 40-50 ppb range in the UGRB and UB (Oltmans et al., 2014a).
It should be noted that the impact of the oil and NG production on ambient air quality goes beyond the wintertime.Kemball-Cook et al. (2010) modeled the impact of the Haynesville Shale development on ambient O 3 levels in Texas and Louisiana.Their photochemical modeling of the year 2012 showed increases in 8 h average ozone values of up to 5 ppb within northeastern Texas and northwestern Louisiana resulting from development in the Haynesville Shale.A model study by Rodriguez et al. (2009) found that the rise in oil and NG production in the Intermountain West could increase maximum 8 h average O 3 levels by 9.6 ppb in southwestern Colorado and northwestern New Mexico in summer.Cooper et al. (2012) analyzed long-term surface O 3 measurements at rural sites from 1990 to 2010 across the US, observing that eastern US sites exhibited a decrease in summertime O 3 over that period, while the western sites did not show a decrease in summertime O 3 .These studies demonstrate the importance of future research needed to quantify accurately the role of rising oil and NG emissions as they relate to ambient O 3 levels in the western US.
Researchers from NOAA and other partnering organizations conducted intensive measurement campaigns in the UB (Fig. 1) during the winters of 2012, 2013 and 2014.The full review of the Uinta Basin Winter Ozone Study (UB-WOS) field study in 2013 can be found in the report available at http://www.deq.utah.gov/locations/U/uintahbasin/studies/UBOS-2013.htm.The topography of the UB is shown in Fig. 1.Here, we focus on measurements from the 2012 and 2013 field campaigns only, with wintertime meteorological and chemical conditions sharply different between the 2 years.During January and February 2012, the UB experienced relatively warm weather (above freezing temperatures during daytime), and the ground was mostly snow free.Thus, the meteorological conditions, including wind speed and mixing layer heights, were not favorable for O 3 production.In contrast, during the winter of 2013, the ground was covered by a layer of snow (∼ 20-30 cm), and air temperatures were typically 10-15 • C lower than in 2012, with high O 3 episodes experienced frequently.Helmig et al. (2014) analyzed observations of VOCs from both the UBWOS 2012 and 2013 field campaigns, showing that the 2013 concentrations of air pollutants such as benzene and toluene were a factor of 5 to 10 times higher than reported values over major US cities.
During UBWOS-2013, the UB was mostly under the influence of synoptic-scale surface high-pressure systems.These conditions resulted in frequent stagnation events in the UB, with light wind and shallow boundary layers.The stagnation episodes were interrupted by occasional storms passing over the UB.The snow cover prevented the heat exchange of air with the ground.Consequently, the boundary layers were shallow (∼ 50-200 m) during daytime.Thus, persistent stagnant meteorological conditions were present throughout most of the 2013 study period in the UB.During the UBWOS field campaigns, a large variety of meteorological, chemistry, and flux measurements were conducted at the Horse Pool ground site (Fig. 1).In addition, frequent tethersonde balloon launchings took place at three sites during the winter of 2013, producing high-resolution vertical profiles of O 3 , temperature, and humidity (Oltmans et al., 2014b).These sites include the Horse Pool, Ouray and Fantasy Canyon locations shown in Fig. 1.
The high wintertime O 3 observations have motivated researchers to attempt to model the photochemical processes driving the O 3 production resulting from the oil and NG production-related emissions.Carter and Seinfeld (2012) used a detailed chemical mechanism in a box model to study high O 3 production observed at two sites in the UGRB in 2008 and 2011.The authors performed multiple sensitivity analyses to assess the effects of ambient VOC speciation, air temperature and atmospheric radiation conditions on the O 3 production.The study found that one site for the 2008 episode was highly NO x sensitive and insensitive to VOCs and nitrous acid (HNO 2 ), while the other 2008 case and both 2011 cases were highly sensitive to changes in specified VOC and HNO 2 concentrations.The authors stressed the need for full three-dimensional (3-D) air quality models to address the high O 3 episodes in the UGRB and similar regions due to regional variability in photochemical and meteorological conditions.Edwards et al. (2013) conducted a box modeling study of O 3 production in the UB in winter 2012 using a near-explicit gas chemistry scheme -Master Chemical Mechanism v3.2 for data collected at the Horse Pool site during the UBWOS-2012 field campaign.According to their box model study, the photochemical regime during that period was characterized as radical limited.Edwards et al. (2014) used a similar chemical box model for the conditions observed in the winter of 2013 in the UB.The study concluded the importance of the build-up and photolysis of carbonyls in deriving high O 3 in the UB during the winter of 2013.
The stagnant meteorological conditions during the winter of 2013 in the UB are characterized as a "cold pool".Cold-pool-type inversion layers, occurring within basins and valleys in the western US, have been studied extensively (Banta and Cotton, 1981;Clements et al., 2003;Lareau et al., 2013).Such valley cold pools (VCP) are identified when a radiosonde indicates the presence of an inversion below the maximum crest height of the surrounding mountains, and average wind speeds beneath the inversion top that are less than 5 m s −1 (Reeves and Stensrud, 2009).The VCPs form as an upper-level ridge approaches the western US and in response to strong mid-level warming, and they can exist several days until being removed, usually by advection or turbulent erosion processes (Reeves and Stensrud, 2009).These "cold pool"-like conditions can result in high air pollution episodes during wintertime.It has been known that meteorological models traditionally have difficulties in accurately simulating such meteorological conditions (Zhong et al., 2001;Baker et al., 2011). Neeman et al. (2014) simulated cold pool situations in the UB during the winter of 2013 and examined the sensitivity of the boundary layer structure to the treatment of the cloud microphysics and snow cover in their meteorological model settings.These studies, and the challenges related to the modeling of wintertime O 3 , are a motivation for investigating the wintertime O 3 problem by means of a coupled 3-D meteorologychemistry atmospheric model.Such models treat meteorology, anthropogenic emissions, advection and vertical mixing of chemical species, photochemical and removal processes in a unified modeling framework.Here, we present results from the Weather Research and Forecasting with Chemistry (WRF-Chem) meteorology-chemistry model, focusing on simulations for the winter of 2013, but also presenting results for 2012 using identical emission scenarios for both years in the comparisons.
Results are presented using two emission data sets for the oil and NG operations located in the UB.One set of emissions is based on an anthropogenic emission inventory used for regulatory purposes, and the other set is derived from in situ atmospheric measurements of reactive odd nitrogen (NO y ), VOCs and CH 4 .We present evaluations of model simulations using observations from the fully instrumented ground site as well as tethersonde profiles from three locations within the UB (Fig. 1).The paper focuses primarily on ozone-related photochemistry within the simulations, but important meteorological features are discussed as well.A sensitivity analysis is also presented that illustrates the im-portance of various physical and chemical processes to elevated wintertime O 3 formation in the UB.Table 1 lists all the acronyms used in this paper.

Emission data sets used in the model
Two emission data sets are used in this study: a bottom-up inventory based on information provided by the US EPA, and a top-down emission scenario for the oil/gas sector (emissions related to oil and NG operations) derived from CH 4 and a suite of VOC and NO y observations at the Horse Pool site during the UBWOS 2012 and 2013 field campaigns.Four major source categories are treated in each emission case: point, mobile onroad, mobile nonroad, and area (or nonpoint).The mobile sector emissions are identical in both cases, and are based on the US EPA NEI-2005, as described in other regional modeling studies, e.g., Kim et al. (2011), Ahmadov et al. (2012), and Brioude et al. (2013).Point source emission estimates within the top-down scenario are also based on the NEI-2005, while the bottom-up point source emission estimates are specified according to the NEI-2011 national inventory.Since the NEI-2005 inventory does not include any oil/gas sector emissions for the UB, double counting of emissions in the top-down emission data set is avoided by using NEI-2005 for non-oil/gas sector emission estimates.
The coal-fired Bonanza Power Plant (Fig. 1) located in the southeastern part of the UB is the predominant point source of NO x and sulfur dioxide (SO 2 ) in the UB.Measurements during the winter 2013 study showed that when shallow boundary layers were present, the power plant effluents were injected above the boundary layer into the free atmosphere.For the UB domain used here, CH 4 and VOC emissions are dominated by the oil and gas production sector within the area (or nonpoint) source category.Details of the emission processing for this sector and for the two emission data sets are presented below.Due to the geographical location of the UB (Fig. 1) and the locations of the emission sources in the vicinity, the upwind transport of air pollution to the UB is small.The major pollution source outside of the basin is Salt Lake City, which is located more than 100 km away.It should be noted that the UB is sparsely populated; thus, the urban emissions are low.

EPA NEI-2011 oil/gas sector emissions (bottom-up)
The bottom-up emission inventory for the UB oil/gas sources is based on the US EPA NEI-2011 (version 1) national inventory released in August 2013 (US EPA, 2013, available at ftp://ftp.epa.gov/EmisInventory/2011v6/flat_files).The national inventory contains annual emission estimates of total VOCs from the oil/gas sector at a county level for nonpoint sources, or at specific locations for point sources.Ancillary information for temporal partitioning, VOC speciation, and spatial distribution of nonpoint sources is available through data files in the version 1 modeling platform (ftp://ftp.epa.gov/EmisInventory/2011v6/v1platform, release date 11/8/2013).The EPA SPECIATE 4.3 database (e.g., Simon et al., 2010) is used for default VOC profile species assignments and the VOC to total organic gases (including CH 4 ) emission conversions.Within the modeling platform, the oil/gas sector VOC speciation for nine distinct oil-and natural gas-producing basins are specified according to Western Regional Air Partnership recommendations (http: //www.wrapair.org/forums/ogwg/PhaseIII_Inventory.html).Four distinct VOC speciation profiles are provided for the UB, and mapped with EPA assignments (gsref_cmaq_cb05_soa_2011ec_v6_11f_onroad_06sep2013 .txt) to the source classification code or activity that is provided with the state-supplied VOC emission estimates.VOC profile to VOC species assignments are available for the Carbon Bond version 5 (CB-05) chemical mechanism within the file gspro_cmaq_cb05_soa_2011ec_v6_11f_beis_nf.txt.CB-05 VOC speciation assignments to the Regional Atmospheric Chemistry Mechanism (RACM) (Stockwell et al., 1997) are straightforward for non-alkane species (mostly formaldehyde, toluene, xylenes and ethylene).The weight fraction of "PAR", representing the lumped alkane class within CB-05, was partitioned into the RACM alkane classes according to the alkane partitioning observed at the Horse Pool site during 2012 and 2013.We refer to this emission inventory as "bottom-up" further in the text.
Primary NO x emissions are specified as 88 % nitric oxide (NO), 10 % nitrogen dioxide (NO 2 ) and 2 % HNO 2 , which are ratios used within the California Air Resources Board nonpoint sources (http://orthus.arb.ca.gov/calnex/data/discl_ calnex2010.html).This partitioning was applied to both bottom-up and top-down emission data sets, though model results are shown to be insensitive to this partitioning in Sect.4.2.3 below.The hourly emissions of the gaseous species were added to the chemical species mixing ratios in the model at every time step during the model integration using temporal profile assignments from cross-reference files in the NEI-2011 version 1 model platform database.The CH 4 emissions were indirectly determined from total VOC estimates within NEI-2011 and the total organic gas to reactive organic gas ratios specified within the SPECIATE 4.3 software.

Top-down oil/gas sector emission estimates
The top-down oil/gas emission estimates for the UB region are based on regression slopes of NO y and 59 individual VOCs with CH 4 observed during daylight hours at the Horse Pool site during the winters of 2012 and 2013.Regressions are determined for observations taken between 10:00 and 16:00 Mountain Standard Time (MST) for two reasons: (1) a broader, more regional representative sample is available during daylight hours, when the boundary layer is well mixed, (2) large transient spikes in NO y , VOC and CH 4 observed mostly during stable nighttime conditions that can decrease correlations and influence regression slopes are removed.Absolute emission values are specified by scaling regression slopes with the Uintah County (located in the eastern part of the UB) mean CH 4 flux estimate of 55 × 10 3 kg h −1 as determined from upwind and downwind aircraft transects during one flight in February 2012 (Karion et al., 2013).We distribute this estimated emission over the entire UB, which includes both Uintah and Duchesne (located in the western part of the UB) counties, even though the aircraft measurement used by Karion et al. (2013) covered only the eastern part of the UB.We used this estimate, as it is the only available top-down estimate of CH 4 flux for the oil/gas sector in the UB at present.Here, we assume that this emission estimate is representative of 2013 as well.A summary of the regression slopes, r 2 Pearson correlation coefficients, and partitioning within the photochemical mechanism is presented in the Supplement (Table S1).It should be noted that the VOC to CH 4 ratios reported here agree quite well with the independent estimates carried out by Helmig et al. (2014).As explained in that study, VOC emission ratios determined by this regression technique do not account for removal by the OH radical.Thus, emission estimates, particularly for more reactive VOCs, should be considered lower limits.
No hourly variation is assumed in the top-down emission scenario for the oil/gas sector.Here, the oil/gas emissions are spatially allocated according to well numbers in a given model grid cell, using well location information from the Department of Oil, Gas and Mining, Utah Department of Natural Resources (http://oilgas.ogm.utah.gov/Maps/OG_Maps.htm), valid as of May 2012.VOC speciation is assumed to be the same for both oil and NG wells.Only wells within Duchesne and Uintah counties are considered.Here, the spatial patterns of NO x , VOCs, and CH 4 emissions from oil and NG operations in both counties are assumed to depend only on well density as a simple first-order approximation.Additional information would be needed to more accurately locate significant isolated emission sources associated with combustion and fugitive processes within the UB.These sources include, but are not limited to, evaporation ponds, diesel generators used in the oil and NG well drilling, oil storage tanks, compressors, NG compressor stations, processing plants and venting operations.We name this emission set "top-down", because it relies on the atmospheric measurements made by an aircraft (Karion et al., 2013) and at the surface site.
Figure 2a and b show the spatial distribution of 24 h averaged total toluene emissions over the UB in the 4 km resolution model grid from both emissions sets, with the emissions in Fig. 2b being proportional to the spatial density of the oil and NG wells.There are two high emission spots in the region, one in the western part of the UB, and another to the east, south of Horse Pool.Total fluxes of CH 4 , VOCs and NO x from both emission scenarios for the oil/gas and other sectors in the UB are provided in Table 2. CH 4 and total VOC emissions for the oil/gas sector within the UB in the bottom-up inventory (NEI-2011) are lower by factors of 4.8 and 1.8 than the top-down estimates, respectively.Conversely, NO x emissions are about 4 times higher in the bottom-up inventory.The large difference in VOC/NO x emission ratios between the two products is shown below to have a profound impact on modeled O 3 formation within the UB.Table 2 also shows that VOC emissions from the oil/gas sector dominate emissions from all other sources (vehicle, residential, etc.) in the UB, and that NO x emissions are dominated by the oil/gas sector in the bottom-up inventory, but not in the top-down emission data set.We assumed NO y as an inert species in deriving the NO x emissions.

Air quality model description
The WRF-Chem model is a fully coupled meteorologychemistry model (Grell et al., 2005), and a recent version of the model (version 3.5.1)was used in this study (http://ruc.noaa.gov/wrf/WG11/).The main WRF-Chem model settings are presented in the Supplement (Table S2).The gas-phase chemistry mechanism used here is called "RACM_ESRL" within the WRF-Chem model.The RACM_ESRL gas chemistry scheme is based on the RACM mechanism (Stockwell et al., 1997), which is designed to be suitable for modeling full tropospheric chemistry in remote and polluted re-gions.The RACM_ESRL mechanism contains additional chemical reactions, species and updated rate coefficients as discussed within Kim et al. (2009).The mechanism contains a wide range of inorganic and organic species and their intermediates, which are advected and vertically mixed at every time step during model integration, and includes 23 photolysis and 221 chemical reactions.It has been successfully used in previous air quality model studies, e.g., Kim et al. (2009), Kim et al. (2011), andAhmadov et al. (2012).The advantage of the WRF-Chem model is that the same 3-D model grid, time step and advection scheme are used for both meteorological and chemical variables.Below, we describe some specifications and modifications to the WRF-Chem 3.5.1 code that are needed for the wintertime UB simulations of 2012 and 2013.
The dry deposition of gas species is parameterized following Erisman et al. (1994).As the default WRF-Chem code was intended to simulate summertime O 3 , the deposition scheme uses the "snow or ice" type of land use category solely for the areas permanently covered by ice, e.g., the Arctic.The deposition scheme was modified to treat model grid cells as "snow or ice" whenever they are covered by snow.This helps to properly simulate dry deposition fluxes of O 3 and other species to the snow-covered ground for the 2013 simulations.
We used the Tropospheric Ultraviolet and Visible (TUV) photolysis scheme (Madronich, 1987) within WRF-Chem.The column density for O 3 in the photolysis scheme was set to 270 Dobson units (DU) based on the OMI satellite data set (http://mirador.gsfc.nasa.gov/)for the simulation time period (February 2013) and latitude.Additionally, the photolysis scheme was modified to include the effect of snow on surface albedo.The snow effect on photolysis is not considered in the default WRF-Chem model developed primarily for summertime photochemistry, similar to other air quality models.Based on the measured albedo values during the UBWOS field campaign in 2013, a fixed value of 0.85 was set for the surface albedo in the photolysis module when a given grid cell was covered by snow.The default values of the surface albedo appropriate for bare ground conditions in the TUV scheme lie within the range of 0.05-0.15,depending on the wavelength.We note that the photolysis scheme in WRF-Chem takes into account surface elevation in calculating the effect of Rayleigh scattering.This could be important for high-elevation sites in the western US, such as the UB and UGRB, which are located at about 1400 m and 1800 m a.s.l, respectively.
North American Mesoscale analysis fields (www.emc.ncep.noaa.gov)were used as boundary and initial conditions for the meteorological fields.Idealized vertical profiles were used as initial and boundary conditions to assign the background mixing ratios for some of the long-lived chemical species (McKeen et al., 2002).These idealized profiles are based upon results from the NOAA's Aeronomy Laboratory Regional Oxidation Model.The detailed description of the initial and boundary conditions can be found in the WRF-Chem user's guide available from http://ruc.noaa.gov/wrf/WG11/Users_guide.pdf.
The meteorology and chemistry were simultaneously simulated by the WRF-Chem model, initially on the 12 km resolution domain covering a large part of the western US.For this domain, the bottom-up anthropogenic emissions inventory was used in the simulations.Model simulations were conducted using the same settings and emissions for both the January-February 2012 and 2013 time periods.The second stage of WRF-Chem simulations were conducted on a 4 km resolution domain by performing one-way downward nesting from the 12 km domain simulations whereby the output from the coarse domain simulations was used to provide initial and boundary conditions for the meteorological and chemical variables (mixing ratios of the gas species) in the 4 km model domain.Elevation maps for both model domains are shown in the Supplement (Fig. S1).Hereafter, we present the modeling results solely from simulations done on the 4 km model grid, as this domain can better represent the "cold pool" conditions and the complex topography in the region.

Base case simulations
First, we present O 3 modeling results for UBWOS-2012. Figure 3a shows time series of measured and modeled O 3 concentrations during 31 January-28 February 2012 at the Horse Pool site.There are three modeling cases here, which were performed in the same way, except for different emission scenarios -bottom-up, top-down, and top-down without any oil/gas emissions in the UB.Observed hourly O 3 mixing ratios at the site reach ∼ 40 ppb during afternoon hours, while the variability of daytime O 3 from one day to another is small, and none of the emission cases differs markedly.The variability of O 3 within the UB is dominated by transport of background O 3 into the basin rather than by local chemistry.As noted before, the meteorological conditions did not result in any O 3 exceedances during the 2012 field campaign.Maximum O 3 values during daytime are reasonably well simulated by both the bottom-up and top-down cases, while the bottom-up inventory shows stronger O 3 depletion due to NO x titration during nighttime.The model case without the oil/gas emissions is slightly different than the topdown case, indicating a small contribution of the oil/gas sector to O 3 production in the UB for 2012.3. Statistical evaluations for meteorological variables at Horse Pool for the same time periods are provided in the Supplement (Table S3).Hereafter, we report the model evaluation statistics for daytime measurements only, between 09:00 and 17:00 MST, since the atmospheric models have difficulties in accurately simulating stable nocturnal planetary boundary layers (PBL) when the PBL depth could be a few meters.These tables also provide the median of observed values for several chemical species at Horse Pool during the 2012 and 2013 field campaigns.The detection methods used to obtain these measurements are described in detail by Warneke et al. (2014).The median of the measured mix- ing ratios of primary and secondary chemical species highlight the contrast in pollutant levels in the UB between the two consecutive years.
The evaluation statistics reveal that the model captures the temporal variability of primary species better in 2012 than in 2013.This could be caused by the fact that, during 2013, turbulent mixing and advection were much weaker than in 2012, and uncertainties in the spatial representation of emissions would affect the correlations for primary species more strongly in the 2013 evaluations.This would likely be the case for both the bottom-up and top-down emission cases.Primary species such as NO y and some non-methane hydrocarbons are somewhat underestimated during daytime for the top-down emission case in both 2012 and 2013.Table 3 clearly demonstrates that the bottom-up inventory leads to a stronger model underestimation of CH 4 in 2012 compared to the top-down emission estimates.The evaluation of CH 4 simulations at a ground site provides an independent validation of the CH 4 flux estimates from the oil and NG operations in the UB determined by using aircraft measurements.The discrepancy between the emission scenarios in terms of CH 4 simulations widens in 2013 simulations (Table 3), when high CH 4 levels were observed at Horse Pool.However the top-down emission case still underestimates CH 4 levels by ∼ 40 %.This model underestimation may be caused by two sources of error that could potentially offset each other.First, as stated above, the CH 4 flux estimate of Karion et al. (2013) used to scale the primary emissions is likely a lower limit, since the estimate is limited to a single day of aircraft sampling that only sampled emissions from the oil and NG operations emissions in the eastern part of the UB.Second, biases in the model meteorological fields, such as wind speed and mixing layer height, could be affecting either year's statistics.
Evaluation of the model for other gaseous species at Horse Pool for the two emission cases reveals that the bottom-up emissions (NEI-2011) lead the model to strongly overestimate the NO y concentrations, while the VOCs are greatly underestimated (Table 3).Using the top-down emission scenario, the model explains the temporal variability in the primary species such as NO y , alkanes (ethane), and aromatics (toluene and xylene) better than the bottom-up inventory in 2013, while the two cases are not that different in 2012.This shows the advantage of the top-down emission estimates compared with the bottom-up (NEI-2011) case for the oil/gas emissions in the UB.The model captures both the temporal variability and the daytime values of secondary species such as O 3 , peroxyacyl nitrates (PAN) and acetaldehyde (CH 3 CHO) in 2013 quite well.
Table 3 reveals that, for 2012, both emission cases show very similar correlations for the daytime O 3 concentrations, as the O 3 concentrations were slightly affected by local emissions and photochemistry.However, in 2013, the top-down emission case exhibits a profoundly better correlation than the bottom-up case.The top-down emission case captures both the magnitude and variance of the daytime O 3 values at Horse Pool well.In Table 3, the statistics for odd oxygen (O x = O 3 + NO 2 ) are also shown.This helps to identify the role of O 3 titrated by freshly emitted NO x in the model and observation.The bias statistics show that the topdown emission scenario allows the model to capture the secondary species, which are products of the photochemistry.Conversely, the bottom-up inventory results in strong underestimation of the secondary species during 2013.
The statistical comparisons in Table 3 for reactive nitrogen species show a marked difference between 2012 and 2013 for observed levels of NO y .Median NO y levels are more than a factor of 3.5 higher in 2013 compared to 2012, yet median NO x levels are nearly identical.The differences in NO x /NO y partitioning between the 2 years is qualitatively reproduced in the top-down inventory results, but difference for individual NO y species are apparent.The statistical comparisons in Table 3 for reactive nitrogen species show a marked difference between 2012 and 2013 for the top-down emission case.Both years show a 25 to 30 % low bias in NO y .One would expect biases in individual NO y species to reflect the bias of NO y itself, which is the case for NO x , nitric acid (HNO 3 ), and PAN in 2012.However, in 2013, NO x shows a positive bias, and HNO 3 is significantly underpredicted relative to NO y .The HNO 3 underprediction can be explained by high HNO 3 deposition velocities in the default version of the WRF-Chem model used here.Measurements of HNO 3 deposition to snow indicate a low deposition velocity at cold temperatures but increases with increasing temperature (Johansson and Granat, 1986), and even HNO 3 emissions from snow due to heterogeneous conversion of other reactive nitrogen species to nitrate at the snow surface (Dibb et al., 1998).The overprediction of NO x may likewise be related to uncertainties in the deposition of NO 2 to the snow-covered surface, but as discussed further below in the model sensitivity section, nighttime heterogeneous conversion of N 2 O 5 to aerosol, which is not included in our base model configuration, could explain part of the NO x discrepancy during 2013.
In order to demonstrate how well the model parameterizes the photochemistry in both emission cases, we plot measured and modeled O x vs. PAN during the daytime for 2013 (Fig. 4).As seen from the scatter plot, the base case (topdown emission scenario) is able to predict the observed relationship between O x and PAN.The model case with the bottom-up inventory also matches the observed O x -PAN relationship, but over a much smaller concentration range.As only the top-down emission inventory can explain the high O 3 in the UB in 2013, we analyze this base case model scenario in more detail below.
The time period shown in Fig. 3b consists of three distinct episodes with high O 3 measurements.We group these episodes as follows: the first and longest episode covers the 29 January-8 February time period.The other two episodes cover 12-17 February and 19-21 February 2013.Between the episodes, synoptic-scale surface low-pressure systems were present, which led to the partial clearing of O 3 and other pollutants out of the UB.Such meteorological conditions occurred prior to 29 January and between the stagnation episodes.Weather maps in Supplement Fig. S2 show synoptic situations associated with a frontal passage and stagnation episode that occurred in the UB during UBWOS-2013.Hereafter, we focus on the 29 January-8 February stagnation period.Meteorological conditions associated with O 3 buildup for other time periods are very similar to this episode.We also provide evaluations for some of the meteorological variables over the same time period in the Supplement.
Figure 5a and b illustrate the simulated O 3 distribution within the UB during a high ozone event.Modeled O 3 mixing ratios and horizontal wind vectors over the vertical plane across the Ouray and Horse Pool sites (Fig. 1) are shown at night (05:00 MST) and in the afternoon (15:00 MST) on 5 February 2013, a typical stagnation day when O 3 levels reached 124.2 ppb at Horse Pool (Fig. 3b).The plots show the very low wind speed inside the UB, demonstrating the wind regime during a VCP type situation.Figure 5a shows the high O 3 from the previous days' photochemistry in early morning above the western slope within a residual layer.The highest O 3 mixing ratio is simulated at ∼ 100 m above the ground at night.The downslope terrain flows converge at the bottom of the UB at night.Later, the layer enriched with O 3 and other pollutants aloft is moved to the eastern part of the UB during the morning hours.As we show in the Supplement (Fig. S3c), the wind speed of the nocturnal drainage flows over the UB are overestimated at Horse Pool and possibly elsewhere.This overestimation likely causes a stronger and faster drainage of the air masses near the ground at night, than what was observed.These strong nighttime drainage flows also displace primary emitted species with background concentrations at night.
Then, the shallow vertical mixing layer that developed along the slopes of the UB dilutes the pollutants aloft within itself.The afternoon O 3 distribution for the same vertical plane shows the horizontal and vertical extent of the O 3 build up in the UB (Fig. 5b).O 3 mixing ratios reached ∼ 120 ppb over Horse Pool and Ouray at 15:00 MST.The highest O 3 buildup is simulated between these two sites.The mixed layer with high O 3 (> 100 ppb) is ∼ 100 m thick during the daytime.High O 3 buildup (∼ 100 ppb) is simulated over the western part of the UB, reaching an elevation as high as 1900 m a.s.l. on the western slope.Another prominent feature on this particular day in the vertical O 3 distribution is the depletion of O 3 at ∼ 1800 m a.s.l.apparent in the eastern part due to O 3 titration caused by high NO x emissions from Bonanza Power Plant.Due to shallow mixing layers at the surface, the plume from the power plant's stack does not mix with the air near the ground.Photographic evidence during UBWOS-2013 shows thermally buoyant plumes from Bonanza Power Plant extending well above the shallow PBL during the stagnant meteorological situations (Oltmans et al., 2014b).The WRF-Chem simulation appears to underpredict the Bonanza plume rise by ∼ 150 m and overpredict the vertical spread of the emissions from the power plant.Later, we quantify the contribution of the NO x emissions from this power plant to O 3 at the Horse Pool site within a sensitivity simulation that removes Bonanza Power Plant emissions.It is apparent from Fig. 5a and b that the O 3 mixing ratios transported above the UB, advected predominantly by westerly winds, are low, close to background O 3 levels.and its recirculation in the west-east direction within the UB is qualitatively confirmed by the tethersonde measurements taken at Horse Pool.Moreover, the tethersonde comparisons confirm that the Horse Pool surface O 3 measurements are quite representative of the overall O 3 production in the eastern UB, and that the diurnal pattern of O 3 recirculation to Horse Pool is not localized to just that site.Figure 6a and b also show PBL heights determined by using the measured and simulated virtual potential temperature vertical profiles on 5 February 2013.The PBL height estimates for all the daytime tethersonde launchings during the UBWOS-2013 and their comparison with the model output are provided in the Supplement (Fig. S5).
In order to verify how well the base model case with the top-down emission scenario simulates O 3 at other locations, Figure 7a and b show comparisons of O 3 measurements near the ground at the Ouray and Fantasy Canyon locations (shown in Fig. 1).The observed O 3 concentrations are derived from the tethersonde launches conducted during daytime (09:00-17:00 MST), windowed for the evaluation time period (29 January-8 February 2013).The figures present model O 3 mixing ratios at the times when the tethersonde measurements took place near the ground.The daytime O 3 time series at both sites exhibit a day-to-day variability similar to the Horse Pool measurements.The observed daily maximum O 3 mixing ratios reached ∼ 160 ppb at Ouray and ∼ 140 ppb at Fantasy Canyon on 5 and 6 February 2013, respectively.Later on, maximum daytime O 3 values started decreasing at both sites.The base model case captures the elevated afternoon O 3 concentrations at Ouray and Fantasy Canyon quite well.The time series comparisons at all three surface sites show that the WRF-Chem model is able to simulate the spatial and temporal distribution of daytime O 3 in the eastern portion of the UB reasonably well.

Sensitivity simulations
Here, we present different sensitivity simulations to parse out the relative contribution of various factors to O 3 production within the base case simulation.For this purpose, we performed multiple model simulations for the same stagnation episode (29 January-8 February 2013) using the same model settings and boundary conditions in the 4 km resolution domain as the base case.These simulations were conducted by applying perturbations to the emissions, photochemical mechanisms and physical processes, while all other settings were kept the same as for the base case simulation.We confine our analysis to O 3 at the Horse Pool site.It should be noted that the sensitivities of O 3 to various perturbations are not additive among the various cases, as the O 3 chemistry is very nonlinear.
In order to quantify the impact of the perturbation cases, we use the difference between the base case and the case without the oil/gas emissions (shown in Fig. 3b) as a normalizing constant.We define the impact ratio (IR) for a given simulation case as follows: no oil/gas 3 . (1) Here, we use the median bias (MB) for O 3 mixing ratios for a given model case as the diagnostic variable.Table 4 presents both MB and IR values for all the model cases discussed hereafter.The IR for the base case (B0) is equal to 0, and the IR for the no oil/gas emission case is equal to 1 (B1), by definition.

Sensitivity to the photolysis rates
As discussed in Sect.3, the WRF-Chem model was modified to include the effect of surface albedo due to surface snow.We tested the model with the same settings as in the base case, but with default bare ground surface albedo conditions used.The test simulation (P1) shows that, in this case, the IR value is ≈ 1 (Table 4), and this case fails to capture the O 3 production in the UB.Hence, we cannot simulate the high O 3 levels without taking into consideration the enhancement in photolysis rates due to the reflected solar irradiance from the snow cover.This result is similar to the box model cases reported for VOC sensitive regimes for the regions with high oil and NG emissions (Carter and Seinfeld, 2012;Edwards et al., 2013).We also checked the impact of overhead O 3 column density on the simulated O 3 concentrations at the surface.The case (P2) includes a test simulation where the column O 3 density was set to 350 DU in the TUV scheme, as in the default WRF-Chem code.This 30 % increase reduces modeled O 3 , with an IR for this case of 0.22 (Table 4).Thus, model O 3 at Horse Pool is somewhat sensitive to the overhead O 3 column, but the model can still simulate the strong O 3 buildup in the UB.

Sensitivity to the surface deposition processes
As noted in Sect.3, the presence of a snow layer on the ground also modifies the dry deposition of the gas-phase species.To quantify the contribution of this process to surface O 3 , we performed a simulation where the modeled dry deposition velocities for all the gas species are the same as on bare ground, i.e., the default WRF-Chem model.The IR for this case (D1 in Table 4) is significant (0.48).The model in this case still predicts a strong diurnal cycle in O 3 at the surface, but with daytime maximums reduced: 20 to 35 ppb.
Another simulation (D2) was done by setting the dry deposition velocity of O 3 to zero in the model.During the UBWOS-2013 field campaign, the measured daytime deposition velocities were ∼ 0.002-0.005cm s −1 over the snowy surface.These values are lower than what the model simulations show for O 3 dry deposition velocities (∼ 0.02 cm s −1 )

Sensitivity to the emission perturbations
Since emissions from the oil/gas sector play a crucial role in driving up O 3 in the UB, we further investigate the influence of various emission scenarios on O 3 at Horse Pool.First, we check the sensitivity of O 3 to possible NO x (E1) and VOC (E2) emission reductions.This potentially has an important implication for air quality regulations, to find out what emission reduction strategies are more efficient and cost effective to mitigate O 3 exceedances of NAAQS.The E2 simulation in Table 4 reveals that the reduction of VOC emissions by 30 % reduces O 3 nearly one-to-one on a percentage basis (IR = 0.33).Reducing the NO x emissions from the oil/gas sector by the same amount (E1, IR = 0.01) results in essentially no reduction in O 3 .The simulation reducing NO x emissions by a factor of 3 (E4) slightly reduces model O 3 production (IR = 0.14), but is still significantly less effective than the 30 % VOC reduction case E2.It is only when NO x emissions from the oil/gas sector are completely removed (E5, IR=0.45) that a significant reduction in O 3 occurs without eliminating any VOCs.We additionally note that, when all oil/gas sector NO x emissions are removed, the model still predicts 2 to 3 ppb daytime NO x .The remaining NO x at Horse Pool is 57 % from the towns and highways on the northern edge of the basin, with Bonanza Power Plant accounting for the remainder.Thus, under the high VOC emissions within the UB these levels of NO x are sufficient to sup-port more than half the O 3 buildup seen in the observations and base case model.The model case (E6, IR = 0.03) shows that the impact of NO x emissions from Bonanza Power Plant on O 3 concentrations at Horse Pool is very small.This IR is an upper limit, since the model appears to underpredict the height of the Bonanza source plume, as discussed earlier.A model budget of NO y in the UB deduced from simulations E5 and E6 shows that 66 % of simulated NO y is due to oil and gas activity, 10 % is due to Bonanza Power Plant, and 24 % is due to vehicle and urban emissions (Vernal and Roosevelt, UT) from the edge of the basin.Again, this 10 % contribution from Bonanza can be considered an upper limit to its actual contribution.
Simulations E7 and E8 quantify the contribution of alkane (> C2) and aromatic VOC species to model O 3 by zeroing out the oil/gas sector emissions for each of these VOC classes.The results show their relative contributions to be nearly the same, and comparable to case E5 where all the oil/gas sector NO x emissions are removed.Yet, the VOC measurements (Table 3) and top-down inventory (Table S1 in the Supplement) show that aromatic emissions are much smaller than alkane emissions.On a molar basis total aromatic emissions are only ∼ 8 % of the emissions of all alkanes having greater than or equal to 5 carbon atoms.The disproportionate impact of oil/gas sector aromatics on wintertime O 3 production has been documented by a box modeling study for another basin, the UGRB in Wyoming (Carter and Seinfeld, 2012).
Another uncertainty is the role of primary formaldehyde (CH 2 O) emissions.CH 2 O is largely produced by photochemistry, and has a lifetime of less than a couple of hours during midday.Therefore, it is hard to segregate the fraction of CH 2 O due to primary emissions using atmospheric measurements.The high correlation of daytime CH 2 O with CH 4 (Table S1 in the Supplement) could simply be due to pho-tochemical CH 2 O formation of correlated primary species.The test simulation (E9, IR=0.18) shows a moderate impact of the assumed CH 2 O emission rate.However, CH 2 O is underpredicted by 50 % (Table 3) in the base case simulation that includes the primary emissions.
A modification to the base case simulation was performed to check the impact of the assumed partitioning of oil/gas NO x emissions between NO, NO 2 and HNO 2 by assuming all NO x emissions as NO only (E10 case in Table 4).The E10 case results (IR = 0.23) are close to the E9 case (IR = 0.18), suggesting a small 5 % impact related to this assumption about NO x emission partitioning.This simulation also indicates the insignificant role of primary HNO 2 emissions in simulating the high O 3 episode at Horse Pool.

Sensitivity to photochemical processes
It is known that the major hydroxyl radical (OH) production in urban photochemical smog comes from the reaction of atomic oxygen (O 1 D) with water vapor (H 2 O) molecules.Both H 2 O and O 3 (the photolytic source of O 1 D) are abundant in polluted urban air in summertime.We checked the importance of this pathway in O 3 production in the UB, by shutting off this reaction in the model.The results indicate that elimination of this pathway (C1 in Table 4) in the model reduces O 3 concentrations to some degree, with an IR for this case of 0.25.The model therefore forms significant O 3 through other radical-producing pathways.
Similarly, the model sensitivity to the radical formation channel from CH 2 O photolysis was determined by turning off this photolysis reaction (C2 in Table 4).This pathway has twice the impact of the case C1, illustrating the importance of primary and secondary CH 2 O as a radical source term within the calculations.Additional radical sources, primarily dicarbonyls and hydroxy ketones, are the remaining peroxy radical sources contributing to the high modeled O 3 .The radical production from photolysis of various carbonyls for the UBWOS-2013 are studied in detail by Edwards et al. (2014).
Another uncertainty is the loss of NO x species to particle surfaces.This process depends on aerosol surface area, humidity and other factors.Dinitrogen pentoxide (N 2 O 5 ) formed during the night can participate in heterogeneous aerosol reactions to form nitrate, effectively removing NO x from the system (Brown et al., 2006).The base case simulation does not include aerosols or any treatment of N 2 O 5 conversion to nitrate.In order to estimate an upper limit for the impact of this process, we performed a test simulation in which the chemical reaction between nitrogen trioxide (NO 3 ) and NO 2 is assumed to yield two HNO 3 molecules directly, rather than through the N 2 O 5 intermediate.The results for this case (C3 in Table 4) show that the model could be somewhat (IR = 0.19) sensitive to the treatment of this heterogeneous pathway, but can still simulate high O 3 concentrations under such upper limit assumptions.We note that adding this heterogeneous pathway reduces the positive bias in NO x seen in the top-down model case in Table 3.For the time period of the sensitivity runs, the median of model to observation ratio (MMO) of NO y is 0.64.By adding the heterogeneous pathway the MMO for NO x decreases from 1.06 to 0.88, and the MMO for HNO 3 increases from 0.23 to 0.41, which is more consistent with the magnitude of the 2013 NO y underprediction.

Summary
Our study using the fully coupled meteorology-chemistry WRF-Chem model provides a means of examining the different factors driving high O 3 levels during the wintertime over the UB from a regional modeling perspective.Two highly contrasting wintertime periods simulate snow-free 2012 and snow-covered 2013.The simulations were performed with two different emission scenarios for O 3 precursors from the oil/gas sector, the bottom-up EPA NEI-2011 (version 1) inventory and a top-down emission estimate based on observed ratios of VOCs and NO y to CH 4 combined with the observed CH 4 emission rate from Karion et al. (2013).The same meteorological options and domain configuration within the WRF-Chem model were used.They adequately simulate conditions during both 2012 and 2013 and, most importantly, the low wind speeds, low temperatures, and shallow mixing heights associated with cold pool episodes during the winter of 2013.The high-resolution 3-D model grid and tight coupling between meteorology and tracer transport in WRF-Chem enables a proper simulation of the pollutant accumulation in shallow mixed layers and their transport aloft over the UB.
Comparisons of results contrasting the 2012 and 2013 winter simulations show that model O 3 is very insensitive to differences between the bottom-up and top-down scenarios during low O 3 conditions in 2012, but highly sensitive to these differences during 2013, with only the top-down emission data set able to reproduce the observed high O 3 events in 2013.Total oil/gas sector emissions of NO x are a factor 4 higher, and VOC emissions are 56 % lower, in the bottomup inventory compared to the top-down emission data set in the UB.Statistics for NO x and VOC species are consistent for both years, showing that the bottom-up inventory yields a high bias in simulated NO y and nitrogen containing compounds and a low bias for several VOCs.Using the top-down emissions results in much better agreement with the observed precursor levels, though low biases in VOCs and NO y on the order of 10-30 % are characteristic of the top-down simulations for both years.Under the snow-free conditions of 2012, the meteorological conditions led to effectively coupling air within the UB to air masses further upwind, making O 3 levels less dependent on the local oil/gas sector emissions.In contrast, the cold pool conditions of 2013 confined locally emitted NO x and VOCs to very shallow (50-200 m) PBLs, mak-ing ground level O 3 quite sensitive to the adopted emission estimates and products, particularly the VOC/NO x emission ratio.
The CH 4 emission estimate by Karion et al. (2013), which forms the basis of our top-down inventory, is based on a single flight in 2012, which only sampled CH 4 emissions downwind of operations within the eastern part of the UB.Consistent with the NO x and VOC emissions, the top-down model results show a 10 and 40 % low bias for monthly CH 4 statistics in 2012 and 2013, respectively (Table 3).Assuming that meteorology is adequately represented in the model, one would conclude that the Karion et al. (2013) CH 4 flux estimate of 55 × 10 3 kg h −1 (±30 %), when applied over the entire UB, is a reasonable lower limit.Moreover, NO x and VOC low biases in the top-down emission case could be eliminated by simply increasing the CH 4 basin-wide emission specification by ∼ 25 %.By way of contrast, the bottom-up inventory requires significantly larger modifications in order to eliminate NO y and VOC biases, suggesting a large uncertainty with the methods and numbers used to derive the oil/gas sector emissions found in the NEI-2011 inventory.
A number of perturbation cases to the base model simulation were performed for the 29 January-8 February 2013 high O 3 episode in order to quantify the impact of various model settings, emission reduction scenarios, and photochemical pathways.The impact of meteorological settings applied within the WRF model is not addressed, and is left to future work.The largest O 3 impacts are found to be associated with snow on the ground.Without a high surface albedo and reduced O 3 deposition velocity induced by the snow cover, the model fails to reproduce the high daytime O 3 observed at a surface site.The 3-D transport and photochemistry formulation afforded by the WRF-Chem model allows the relative importance of snow cover to O 3 deposition to be calculated meaningfully, and this is found to be a necessary condition for the high O 3 events.
The NO x and VOC emission reduction cases show that high O 3 episodes in the UB are VOC limited.The impact of NO x reductions is found to be very nonlinear, with reductions showing significant O 3 impacts only after the first 2/3 of the specified oil/gas sector NO x emissions are removed.Our results suggest that efforts to reduce wintertime O 3 from the oil/gas sector through VOC reductions would benefit the most by targeting the less abundant aromatic compounds, as their contribution to high O 3 formation is nearly as significant as the alkanes that comprise the major fraction of the total VOC emissions in the UB.It should be noted that total VOC emissions, especially VOC/NO x emission ratios over urban areas (e.g., Salt Lake City), are much lower than those in the UB; therefore, such urban areas are not expected to experience high wintertime ozone production when cold-pooltype wintertime conditions are present.
Similar to the finding of Edwards et al. (2014), our sensitivity analysis shows that it is important to quantify the primary and secondary sources of CH 2 O in the UB during wintertime.Despite the underprediction of CH 2 O, its photolysis is a dominant source of ozone-producing radicals within the model.Another important conclusion from the sensitivity analysis is that high O 3 concentrations can be simulated without direct emissions of HNO 2 , which was previously considered a potential radical source under snowy conditions.
Several future refinements may be necessary in order to better simulate the meteorological and photochemical processes pertinent to the wintertime conditions.Although using emissions based on observations offers significant improvement over the bottom-up emissions inventory, there are potentially important details missing in the top-down emission estimates.VOC speciation profiles for oil vs. NG wells, and other stationary sources, are known to be different (Warneke et al., 2014), and the treatment of NO x emissions in the topdown approach does not consider the heterogeneity of the various source categories.Thus, a synergy between bottomup and top-down approaches will be needed in order to develop more accurate emission inventories for the oil/gas sector in the UB.Further refinements in gas-phase chemical mechanisms may be needed to better characterize the conditions pertinent to oil/gas emissions in wintertime beyond the chemistry associated with summertime urban photochemistry.
The 3-D meteorology-chemistry model framework and top-down emission estimates presented here provide a unique set of tools and methods to help better understand and quantify the major factors determining high wintertime O 3 levels within the Uinta Basin, UT, and in other oil and natural gas production regions with potentially similar conditions.
The Supplement related to this article is available online at doi:10.5194/acp-15-411-2015-supplement.

Figure 2 .
Figure 2. 24 h averaged anthropogenic emissions of toluene in the Uinta Basin and surroundings: (a) bottom-up (NEI-2011) inventory; (b) top-down emission data set.The letters point to the surface measurement sites: o -Ouray, h -Horse Pool and f -Fantasy Canyon; power plant: B -Bonanza; towns: R -Roosevelt and V -Vernal.The solid line is a border between the states of Colorado and Utah.

Figure 3 .
Figure 3.Time series of the measured and modeled hourly O 3 mixing ratios at Horse Pool in (a) 2012 and (b) 2013.

Figure 5 .
Figure 5. Simulated O 3 distribution and wind vectors over the UB (west-to-east direction in the WRF grid).The Horse Pool and Ouray surface stations along the cross section are shown; (a) early morning (05:00 MST), and (b) afternoon (15:00 MST) on 5 February 2013.The vertical wind components were multiplied by 100 for illustration of the wind vectors.
Figure 6a and b present time series of the observed and modeled O 3 vertical distributions over Horse Pool.The observed distribution was obtained by interpolating the eight O 3 tethersonde measurements for 5 February 2013.The modeled O 3 values were extracted at the time of the vertical profile measurements and then interpolated.The observed temporal variability of O 3 over Horse Pool (Fig. 6a) shows that the strong O 3 buildup starts at ∼ 10:00 MST, and it is confined to a shallow mixing layer (∼ 100 m thick).The O 3 aloft is found over Horse Pool around 10:00-15:00 MST.The model shows similar characteristics for vertical O 3 profiles during the day.A layer aloft with high O 3 is seen at ∼ 350 m above the ground in the model.The depletion of O 3 above is caused by titration of O 3 by Bonanza Power Plant NO x emissions at higher altitudes.Figures 5a, b and 6a, b provide insight into the timing and horizontal and vertical distributions of high O 3 observed near the surface and above in the UB.The model predicted vertical distribution of O 3 , R. Ahmadov et al.: Understanding high wintertime ozone pollution events

Figure 6 .
Figure 6.Vertical O 3 distribution above the Horse Pool site, 5 February 2013: (a) measured by the tethersonde; (b) modeled.There are data from eight tethersonde profiles in panel (a).The O 3 mixing ratios were interpolated for other times of the day.Black circles show PBL heights determined using the vertical gradient of the virtual potential temperature.For the third and fourth tethersonde launchings, the meteorological data are not available to calculate the PBL heights.

Figure 7 .
Figure 7. Time series of the daytime (09:00-17:00 MST) O 3 mixing ratios measured near the surface by tethersondes and modeled at the (a) Ouray and (b) Fantasy Canyon sites in the winter of 2013.The locations of these tethersonde launchings are shown in Fig. 1.Here, the modeled O 3 values are shown solely for the times when the tethersonde measurements were conducted.

Table 1 .
List of acronyms used in the text.

Table 2 .
Anthropogenic emission estimates (in metric tons) for Duchesne and Uintah Counties based on the EPA NEI-2011, NEI-2005 and the top-down estimates for oil and natural gas operations alone."All other activity" excludes the Bonanza Power Plant emissions.NO x emissions are in metric tons/year of NO 2 equivalent.

Table 3 .
Statistics for observed and simulated chemical species mixing ratios at Horse Pool, daytime 09:00-17:00 MST hourly average observations for the UBWOS 2012 and 2013 campaigns; N -number of used hourly observations, MO -median of observations, r-Pearson correlation coefficient, MB -median model-observation bias, MMO -median of model over observation.

Table 4 .
Median bias (MB) and impact ratio (IR) for O 3 at Horse Pool for various perturbations to the base model.Model results are compared against hourly average observations between 09:00 and 17:00 MST, 29 January-8 February 2013.