Measurement data were collected at the Horsepool atmospheric monitoring station in the central Uinta Basin, Utah (40.143° N, 109.469° W; 1569 m above sea level). This remote desert site has minimal influence from urban emissions and is primarily impacted by nearby oil and gas operations, making it an ideal location for investigating their effects on regional air quality and wintertime O3 formation (Lyman et al., 2015; Neemann et al., 2015; Mansfield et al., 2020).

The simulation period (24–27 February 2019) was notable for a strong thermal inversion and the buildup of winter O3. These inversion episodes were characterized by persistent snow cover from January through early March 2019, which contributed to multiday stagnation episodes and the accumulation of O3 near the surface. The inversion layer acted as a barrier, limiting vertical mixing and trapping O3 precursors close to the ground, thereby enhancing local photochemical O3 production (Lyman et al., 2022).

Trace gases were measured at the Horsepool site by drawing ambient air through an unheated PTFE filter pack inlet into an indoor PFA manifold at a flow rate of 10 Lmin-1. O3, NOx, NOy, and CO were measured using Ecotech analyzer models 9810, 9841, 9843, and 9830, respectively; methane and total nonmethane hydrocarbons were measured using a Chromatotec Chroma THC Analyzer; and PM2.5 was measured using a Met One BAM 1020. Weekly calibrations were performed using an Ecotech GasCal system and a Thermo 701H zero air generator. Meteorological parameters, including snow depth (MaxBotix MB7092), total solar radiation (Kipp and Zonen CNR4), wind speed and direction (RM Young 05108-45L), temperature and humidity (Vaisala HMP155), and barometric pressure (Vaisala PTB101B) were recorded with a Campbell CR1000 datalogger, which was checked annually against NIST-traceable standards (Lyman et al., 2022).

Speciated nonmethane organic compounds were also measured, including C2–C10 hydrocarbons, C1–C3 alcohols, and 12 carbonyl compounds. Carbonyl measurements were not available for 2019; therefore, average values from inversion periods in other seasons were used (Lyman et al., 2022). Air samples were collected in Silonite-coated whole-air canisters for hydrocarbons and alcohols and on DNPH (2,4-dinitrophenylhydrazine) cartridges for carbonyls. One 3 h sample was collected daily, with start times alternating between midnight and noon to capture the diurnal extremes of mixing ratios, as midday typically reflects higher concentrations driven by emissions and photochemical activity, while midnight represents lower values when these processes are reduced. DNPH cartridge samples were analyzed by eluting the absorbed carbonyl compounds using a mixture of 75 % acetonitrile and 25 % dimethyl sulfoxide. The resulting extracts were then injected into a high-performance liquid chromatography (HPLC) system equipped with a UV detector to separate, identify, and quantify the carbonyl–DNPH derivatives based on retention times and peak areas. Additional methodological details are provided in Lyman et al. (2020). Whole-air canister samples were preconcentrated using an Entech 7200 with a cold-trap dehydration method and analyzed by gas chromatography with flame ionization detection for C1–C3 hydrocarbons and by gas chromatography–mass spectrometry for the remaining compounds (Lyman et al., 2022, 2020).

The Framework for 0-D Atmospheric Modeling (F0AM) box model version 4.1 (Wolfe et al., 2016; Xiong et al., 2023) was used with an average diurnal cycle configuration to simulate winter O3 production, similar to Lyman et al. (2022). Average hourly meteorological data, CO, and total NOx from the 4 d modeling period were used as inputs for each modeled day, with NOx speciation determined internally by the model. Model inputs included study period average mixing ratios of all organic compounds, including methane, as well as individual hydrocarbons, alcohols, and carbonyls. An albedo of 0.7 was applied for winter conditions, consistent with observations during the study period, and an O3 column of 275 Dobson units was used based on data from the OMI satellite (OMDOAO3e v003) accessed through NASA's Giovanni application (NASA Giovanni, 2024). Variable boundary layer heights were prescribed following Edwards et al. (2014). The same variable dilution constant (kdil) as in Edwards et al. (2014) was initially applied and subsequently adjusted until modeled O3 concentrations were consistent with observed values, following Ninneman et al. (2023).

A subset of the MCMv331 was used, consisting of 3423 chemical species and 10 309 chemical reactions (Saunders et al., 2003; Bloss et al., 2005). This explicit mechanism represents detailed oxidation pathways of measured organic compounds, including stepwise degradation of NMOCs, formation of intermediate carbonyls, radical cycling involving OH, HO2, and RO2, and their interactions with NO_x that drive winter O3 production. HONO was not prescribed as a model input, as its mixing ratio in the Uinta Basin is relatively low (∼0.1-0.2ppb) during winter (Li, 2024; Edwards et al., 2014). Its contribution to HOx production is captured within the MCMv331 mechanism through endogenous gas-phase chemistry.

For a subset of model runs, we introduced heterogeneous chemistry to understand the impact on winter O3 formation due to OH and HO2 uptake onto aerosol surfaces, following the method of Ninneman et al. (2023). In this approach, the uptake of OH and HO2 was represented using a fixed uptake coefficient (γ) of 0.2. The heterogeneous loss rates were calculated as a function of aerosol surface area using observed PM2.5 concentrations and an assumed specific surface area, consistent with the parameterization described in Ninneman et al. (2023).

The output of MCMv331 was compared with several lumped mechanisms, including the RACM2, SAPRC07, and CB6, which consist of fewer species and reactions. Lumped mechanisms simplify atmospheric chemical modeling by aggregating similar species and reactions into fewer representative categories, which reduces computational complexity and enhances efficiency, making them ideal for large-scale air-quality models. Lumped mechanisms allow for faster computation while still capturing essential chemical processes, which is very important for the computational efficiency of large 3-D photochemical models (Zaveri and Peters, 1999).

SAPRC07 and RACM2 use a molecule-based lumping approach. In these mechanisms, organic compounds with similar molecular structure, functional groups, and gas-phase reactivity are grouped into a single surrogate species. The reaction rate constants of each surrogate are typically calculated as weighted averages of the individual compounds included in the lump, based on their chemical reactivity and atmospheric importance. As a result, each surrogate species represents the overall chemical behavior of a group of similar compounds rather than any individual molecule. SAPRC07 consists of 134 surrogate species and 698 chemical reactions, while RACM2 includes 124 surrogate species and 363 chemical reactions. In contrast, the CB6 mechanism follows a structure-based lumping approach. Instead of representing whole molecules, CB6 conserves carbon–carbon bond types and functional carbon groups. Organic compounds are decomposed into carbon bond fragments such as alkyl groups, double bonds, and aromatic rings, which are then assigned to surrogate species based on bond type. Chemical reactions track the transformation of these carbon bonds during oxidation processes, allowing CB6 to represent organic carbon chemistry with fewer species while maintaining carbon conservation. The CB6 mechanism includes 77 surrogate species and 216 chemical reactions. All three of these lumped mechanisms retain some explicit organic species, including formaldehyde. SAPRC07 and RACM2 retain more explicit species than CB6. The version of CB6 used in this study was CB6r2 (Cao et al., 2021). In CB6r2, HO2NO2 (denoted PNA) is represented explicitly, forming from HO2 and NO2 and removed by thermal decomposition and photolysis, in the same way as in the other mechanisms. HO2NO2 decomposition is the largest single contributor to HOx production in CB6 (see Sect. 3.1), consistent with its role in MCMv331, SAPRC07, and RACM2.

To compare these mechanisms with MCMv331, species mapping was performed to aggregate and convert measured organic chemical species into the corresponding species in the lumped mechanisms. Species properties from the United States Environmental Protection Agency (EPA) SPECIATE database version 5.2 were used, along with a species conversion database provided by Ramboll (EPA, 2024). The species conversion database is provided as Data File S1 in the Supplement. When more than one MCM species corresponded to a single lumped mechanism species, their mixing ratios were summed to obtain the lumped species concentration.

Hydrocarbons in the Uinta Basin are primarily emitted from oil and gas activities (Lyman et al., 2015; Mansfield et al., 2020; Warneke et al., 2014). Because the hydrocarbon dataset was temporally limited, their mixing ratios were held constant throughout the simulation period. On the other hand, carbonyl compounds can be emitted directly or formed secondarily through photochemical reactions of hydrocarbons. To accurately represent atmospheric conditions and allow for realistic secondary carbonyl production, optimized emission fluxes (EFs) were developed for carbonyl compounds. Initial mixing ratios were set to measured values, while subsequent modeled mixing ratios were allowed to vary over time. Emission flux values were iteratively adjusted to achieve minimal increasing or decreasing trends in mixing ratios across the four modeled days. If the Day 4 mixing ratio of a given carbonyl compound was lower than the Day 1 value, the initial EF was considered too low and was incrementally increased until the trends aligned as closely as possible. Conversely, if mixing ratios increased throughout the simulation, the EF was considered too high and was reduced, in some cases to zero. An EF of zero implied no local emissions and that all carbonyls, except those initially present, were formed from atmospheric photochemical reactions (Maasakkers et al., 2019). Through systematic adjustment of EF values, the modeled carbonyl mixing ratios were constrained to reflect observed atmospheric behavior, thereby improving the reliability of the simulation results.

To investigate the influence of individual carbonyl compounds on winter O3 production, the F0AM box model was initially run without any modifications to establish baseline O3 production. The initial mixing ratio of a selected carbonyl compound was then set to zero and constrained to remain at zero in a subsequent model run. The average O3 production rate (ppbh-1) between 09:00 and 16:00 local time, MST on the fourth day of the simulation was compared between the two scenarios. This time window was selected to capture periods of consistent photochemical activity, during which carbonyl compounds contribute to radical production and daytime O3 formation (Pang and Mu, 2006). Differences in O3 production rates resulting from forcing the mixing ratio of a given carbonyl compound to zero were used to quantify its influence. This procedure was systematically repeated for all measured carbonyl compounds, and the resulting outputs were compared to determine compound-specific contributions to O3 production.

A sensitivity analysis was conducted to assess the impacts of primarily emitted organic groups (alkanes, alkenes, alkynes, alcohols, and aromatics) on carbonyl production. The analysis focused on selected carbonyl compounds, including formaldehyde, acetaldehyde, methacrolein, benzaldehyde, and methyl ethyl ketone, which were chosen based on their significant influence on winter O3 formation. Initial mixing ratios of the primary organic groups were varied by ±50%, while maintaining constant speciation within each group, and the corresponding percentage changes in average mixing ratios of carbonyls were evaluated between 13:00 and 17:00 local time, MST on the fourth day of the simulation. This time window was selected to avoid rapid morning changes in boundary layer height that could introduce variability in pollutant mixing ratios (Lapworth, 2006; Mahrt, 1981). During the afternoon period, the atmosphere is more homogeneous, allowing for an even distribution of precursors and reaction products. It also represents a steady-state condition for radical species (⋅OH, HO2⋅, and RO2⋅), ensuring a balanced production and loss cycle (Lew et al., 2020; Heard et al., 2004). This approach allowed us to evaluate the sensitivity of carbonyl formation to fluctuations in levels of primary organic groups.

A similar sensitivity analysis was performed to determine the influence of primarily emitted organic groups on the O3 production rate, which allowed us to identify the individual impact of each organic compound group on winter O3 production.

The study period (24–27 February 2019) was characterized by a persistent thermal inversion, continuous snow cover, and the progressive accumulation of O3 near the surface, as shown in Fig. S1 in the Supplement. On the fourth day, the daily maximum 8 h average O3 mixing ratio reached 94 ppb and the maximum hourly value reached 102 ppb, both exceeding the U.S. Environmental Protection Agency standard of 70 ppb. These conditions are representative of severe wintertime O3 episodes in the Uinta Basin, where high NMOC emissions from oil and gas operations are trapped by stagnant boundary layers and limited sunlight. The negative relationship between O3 and relative humidity is shown in Fig. S2. In comparison, the F0AM Box model with the explicit chemical mechanism MCMv331 estimated a maximum O3 mixing ratio of 107 ppb on the fourth day of the simulation (Fig. 1). After incorporating heterogeneous chemistry based on MCMv331, the O3 level decreased by 1 ppb. The SAPRC07 lumped mechanism estimated the O3 level at 111 ppb, while RACM2 predicted slightly lower than the measured value at 100 ppb, and CB6 provided the highest estimation of 122 ppb. Liu et al. (2023) conducted a Box model study with commonly used chemical mechanisms during summertime in multiple cities of China and identified that RACM2 showed the best agreement with observation during the polluted period, followed by MCMv331 and SAPRC07. These findings aligned with our studies. On the other hand, Shareef et al. (2022) evaluated different lumped chemical mechanisms with CMAQ (3D chemical transport model) on winter O3 prediction in Alberta, Canada, and found no significant differences in SAPRC07, RACM2, and CB6 output.

The O3 budget on Day 4 shows a consistent pattern across all four mechanisms (Tables S8.1–S8.4 in the Supplement). O3 production is controlled almost entirely by the reaction between ground-state oxygen atoms and molecular oxygen, contributing 99 %–100 % of total formation, while reactions involving organic peroxy radicals and HO2 contribute less than 0.01 %. O3 loss is driven mainly by photolysis, which accounts for 75 %–78 % of total removal, followed by the reaction of NO with O3, which contributes 18 %–21 %. This budget structure reflects the mechanistic representation of O3 formation in F0AM, in which O+O2→O3 is the final step common to all pathways. A more process-oriented view frames O3 production in terms of the HO2+NO and RO2+NO reactions that convert NO to NO2, since each such conversion yields an O3 molecule upon subsequent NO2 photolysis. The integrated rates of these radical-driven conversions (Table S8.5) track the simulated O3 levels across mechanisms, being lowest in RACM2 (26.6 molec.cm-3) and highest in CB6 (36.7 molec.cm-3). HO2+NO contributes 43 %–51 % and RO2+NO 49 %–57 % of this radical-driven conversion across the four mechanisms. This conversion is sustained by radical production, whose total rate differs substantially across mechanisms (Fig. 3 and Tables S9.1–S9.4). CB6's HOx production (38.7 molec.cm-3) is roughly 1.6–1.8 times that of MCMv331, SAPRC07, and RACM2 (21.2, 23.7, and 22.1 molec.cm-3), driving faster NO to NO2 cycling and ultimately higher simulated O3.

Carbonyl compounds are examined here as intermediaries in O3 production, as their photolysis and oxidation reactions generate HO2 and RO2 radicals that drive NO to NO2 conversion and ultimately O3 formation. Figure 2 shows that O3 was more sensitive to changes in carbonyl mixing ratios in the lumped mechanisms compared to MCMv331. Master Chemical Mechanism (MCM) employs an explicit representation of reactions, wherein carbonyl compounds degrade through multi-generational, branched pathways that distribute radical formation across several intermediate species and slower reaction steps (Saunders et al., 2003). On the other hand, with simplified chemical mechanisms, carbonyl chemistry is formulated to produce O3, forming radicals like HO2 rapidly and directly, often in a single reaction step. This fundamental difference in mechanism structure may explain the higher O3 sensitivity to carbonyls observed with the lumped mechanisms compared to MCMv331 (Goliff et al., 2013; Yarwood et al., 2010).

To understand the mechanistic basis for the differences in carbonyl-driven O3 sensitivity observed across mechanisms in Fig. 2, the modeled day 4 HOx production and loss budgets across the four chemical mechanisms are illustrated in Fig. 3, expressed as percentages of each mechanism's total (Table S9 provides more information). Across all mechanisms, thermal decomposition of peroxynitric acid (HO2NO2) is the single largest HOx production pathway, accounting for 23 %–38 % of total production, with CB6 at the upper end of this range (37.8 %). The dominance of HO2NO2 thermal decomposition is physically consistent with the cold wintertime conditions of the Uinta Basin, where low temperatures (∼0°C) allow HO2NO2 to accumulate overnight as a radical reservoir that is released during daylight hours, consistent with findings by Edwards et al. (2014) at the same site. Organoperoxy radical (RO2)+NO reactions were far more important for the lumped mechanisms compared to MCMv331, but this is likely an artifact of collapsing multi-step reaction sequences in lumped mechanisms, leading to higher budget percentages from fewer reaction pathways. In the HOx loss budget, HO2+NO2 termination dominates in all mechanisms (25 %–40 %), and OH reactions with alkanes represent the second-largest sink (9 %–33 %), consistent with the dominance of alkane emissions in the Uinta Basin. OH-alkane reactions appear less important for MCMv331 in the figure because it catalogs reactions with specific alkanes, so most OH-alkane reactions were not among the top five most important reactions and were instead included as Other. Notably, the OH + benzaldehyde loss term appears in the MCMv331 top-5 reactions (approximately 3 % of total HOx loss) but is absent from the RACM2 and CB6 budgets, providing a direct mechanistic basis for the negative O3 sensitivity to benzaldehyde observed in MCMv331 and SAPRC07 but not in the lumped mechanisms (Fig. 2 and Table 3).

Notwithstanding these caveats, all four mechanisms have fairly similar HOx production budgets. Total HOx production rates were similar for MCMv331, SAPRC07, RACM2 (21.2, 23.7, and 22.1 molec.cm-3, respectively), while CB6 radical production was notably higher, at 38.7 molec.cm-3. Higher radical production in CB6 may explain the higher ozone mixing ratios modeled by CB6, as shown in Fig. 1.

The F0AM box model with MCMv331 identified formaldehyde as the dominant contributor to the O3 production rate, accounting for a change of 0.20 ppbh-1, or approximately 50 % of the total contribution from carbonyl compounds. This finding aligns with Edwards et al. (2014), who reported formaldehyde as the primary daytime radical source in the Uinta Basin, contributing 30 % of radical formation from carbonyls, with the remainder attributed to larger carbonyl species. Across all chemical mechanisms examined, both the production and loss of formaldehyde lead to substantial HO2 radical formation through Reactions (R6) and (R7), which play a critical role in O3 production. The combined HO2 production from formaldehyde accounts for approximately 8 %–20 % of total hydrogen oxide radicals (HOx) production across the mechanisms considered (Table 1). In MCMv331, acetaldehyde was found to be the second-highest contributor in our study, causing a change of 0.06 ppbh-1 in the O3 production rate, with SAPRC07 and RACM2 following a similar trend. Overall, acetaldehyde contributes roughly half of the HO2 produced by formaldehyde (Tables 1 and 2). In addition, reactions with acetaldehyde lead to a substantially greater loss of OH, an important radical in O3 production, compared to formaldehyde. Oxidation of acetaldehyde by OH forms the acetyl peroxy radical, which can be sequestered as peroxyacetyl nitrate under cold conditions. This compound is thermally stable at low temperatures and can transport NO_x and radicals over long distances, thereby reducing their availability for local O3 production (Fischer et al., 2014).

In CB6, most medium and high molecular weight carbonyls are lumped into the single “KET” species (Yarwood et al., 2010), so KET represents a much larger carbon pool than any individual carbonyl. As shown in Table S10.4, almost all KET is formed through the alkoxy-radical pathway (ROR → KET, ∼99%), which strongly promotes radical propagation. At the same time, formaldehyde in CB6 undergoes a substantial HO2 removal reaction (FORM + HO2, ∼31% of its loss; Table S10.4), limiting its radical yield. As a result, KET has a much stronger effect on O3 production than formaldehyde in CB6.

The relative contributions of each carbonyl compound remained largely unchanged after introducing heterogeneous chemistry into the model (Fig. S4). This is likely due to low specific humidity during the simulation period, which limits the uptake of HOx radicals by aerosols. As a result, the radical budget and hence O3 production potential were not significantly altered (Fig. 1), preserving the dominance of formaldehyde and acetaldehyde among carbonyl precursors.

The impact of benzaldehyde on O3 production differs across chemical mechanisms. In MCMv331, benzaldehyde shows a negative contribution of -0.03ppbh-1, indicating that increasing benzaldehyde reduces the O3 production rate, with a similar negative response observed in SAPRC07. In both MCMv331 and SAPRC07, benzaldehyde undergoes detailed aromatic degradation that consumes radicals or forms less reactive products. As a result, benzaldehyde competes for OH and acts as a radical sink, leading to increased HOx loss (Table 3). In MCMv331, phenoxy radicals formed from benzaldehyde directly consume O3, further contributing to the negative impact on O3 production. In contrast, in RACM2, the OH + benzaldehyde reaction is much less important relative to the total HOx budget, so increasing benzaldehyde does not significantly increase HOx loss. Instead, oxidation of benzaldehyde in RACM2 forms peroxybenzoyl radicals that enhance NO to NO2 cycling, resulting in net positive O3 production.

Edwards et al. (2014) attributed 85 % of radical production to the photolysis of carbonyl compounds during a 2013 winter O3 episode at the same Uinta Basin location. However, studies have shown that the emissions in the Uinta Basin have declined since the Edwards et al. study (Mansfield et al., 2020; Lin et al., 2021), and during our study, we found that the initial mixing ratios of several carbonyls were lower compared to those of Edwards et al. on the fourth day of the simulation period (Table S1). Furthermore, the emission flux required to simulate representative carbonyl mixing ratios was minimal or zero. The addition of carbonyl emissions in the box model led to an overestimation of their mixing ratios, indicating that carbonyls were mostly secondary pollutants, created in the atmosphere from photochemistry, rather than emitted directly from sources. Stockwell et al. (2011) and Nogueira et al. (2017) also identified carbonyl compounds mainly as secondary pollutants generated through the oxidation of hydrocarbons from anthropogenic or biogenic emissions.

Since carbonyls were found to be largely secondary pollutants, their formation is traced here to primarily emitted hydrocarbon groups. Understanding which hydrocarbon groups drive carbonyl production is essential for identifying the key precursors of wintertime O3. Figure 4 shows the production pathways for formaldehyde and acetaldehyde, the two dominant carbonyl O3 precursors, across the four mechanisms (additional carbonyl production and loss budgets are available in Table S10). For formaldehyde, the methyl peroxy radical + NO reaction (CH3O2+NO→HCHO+HO2) accounts for 55 %–60 % of production in MCMv331, SAPRC07, and RACM2. CB6 shows a notably different pattern, with the CH3O2+NO contribution falling to 36 % while alkene oxidation rises to nearly 20 %. For acetaldehyde, all four mechanisms show strong dominance of alkyl peroxy, and alkoxy radical pathways derived from alkane oxidation, with ethoxy radical decomposition (C2H5O⋅→CH3CHO+HO2) accounting for approximately 75 % of production in MCMv331 and similar lumped pathways dominating in SAPRC07 and RACM2. CB6 is again distinctive, with a significant contribution from its lumped ketone species (KET → ALD2, approximately 17 %; denoted as aldehyde photolysis in the figure), reflecting the broader lumping of C3+ alkane oxidation intermediates into a single ketone pool in that mechanism. Also, Fig. 4 shows 99 % of acetaldehyde production in the SAPRC07 mechanism comes from ethoxy radical decomposition, while the other mechanisms show a significant contribution from larger organic oxy or peroxy radical species. SAPRC07 actually uses a system of “yield tokens” as outcomes of upstream reactions, obscuring the radical species that generated acetaldehyde. In the absence of information about radical species, we assigned all to ethoxy radical decomposition in the figure. Total formaldehyde production rates were 3.31, 3.25, 3.88, and 4.28 molec.cm-3 for MCMv331, SAPRC07, RACM2, and CB6, respectively; total acetaldehyde production rates were 2.12, 3.77, 4.02, and 2.86 molec.cm-3 for the same mechanisms.

Figure 5 illustrates the contribution of various hydrocarbon groups to the formation of key carbonyl compounds, including alkanes, alkenes, alkynes, aromatics, and alcohols. The MCMv331 results indicated that formaldehyde formation is influenced by a wide variety of NMOC groups (4 %–7 % each) because, as shown in Fig. 4, its dominant production pathway (R6, 55.3 %) relies on the methyl peroxy radical (CH3O2) (Reactions R1–R3), a universal intermediate generated from oxidation of multiple organic classes, including alkane oxidation via fragmentation of larger alkoxy radicals, alkene ozonolysis through Criegee intermediate decomposition, and aromatic ring-opening reactions. In addition, formaldehyde is produced through several direct pathways independent of CH3O2, such as terminal alkene ozonolysis, methanol oxidation, and glyoxal photolysis (Saunders et al., 2003).

Alkanes are generally less reactive than other NMOC groups; however, due to their prominence in the Uinta Basin atmosphere, they still play a dominant role in the formation of formaldehyde and other carbonyl compounds. The mixing ratios of acetaldehyde, benzaldehyde, and methyl ethyl ketone increased by 24 %, 28 %, and 20 %, respectively, in response to a 50 % increase in the initial mixing ratio of total alkanes. Ethane reacts with the OH radical to form the ethoxy radical (C2H5O⋅) (R9), which is then oxidized to produce acetaldehyde (Reaction R10). This reaction pathway is the dominant source of acetaldehyde, accounting for 74.9 % of its total formation. Isopentane also acts as a significant contributor to acetaldehyde formation (Table S4), feeding into the same pathway. Its branched structure promotes alkoxy radical decomposition, which releases ethyl radicals that subsequently produce acetaldehyde through the same ethoxy radical pathway (Reactions R9–R10) as ethane. There are no such comparable pathways available for other NMOC groups, making alkane oxidation the primary contributor to acetaldehyde production. Ethanol oxidation forms some acetaldehyde, but ethanol mixing ratios were low in the model, and Figs. 4 and 5 show this to be a minor pathway. R9C2H6+OH⋅→C2H5O⋅R10C2H5O⋅+O2→CH3CHO+HO2⋅R11Isoprene+OH→MethacroleinR12C6H5O⋅+O3→C6H5O2 Methacrolein production is typically associated with isoprene oxidation (Reaction R11); however, isoprene mixing ratios are negligible in the wintertime Uinta Basin due to plant dormancy. Budget analysis confirms that methacrolein is introduced almost entirely through primary emissions (Table S10.1). A 50 % increase in alkane mixing ratios resulted in an 83 % increase in methacrolein mixing ratio in MCMv331, the largest response among all carbonyl compounds (Fig. 5). This sensitivity arises because OH radicals account for 80 % of total methacrolein destruction, and alkane oxidation is simultaneously the largest OH sink in the system. Increasing alkane concentrations enhances OH consumption, reducing steady state OH and suppressing methacrolein destruction. Other carbonyls such as formaldehyde and acetaldehyde are less affected because their chemical production also decreases when OH is reduced, partially offsetting the slower destruction. Methacrolein, with no chemical production source, lacks this counterbalance. A 50 % increase in aromatics reduced methacrolein by 9 % due to phenoxy radical-driven HOx suppression. The O3 budget reveals that phenoxy radicals directly destroy O3 (R12, 2.31 moleculecm-3, 0.19 % of total O3 loss), reducing the primary OH source (O3+hv→O(1D)→2OH). Additionally, phenoxy radicals react with NO2 to form nitrophenols (C6H5O+NO2→HOC6H4NO2), which terminate radical chains and sequester NOx, thereby suppressing the HO2+NO→OH+NO2 recycling pathway. These combined effects reduce OH availability for isoprene oxidation, resulting in decreased methacrolein production.

Sensitivity analyses following the same method with the SAPRC07 and RACM2 (Figs. 6 and 7) chemical mechanisms showed similar trends as MCMv331, except aromatics had a more substantial positive effect on benzaldehyde formation than alkanes, according to RACM2. Reactions (R13) and (R14) represent the formation of benzaldehyde from benzene in RACM2. Here, benzene reacts with OH to produce an epoxy muconate intermediate, which reacts with O3 to directly produce benzaldehyde (R14, accounting for 72.1 % of benzaldehyde production). In contrast, in MCMv331, ring-opening chemistry produces dicarbonyls (glyoxal, butenedial) that cannot form benzaldehyde because the aromatic ring is destroyed (Reaction R15). In MCMv331, benzaldehyde production requires the intact benzene ring and occurs only through the minor side-chain oxidation pathway (R16–R17, 0.7 % of production), with 99.3 % coming from direct emissions. This mechanistic difference explains why increasing aromatics has a stronger positive effect on benzaldehyde in RACM2 than in MCMv331. R13Benzene+OH→Epoxy MuconateR14Epoxy Muconate+O3→Benzaldehyde+ProductsR15Benzene+OH→Glyoxal+ButenedialR16Toluene+OH→Benzyl RadicalR17Benzyl Radical→Benzaldehyde+HO2 The CB6 mechanism is based on carbon bond types rather than explicit chemical species, resulting in fundamentally different formaldehyde sensitivity compared to MCMv331 (Fig. 8). In MCMv331, alkane oxidation produces methoxy radicals (CH3O) through sequential fragmentation, which decompose directly to formaldehyde (Reaction R6). In contrast, CB6 represents alkanes as paraffinic carbon bonds (PAR), and their oxidation produces a lumped alkoxy intermediate (ROR) that decomposes exclusively to ketones (KET), acetaldehyde, acetone, and higher aldehydes but not formaldehyde. Meanwhile, alkenes in CB6 directly produce formaldehyde with high yields (Ethene+OH→1.56 Formaldehyde). Since alkane oxidation consumes 34 % of OH without producing formaldehyde, increasing alkanes reduces OH availability for alkene oxidation, resulting in decreased formaldehyde production. Acetaldehyde and ketone formation remain alkane-dominated in CB6, consistent with other mechanisms. This carbon-bond lumping approach fundamentally alters formaldehyde's NMOC sensitivity in CB6 compared to more explicit mechanisms like MCMv331 and SAPRC07.

The overall contribution of primarily emitted hydrocarbon groups to wintertime O3 production was quantified through a sensitivity analysis using the same chemical mechanisms as before. This analysis completes the chain from primarily emitted hydrocarbons through their influence on carbonyl formation to O3 production. It also provides the basis for identifying emission reduction targets. Figure 9 reveals that both MCMv331 and SAPRC07 chemical mechanisms exhibit similar chemical behavior, with alkanes having the most significant influence on winter O3 formation. A 50 % increase in the mixing ratio of alkanes resulted in approximately a 0.29 and 0.35 ppbh-1 increase in O3 production rate on the fourth day of the simulation with the MCMv331 and SAPRC07 mechanisms, respectively. The dominant role of alkanes in winter O3 formation can be attributed to their high ambient mixing ratio and efficient radical propagation chemistry. Although alkane reactions with OH radicals have moderate rate constants, their elevated mixing ratios enable efficient production of alkyl peroxy radicals. These peroxy radicals subsequently propagate the NO to NO2 conversion cycle, ultimately contributing to O3 production (Reactions R1–R5). While this fundamental chemistry is represented in all mechanisms, RACM2 underestimates the full impact of alkane degradation due to over-simplification through species lumping. Compared to MCMv331, RACM2 showed a 76 % reduction in parent alkane species and an 86 % reduction in peroxy radical species after lumping. This simplification resulted in a 34 % reduction in HO2 radical yields from alkane oxidation, which consequently reduced the apparent contribution of alkanes to O3 production relative to MCMv331.

In MCMv331 and SAPRC07, aromatics emerged as the second most important precursor for winter O3 formation due to their efficient sequential multi-step oxidation chemistry. The initial OH-addition to the aromatic ring produces bicyclic peroxy radicals, which subsequently undergo ring-opening to form highly reactive dicarbonyl compounds such as glyoxal and methylglyoxal. These dicarbonyls photolyze to produce additional HOx radicals, creating a radical amplification chain that significantly enhances O3 production per aromatic molecule oxidized. Simulations using RACM2 showed a similar magnitude of aromatic-induced O3 production change compared to MCMv331 and SAPRC07. The CB6 mechanism identified alkanes as the major contributor, similar to MCMv331 and SAPRC07. However, it ranked alkenes and alkynes, rather than aromatics, as the second-highest contributors to winter O3 production, diverging from the trends observed in the other chemical mechanisms.

Although alkanes showed the largest contribution to O3 production, a 50 % increase in alkane mixing ratios resulted in only a modest change of about 0.2–0.4 ppbh-1 in the O3 production rate on day four of the model run across the chemical mechanisms. This limited response likely occurs because both physical and chemical constraints begin to develop by the fourth day of the simulation (Fig. S1). As the inversion persists, the accumulation of precursors approaches a balance between emissions into the basin and loss through air exchange at the top of the inversion, which limits further increases in O3 forming compounds (Edwards et al., 2014). At the same time, highly reactive NMOCs are consumed during the early days of the episode, and the remaining NMOC mixture is dominated by slower reacting species that are less efficient at producing O3 (Koss et al., 2015; Lyman et al., 2018). Further, more NO_x is consumed as photochemical activity is enhanced day upon day, leading to less NO_x available to produce O3 (Lyman et al., 2018). Thus, increasing precursor mixing ratios by 50 % has a much smaller effect on O3 production on day four than during the earlier days.

Previous box model and observational studies indicate that alkanes can play a critical role in wintertime O3 production in oil and gas–influenced regions (Chen et al., 2020; Koss et al., 2015). In Utah's Uintah Basin, light alkanes are the most abundant in NMOC (Table S2). Although light alkanes react more slowly with OH radicals than aromatics, their high ambient mixing ratios cause them to dominate the NMOC mixture by volume and account for approximately 70 % of OH-initiated hydrocarbon oxidation (Koss et al., 2015). Studies in other oil and gas producing regions similarly report that light alkanes contribute about 60 % of total OH reactivity and play an important role in the formation of carbonyl compounds, which serve as major radical sources for O3 production (Gilman et al., 2013; Edwards et al., 2014; Field et al., 2015).

The contribution of different NMOC groups to the O3 production rate remained largely unchanged after incorporating heterogeneous chemistry in the box model simulation using MCMv331, consistent with the explanation provided in Sect. 3.2 (Fig. S5). Figure S5 shows that a 50 % increase in the mixing ratio of each NMOC group with heterogeneous chemistry resulted in nearly similar magnitudes of change in O3 production rate as without heterogeneous chemistry.