The NOAA Twin Otter sampled nine wildfires with 39 flights between 3 August and 5 September 2019 in the western US. The aircraft was based mainly in Boise, ID, and briefly in Cedar City, UT. The NOAA Twin Otter payload limited flight duration to 3.0 h or less, and the aircraft typically flew 2–3 times in a day to achieve plume sampling from midafternoon into the night. Aircraft speed was 71.8±3.8ms-1 (average ±1-σ), which yields a horizontal resolution of ∼70 m for the in situ 1 s measurements. Attempts to probe the same air mass downwind, known as Lagrangian sampling, proved difficult to achieve due to complex plume structure, terrain, and airspace. Therefore, we define the sampling strategy as semi-Lagrangian. Even so, estimated emission times (calculated from estimated plume ages) suggest that smoke sampled on successive intercepts at the Castle fire and 204 Cow fire (simply referred to as Castle and Cow, respectively, from here on) plume centers was emitted within 3 and 10 min time periods, respectively. However, plume age uncertainties for the Cow plume are large (Table S2 in the Supplement).

This analysis uses NOAA Twin Otter observations of BBVOCs and HONO from a University of Washington iodide high-resolution time-of-flight chemical ionization mass spectrometer (UW I- HR ToF CIMS, 2 Hz; Lee et al., 2014) as well as a Tenax cartridge sampler with subsequent GC×GC analysis for speciated BBVOCs (intermittent transect integrations; Hatch et al., 2015; Mondello et al., 2008), which we use to support mass assignments from the UW I- HR ToF CIMS for some phenolic compounds (see the Supplement).

We use data from a commercial cavity ring-down spectrometer (Picarro G2401m) for measurements of CO, CO2, and CH4 (0.5 Hz; Crosson, 2008). We use measurements from a custom chemiluminescence instrument (CL, 1 Hz) for NO, NO2, and O3 (Sparks et al., 2019). Aerosol surface area measurements were collected by an ultrahigh-sensitivity aerosol spectrometer (UHSAS, 1 Hz; Kupc et al., 2018). The UHSAS data were corrected for coincidence up to a factor to 1.4, following the method described in Kupc et al. (2018). The sample for the UHSAS was diluted up to a factor of 2.9 for part of the flights to increase accuracy at higher concentrations. The aircraft had a standard meteorological probe (Aventech ARIM 200) for temperature, pressure, relative humidity, wind speed, and direction. We use NO2 photolysis rates (jNO2) collected by upward- and downward-facing jNO2 filter radiometers (Metcon GmbH, 1 Hz; Kupc et al., 2018; Warneke et al., 2016).

The NASA DC-8 aircraft sampled 14 wildfires in the western US while based in Boise, ID, as well as about 90 prescribed agricultural southeastern US fires while based in Salina, KS, between 22 July and 5 September 2019. Aircraft speed was 167.2±3.4ms-1, which yields a horizontal resolution of ∼167 m for the in situ 1 s measurements. Similar to the NOAA Twin Otter, sampling was semi-Lagrangian. However, smoke emission times for the plume centers of Williams Flats fires 1 and 2 (referred to as WF1 and WF2 from here on) covered a larger time period (∼30–60 min) compared to the NOAA Twin Otter (Table S2 in the Supplement).

In this analysis we use measurements of CO from a tunable diode laser spectrometer (1 Hz; Sachse et al., 1991) when available and from a cavity enhanced spectrometer (CES, 1 Hz; Eilerman et al., 2016) when unavailable. In the fires investigated here, both instruments agree well within <1 %. Measurements of NO2, NOy, and O3 are provided by a NOAA chemiluminescence (1 Hz; Pollack et al., 2010; Ridley et al., 1992; Stedman et al., 1972) instrument. When measurements of NO2 by the NOAA CL instrument are unavailable, we use measurements by a NOAA CES (1 Hz; Min et al., 2016). These two measurement methods of NO2 agree within 12 % for the fires we investigate. We use measurements of NO by a laser-induced fluorescence instrument (1 Hz; Rollins et al., 2020). Measurements of BBVOCs and HONO are taken from the NOAA I- ToF CIMS (1 Hz; Neuman et al., 2016; Veres et al., 2020) as well as the University of Innsbruck proton transfer reaction time-of-flight mass spectrometer (UIBK PTR ToF MS; Müller et al., 2014). Peroxyacetyl nitrate measurements were performed by a thermal dissociation CIMS (1 Hz, Ro Lee et al., 2020). Aerosol surface area measurements are taken from a scanning mobility particle sizer and laser aerosol spectrometer (SMPS and LAS, 1 Hz, LAS, 2021; Moore et al., 2021; SMPS, 2021). Spectrally resolved actinic flux was measured with separate upward- and downward-facing actinic flux optics (CAFS, 1 Hz; Shetter and Müller, 1999). These fluxes were used to calculate photolysis rates using the photochemistry routine contained in the NCAR TUV model (v5.3.2, https://www2.acom.ucar.edu/modeling/tropospheric-ultraviolet-and-visible-tuv-radiation-model, last access: 24 October 2021).

Plume age estimates are made by air parcel trajectories computed in the HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) model with multiple high-resolution meteorological datasets (HRRR 3 km, NAM CONUS nest 3 km, and GFS 0.25∘). These estimates account for buoyant plume rise as well as horizontal advection. Uncertainties in plume age are determined from spread between the meteorological datasets, mismatch between observed and archived winds, and trajectory spatial error in missing the known fire source. Typical uncertainties are 25 % of the estimated age (Holmes et al., 2020).

Our model uses the Master Chemical Mechanism (MCM, v3.3.1 via http://mcm.york.ac.uk, last access: 23 September 2021), in conjunction with a NOAA biomass burning mechanism included in F0AM v4.0 (Bloss et al., 2005; Coggon et al., 2019; Decker et al., 2019; Jenkin et al., 1997, 2003, 2012, 2015) and updates to OH- and NO3-initiated oxidation of phenolic compounds (Bolzacchini et al., 2001; Calvert et al., 2011; Finewax et al., 2018; Nakao et al., 2011; Olariu et al., 2002, 2013; Schwantes et al., 2017). Briefly, we update the phenolic oxidation product yields of catechol, methylcatechol, and three dimethylcatechols reacting with NO3 and OH. Further, we expand the phenolic oxidation pathways in the MCM from 50 to 140 reactions by extrapolating analogous branching ratios, rate coefficients and products from studies of phenol and cresol oxidation (see the Supplement).

We initiate the model, as discussed in Sect. 2.3.3, using an emissions inventory of 302 BBVOCs in the form of emission ratios (ERs). 1ERx=XppbvCOppmv, Note that an ER is used to describe an emission (when smoke age=0) and is different than a normalized excess mixing ratio (defined in Sect. 2.4.1) used to describe observations when smoke age>0. The ER inventory is described in detail in Decker et al. (2019) and uses an average of BBVOC emission ratios of ponderosa pine fuel from the Fire Lab at Missoula Experiment (FLAME-4) (Hatch et al., 2017) and the Fire Influence on Regional and Global Environments Experiment (FIREX lab) (Koss et al., 2018) with rate coefficients taken from literature when available or estimated when unavailable. Approximately 250 BBVOCs in the inventory are not included in the MCM and do not have published mechanisms. Therefore, reactions of those compounds with NO3, OH, and O3 lead to a generic product.

The model includes heterogeneous NO3 and N2O5 uptake onto aerosol, calculated for NO3 heterogeneous reactivity, as 2kNO3aerosol=Keq[NO2]kN2O5+kNO3+aerosol, where kNO3aerosol is a first-order rate coefficient, Keq is the equilibrium rate constant for Reaction (R4), and kNO3+aerosol is a first-order rate coefficient for uptake expressed below. Note, however, that the following equation applies for small uptake coefficients and small aerosol diameters where gas-phase diffusion does not limit uptake. For large particle diameters or large uptake coefficients, the simplified heterogeneous uptake equation requires a correction for gas-phase diffusion (Fuchs and Sutugin, 1970; Kolb et al., 2010). For accumulation mode particles on the order of 100 nm and uptake coefficients on the order of 0.01, this correction is not important. 3kx+aerosol=γc‾SA4 Here γ is the aerosol uptake coefficient, c‾ is the mean molecular speed, and SA is the measured aerosol surface area at plume center. We use γN2O5=10-2 and γNO3=10-3 (McDuffie et al., 2018).

Our model is constrained to plume-center and background measurements of aerosol surface area, photolysis rates, O3, CO, NOx, HONO, and total oxidized nitrogen (NOy). Measurements of NOy are only available from the DC-8 measurements. We also constrain our models to the meteorological conditions pressure, temperature, and relative humidity. Fire emissions and photolysis conditions can change rapidly; therefore, we constrain the model to a subset of plume transects. We chose transects that showed a monotonic decrease in CO with distance from the fire, cover an age range of at least 1 h, and have similar emission times as shown in Figs. S2–S3 and Table S2.

All model runs included a constant first-order plume dilution rate coefficient (kdil) determined by applying an exponential fit to observed CO as a function of plume age (Fig. S3 in the Supplement). We fit only points used to constrain the model and fixed the exponential fit offset to the observed CO background. We applied kdil to all species in the model. We find values of kdil that range between 1.6–46×10-5s-1 (Table S3 in the Supplement), equivalent to a lifetime (τdil=1/kdil) of 0.6–17.3 h.

Plume-center observations were determined using a “top 5 %” method as described by Peng et al. (2020). Briefly, within a transect we determine the location of the greatest 5 % of observations for CO and use that location of the plume for analysis of other compounds. This method obtains an average observation for the center, or most concentrated, parts of the plume. Reported uncertainties are the 1-σ variability of the top 5 % region and instrument uncertainties added in quadrature.

Particulate matter in BB plumes attenuates sunlight, thus photolysis rates, in a process we refer to as plume darkening. In WF plumes we use plume-center observations of 20 photolysis rates (listed in Table S4 in the Supplement), but for the Castle and Cow plumes only jNO2 is available due to the limited instrument payload on the NOAA Twin Otter. Average attenuation of jNO2 within the WF1 and WF2 plumes was 96 % (meaning jNO2 at plume center was 4 % of jNO2 outside of the plume). Plume-center attenuation of jNO2 was 29 % for the Castle plume. We sample the Cow plume after sunset and therefore do not have observation of jNO2 while the smoke was under sunlight (0–2 h). We estimate that plume-center jNO2 attenuation was 34 %. This estimate was made by comparing jNO2 attenuation to plume size (by CO) in the WF and Castle model runs and is consistent with jNO2 attenuation in plumes emitted from the Cow fire sampled on other days. All other photolysis rates were estimated using a ratio of the observed jNO2 to calculated photolysis rates using an MCM trigonometric solar zenith angle (SZA) function below. 4J=l⋅cos⁡(SZA)m⋅e-n⋅sec⁡(SZA), where l,m, and n are derived from least squares fits to j values from a radiative transfer model and literature cross sections and quantum yields. This calculation is a standard photolysis value method in F0AM and is described by Jenkin et al. (1997). However, this method does not account for overhead O3 column, surface albedo, aerosol, or other effects.

In all of the plumes studied here, observed jNO2 rates are below 10-3s-1 excluding the first few minutes of the Castle plume (see Fig. 2). Values of jNO2 in the WF2 plume remained low, near 10-4s-1 during the sampling time. In contrast, the WF1 plume exhibits increasing jNO2 rates, which eventually reach 8×10-4s-1. Differences in the photolysis rates between the first and second pass are likely due to the setting sun. Finally, observations of photolysis rates are negligible in the Cow plume as it was sampled after sunset.

In all plumes except the Castle plume, our first transect sampled smoke 36 min–2 h old; therefore, we implemented an iterative method (McDuffie et al., 2018; Wagner et al., 2013) to estimate initial emissions (at age=0). We began with best-guess estimates of CO, NO, NO2, HONO, O3, and all BBVOCs (determined by CO and our emissions inventory by Eq. ) and then systematically changed these initial conditions to minimize the differences between model output and observations downwind. Initial conditions in the Castle run were taken directly from observations of NO, NO2, O3, CO, HONO, phenol, catechol, cresol, and methylcatechol in the first transect where the plume age was 3±1 min and therefore was close to age=0. We initiated the remaining 298 BBVOCs by using CO and Eq. (). Initial conditions for all cases are shown in Table S5 in the Supplement. In all cases, backgrounds of NO, NO2, O3, CO, and HONO were taken as an average outside of the plume, and BBVOC backgrounds were assumed to be zero. Background mixing ratios used in all cases are shown in Table S3.

We determined best-guess estimates of CO and HONO directly from observations of the first transect. To determine a best-guess estimate for NOx, we used the sum of observed NO and NO2 for the Cow run or NOy minus HONO (as NOy will contain HONO) for the WF runs. Best-guess estimates of O3 were determined using an average of background O3 observations from a flight leg upwind of the fire and outside of the plume transects, which can vary (Table S6 in the Supplement).

We began iteration with CO and kdil by increasing best-guess estimates of CO and varying kdil within the fit errors until we minimized the differences between observed and modeled CO. This, in turn, determines the emissions of BBVOCs by Eq. (). Next, we iterated NOx, HONO, and the NO/NOx ratio such that the sum of NOx and HONO did not exceed the observed NOy, and the initial NO/NOx ratio remained between 0.6–1 (Roberts et al., 2020). Lastly, we iterated the initial and background O3. As explained in Sect. 2.4, we were required to iterate on background O3 in some model runs in order to achieve agreement between model and observations. We repeated the above process to minimize the differences between model and observations. In an attempt to avoid finding a local solution, as opposed to the “best” solution, we reversed the order of iterating O3, NOx, and HONO when repeating the above process.

The WF fire emissions were significantly greater than the Castle and Cow fire emissions as is seen in the observed CO data (Fig. 2). Initial plume-center CO was 8.26 and 8.33 ppmv in WF1 and WF2, respectively, but 2.62 and 1.95 ppmv for Cow and Castle, respectively.

We report our observations for each species (X) relative to CO in the form of normalized excess missing ratios (NEMRs) following Yokelson et al. (2013) and shown in Fig. S4. 5NEMR=XPlume-XBackground(ppbv)COPlume-COBackground(ppmv) Ozone depression and negative NEMRs at the plume center were observed in all of the sunset, nighttime, or darkened fire plumes analyzed here. Observations of ΔO3/ΔCO (where Δ indicates background-corrected) in the Castle plume remains at just below background levels of O3 in all observations likely due to the small plume size and large O3 background (82.5±2.1 ppbv). Generally, ΔO3/ΔCO increases with plume age due to photochemical O3 production and mixing with background O3. Ozone in the midday WF1 plume reaches 44.8±3.4ppbvppmv-1 of CO or 67.4 ppbv above background after 3.8±0.5 h of transport.

Referring to Fig. S4, we find that observed ΔNO/ΔCO, ΔNO2/ΔCO, and ΔHONO/ΔCO have variable trends in all plumes. Observations of ΔNO/ΔCO are near-zero (≤0.1 ppbvppmv-1) in the Castle and WF1 plumes and elevated in the WF2 and Cow plumes (0.21±0.02–1.21±0.13 ppbvppmv-1). Observed ΔNO/ΔCO in the WF2 plume changes sharply between the first four and last five transects, suggesting changes in fire emissions or photolysis near emission. In order to avoid these changes, we use only observations from the latter to constrain our model, as discussed in Sect. 2.3.2.

There is a general decrease in ΔNO2/ΔCO and ΔHONO/ΔCO over 4 h of aging. Observations of ΔNO2/ΔCO in the WF1 plume decrease at a faster rate than those in the WF2 plume; however, both plumes exhibit about 8.6 ppbvppmv-1 in the youngest smoke (35±8 min old).

Emissions of NOx from biomass burning plumes provide a source of NO3 that is suggested to be a major oxidant for BBVOCs (Kodros et al., 2020). The instantaneous NO3 production rate, P(NO3), is a common metric of the potential for NO3 chemistry (Brown and Stutz, 2012). 6P(NO3)=kNO3[NO2][O3] At the center of the plumes presented in this study, NO3 production rates were between 0.1 and 1.5 ppbvh-1 as seen in Fig. 2. These NO3 production rates are consistent with those found in a nighttime agricultural smoke plume measured above a rural area at the border of Missouri and Tennessee during the Southeast Nexus campaign (SENEX), which varied between 0.2 and 1.2 ppbvh-1 (Decker et al., 2019). These values of P(NO3) are also similar to those found in urban plumes and forested areas. Production rates of NO3 in urban plumes typically range within 0–3 ppbvh-1 at night but can be larger. In forested regions, P(NO3) is typically below 1 ppbvh-1 at night (Brown and Stutz, 2012).

Agreement between the modeled P(NO3) and observed P(NO3) reflects agreement between observed and modeled NO2 and O3. The WF1 model run slightly overpredicts NO2 after 3 h of aging and therefore overpredicts P(NO3). Similarly, the Cow model run slightly underpredicts NO2 compared to observations; therefore, the trend in P(NO3) is slightly underpredicted.

Our work focuses on the role of phenolics in BB plumes and includes updated and expanded phenolic oxidation mechanisms as described in the “Expansion of Phenolic Mechanism Description” section in the Supplement. Therefore, capturing the phenolic evolution in our models is critical to understanding the importance of phenolics in BB. In the Castle case, which is initiated with observations of phenolics, we find excellent agreement for catechol, methylcatechol, phenol, and cresol (Figs. S5 and S9). Further, we find that the model run lies on the upper edges of nitrocatechol errors and the lower edge of nitrophenol errors. The model run underpredicts nitrocresol by a factor of 60. Note that we do not have available calibrations for nitromethylcatechol but do provide observations in arbitrary units for the purpose of comparing the time evolution of this compound.

Overall the model recreates the relative time evolution of nitrophenolics well. Disagreement between the model and observed compounds could be caused by many factors including, but not limited to, interfering isomers measured by the UW I- HR ToF CIMS or the NOAA I- ToF CIMS, variable fire ERs, and loss or production of nitrophenolics not captured by our mechanism. The MCM includes several gas-phase loss processes of nitrophenolics but no gas to particle partitioning. Nitrophenolics readily partition to the aerosol phase (Finewax et al., 2018). Further, the MCM does not include photolytic loss of nitrophenolics, despite some evidence to the contrary (Sangwan and Zhu, 2016, 2018). Omitting the aerosol loss pathway may be the cause for these discrepancies. However, precisely how these differences affect the model and observation comparison is uncertain. Therefore, when analyzing gas-phase nitrophenolic evolution, we only consider integrated formation, as discussed in Sect. 3.3.2.

All other model runs were not initiated to observations of phenolics due to the older age of smoke during the first transect. Even so, in the Cow model run (Figs. S6 and S10) we find agreement with catechol and methylcatechol within observation errors. Modeled phenol is about a factor of 3 (Δ 1.4–2.0 ppbv) greater than the observations. Modeled cresol is about a factor of 10 greater than observations, while its oxidation product, nitrocresol, is 7 times less than the observations. Models are thus able to reproduce some, but not all, phenolic observations in the Cow plume.

Observations of phenolics in the WF plumes are limited to uncalibrated catechol and nitrocatechol observations from the NOAA I- ToF CIMS (Figs. S7–S8 and S11–S12). In the WF1 model run, catechol and nitrocatechol appear to deplete faster than the model would suggest. The time evolution of nitrocatechol in the WF2 plume agrees well with the model, and in the WF1 model run the model matches the rough timing of the observed maximum signal.

The integrated oxidation rate or the oxidation budget (Fig. 4) is similar to initial reactivity shown in Fig. 3 for OH oxidation, suggesting initial reactivity may be a good indicator for integrated reactivity. However, this does not hold true for NO3 or O3.

The initial NO3 reactivity differs substantially from the oxidation budget. For example, 20 % of initial NO3 reactivity is due to NO, but NO accounts for ≤1 % of integrated NO3 loss. Further, photolysis of NO3 accounts for <1 % of NO3 loss in all model runs and is greatest in the Castle plume (0.6 %) where measured jNO2 and calculated jNO3 reached maximum values of 4×10-3 and 0.14 s-1, respectively. Although daytime NO3 oxidation of reactive VOCs has been found for heavily polluted urban air (Brown et al., 2005; Geyer et al., 2003; Osthoff et al., 2006), the dominant NO3 loss processes in urban plumes is NO reaction and photolysis (Brown and Stutz, 2012; Wayne et al., 1991). The different controlling NO3 loss pathway here highlights the unique and highly reactive environment of BB plumes. Further, 67 %–70 % of integrated NO3 reaction is due to phenolics, which is larger than initial total NO3 reactivity (56 %). Integrated alkene, terpene, and furan oxidation by NO3 are all lower than their initial reactivities.

The production of NO3, by Reaction (R3), and subsequent loss to BBVOCs is a significant (8 %–21 %) loss of O3 and much greater than the initial O3 reactivity to NO2 of 0.4 %. Similarly, integrated loss of O3 to alkenes (40 %–49 %) and terpenes (16 %–23 %) is much less than initial reactivity would suggest (54 % and 39 %, respectively). Conversely, phenolics and furans account for 4 %–11 % and 13 %–20 % of O3 loss, respectively, even though their relative initial reactivity is <1 % and 7 %, respectively. Overall, the differences between initial reactivity and integrated oxidation rate are explained by changing reactivity as BBVOC are oxidized with plume age.

An example is seen in Fig. 5 for O3 in the Castle model run, which has a large O3 background (72±1–82±2 ppbv), is a relatively small plume, and is sunlit at emission. As a result, alkenes and terpenes are depleted quickly through oxidation by O3 and OH. The combined O3 reactivity of alkenes and terpenes reduces from 82 % to 44 % after 2 h, during which time phenolic reactivity increases from <1 % to ∼40 %. In other words, as BBVOCs are depleted, the reactivity profile of each oxidant will change and can result in significant differences between the initial reactivity and oxidant budget.

In contrast to NO3 and O3, loss of OH by each BBVOC group is within 1 % of that predicted by the initial reactivity, except for terpenes. Initial reactivity of terpenes is about 13 %, while actual destruction of OH by terpenes averaged to 8 %. While terpene oxidation by OH is lower than its reactivity in all model runs, it is especially low (2 %) in the WF1 model run, which is likely due to the large concentration of O3 from photochemical production.

Losses of OH are not only due to highly reactive BBVOCs. HCHO, CO, and NO2 are responsible for 12 %–14 % of OH destruction. This is consistent with an OH reactivity analysis from North American fuel types burned during the FIREX laboratory study, which found 13%±1% of OH reactivity was due to HCHO, CO, and NO2 (Gilman et al., 2015). The fraction of OH reactivity toward CO and NO2 are similar to those found in a tropical rainforest (Fuchs et al., 2017) but much smaller than the fraction of OH reactivity toward CO (7 %) and NO2 (18 %) found at an urban site (Gilman et al., 2009) and the fraction of OH reactivity toward CO (20 %–25 %) and NOx (12 %–22 %) at a rural site (Edwards et al., 2013).

To study the competition between all oxidants, we focus on three main BBVOC groups: phenolics, furans/furfurals, and alkenes/terpenes. Generally, furans/furfurals and alkenes/terpenes groups are mainly oxidized by OH and O3, while NO3 plays a small role (Fig. 6). Oxidation of furans/furfurals and alkenes/terpenes by OH (18 %–55 %, 11 %–43 %, respectively) and O3 (39 %–76 %, 54 %–88 %, respectively) can vary widely depending on the plume. We find this is due to the variability of actinic flux. In model runs with less photolysis at emission, OH oxidation is low compared to model runs that are more optically thin. This reduction of oxidation by OH appears to be replaced by O3 rather than NO3. For example, relative furan/furfural oxidation by OH in the WF1 model run (relatively large integrated jNO2) is 31 % less than that in the Cow model run (comparatively lower integrated jNO2), yet O3 oxidation is 32 % greater.

This relationship does not hold for phenolics, which are subject to significant NO3 oxidation (26 %–52 %) (Fig. 6). Phenolic oxidation by OH (22 %–43 %) and O3 (16 %–33 %) are slightly less than NO3. As a result, phenolic oxidation by NO3 dominates in the WF1 and Dark model runs, while OH dominates in the Castle model run. In the WF2 and Cow model runs, NO3 and OH oxidation is roughly equal.

Generally, NO3 oxidation of phenolics increases with O3 availability and decreases with available actinic flux, but these relationships are coupled and complex. One example is seen in the WF2 model run, which has the second lowest integrated jNO2 value, and large emissions of NO that keep O3 low during sunlit hours. Therefore, P(NO3) is reduced, NO3 is present at lower mixing ratios within the first hour of oxidation, and phenolics are less subject to NO3 oxidation when compared to the other model runs.

As actinic flux increases so does OH and O3 production, and therefore oxidant competition. One example is shown by the Castle model run where OH leads phenolic oxidation (41 %) with O3 second (33 %). The Castle model run demonstrates the greatest observed background O3 (90 ppbv). Further, the Castle model run has significantly smaller total emissions (based on CO) than the other model runs and the greatest integrated jNO2. Due to the increased background O3 and photochemical production of OH, NO3 plays a smaller role in the oxidation of phenolics.

Generally, the modeled total phenolic oxidation rate varies between 1–10 ppbvh-1 at emission (Fig. 7a–d), but the change in oxidation rate is not constant and trends with available actinic flux. Model runs with active initial photochemistry (Castle, WF2, and Cow) exhibit decreasing total oxidation rates, while model runs with little to no photolysis (WF1 and Dark) reach a local maximum rate after ∼2 and ∼5 h, respectively. These increases in oxidation rate are due to increases in O3 and NO3 oxidation once NO is depleted. Generally, the phenolic lifetime increases with decreasing actinic flux. The contrast between day and night phenolic oxidation is best seen by comparing the WF2 and Dark model runs. Phenolic lifetimes in the Dark model run are, on average, a factor of ∼2 greater than phenolic lifetime in the WF2 model run.

Before sunset and in early stages of plume oxidation, the major channel of phenolic oxidation is via OH. However, in the WF1 model run NO3 oxidation dominates after only 12 min (Fig. 7a). As the WF1 model run dilutes, photolysis rates increase and O3 is entrained promoting O3 and NO3 production. This increase in oxidant concentration keeps phenolic oxidation >1 ppbvh-1 for at least 4 h before the end of the model (see Sect. 2.3), unlike other model runs that drop below 1 ppbvh-1 of total phenolic oxidation within 0.5–3 h. After 2.6 h, in the WF1 model run, all oxidants contribute equally to phenolic oxidation; thereafter, OH and O3 equally split oxidation, while the influence of NO3 decreases. At the end of the WF1 model run, 69 % of initial phenolics remain unoxidized (Fig. S15 in the Supplement).

As the sun sets in our sunset model runs (WF2, Castle, and Cow), a transition from OH-controlled to a mixture of NO3- and O3-controlled oxidation occurs when OH production and total oxidation rate decrease rapidly. Interestingly, OH dominates phenolic oxidation in the Dark model run (initiated after sunset) for the first 1.8 h before NO3 oxidation takes over. During this time, OH is produced by decomposition of Criegee intermediates formed through ozonolysis of unsaturated hydrocarbons, primarily catechol (Fig. S14), methylcatechol and limonene. In other sunset model runs, OH plays a smaller role after sunset. Even so, this suggests that all BBVOC oxidation after sunset is driven by O3 chemistry, either through direct oxidation by O3, NO2+O3 to form NO3, or by formation and decomposition of Criegee intermediates to form OH.

The WF2, Dark, and Cow model runs all contain unreacted phenolic emissions at sunrise the following day (48 %, 61 %, and 8 %, respectively, Fig. S15). The WF2 and Dark model runs have significantly more phenolics that remain at sunrise because of their larger (∼3 times) emissions compared to the Cow model run. Further, the WF2 and Dark model run conditions differ only by the presence of photolysis; therefore, the difference in remaining phenolics between the WF2 and Dark is due to the time of day the smoke was emitted. In contrast to these three model runs, the emissions in Castle are depleted within 2.6 h due to its small size.

Nitrophenolic formation increases with O3 and photolysis, which promotes formation of NO3 and OH. For example, the Castle and Cow model runs have relatively large O3 and jNO2 at emission and therefore form nitrophenolics rapidly (0.6–1.4 ppbvh-1 within the first 15 min). In contrast, the WF and Dark model runs have near zero O3 due to large emissions of NO and relatively low or zero jNO2 and therefore form nitrophenolics more slowly (<0.1–0.7 ppbvh-1 within the first 15 min).

Despite the rapid formation of nitrophenolics in the Castle model run, it has the least (excluding WF1) total nitrophenolic formation relative to total emissions as seen in Fig. 8. Figure 8 shows integrated nitrophenolic formation per emitted ppmv of CO, which allows us to compare total nitrophenolic formation across varying plume sizes. In contrast to the Castle model run, the Cow model run has the greatest nitrophenolic formation. These differences are the result of differing phenolic oxidation pathways. The Castle model run has a large (90 ppbv) O3 background, which results in O3 accounting for ∼40 % of phenolic oxidation between 30 min–2 h of age (Fig. 7c). At the end of the Castle model run (2.6 h), O3 oxidation accounts for 33 % of total phenolic loss, the largest of any model run (Fig. 6). This is markedly different than the Cow model run where OH and NO3 chemistry control phenolic oxidation before sunset and NO3 after. While O3 accounts for only 16 % of phenolic loss at the end of the model run (∼12 h). In our model, the reaction of O3 + phenolics forms a ring-opening product (Fig. S14), but the rate coefficients and mechanisms are largely uncertain as discussed in the following section.

We include 157 phenolics in our above analysis, but only a few phenolics account for large fractions of nitrophenolic formation. At the end of our model runs, catechol and methylguaiacol account for the largest fraction of phenolic oxidation. Both compounds are mostly oxidized by NO3. Catechol + NO3 alone accounts for 10 %–16 % of total phenolic oxidation rate or 30 %–32 % of NO3 + phenolic oxidation. Similarly, methylguaiacol accounts for 22 %–26 % of NO3 + phenolic rates and is the largest fraction of phenolic oxidation by OH (17 %–18 % of OH + phenolic rates). However, to our knowledge, oxidation products of methylguaiacol by OH and NO3 are unknown but likely lead to nitrophenolics; therefore, our nitrophenolic formation rates are likely underestimated.

The reaction of OH and NO3 with catechol to form nitrocatechol accounts for the largest fraction (32 %–33 %) of total nitrophenolic formation. Therefore, here, we focus on nitrocatechol and detail the nitrocatechol yield from NO3 and OH + catechol. Understanding nitrocatechol yield and its sensitivities is important to understanding the fate of NOx and NOx lifetime discussed in the final sections. However, the nitrocatechol yield depends on many variables such as the concentrations of NOx, BBVOC, O3, and the NOx/BBVOC ratio as well as the certainty in our chemical mechanisms. Therefore, we discuss the sensitivity of all of these factors on nitrocatechol yield below.

Yields of nitrocatechol vary between 33 %–45 % depending on the model run, where NO3 is responsible for 72 %–92 % of nitrocatechol (Fig. 9a). Figure 9 explores factors that govern nitrocatechol yield, defined as the molar ratio of nitrocatechol production to catechol destruction. Yields of nitrocatechol from OH are low relative to NO3 yield due to the formation of trihydroxybenzene and benzoquinones (Fig. S14), which account for 10 %–32 % and 4 %–5 % of total catechol loss, respectively.

The largest yield (45 %) is from the Dark model run, where NO3 oxidation accounts for more than 52 % of phenolic oxidation. In contrast, the lowest yield of nitrocatechol is from the Castle model run (33 %), which has the lowest emissions of NOx compared to the other model runs. A similar yield (34 %) is found in the WF1 model run; however, this model ends after only 4 h when 69 % of phenolics still remain. In short, nitrocatechol yield increases with increasing fraction of phenolic oxidation by NO3.

To understand the dependence of nitrocatechol formation on O3, NOx, total BBVOC emissions (defined by the sum of ERs in our BBVOC inventory), and BBVOC/NOx, we ran a sensitivity analysis on the nitrocatechol yield (Fig. 9b–e). Based on emitted NOx and CO, BBVOC/NOx ratios in plumes we sampled range from 11–35. However, due to fire variability, BBVOC emissions can vary by at least a factor of 2 and for many BBVOCs by more than a factor of 10 from our emission ratios (Decker et al., 2019). Furthermore, we only account for BBVOCs that are most reactive to O3, OH, and NO3, which is smaller than total emitted BBVOCs.

The nitrocatechol yield generally decreases with increasing BBVOC/NOx (color scale and white lines in Fig. 9b). As expected, nitrocatechol yields increase with increasing NOx (Fig. 9c). Across all model runs, the nitrocatechol yield increases to 43 %–57 % over a NOx range of 4.2–91.2 ppbv. Further, the nitrocatechol yield changes to 27 %–50 % (Fig. 9d) when varying total BBVOC emissions by a factor from 4 to 0.5. Finally, we investigate the sensitivity of nitrocatechol yield to initial O3 and find that all model runs have little sensitivity to O3 (Fig. 9e) with an absolute change in nitrocatechol yield <3 % for all model runs when varying initial O3 over a range of 0–113 ppbv.

The low sensitivity of nitrocatechol yield to O3 may be partially explained by competition between O3 and NO3 + phenolic reactions after sunset. To explore this, we use a framework developed by Edwards et al. (2017). Briefly, as stated in Sect. 3.2.1, BBVOCs are the main sink for NO3; therefore, the NO3 loss rate is controlled by the NO3 formation rate. As a result, NO3 can be considered to be in approximate steady state between production by NO2 + O3 and loss by NO3 + BBVOC. Further, according to Fig. 4, the majority of NO3 is lost to phenolics. As a result, the rate of phenolic oxidation after sunset (when OH oxidation of phenolics is minimized) can be approximated as 10-d[phenolics]dt≈(kO3[phenolics]+kNO2+O3[NO2])[O3], which shows that the dominant oxidant is determined by the ratio of NO2 and phenolics. We find that the ratio of phenolics to NO2 at which NO3 and O3 oxidation is equal to be ∼10 (at 298 K, using an ER-weighted average kO3=2.6×10-18cm3molecule-1s-1) with NO3 oxidation more important below this ratio and O3 oxidation more important above it. Modeled phenolics/NO2 ratios at sunset range between 0.7–1.2, and in all model runs, except the Castle model run, the ratio decreases with age. This suggests that in all model runs NO3 oxidation is expected to control phenolic oxidation after sunset.

The phenolic oxidation analysis above relies on phenolic mechanisms and rate coefficients that are highly uncertain. For example, the above-calculated ratio could be much lower in cold lofted plumes, but knowledge of temperature-dependent O3 + phenolic rate coefficients (kO3) are unavailable. Using temperatures observed in the WF2 plume (∼268 K) for kNO2+O3 (but using kO3 at 298 K), the phenolics to NO2 ratio at which NO3 and O3 oxidation is equal would be ∼4.

The rate coefficient and products for the reaction of catechol + O3 that we use are generated using the MCM methodology (Jenkin et al., 2003; Saunders et al., 2003). An experimental study on the gas-phase reaction of catechol + O3 finds an RH-dependent rate coefficient that decreases nonlinearly from 1.3×10-17 to 1.2×10-19cm3molecule-1s-1 with increasing RH (El Zein et al., 2015). The MCM uses a rate coefficient of 9.2×10-18cm3molecule-1s-1. Further, to our knowledge there are no experimental kinetic or mechanistic studies of phenol + O3. In the plumes we investigate, RH varied between roughly 20 %–60 %. Using an RH-dependent rate coefficient for O3 + catechol, we find that the nitrocatechol yields range between 31 %–58 % with little change in yield for the Castle model run (-2 %) and larger change for the Dark model run (+13 %).

The evening emitted plumes modeled in this paper exhibit photochemical loss of NOx initially. In the period prior to sunset, PANs and PNA (peroxynitric acid, HO2NO2) dominate NOz, and PANs alone account for 51%±6% of NOz by sunset. The WF2 plume is lofted and therefore cold (∼267 K), which results in a long peroxyacetyl nitrate and PNA lifetime (∼150 h, and ∼0.4 h, respectively, calculated from the model directly; Atkinson et al., 2006). Even so, as these plumes continue to age, PANs and PNA decompose slowly (Fig. 10) to provide NO2 that promotes nitrophenolic formation and increases nitrophenolic yield (see Sect. 3.3.3). The increase in NO2 after sunset promotes methyl peroxynitrate (CH3O2NO2) as well as NO3 chemistry products, which grow steadily overnight. The contribution of PANs and PNA to NOz decreases from 71%±6% at sunset to 17%±2% at sunrise. Relative NOx loss to PANs and PNA is mostly replaced by the formation of nitrophenolics (Δ 19%±1%), HNO3 by NO3 chemistry (Δ 22 %), and other or unknown NO3 products (Δ 11 %) overnight.

After sunset NO3 chemistry takes over, and by sunrise NO3 chemistry products lead the (66%±2%) NOz budget. Nitrophenolic formation accounts for 56%±2% of NOz in the form of HNO3 and nitrophenolics where nitrophenolics alone account for 29%±1% of NOz. Total HNO3 formation accounts for 31%±1% of NOz; however, most (88 %) of the HNO3 results from NO3 chemistry. Despite accounting for only 9 % (by mole) of initial emissions in our model runs, phenolics have a large and disproportionate effect on NOx loss at night.

A similar example is seen in the Dark model run (Fig. S18 in the Supplement), where PANs and PNA dominate NOz budget for 2.3 h until NO is depleted. At this time, PNA and PANs steadily decrease, while NO3 products steadily increase throughout the night. By sunrise the next day, NO3 chemistry products (including unknown products) account for 80 % of NOz. In all model runs there is a significant (12 %–16 %) NOz formed through NO3 chemistry that leads to unknown products. These unknown products are primarily the result of NO3 + heterocycles such as furans and pyrroles, which have published rate coefficients but little mechanistic work in the literature.

Our NOz budget generally agrees with the NOy budget of western US wildfire smoke sampled during the 2018 Western wildfire Experiment for cloud Chemistry, Aerosol absorption and Nitrogen (WE-CAN) presented by Juncosa Calahorrano et al. (2020). Generally, the maximum fraction of PANs in our budget (∼50 %) agrees with Juncosa Calahorrano et al. (∼40 %) within our model uncertainties. Comparisons of particulate nitrate and organic nitrogen (gas or particulate) between our model run and the analysis of Juncosa Calahorrano et al. (2020) are uncertain since our model does not account for gas–particle partitioning of nitrophenolics. Our model begins to deviate from the NOy budget trend seen by Juncosa Calahorrano et al. (2020) once the sun sets, as expected.

The availability of O3 and sunlight at emission strongly affects NOx lifetime (τNOx, Fig. 11) defined below 11τNOx=1∑iki, where ki is a unimolecular rate coefficient for Reactions (R3) and (R8)–(R10). Model runs with relatively large photolysis and O3 at emission (Castle, Cow, and WF1) have near-emission τNOx values that range from 1–3 h (Fig. 11), which are accompanied by larger total oxidation rates for all BBVOCs (Figs. S15–S17). These model runs also exhibit the fastest nitrophenolic formation rates (Sect. 3.3.2 and Fig. 8). In contrast model runs with low or zero photolysis and near-zero O3 (WF2 and Dark) exhibit near-emission values of τNOx=∼10–16 h and τNOx=20–150 h, respectively. The absence of photolysis in the Dark model run explains the large difference in τNOx between the WF2 and Dark model runs as the WF2 model run has greater O3 and P(NO3) that promotes NO3 chemistry as well as OH radical that promotes PANs formation. In short, we find that “daytime” conditions have shorter NOx lifetimes, greater rates of BBVOC oxidation, and greater rates of nitrophenolic formation when compared to “nighttime” conditions.

Once NO is depleted in both model runs, NOx chemistry changes. The BBVOC oxidation rate rapidly increases (Figs. S15–S17), and NOx loss switches from primarily PANs and PNA to nitrophenolic production as the sun sets (Fig. 10), and O3 is entrained from the background. As such, τNOx decreases markedly to ∼0.5 h.

Due to their reduced oxidation rates at emission, the WF2 and Dark model runs retain about half (46 % and 57 %, respectively) of the emitted NOx by sunrise the next day. Here, we calculate remaining NOx as the fraction of NOx remaining at the end of our model divided by the amount of NOx that was reacted, excluding dilution. This is about a ΔNOx/ΔCO ratio of ∼4 ppbvppmv-1 at sunrise, which is similar to the initial emissions of Castle (∼6 ppbvppmv-1) and WF1 (∼5 ppbvppmv-1). Further, at sunrise, we expect the WF2 and Dark plumes to be more optically transparent and free of NO and thus oxidation rates to increase rapidly as they both still contain NOx. An increase in oxidation at sunrise will likely be more important for the Dark model run, as it retains 61 % of the emitted phenolics as opposed to 48 % in the WF2 model run. Plumes emitted after sunset have slower oxidation rates compared to daytime plumes (Sect. 3.2) but undergo additional oxidation from evening to morning. However, outside of the plume center, where O3 is less effected by reaction with NO and is more likely to be generated by photochemical production, NOx loss rates may be much larger. Therefore, NOx away from the plume center will likely be depleted more rapidly.