Modelling the impact of gas-phase pyruvic acid on acetaldehyde and peroxy radical formation in the boreal forest

Based on the first measurements of gas-phase pyruvic acid (CH3C(O)C(O)OH) in the boreal forest, we derive effective emission rates of pyruvic acid and compare them with monoterpene emission rates over the diel cycle. Using a data-constrained box-model, we determine the impact of pyruvic acid photolysis on the formation of acetaldehyde (CH3CHO) and the peroxy radicals CH3C(O)O2, CH3O2 and HO2 during an autumn (IBAIRN) and summer (HUMPPA) campaign at the same site. The 15 results are dependent on the photodissociation mechanism of pyruvic acid and we examine different scenarios in which the main photolysis products are either acetaldehyde or the CH3C(O)O2 radical, with different overall quantum yields. If CH3CHO is taken to be the main product (as presently recommended by evaluation panels) we find that pyruvic acid photolysis can be a dominant source of this aldehyde in the boreal forest with a contribution of 79 % (IBAIRN) and 94 % (HUMPPA) and may help explain the high acetaldehyde levels observed during HUMPPA. On the other hand, if photolysis leads mainly to the 20 formation of radicals, the emission of pyruvic acid has a profound impact on the rates of formation of peroxy radicals (with a contribution of ~20–50 %) and shifts the onset of radical production to earlier in the morning when actinic flux is dominated by wavelengths that are too long to initiate efficient ozone photolysis but which are absorbed by pyruvic acid.


Introduction 25
Organic acids play a crucial role in tropospheric chemistry, impacting secondary organic aerosol formation, air quality and climate (Kanakidou et al., 2005;Hallquist et al., 2009). Pyruvic acid (CH3C(O)C(O)OH), an organic acid that is central in plant metabolism as part of the Krebs cycle (Walker, 1962), is found in tropospheric air in the gas phase as well as in the aerosol phase, especially in the boundary layer of vegetated regions. Gas-phase mixing ratios ranging from a few to several hundred parts per trillion (pptv) have been reported in various locations around the world, including the tropical rain forest, 30 https://doi.org /10.5194/acp-2020-975 Preprint. Discussion started: 20 October 2020 c Author(s) 2020. CC BY 4.0 License. boreal forest, rural areas with temperate forest, and regions influenced by urban outflow. A recent overview of existing measurements of gas-phase pyruvic acid is given by Eger et al. (2020).
A major source of pyruvic acid is understood to be the photo-oxidation of isoprene, via the ozonolysis of methyl vinyl ketone and subsequent hydrolysis of the Criegee intermediates (Jacob and Wofsy, 1988;Grosjean et al., 1993;Paulot et al., 2009).
Further potential sources are the photolysis of methylglyoxal (Raber and Moortgat, 1995), the gas-phase photo-oxidation of 5 aromatics in the presence of NOx (Grosjean, 1984;Praplan et al., 2014), the aqueous-phase oxidation of methylglyoxal (Stefan and Bolton, 1999) and reactions taking place within biomass burning plumes (Andreae et al., 1987;Helas et al., 1992). In addition, pyruvic acid has been reported to be directly emitted from vegetation (Talbot et al., 1990;Jardine et al., 2010a;Jardine et al., 2010b;Eger et al., 2020). Compared to acetic acid, the presence of a second (non-acidic) carbonyl group imparts on pyruvic acid an absorption spectrum that extends from ultraviolet to visible wavelengths (see Fig. 1) and photolysis is a 10 major sink of pyruvic acid in the boundary layer, with deposition and heterogeneous uptake to the aerosol phase also contributing to its removal. Photolysis of pyruvic acid in air results in a number of different radical and stable products (see Global models have recently revealed discrepancies between simulated and measured acetaldehyde concentrations (Millet et al., 2010;Wang et al., 2019;Wang et al., 2020). Wang et al. (2020) reported CH3CHO mixing ratios that were up to a factor of 10 higher than global chemistry-transport model (EMAC) results in the marine boundary layer around the Arabian Peninsula, implying missing sources in remote and polluted regions. Wang et al. (2019) also found that models systematically underestimate CH3CHO compared to observations in the remote troposphere, implying a missing source of acetaldehyde. This 20 finding was supported by the simultaneous measurement of PAA (for which acetaldehyde is a precursor in remote environments) and by the organic aerosol source of CH3CHO being insufficient to explain the results. Instead, Wang et al. (2019) suggested that CH3CHO arises from the degradation of gas-phase organic compounds. Pyruvic acid, among other organic acids in the gas and aerosol phase, might be one of the compounds that transport and release acetaldehyde to the remote troposphere and its integration into global models might contribute to resolve discrepancies, especially in forested regions. 25 Generally, field measurements as well as modelling and laboratory-based kinetic studies on pyruvic acid are limited and its impact on atmospheric chemistry is still poorly understood. In this study we highlight the potential role of pyruvic acid in the boreal forest, one of the largest terrestrial biomes on Earth.

Experimental data and box model description
The goal of this study is to quantify the impact of pyruvic acid on acetaldehyde and radical formation rates by using a data-30 constrained, chemical box-model. For this purpose we make use of experimental data from two field studies, which were both performed in the Finish boreal forest at the "Station for Measuring Forest Ecosystem-Atmosphere Relations II" (SMEAR II) https://doi.org /10.5194/acp-2020-975 Preprint. Discussion started: 20 October 2020 c Author(s) 2020. CC BY 4.0 License.
in Hyytiälä (61.846 °N,24.295 °E, 180 m above sea level, see Hari and Kulmala (2005)), an area that is mainly characterised by large biogenic (monoterpene-dominated) emissions and low NOx concentrations (Williams et al., 2011). An overview of the instruments used to measure trace-gases and other parameters relevant for this study is presented in Table S1 of the supplementary information.

The IBAIRN and HUMPPA campaigns 5
The IBAIRN campaign (Influence of Biosphere-Atmosphere Interactions on the Reactive Nitrogen budget) took place in September 2016, during the summer-autumn transition, and was characterised by frequent temperature inversions near ground level during night-time (Liebmann et al., 2018), which led to the accumulation of nocturnally emitted trace gases from vegetation. A detailed description of the campaign and the instrumental setup can be found elsewhere (Liebmann et al., 2018;Eger et al., 2020). Briefly, pyruvic acid was measured by a chemical ionisation quadrupole mass spectrometer with a detection 10 limit (LOD) (10 s, 2σ) of 15 pptv (Eger et al., 2020). The sum of monoterpenes (henceforth referred to as MT) was measured by a PTR-ToF-MS and single monoterpenes were monitored by a GC-AED, described in detail by Liebmann et al. (2018).
Despite some discrepancies related to instrument location and inhomogeneity in terpene emissions within the forest, both instruments were in reasonably good agreement throughout the campaign. Since we require a high temporal resolution for our simulation, we have used the PTR-ToF-MS dataset. Photolysis rate coefficients (Jpyr, JNO2, JO(1D) and JHCHO) were derived using 15 actinic flux measurements from a spectral radiometer (METCON GmbH) and evaluated cross sections and quantum yields (Burkholder et al., 2015). Mixing layer (MXL) heights were derived by combining in-situ measurements made by a scanning Doppler lidar (Hellén et al., 2018) with results from the ECMWF ERA-Interim reanalysis (Dee et al., 2011) with a spatial resolution of ~80 km. Since the lidar was unable to resolve MXL heights < 60 m (as regularly experienced during nocturnal inversions), all values below this threshold have been set to 60 m, representing an upper limit. 20 The summertime boreal forest field measurement intensive HUMPPA-COPEC took place at the same location in July and August 2010 and was characterised by unusually warm temperatures for this time of the year with exceptionally large emissions from the biosphere. A detailed description of the campaign and meteorological situation can be found in Williams et al. (2011).
Insights into HOx and ROx chemistry are presented by  who suggested that di-carbonyl compounds, including pyruvic acid, could help explain the high production rates of radicals necessary to explain the observations of H2O2 25 and PAA, especially during periods when the measurement site was impacted by biomass burning in Russia.

Box model description
The box model was developed with the goal of simulating the impact of pyruvic acid photolysis on formation rates of OH, HO2, CH3C(O)O2, CH3CHO, CH3O2, HCHO and CH3OOH over several diel cycles during the two field campaigns in the  Table S2 of the Supplement. Rate coefficients were taken from the IUPAC evaluations (IUPAC, 2020). The atmospheric methane concentration was set to a constant value of 1.8 ppmv. Non-methane alkanes, the degradation of which represents ~ 30-45 % of the acetaldehyde source globally (Millet et al., 2010) were constrained to 1000 pptv of ethane, 250 pptv of propane and 150 pptv of n-butane, as found in similar environments in Finland (Hakola et al., 2006;Hellén et al., 2015). 5 For the box model, programmed in FACSIMILE code (Curtis and Sweetenham, 1987), three different scenarios were investigated, in which the pyruvic acid chemistry listed in Table S2 was modified (see below) in order to examine the sensitivity of the model output to photolysis quantum yields. The box model simulated the field data at 10 min temporal resolution.

Pyruvic acid loss rate and model scenarios
Photolysis and dry deposition are considered the dominant loss terms for gas-phase pyruvic acid, as reaction with OH is slow 10 (1.2 × 10 -13 cm 3 molecule -1 s -1 at 298 K) (Mellouki and Mu, 2003). Heterogeneous uptake to atmospheric aerosols is also calculated to be inefficient at this site, where particle surface area densities were of the order of 2  10 -7 cm 2 cm -3 and the particles contained a large organic fraction (Liebmann et al., 2019) that is likely to reduce the uptake coefficient compared to that measured for purely aqueous particles ( = 0.06, Eugene et al. (2018)).
In Fig. 1 we present the wavelength resolved photolysis rates across the UV-absorption spectrum of pyruvic acid (assuming 15 the overall photolysis quantum yield is 0.2) and a scheme showing possible routes to formation of acetaldehyde and radicals following its photolysis. The three most important photolysis channels are: where i are branching ratios with 1 + 2 + 3 = 1.
CH3CO and HOCO react rapidly in air to form peroxy radicals: The methylhydroxycarbene product of R1 (CH3COH) is an unstable intermediate which can rearrange to form CH3CHO (R6) 25 with branching ratio  or react with O2 to form HO2 and CH3CO (branching ratio 1-), which via R4, results in formation of The quantum yield of acetaldehyde formation (1) is thus equal to 1, where  is the overall quantum yield. The quantum 30 yield of formation of CH3CO3 + HO2 (via R1/R7 and R2) is 2 + 1(1-). Direct formation of CH3C(O)OH in R3 has a quantum yield of 3.
For the calculation of the pyruvic acid photolysis rate constant (Jpyr) and photolysis products we used different scenarios A-C that reflect the large variability in quantum yields and branching ratios reported in the literature. They are outlined in the following and summarised in Table 1 along with the dominant photolysis products in air. In scenario A, which represents a reference for the other scenarios, we assumed that the overall quantum yield for pyruvic acid photolysis () is zero. As its reaction with OH is slow, setting Jpyr to zero effectively removes any contribution of pyruvic acid to acetaldehyde or radical 5 generation. In scenario B we used pyruvic acid cross sections and quantum yields according to the UPAC recommendations The deposition rate of pyruvic acid was calculated from kdep = vdep hMXL -1 during day and kdep = 2 vdep hMXL -1 during night (Shepson et al., 1992a), with the transition following the diel variation in the mixing layer height hMXL (see Fig. S1 in the Supplement). Further, as the two trace-gases have similar solubilities, we assumed that the deposition velocity of pyruvic acid 15 is equal to that of H2O2, so that vdep = 8.4 cm s -1 during day and vdep = 0.8 cm s -1 during night, as derived by  for the same site. This resulted in a minimum dry-deposition loss rate constant of kdep = 0.9 × 10 -4 s -1 during day and a maximum of kdep = 1.8 × 10 -4 s -1 during night.

IBAIRN simulation
The simulation was initiated on 05.09.2016 but we analysed the model output only in the period 09-21.09.2016, as high-20 resolution MT data was not available for the first campaign days. In addition to the parameters generally constrained in all model runs (see above), for IBAIRN the concentration of pyruvic acid was determined by the observations. The concentration of PAN, which is generally the most abundant of peroxy acetyl nitrates (PNs), was calculated from [PAN] = 0.9 × [PNs] (estimation based on observations by e.g. Shepson et al. (1992b), Roberts et al. (2004) and Roiger et al. (2011)). OH was calculated from the correlation of ground-level OH measurements with UVB radiation intensity at the Hyytiälä site (Petäjä et 25 al., 2009;Hellén et al., 2018) with [OH] = 5.62 × 10 5 [UVB] 0.62 molecule cm -3 when UVB is in units of W m -2 .
We added a constant value of 1.5 s -1 (accounting for unmeasured oxygenated VOCs) to the total OH reactivity (ROH), so that the simulated OH concentration approximately matched the OH concentration derived from the UVB measurement.

HUMPPA simulation
The simulation was initiated on 14.07.2010, the output was only used from 21.07.-08.08.2010 due to missing PAN data at the 30 beginning of the campaign. In addition to the parameters generally constrained in all model runs (listed above), for HUMPPA https://doi.org/10.5194/acp-2020-975 Preprint. Discussion started: 20 October 2020 c Author(s) 2020. CC BY 4.0 License. the concentrations of HCHO, HONO and CH3CHO were determined by the observations. However, due to a lack of experimental data, we estimated the pyruvic acid concentration based on measured MT mixing ratios and the relationship between pyruvic acid and MT emission rates derived from the IBAIRN analysis (which is discussed later in Sect. 3.1). During HUMPPA, the site was impacted by a biomass burning plume on two occasions (26.07.-30.07.2010 and 07.08.-09.08.2010) and levels of CH3CN, CO, PAN, H2O2, PAA and several oxidised organics were considerably enhanced over normal 5 conditions. We present results both including and excluding the biomass burning periods.

Results and discussion
In the following, we discuss the model output for the IBAIRN and HUMPPA campaigns with a focus on pyruvic acid emission rates and its impact on acetaldehyde and radical chemistry in the boundary layer of the boreal forest.

Autumn campaign (IBAIRN): Pyruvic acid emission rate relative to monoterpenes 10
In order to quantify the pyruvic acid emission rate (Epyr) during IBAIRN we assume that only photolysis and dry deposition contribute significantly to its overall loss rate (Sect. 2.2.1) and that pyruvic acid is in steady-state. The latter assumption is reasonable as its mean lifetime was (2 ± 0.5) h (for scenario B; shortest at night) and changes in the mixing ratio could reliably be reproduced by the simulation. Due to a homogeneous fetch at the measurement site we can neglect transport processes and Epyr is defined by Eq. (1), where [pyr]ss is the measured concentration, Jpyr is the photolysis rate constant of pyruvic acid, kdep 15 is the first-order loss rate constant for its dry deposition, and hMXL is the well-mixed boundary layer height.
Epyr is effectively an emission rate normalised to the MXL height (hMXL) and has units of pptv s -1 m. As the photolysis is a substantial fraction of the overall losses of CH3C(O)C(O)OH, the choice of quantum yield  (scenarios A, B and C) directly impacts the calculated emission rate. The same calculation is performed for the monoterpenes (EMT) over the same period (and 20 thus for the same MXL height). We note that hMXL controls not only the value of kdep but also directly affects the mixing ratios of both MTs and pyruvic acid for a given emission rate. The relative emission rate (Epyr / EMT) can be calculated from Eq. (2) where terms in square brackets are concentrations.
Note that the calculated emission rates are subject to substantial uncertainties arising from the measurement of pyruvic acid, MT, OH, O3, NO3, hMXL and Jpyr. In particular, the results are very sensitive to the estimated deposition velocity (vdep) of pyruvic acid which is an estimate based on the deposition velocity of H2O2 which itself has an uncertainty of ~ 90 % (Fischer et al., 2019). Further, our calculations are based on the assumption that the sources for pyruvic acid and MT emissions are evenly distributed and measurements made at ~ 8.5 m above the ground are representative of the entire boundary layer (i.e. 5 that the boundary layer is well-mixed, including the very shallow boundary layer at night). A gradient in pyruvic acid mixing ratios at night cannot be ruled out, which would impact on our results. We estimate that the emission ratio (Epyr / EMT) in Eq.
(2) is associated with an overall uncertainty of a factor ~2.
A time series of pyruvic acid and MT mixing ratios along with the MXL heigth (hMXL) derived from a lidar measurement and from the ERA-Interim renanalysis is shown in Fig. S2 of the Supplementary Information. Whereas both MXL height datasets 10 agree very well during the night when the MXL is shallow (usually < 100 m), the lidar data is on average a factor of ~2 lower during day and characterised by a much higher variability. For the derivation of the diel profile of hMXL (Fig. S1) we took an average of both datasets. The diel variation displayed in Fig. S2, with highest MT mixing ratios at night, is characteristic for this boreal forest site and has been observed in earlier studies (Hellén et al., 2018).
In the following, we focus on the mean, diel profiles of Epyr, EMT, J-NO2, T and hMXL for the IBAIRN campaign, which are 15 presented in Fig. 2. During September, the emission rate of pyruvic acid (Epyr) reaches its maximum a few hours after solar noon when the temperature peaks, similar to EMT. However, the amplitude of the day-to-night difference in Epyr is a factor of ~3 smaller than observed for EMT. This could indicate that pyruvic acid emissions are less temperature-dependent than MT emissions (see below) and that other environmental factors might additionally play a role at this time of year.
The emission rates of the MTs derived as described above show a large day-night variation with a factor ~20 larger values 20 around noontime compared to midnight. This is significantly larger than the expected variation (factor 2-3) based on the average noon-to-midnight temperature difference of 10 K and the parameterisation of Guenther et al. (1993) whereby EMT ∞ exp(β(T -297 K)) with β = 0.1 K -1 (which is in line in with the empirical value of β = 0.12 K -1 that was derived for this site in September by Hellén et al. (2018)). One potential reason for this discrepancy may be related to emissions in autumn from fresh leaf litter that significantly contribute to the observed mixing ratios (Hellén et al., 2018) and that the assumption of evenly 25 distributed sources and a well-mixed boundary layer is not necessarily valid during night, especially during strong temperature inversions.
We note that (like the values of Epyr) the temperature dependence derived is strongly influenced by the diel variation of the MXL height and thus carries significant uncertainty and may not be transferable to other locations or even times of the year.
As our measurements of pyruvic acid in the boreal forest are the first to have been made, we cannot compare our relative emission ratio (Epyr / EMT) with previous measurements in the boreal environment. Instead, where possible, we derive the 5 emission ratio from measurements of MTs, isoprene and pyruvic acid in warmer climates. Jardine et al. (2010b) performed measurements in an enclosed (glass dome) tropical forest biome at Biosphere 2 in Arizona, US, where they found maximum concentrations of 120 ppbv isoprene, 6 ppbv monoterpenes and 15 ppbv pyruvic acid. As the glass dome absorbed actinic wavelengths and prevented active photochemistry, the chemical loss processes for pyruvic acid, isoprene, and MT (including photolysis and reactions with OH, O3 and NO3 ) are negligible. Initially disregarding the deposition of isoprene and MT, we 10 derive lower limits of (Epyr / Eiso) ~ 0.17 and (Epyr / EMT) ~ 4 (see Table 2). However, due to the presence of large concentrations of isoprene-consuming microbes in the soil of Biosphere 2, the isoprene loss rate via deposition may be enhanced, which will decrease the effective emission ratio (Epyr / Eiso). In addition, branch enclosure studies were performed on a mangifera indica (mango) tree within Biosphere 2, yielding mean fluxes (in nmol m -2 s -1 ) of 3.2 for isoprene, 0.09 for MT and 0.15 for pyruvic acid. Pyruvic acid emissions peaked during the day when temperature and photosynthetically active radiation (PAR) were 15 highest and correlated very well with isoprene emissions and (to a certain extent) with MT emissions. Assuming that a mango tree is representative for the tropical vegetation, we derive emission ratio of (Epyr / Eiso ~ 0.05 and (Epyr / EMT) ~ 1.7 (see Table   2), which is consistent with our estimations for the IBAIRN campaign. However, given that Talbot et al. (1990) observed great variability in pyruvic acid emission fluxes among five different tree species during measurements in the tropical Ducke Forest Reserve close to Manaus, Brazil, this agreement may, to some extent, be coincidental. Talbot et al. (1990) also reported a mean 20 emission flux (derived from enclosure experiments) relative to isoprene of (Epyr / Eiso) ~ 0.003, which is about one order of magnitude smaller than in the study of Jardine et al. (2010b). In a further branch enclosure study by Jardine et al. (2010a) emissions from a creosotebush (Larrea divaricata), which is typically found in US drylands, were investigated. Average noontime branch emission rates (in µg C gdw -1 h -1 ) of 7.5, 10.4 and 0.2 for isoprene, MT and pyruvic acid resulted in relative emission ratios of (Epyr / Eiso) ~ 0.05 and (Epyr / EMT) ~ 0.07 for this mixed isoprene-MT-emitting species. 25 The comparison with the few datasets available in the literature indicates that the variability of the emission factors (Epyr / EMT) and (Epyr / Eiso) among different plant species and different environments is large. In addition, a lack of pyruvic acid measurements over different seasons in the boreal forest means that we cannot exclude that the value we derive is biased by emissions (e.g., from ground-level, decaying plant-litter in September) that are peculiar to this season and environment. The conclusions we draw from our following analysis are therefore relevant for the autumnal boreal forest but require validation 30 before being extended to other regions and seasons with confidence. https://doi.org /10.5194/acp-2020-975 Preprint. Discussion started: 20 October 2020 c Author(s) 2020. CC BY 4.0 License.

Contribution of pyruvic acid to acetaldehyde and radical formation
To quantify the impact of pyruvic acid photolysis during IBAIRN, we identified the relative contributions of the processes forming OH, HO2, CH3C(O)O2, CH3CHO, HCHO and CH3OOH from our model runs, and compared scenario A (the reference run with no impact from pyruvic acid) with scenarios B and C. The campaign-averaged production rates are summarised in Table 3. 5 Figure 3 illustrates the relative importance of the various pathways to OH formation and the resulting OH concentrations for the three different scenarios. On average, the main source of OH radicals was the direct formation from O3 photolysis (39 % in scenario B), followed by the reaction of HO2 with NO (30 %) and O3 + MT (25 %). The inclusion of pyruvic acid photolysis in the model increased the relative importance of the HO2 + NO reaction by 1 % (scenario B) or 6 % (scenario C) due to increased HO2 production. The overall OH production rate (13.1 × 10 5 molec cm -3 s -1 in scenario A, see Table 3) increases by 10 2 % in scenario B or 13 % when considering scenario C.
For HO2 (Fig. 4) the three major sources during IBAIRN were the reactions OH + CO (39 %), RO2 + NO (36 %), and CH3O2 + NO (19 %). In scenario B, pyruvic acid accounts for just 2 % of the total HO2 production (via R2 and R5, see Fig. 1) whereas in scenario C this increases to 14 % (via R1 and R7). The overall HO2 production rate (5.8 × 10 5 molec cm -3 s -1 in scenario A, see Table 3) increases by just 2 % in scenario B but by as much as 36 % in scenario C. Clearly, the combination of a high 15 quantum yield of pyruvic acid photolysis and direct formation of radicals can have a significant impact on the HOx budget. As the photolysis of pyruvic acid occurs at wavelengths that are red-shifted compared to O3 photolysis, the model predicts that the onset of photochemical radical formation in scenario C is about 1 hour earlier than without pyruvic acid photolysis. In the absence of pyruvic acid photolysis (scenario A), the production of acetaldehyde is dominated by the degradation of alkanes (n-butane, ethane, and propane). When including formation of CH3CHO via pyruvic acid photolysis in scenario B (Fig. 6), the contribution of the alkanes drastically decreases, and pyruvic acid becomes the dominant source of acetaldehyde (79 %) in this environment. The fractional contribution remains high (53 %) in scenario C even though CH3CHO is formed 30 only at low yield (5 %) in this case. The pyruvic acid mixing ratios observed during IBAIRN would result in mean simulated acetaldehyde mixing ratios of ~115 pptv (Fig. 6d), which is reasonable for this site as shown in the HUMPPA dataset (Sect.
The three main sources of the CH3O2 radical in air are the reactions CH4 + OH, CH3C(O)O2 + HO2 and CH3C(O)O2 + NO, all preceding via formation of the CH3 radical, in the last two cases a result of decomposition of the unstable intermediate CH3CO2.
The CH3O2 radical is a source of both the most abundant aldehyde (HCHO) and methyl hydroperoxide (CH3OOH) in the troposphere, the former preferentially formed at high NOX levels, the latter under low NOX conditions. As shown in Fig. 7, pyruvic acid photolysis increases the fraction of CH3O2 formed from CH3C(O)O2 to 5 % (scenario B) and 18 % (scenario C), 5 with methane oxidation still accounting for most of the total production rate. In scenario C, the summed production rates of HCHO and CH3OOH are increased by 42 % each (Table 3).

Summer campaign (HUMPPA): Contribution of pyruvic acid to acetaldehyde and CH3C(O)O2 formation
As pyruvic acid was not measured during HUMPPA, we use the calculated mean emission rate of pyruvic acid relative to MT during IBAIRN (derived from Eq. 2, see Table 1) as an input parameter for the HUMPPA simulation to estimate the pyruvic 10 acid mixing ratios (in steady state) shown in Fig. 8.
Compared with IBAIRN, where [pyr] / [MT] ~ 0.3 on average, for HUMPPA the mean concentration ratio is ~ 1.6, which is mainly a consequence of the higher temperatures and larger MXL heights during the summertime HUMPPA campaign (especially at night) which result in lower pyruvic acid deposition rates. In addition, it is also influenced by the higher OH and O3 levels in summer (HUMPPA) which contribute to MT losses. The highest pyruvic acid mixing ratios are modelled for the 15 night-time (lowest hMXL) and in periods impacted by biomass burning from Russia, which is indicated by elevated CO and HCN levels (periods highlighted in light blue). Coincidentally, the biomass burning periods were accompanied by high temperatures and high MT levels also. When excluding the biomass burning periods from the analysis, simulated pyruvic acid mixing ratios occasionally reach values up to ~ 2 ppbv with a mean campaign value of ~ 0.5 ppbv.
Similar to the autumn campaign IBAIRN we can now quantify the contribution of pyruvic acid photolysis to the formation of 20 acetaldehyde, peroxy radicals and PAA. Therefore, we compare Scenarios A with B and C (as for the IBAIRN simulation in Sect. 3.2), in this case focussing on the CH3C(O)O2 radical and on acetaldehyde as both are precursors to CH3C(O)OOH, which was measured during HUMPPA.
In the absence of pyruvic acid photolysis (Scenario A), the production of acetaldehyde is dominated by n-butane, ethane, and propane. However, these globally important CH3CHO precursors are not sufficient to explain the acetaldehyde concentrations 25 that were observed (Fig. 8). Another potential source is the direct biogenic emission from vegetation (see e.g. Rissanen et al. (2020), making up ~15 % of the total CH3CHO source (Millet et al., 2010). However, on a global scale (and also in Finland) these emissions are a factor of ~5 smaller than the photochemical production term and have thus been neglected in the simulation. By enabling formation of CH3CHO via pyruvic acid photolysis in Scenario B (see Fig. 9), the contribution of the alkanes reduces to 6 % and pyruvic acid becomes the dominant source of acetaldehyde (94 % in scenario B, 76 % in scenario 30 C) during HUMPPA. Removing periods impacted by biomass burning plumes does not significantly change the overall conclusions (pyruvic acid contribution decreases by ~ 2 %). When comparing the time series of measured acetaldehyde mixing ratios with the model output (Fig. 8), we note that scenario B more closely reproduces the observations than scenario A which https://doi.org /10.5194/acp-2020-975 Preprint. Discussion started: 20 October 2020 c Author(s) 2020. CC BY 4.0 License. greatly underpredicts the observed CH3CHO for much of the campaign. We conclude that pyruvic acid photolysis represents a potentially important contribution to acetaldehyde mixing ratios in the boreal forest and possibly also on a global scale (above forested regions).
The major source of acetylperoxy radical in Scenario A is the decomposition of PAN (93 %), with a minor contribution from the oxidation of acetaldehyde (7 %) (see Fig. 10). In Scenario B, the direct impact of pyruvic acid photolysis on CH3C(O)O2 5 is 3 %, whereas the indirect impact from the formation of acetaldehyde and its subsequent oxidation via CH3CHO + OH is negligible. In contrast, in Scenario C, the direct contribution from the photolysis of pyruvic acid is 19 %, making pyruvic acid an important contributor to acetylperoxy radical generation in this environment and underlining the importance of dicarbonyls, as suggested by . However, the pyruvic acid mixing ratios alone (without other dicarbonyls) cannot explain the observed large PAA mixing ratios during HUMPPA , especially during the biomass burning 10 episodes.

Conclusions
Using pyruvic acid cross sections and quantum yields ( = 0.2) according to the IUPAC recommendations (scenario B in which the dominant photolysis products are CH3CHO + CO2), we identified pyruvic acid as an important source of acetaldehyde (CH3CHO) in both autumn (IBAIRN) and summer (HUMPPA) with a contribution of 79 % respectively 94 % to the overall 15 production rate. Pyruvic acid may thus have contributed to the source of elevated mixing ratios of acetaldehyde observed during HUMPPA which could not be explained by the degradation of alkanes. Our results might also help to explain the discrepancies between modelled and observed CH3CHO mixing ratios in remote, forested regions where emissions of pyruvic acid may be significant. In contrast, under scenario B, the effect on OH, HO2, CH3C(O)O2 and HCHO formation rates during both campaigns is minor (< 10%). 20 This picture changes drastically under scenario C, whereby the quantum yield is a factor of ~ 4 larger and formation of HO2 and CH3C(O)O2 is preferred over CH3CHO. In both campaigns pyruvic acid now increases the overall HO2 production by ~ 20-35 % and the overall HCHO production by ~ 25-40 %. In scenario C, pyruvic acid photolysis is comparable to PAN decomposition as source of CH3C(O)O2 during IBAIRN, also shifting the onset of radical production to earlier in the morning.
The presence of pyruvic acid can partly explain the elevated PAA mixing ratios observed in HUMPPA, though other 25 oxygenates might play a role as well, especially during biomass burning events .
In general, our results are strongly dependent on the chosen quantum yields and deposition velocities. To minimise the uncertainty in our calculations, there is an urgent need for further experimental work on the photochemistry of pyruvic acid.
In addition, measurements of the deposition velocity of pyruvic acid in different environments are required to better constrain its lifetime and thus the impact of photolysis. Further, more enclosure studies will be necessary to investigate the dependence 30 of pyruvic acid emission rates on different plant types and environmental conditions. https://doi.org/10.5194/acp-2020-975 Preprint. Discussion started: 20 October 2020 c Author(s) 2020. CC BY 4.0 License.

Data availability
The Max Planck Institute data used for the IBAIRN analysis is archived with Zenodo at https://doi.org/10.5281/zenodo.3254828 (Crowley and Fischer, 2019). HUMPPA-COPEC-2010 data can be obtained on request (via John N. Crowley) from the owners.

Author contributions 5
PGE was responsible for the pyruvic acid measurement during IBAIRN and, with contributions from JNC and JL, ran the box model, analysed the model results for the IBAIRN and HUMPPA campaigns and wrote the manuscript. NS was responsible for the CRDS measurements of NO2 and PANs during IBAIRN. JS was responsible for the O3 and J-value measurements during IBAIRN. HF was responsible for the NO and CO measurements during IBAIRN and HUMPPA. EK and JW were responsible for the monoterpene measurements during IBAIRN. VV was responsible for the mixing layer height measurements 10 during IBAIRN. All authors contributed to the paper.

Competing interests
The authors declare that they have no conflict of interest.