Missing peroxy radical sources within a summertime ponderosa pine forest

. Organic peroxy (RO 2 ) and hydroperoxy (HO 2 ) radicals are key intermediates in the photochemical processes that generate ozone, secondary organic aerosol and reactive nitrogen reservoirs throughout the troposphere. In regions with ample biogenic hydrocarbons, the richness and complexity of peroxy radical chemistry presents a signiﬁcant challenge to current-generation models, especially given the scarcity of measurements in such environments. We present peroxy radical observations acquired within a ponderosa pine forest during the summer 2010 Bio-hydro-atmosphere interactions of Energy, Aerosols, Carbon, H 2 O, Organics and Nitrogen – Rocky Mountain Organic Carbon Study (BEACHON-ROCS). Total peroxy radical mixing ratios reach as high as 180 pptv (parts per trillion by volume) and are among the highest yet recorded. Using the comprehensive measurement suite to constrain a near-explicit 0-D box model, we investigate the sources, sinks and distribution of peroxy radicals below the forest canopy. The base chemical mechanism underestimates total peroxy radicals by as much as a factor of 3. Since primary reaction partners for peroxy radicals are either measured (NO) or underpredicted (HO 2 and RO 2 , i.e., self-reaction), missing sources are the most likely explanation for this result. A close comparison of model output with observations reveals at least two distinct source signatures. of reactive hydrocarbons. The maximum magnitudes of these missing sources ( ∼ 120 and 50 pptv min − 1 , respectively) are consistent with previous observations alluding to unexpectedly intense within We conclude that a similar mechanism may underlie many such observations.

Abstract. Organic peroxy (RO 2 ) and hydroperoxy (HO 2 ) radicals are key intermediates in the photochemical processes that generate ozone, secondary organic aerosol and reactive nitrogen reservoirs throughout the troposphere. In regions with ample biogenic hydrocarbons, the richness and complexity of peroxy radical chemistry presents a significant challenge to current-generation models, especially given the scarcity of measurements in such environments. We present peroxy radical observations acquired within a ponderosa pine forest during the summer 2010 Bio-hydroatmosphere interactions of Energy, Aerosols, Carbon, H 2 O, Organics and Nitrogen -Rocky Mountain Organic Carbon Study (BEACHON-ROCS). Total peroxy radical mixing ratios reach as high as 180 pptv (parts per trillion by volume) and are among the highest yet recorded. Using the compre-hensive measurement suite to constrain a near-explicit 0-D box model, we investigate the sources, sinks and distribution of peroxy radicals below the forest canopy. The base chemical mechanism underestimates total peroxy radicals by as much as a factor of 3. Since primary reaction partners for peroxy radicals are either measured (NO) or underpredicted (HO 2 and RO 2 , i.e., self-reaction), missing sources are the most likely explanation for this result. A close comparison of model output with observations reveals at least two distinct source signatures. The first missing source, characterized by a sharp midday maximum and a strong dependence on solar radiation, is consistent with photolytic production of HO 2 . The diel profile of the second missing source peaks in the afternoon and suggests a process that generates RO 2 inde-

Introduction
Peroxy radicals are central components of the tropospheric radical pool. Organic peroxy radicals (RO 2 ) are metastable intermediates in the oxidation of volatile organic compounds (VOC) (R1), while hydroperoxyl radicals (HO 2 ) are generated via photolysis of carbonyl-containing VOC (e.g., formaldehyde) (R2), alkene ozonolysis (R3) and radical cycling reactions (R5, R6). RO 2 and HO 2 typically exhibit lifetimes of 1-1000 s with respect to reactions with nitric oxide (NO) (R4, R5) and other peroxy radicals (R7, R8, R9); larger RO 2 may also undergo isomerization and/or unimolecular decomposition. Together with the hydroxyl (OH) and alkoxy (RO) radicals, these species comprise the RO x radical family. Rapid cycling among RO x and the nitrogen oxide radicals (NO x = NO + NO 2 ) lies at the core of photochemical mechanisms that regulate atmospheric composition and its associated impacts on air quality and climate.
A number of key processes hinge upon the fate of peroxy radicals. For example, the conversion of NO to NO 2 via (R4) and (R5) is a critical step in tropospheric ozone formation (Thornton et al., 2002). Reactions of RO 2 with NO and NO 2 may also form alkyl nitrates and peroxy nitrates, respectively, facilitating the redistribution of pollutant precursors over regional and global scales (Moxim et al., 1996;Paulot et al., 2012;Browne and Cohen, 2012). Conversely, cross-reactions of peroxy radicals ((R7) -(R9)) are responsible for termination of radical cycling in high-VOC, low-NO x regimes. These reactions also form hydrogen peroxide (H 2 O 2 ) and organic hydroperoxides (ROOH), which can induce oxidative stress in vegetation (Hewitt et al., 1990). Transformations of RO 2 also generate oxidized VOC that may contribute to formation and growth of secondary organic aerosol (SOA), a major fraction of the global aerosol burden (Hallquist et al., 2009). SOA precursor production depends strongly on the degree of functionalization versus fragmentation (Chacon-Madrid and Donahue, 2011), which in turn varies with the specific molecular structure of each RO 2 radical. Peroxy radicals themselves can act as a major source of oxidants via (R4). Because this reaction shifts RO x into its more reactive form while generating NO 2 (a precursor for ozone and thus OH), it effectively amplifies atmospheric oxidizing capacity. In many environments, this reaction is the main daytime source of OH (Stone et al., 2012). The structure and abundance of VOC precursors shapes the detailed chemistry of peroxy radicals. In the remote troposphere, methane and CO are the primary reactants. Here, methyl peroxy radical (CH 3 O 2 ) is the major RO 2 species and HO 2 concentrations typically exceed RO 2 . Near VOC sources, however, the RO 2 distribution can be considerably more complex and the prevailing chemistry is less well understood. This is particularly true in regions dominated by biogenic alkenes such as isoprene, 2-methyl-3-butene-2-ol (MBO) and monoterpenes. Numerous field studies have identified gaps in our understanding of photochemistry in these environments (Carslaw et al., 2001;Faloona et al., 2001;Thornton et al., 2002;Lelieveld et al., 2008;Ren et al., 2008;Hofzumahaus et al., 2009;DiGangi et al., 2011;Whalley et al., 2011;Griffith et al., 2013;Mao et al., 2012;Kim et al., 2013;Hens et al., 2013), while theoretical and laboratory work continues to reveal new reaction pathways that can significantly impact oxidant levels and reaction product distributions Peeters et al., 2009;Peeters and Müller, 2010;da Silva et al., 2010;Crounse et al., 2011;Wolfe et al., 2012;Crounse et al., 2012Crounse et al., , 2013Liu et al., 2013). Much of this work has focused on isoprene, which comprises one third of the global biogenic VOC emission budget . Revised isoprene mechanisms can profoundly impact model predictions of ozone, aerosol, radicals and reservoir species (Xie et al., 2013;Mao et al., 2013). In most instances, uncertainties in these reaction mechanisms center on the fate of first-generation RO 2 and on processes that control the balance of OH, HO 2 and RO 2 .
Observations of total peroxy radicals in forested environments are relatively sparse. During the ROSE (Rural Oxidants in the Southern Environment) campaign in rural Alabama, Cantrell et al. (1992Cantrell et al. ( , 1993 reported typical sunny midday RO 2 + HO 2 mixing ratios of 100-150 pptv in an isoprene-rich environment under minimal anthropogenic influence (NO ∼100 pptv at noon). Maximum values of up to 300 pptv (parts per trillion by volume) were observed on several days, and a steady-state model predicted an RO 2 / HO 2 ratio of ∼1 for one prototypical day. Qi et al. (2005) also observed fairly high RO 2 + HO 2 mixing ratios (109-134 pptv at midday) above a Japanese mixed deciduous forest with typical NO mixing ratios less than 200 pptv, and it was noted that peroxy radical concentrations consistently peaked ∼3 h after solar noon. At the PROPHET (Program for Research on Oxidants: PHotochemistry, Emissions, and Transport) site in northern Michigan, Mihele and Hastie (2003) measured midday RO 2 + HO 2 ranging from 20 to 60 pptv. As in Qi et al. (2005), peroxy radicals at PROPHET were found to peak several hours after solar noon, indicating an important role for processes other than primary radical production. More recently, Griffith et al. (2013) reported midday mixing ratios of 20-50 pptv for the sum of HO 2 and first-generation isoprene peroxy radicals at PROPHET during summers of 2008 and 2009, consistent with earlier observations. Observations of HO 2 are more ubiquitous and are reviewed elsewhere (Stone et al., 2012), though some of these measurements may contain positive artifacts due to alkene-derived RO 2 (Fuchs et al., 2011). Investigations of RO x cycling are often constrained with measurements of OH and HO 2 but rarely include a constraint on RO 2 . Since RO 2 can comprise half or more of the total peroxy radical budget, such observations are crucial for identifying and eliminating gaps in chemical mechanisms.
We present an analysis of peroxy radical measurements obtained within a ponderosa pine forest during the 2010 Biohydro-atmosphere interactions of Energy, Aerosols, Carbon, H 2 O, Organics and Nitrogen -Rocky Mountain Organic Carbon Study (BEACHON-ROCS). Using the comprehensive suite of observations to constrain a near-explicit 0-D chemical box model, we examine the diel cycle of sources, sinks and partitioning of peroxy radicals in this biogenic environment. Model underestimation of daytime peroxy radical concentrations leads us to consider potential missing radical sources. We quantify these missing processes and place them within the context of canonical chemistry. Analyzing the temporal profiles of missing peroxy radical mixing ratios and production rates, we identify potential novel mechanisms of radical generation that are consistent with previous anomalous observations at this and other forests.

Field campaign
BEACHON-ROCS took place from 1 to 31 August 2010 at the Manitou Forest Observatory (39 • 06 00 N, 105 • 05 30 W, 2286 m above sea level). Site characteristics and project details are presented elsewhere (Ortega et al., 2014). The research site is situated within a ponderosa pine forest with an average canopy height of 18.5 m and no significant understory. A leaf area index (LAI) of 3 for the tree canopy and a tree cover fraction of 0.38 gives a landscape average LAI of 1.14. The closest major urban areas are Colorado Springs (33 km SE) and Denver (70 km N). The site is occasionally impacted by anthropogenic air masses, but prevailing winds bring relatively clean air from the south and southwest. We have not screened the data for anthropogenic influence, but visual inspection reveals no obvious dependence of peroxy radical mixing ratios on wind direction. Major biogenic VOC emissions at this site include 2-methyl-3-butene-2-ol (MBO), monoterpenes (MT) and methanol (Kaser et al., 2013a;Kaser et al., 2013b). Isoprene mixing ratios are typically less than 300 pptv, as this compound is only emitted from ponderosa pines at very low rates (Kaser et al., 2013a). Further details on the site and observations can be found elsewhere (DiGangi et al., 2011;Di-Gangi et al., 2012;Karl et al., 2012;Kim et al., 2013;Kaser et al., 2013a, b).

Peroxy radical measurements
Two classes of peroxy radicals were measured via Peroxy Radical Chemical Ionization Mass Spectrometry (PeRCIMS), described in detail previously (Edwards et al., 2003;Hornbrook et al., 2011). The instrument was housed in a trailer on the forest floor with the inlet protruding 0.5 m from the trailer wall at a height of 1.6 m and oriented to the southeast. The upper part of the PeRCIMS inlet is isolated from sunlight by shielding it with black felt cloth inside the inlet pylon, thereby minimizing the impact of potential artifacts from solar radiation to negligible levels. The inlet is maintained at a minimum temperature of 10 • C; for the conditions of BEACHON-ROCS, the heater was rarely active. The typical sample residence time is 0.18 s in the chemical reaction region and 0.4 s in the ion reaction region. A picture of the trailer and inlet is included in the supplementary material (Fig. S9).
The principle of detection involves three steps. First, ambient air is diluted with either N 2 or O 2 , followed by addition of varying concentrations of NO and SO 2 . The reaction of HO 2 with NO generates OH via (R4), while the reaction of RO 2 with NO generates an RO radical that can react either with O 2 to form HO 2 via (R6) or with NO to make RONO.
Modulation of the NO / O 2 ratio controls the relative rates of (R6) and (R10), allowing the PeRCIMS to operate in either an HO 2 (high NO / O 2 ) or an RO 2 + HO 2 (low NO / O 2 ) measurement mode. The instrument switches between these two modes every 30 seconds. In the second and third steps, OH radicals react with SO 2 to generate sulfuric acid, which is then ionized via reaction with nitrate ions. Ions are detected via a custom-built quadrupole mass spectrometer. OH radicals generated via Reaction (R4) to react with the excess NO and form HONO. Generally, the duration of ambient and background measurements is equal. Peroxy radical mixing ratios are reported as 1 min averages, with an uncertainty of ± 35 % and a detection limit of 2 pptv (for signal/noise = 2) for data from each measurement mode. HO 2 measurements acquired via titration with NO can contain positive artifacts due to partial conversion of RO 2 , especially when the RO 2 radicals are produced via OH addition to alkenes. The primary type of RO 2 interfering in the HO 2 measurement are likely β-hydroxyalkylperoxy radicals (βRO 2 ), formed via OH addition to alkenes (Fuchs et al., 2011;Hornbrook et al., 2011;Whalley et al., 2013a). Upon reaction with NO in the PeRCIMS inlet, these radicals quickly decompose to make an HO 2 radical that is then detected with nearly the same efficiency as ambient HO 2 . Laboratory experiments have shown that, relative to ambient HO 2 , the PeRCIMS sensitivity to isoprene-derived RO 2 is ∼12 % higher, while that to aromatic RO 2 is ∼12 % lower (Hornbrook et al., 2011). Sensitivities to other βhydroxyalkylperoxy radicals have not been tested, but we assume their chemistry will be similar. For this reason, we define two quantities: HO * 2 represents the "HO 2 mode" observations, which we assume to include HO 2 and βRO 2 . RO * 2 denotes the difference between total peroxy radicals and HO * 2 , representing an operationally defined subset of the RO 2 pool. The partitioning of total RO 2 between HO * 2 and RO * 2 is discussed further in Sect. 2.4 and Sect. 4.

Other measurements
Additional observations used in this analysis include OH, NO, NO 2 , O 3 , CO, PAN (peroxyacetyl nitrate), PPN (peroxyrpopionyl nitrate), formaldehyde, glyoxal, a suite of VOC (MBO, isoprene, α-pinene, β-pinene, limonene, camphene, a group of non-speciated monoterpenes, acetone, methanol, benzene, toluene, methyl vinyl ketone, methacrolein, acetaldehyde, propanal, n-butanal, 1,3-butadiene), total OH reactivity, temperature, pressure, relative humidity and NO 2 photolysis frequencies. Table S1 in the Supplement summarizes the measurement techniques and their associated uncertainties. It should be noted that some measurements were taken at various heights, and the heterogeneity of the forest canopy can lead to strong vertical concentration gradients for some reactive species (Wolfe et al., 2011a). Furthermore, the openness of the MFO canopy can also lead to horizontal gradients in chemistry and composition. Most observations utilized here were taken at a height of 1.6-4 m, within the trunk space of the canopy. These heights are also listed in Table S1. Multi-height measurements of VOC and NO x suggest average vertical concentration gradients of less than 4 % per me-ter in this region (data not shown). NO 2 photolysis frequencies were measured both above the canopy and at 2 m (on top of the instrument trailer, just above the OH inlet). We caution that horizontally averaged in-canopy radiation is likely higher than represented by this measurement, as the canopy is relatively open and the J (NO 2 ) sensor was co-located with the relatively shaded OH inlet. Moreover, due to the spatial separation of ∼4 m between the OH and PeRCIMS inlets, this measurement may not reflect the local conditions at the PeRCIMS inlet. Other photolysis frequencies, notably J (O 3 ), are estimated by scaling measured J (NO 2 ) with clear-sky photolysis frequencies calculated from the Master Chemical Mechanism (MCM) v3.2 parameterization (Jenkin et al., 1997;Saunders et al., 2003).
VOC observations were acquired by four separate instruments, each with particular measurement heights, time resolution and speciation. A detailed comparison of these measurements is presented elsewhere (Kaser et al., 2013b). Due to better temporal coverage, MBO, benzene, toluene and acetaldehyde observations are taken from the University of Innsbruck PTR-TOF-MS (proton-transfer reaction timeof-flight mass spectrometry) measurements at 25 m. The remaining VOCs listed above are taken from the NCAR trace organic gas analyzer (TOGA) measurements (also at 25 m). Furthermore, mixing ratios of MBO and monoterpenes, which are emitted by ponderosa pine, are corrected for in-canopy gradients as described in the Supplement. Vertical gradients of these compounds can be significant (up to 80 % change in mixing ratios between 4 and 23 m), thus this correction is a necessary step towards accurately representing in-canopy reactivity.

Model calculations
Detailed model calculations utilized the University of Washington Chemical Model (UWCM). This model incorporates a subset of the MCM v3.2 (Jenkin et al., 1997; with several updates and additional chemistry as detailed elsewhere (Wolfe and Thornton, 2011). Model setup is similar to that described by Kim et al. (2013) but with a few modifications. All input data are averaged to a 30 min diel cycle. Constraints include all observations listed above and constant mixing ratios of 1770 ppbv (parts per billion by volume) for CH 4 and 550 ppbv for H 2 . The MCM mechanism subset includes all reactions from oxidation of MBO, isoprene, α-pinene, β-pinene, limonene, benzene, toluene, butadiene, acetaldehyde, propanal, n-butanal and methane. The simple MT mechanism of Wolfe and Thornton (2011) is used for camphene and a group of unspeciated monoterpenes (the latter are assumed to have the same chemistry as β-pinene). Isomerization of first-generation isoprene peroxy radicals Peeters et al., 2009) is also included using measured rate coefficients for isomerization (Crounse et al., 2011) and subsequent loss of hydroperoxyaldehydes ; recall that isoprene is, Atmos. Chem. Phys., 14, 4715-4732, 2014 www.atmos-chem-phys.net/14/4715/2014/ however, a relatively minor contributor to VOC chemistry at this site. To account for cloud and forest cover, the ratio of in-canopy observed and MCM-calculated J (NO 2 ) is used to generate a scaling factor, which is then applied to all MCMparameterized photolysis frequencies. Some of our analysis below suggests that certain peroxy radicals may be sensitive to direct sunlight (Sect. 5.1); however, the attenuated light profile better represents photolytic sources of OH at the measurement location, and we use observations of this radical to test proposed mechanisms. Emissions and deposition are not explicitly considered, but an additional first-order loss process with a lifetime of 24 h is given to all species to represent physical losses (deposition and advection/dilution) and prevent buildup of long-lived products. The model is initialized at midnight and integrated for three days with observational constraints updated every 30 min. Three days is sufficient to reach a diurnal steady state, and results are shown from the third day. Specific model scenarios are described in the appropriate sections. Uncertainties in model mixing ratios are estimated from observational uncertainties as described in the supplement.
Comparing model results with HO * 2 requires some assumptions regarding the contribution of organic peroxy radicals to this measurement. To first order, we assume that the HO * 2 measurement includes HO 2 and all first-and second-generation β-hydroxyalkylperoxy radicals produced from OH oxidation of MBO, isoprene, monoterpenes, MVK, MACR, butadiene, benzene and toluene; specific radicals used are listed in Table S2 in the Supplement.

Observations
The majority of peroxy radical measurements were collected during the final 2 weeks of the campaign (16-30 August). Figure S1 of the online supplement shows the full time series for peroxy radical mixing ratios, OH concentrations and meteorology. Meteorological conditions were warm and moderately dry (average midday temperature and relative humidity of 24 ± 2 • C and 27 ± 9%), with scattered clouds and occasional rain after midday. Total peroxy radicals exhibited a regular diel cycle with daytime maxima of 100-180 pptv and nighttime minima of 0-10 pptv, within the range of observations from several other forest sites (Cantrell et al., 1992;Cantrell et al., 1993;Qi et al., 2005). HO * 2 tends to track total peroxy radicals, though the ratio of HO * 2 to total peroxy radicals (not shown) can vary significantly (from 0.4 to ∼1) throughout the day. Daytime OH mixing ratios ranged from 3 × 10 6 to 10 × 10 6 molec cm −3 , while nighttime values were typically below the instrument detection limit of 5 × 10 5 molec cm −3 . Due to the regularity of diel cycles at this site and intermittent temporal overlap for many observations, our analysis will focus mainly on average diel behavior. All times discussed below refer to local solar time. Nighttime OH values were typically below the instrument detection limit (5 × 10 5 cm −3 ) and are thus set to half of this value.
The mean diel cycle of peroxy radicals (Fig. 1) displays several interesting features. After sunrise, HO * 2 and HO 2 + RO 2 rise synchronously with both OH and O 3 photolysis frequency, consistent with photochemical sources. From 11:30 to 14:30, there is a sharp rise in HO * 2 and HO 2 + RO 2 that exceeds the smoother diel cycle seen in OH concentrations. This maximum, along with the brief spike at ∼15:30, are consistent daily features and not an averaging artifact. In the afternoon, the decay of HO * 2 is similar to that of OH, while HO 2 + RO 2 exhibits a shoulder that persists until ∼16:30.
Nighttime peroxy radicals are low but consistently above the PeRCIMS detection limit of 2 pptv, with mean mixing ratios from 00:00 to 05:00 of 4 ± 2 pptv and 6 ± 2 pptv for HO * 2 and HO 2 + RO 2 , respectively. Observations in at least one semipolluted area have suggested that NO 3 (nitrate radical) and/or ozone-driven chemistry can drive significant nighttime peroxy radical production in the presence of alkenes (Andres-Hernandez et al., 2013). For the conditions of BEA-CHON, Fry et al. (2013) have demonstrated that the oxidation of monoterpenes by NO 3 chemistry can be a significant source of particulate organic nitrates. Thus, even though peroxy radical levels are low, they imply significant radicaldriven processes occurring in the absence of sunlight. Figure 2 presents diel cycles for other key observations. Data are averaged over the full campaign (1-31 August). Unfortunately, intermittent data gaps do not permit more refined data selection. Though the degree of coverage varies for different species, most observations demonstrate consistent patterns from day to day. The sharp early morning rise in NO x may be due to entrainment from aloft during the breakup of the nocturnal boundary layer (Seok et al., 2013) or surface emission (Alaghmand et al., 2011). The onset of this feature is synchronous with a rapid rise in ozone, likely attributable to the former mechanism. Observed daytime NO mixing ratios of 100-150 pptv are typical for a rural continental site and are within a transition region where reactions with both NO and other peroxy radicals (HO 2 and RO 2 ) are expected to contribute significantly to total peroxy radical loss. MBO and monoterpenes, the primary emissions of ponderosa pine, dominate the reactive VOC budget. Methanol is also abundant at this site (Kaser et al., 2013b), but it is relatively inert. Oxidation products, including formaldehyde, glyoxal and PAN, build up throughout the day, peaking in the late afternoon and decaying at night due to deposition, thermal decomposition and other losses. OH reactivity, a measure of the total OH loss rate, maintains a fairly constant value of 6-7 s −1 during the day and rises to as much 14 s −1 at night.

Model results
Two initial model scenarios are considered. In the first simulation, all observations other than peroxy radicals are used to constrain the model ("base"), while in a second simulation OH is determined by the model ("ModOH"). Figure 3 compares model output with RO x observations for both scenarios, and additional model results are shown in the Supplement (Fig. S4). In the base case, total peroxy radicals are underpredicted throughout the day, with errors of up to a factor of 3 at midday. Unconstraining OH decreases middaymodeled peroxy radicals by more than 50 %. The diel cycle of modeled peroxy radicals closely tracks that of OH in both scenarios. The morning rise in modeled and measured per-oxy radical mixing ratios differs by several hours, suggesting a role for processes other than OH-driven VOC oxidation in early morning production. Moreover, the model captures neither the strong midday maximum nor the afternoon shoulder. In the ModOH scenario, OH concentrations are underpredicted throughout the day. This missing OH is likely at least partly due to an underprediction of HO 2 , which reduces the rate of OH production via (R4).
The relationship between OH and peroxy radicals is somewhat complicated by photochemical gradients and the role of vertical transport. Above the canopy, increased sunlight likely leads to more OH and thus more peroxy radicals, as demonstrated by a separate set of model runs constrained by above-canopy J (NO 2 ) data (Fig. S5). Some of these radicals will be transported into the canopy, sustaining photochemistry amidst attenuated radiation. To our knowledge, there are no published data comparing within-and above-canopy peroxy radical levels; however, detailed 1-D canopy modeling results predict relatively minor gradients in HO 2 and total RO 2 at other forests (Bryan et al., 2012;Makar et al., 1999;Wolfe and Thornton, 2011;Wolfe et al., 2011a). For example, Wolfe and Thornton (2011) calculate that peroxy radical mixing ratios change by less than 10 % within a ponderosa pine forest similar to that of Manitou forest. This may reflect the fact that decreased in-canopy production is balanced by downward mixing. Thus, model underestimates of peroxy radicals may be partly due to a missing source from downward transport, but only inasmuch as production and destruction rates vary between above-and below-canopy environments. Moreover, the tight coupling of OH and HO 2 suggests that this phenomenon can be mostly accounted for by constraining OH to measurements. In this case, modeled Atmos. Chem. Phys., 14, 4715-4732, 2014 www.atmos-chem-phys.net/14/4715/2014/  Fig. 4. Modeled distribution of organic peroxy radicals from the base simulation. Groups include methyl peroxy radical (red), firstgeneration peroxy radicals of MBO (orange), isoprene (green) and monoterpenes (cyan), total acyl peroxy radicals (blue), firstgeneration radicals from oxidation of other measured VOC (magenta), and "secondary" radicals resulting from oxidation of modelpredicted, unmeasured VOC (gray). The ratio of HO 2 to total peroxy radicals is also shown (black diamonds). peroxy radical mixing ratios are essentially independent of radiation (Fig. S5); in other words, OH-initiated chemistry is the dominant peroxy radical source in the model.
Underprediction of OH reactivity could also lead to disagreement between modeled and measured peroxy radicals. Missing OH reactivity is a common feature of many studies in regions dominated by biogenic VOC (Lou et al., 2010;Edwards et al., 2013). A detailed analysis of OH reactivity at BEACHON-ROCS shows that measured species can only account for 41 % of the total OH reactivity on average, with poorer agreement at night (Nakashima et al., 2014). In the current study, the model underestimates total measured OH reactivity by as much as 20 % during the day and by 40-50 % at night for both scenarios (Fig. S4). It is important to note that unmeasured VOC oxidation products comprise a substantial fraction (up to 45 %) of the modeled OH reactivity, and the abundance of these compounds is highly sensitive to the assumed dilution rate of 1 day −1 (Edwards et al., 2013). For example, increasing the "physical loss" lifetime (Sect. 2.4) from 24 to 48 h increases modeled OH reactivity by ∼1 s −1 . We must also be wary of in-canopy heterogeneity. OH reactivity was measured at 4 m above the ground but likely exhibits a vertical gradient within the canopy (Mogensen et al., 2011). From PTRMS observations of MBO and monoterpenes at 1 m and 4 m (data not shown), we expect total OH reactivity to change by less than 10 % between 4 m and the peroxy radical measurement height of 1.6 m. Even with these caveats, modeled and measured OH reactivity agree to within 20 % during the day, suggesting that the base model adequately represents the overall rate of OH-driven RO 2 production. This implies that other processes must also influence the peroxy radical budget.
Overall, modeled HO * 2 agrees somewhat better with observations than does total peroxy radicals, especially with OH constrained (Fig. 3c). Based on the above assumptions, HO 2 comprises 60-70 % of the modeled HO * 2 , giving peak midday modeled HO 2 mixing ratios of 25 and 12 pptv for the base and ModOH cases, respectively (Fig. S4 in the Supplement). Roughly 50-60 % of total modeled RO 2 is produced from alkene oxidation (Fig. 4) and thus predicted to be detected as HO * 2 . The remainder, which we refer to as RO * 2 , can be compared with observations by taking the difference between measured HO 2 + RO 2 and HO * 2 (Fig. 3d). RO * 2 is generally underestimated, especially in the afternoon where the "shoulder" from HO 2 + RO 2 appears as a distinct peak. This feature is an important clue regarding the nature of missing RO 2 .
The distribution of RO 2 radicals in the model closely follows that of VOC precursors (Fig. 4). MBO-derived RO 2 is the most abundant component during the day, while nighttime chemistry is dominated by monoterpenes, particularly β-pinene and limonene. This trend occurs for two reasons. First, MBO emissions are dependent upon both light and temperature (Harley et al., 1998), while monoterpene emissions at this site scale primarily with temperature (Kaser et al., 2013a), leading to different diel cycles in their concentrations (Fig. 2b). Secondly, MBO reacts almost exclusively with OH, while monoterpenes are reactive towards OH, O 3 and the nitrate radical (NO 3 ). The latter oxidant typically only accumulates at night (Fry et al., 2013). Our model results indicate similar contributions from OH and NO 3 chemistry to monoterpene oxidation in the first half of the night, with OH-driven loss prevailing in the early morning as NO 3 decays alongside its precursor, ozone. "Secondary" RO 2 arising from oxidation of unmeasured VOC comprise as much as 49 % of modeled RO 2 , consistent with our earlier discussion of OH reactivity. HO 2 comprises 35-50 % of the total peroxy radical budget (Fig. 4). The diel cycle of HO 2 /(HO 2 + RO 2 ) is essentially the same for both model scenarios, demonstrating that absolute OH concentrations influence peroxy radical abundances but not partitioning. The midday minimum in this ratio reflects the increased importance of Reaction (R8) as a sink of HO 2 (see also Fig. 6).
In theory, model results can be used to estimate the contribution of HO 2 to HO * 2 observations. Several potential approaches are discussed in the Supplement. All of these methods inherently assume that model output faithfully represents the true peroxy radical distribution; however, without additional constraints on the nature of "missing" peroxy radicals, it is difficult to judge the reliability of this assumption. Thus, we elect to focus our analysis on the measured quantities, HO * 2 and RO * 2 .  Fig. 6. Rates of HO 2 production and loss calculated from the base model scenario. Rates are grouped according to the type of reaction. Production includes reaction of RO 2 with NO (gray), VOC photolysis (orange) and OH oxidation of VOC (green). Losses include reactions with NO (gray), HO 2 (blue), RO 2 (orange) and OH and ozone (green). The thick black line denotes the proposed missing HO 2 source, calculated as described in the text.

Analysis and discussion
Modeled and observed peroxy radicals agree to within their respective uncertainties for much of the day (Fig. 3). Clear systematic discrepancies between average mixing ratios, however, suggest that the model is missing or misrepresenting sources and/or sinks of these species. Two features that stand out are the large midday maximum and the afternoon "shoulder," neither of which are captured by the model.  Fig. 3. In the HO 2 -constrained case (green), HO 2 is constrained to "missing" HO * 2 as described in Sect. 5.1. The same uncertainties are assumed for both model scenarios. (b) Scatter plot of the same results for ModOH (magenta dots) and HO 2 -constrained (green asterisks) scenarios. Thin lines define uncertainties in modeled and measured values. Thick lines represent uncertainty-weighted major-axis-regression fits (ModOH: y = 0.24x + 0.18, r 2 = 0.81; ConHO 2 : y = 1.19x -0.05, r 2 = 0.75).
in HO * 2 arises from the midday maximum and the sharp afternoon satellite peak (Fig. 1), while RO * 2 underprediction persists throughout the day and includes most of the broad afternoon shoulder. These features allude to multiple mechanistic issues, and we examine each observation separately.

Missing HO * 2 : a photolytic HO 2 source?
Understanding the nature of the midday HO * 2 maximum requires that we first determine whether this peak is primarily due to changes in HO 2 or RO 2 . Given that (1) RO * 2 does not display a similar feature, and (2) we expect most RO 2 to exhibit similar diel cycles (Fig. 4), additional HO 2 production is the simplest explanation. Production of specific RO 2 that contributes to HO * 2 , such as from photolysis of some yet-unidentified VOC, is also possible but less likely. This hypothesis implies several testable consequences, as detailed below.
To gauge the magnitude of this putative HO 2 source, we first examine the rates of HO 2 production and loss calculated from the 0-D model (Fig. 6). Reaction of RO 2 with NO comprises ∼60 % of the modeled HO 2 source, with smaller contributions from photolysis and OH reaction of oxidized VOC (mainly formaldehyde, HCHO). NO chemistry also dominates the loss of HO 2 , though model underestimates of peroxy radical concentrations likely lead to an underestimate in the loss rates from reaction with HO 2 and RO 2 . We can quantify the "missing" production rate by assuming that HO 2 is in steady state and calculating the loss rate of the missing HO 2 : Here, [HO * 2 ] miss is the concentration of missing HO * 2 (Fig. 5) and τ HO 2 ,mod is the lifetime of HO 2 calculated from the base model scenario. At its peak, the missing production rate is nearly double the total production rate from all known sources (Fig. 6, black line). For perspective, the maximum missing production rate of 102 ppt min −1 is 8 times the HO 2 production rate from reaction of MBO-derived RO 2 with NO and 24 times the production rate from HCHO photolysis. Reducing the HO 2 lifetime to account for underpredicted reaction rates with HO 2 and RO 2 (estimated using total peroxy radical measurements and the modeled HO 2 / RO 2 ratio) increasesP miss by less than 20 %.
Increased HO 2 will provide an additional source of OH. Thus, as a further test, we incorporate this extra HO 2 source as an additional zero-order reaction in the ModOH scenario and compare model-calculated OH with observations (Fig. 7). This modification significantly increases daytime OH concentrations. Modeled OH generally agrees with observations (to within combined uncertainties) in the morning and afternoon hours but is overpredicted in the early afternoon, concomitant with the sharp HO * 2 maximum. The overall model-measurement agreement improves (slope = 0.25 vs 1.19, Fig. 7b), but the correlation degrades somewhat (r 2 = 0.81 vs. 0.75). These results suggest that such an HO 2 radical source cannot be invoked without additional changes to the mechanism. Overprediction of OH at the peak of extra HO 2 production likely indicates that some fraction of the missing HO * 2 is actually RO 2 , counter to the simple assumption made above. Again, we caution that the magnitude of this missing HO * 2 source is highly uncertain. Furthermore, these results support the conclusion of Kim et al. (2013) that HO 2 can be a major source of OH in the canopy environment.
A closer analysis of the observational data set provides further characterization of the missing source. Figure 8 shows an example of the relationship between peroxy radicals and solar radiation on 22 August; similar correlations were observed on many days of the campaign. Cloud cover regularly reduces direct sunlight at this site, decreasing above-canopy J  (22 August). HO * 2 (red) and RO * 2 (black) are plotted on the left axis, while NO 2 photolysis frequencies are plotted on the right axis for both above-canopy (light gray) and below-canopy (dark gray) measurements. All data are 1 min averages.
of the radiation here is a combination of scattered diffuse light and occasional sun flecks (i.e., direct sun). In-canopy J (NO 2 ) is a factor of 3-6 lower than above-canopy clear sky values (Fig. S5); we again caution that this is not necessarily representative of the "average" in-canopy environment (Sect. 2.3). Indeed, the results in Fig. 8 suggest that direct sunlight penetrates much of the overstory throughout midday. This is consistent with our expectations for this relatively open canopy (LAI = 1.14, tree coverage = 38 %). HO * 2 increases to ∼80 pptv from 11:30 to 14:30, consistent with the midday maximum in the diel average; however, HO * 2 also decreases rapidly during periods of sustained radiation attenuation, down to levels similar to those observed before and after the maximum. RO * 2 exhibits some correlation with above-canopy radiation (e.g., the troughs at times 12:40 and 15:00), but generally the correlation is weaker and variations in RO * 2 are more independent of radiation. These trends are also borne out in the broader statistics of the full data set (Fig. S7). The difference in the radiation dependence of HO * 2 and RO * 2 suggests that fast changes in HO * 2 are not solely driven by the radiation dependence of OH.
The simplest explanation for the observed behavior is production of peroxy radicals from gas-phase oxidation and/or photolysis of VOC. Additional production from OH chemistry is unlikely, as the model is constrained by measured OH and reproduces observed OH reactivity to within 20 % during the day (Fig. S4). Only a handful of VOC are known to exclusively produce HO 2 during OH oxidation, notably formaldehyde (HCHO) and glyoxal (HCOCHO), both of which are constrained by observations in our model. Gas-phase photolysis of an unidentified compound is another potential explanation, though the emission (or production) of such a molecule would need to match the unique profile of the HO * 2 midday maximum (Fig. 5). None of the other 213 meteorological and chemical observations resemble this profile, thus we have no additional clues as to the nature of this source. HO 2 production from reactions of "missing" RO * 2 with NO may also explain some of the missing HO * 2 , but not the midday maximum (see Sect. 5.2 and Fig. 11).
Numerous investigations have inferred the presence of significant unidentified reactive hydrocarbons in biogenic environments. Often this conclusion arises from discrepancies between measured and calculated OH reactivity (Di Carlo et al., 2004;Lou et al., 2010;Sinha et al., 2010;Mogensen et al., 2011;Nölscher et al., 2012;Edwards et al., 2013). It is still debated whether the missing reactivity is due to primary emissions or secondary oxidation products, though this likely varies from site to site. While underrepresented OH reactivity could have a profound impact on peroxy radical chemistry elsewhere, we reiterate that this is likely not a viable explanation for missing peroxy radicals in the present study. Fast downward ozone fluxes (Goldstein et al., 2004;Kurpius and Goldstein, 2003) and high levels of oxidized VOC (Holzinger et al., 2005) have also been taken as evidence for unconventional in-canopy chemistry. Both of these findings originate from observations in a ponderosa pine ecosystem similar to the Manitou Experimental Forest, thus it is conceivable that similar chemistry is at play. Perhaps the most relevant line of evidence for the present study is the observation of unexpectedly large upward fluxes of HCHO during BEACHON-ROCS (DiGangi et al., 2011). Formaldehyde is a major product from the oxidation of nearly every VOC and is thus an excellent tracer for the overall efficiency of hydrocarbon degradation. DiGangi et al. (2011) demonstrated that a simple mass balance model incorporating known chemical and physical processes underpredicted the observed HCHO flux by a factor of 6. The investigators concluded that the missing HCHO source could be attributed to oxidation of unidentified biogenic VOC and/or direct emissions of HCHO from vegetation. In either case, closure of the HCHO flux budget required that the missing process correlate with solar radiation.
It is possible that the missing sources of HCHO and HO * 2 are related. The maximum "missing" HCHO flux of ∼20 pptv m s −1 implies an in-canopy HCHO production rate of 65 pptv min −1 (for a canopy height of 18.5 m), within the range of the missing HO * 2 source (Fig. 6). Candidate precursors for both HCHO and HO 2 are methylperoxy, hydroxymethyl and β-hydroxyalkoxy radicals, which decompose rapidly under normal atmospheric conditions: The lifetimes of these radicals are so short that the above reactions are often assumed to be instantaneous. In conventional chemical mechanisms, these radicals are intermediates The concentration-weighted average RO * 2 lifetime is also shown (dashdotted magenta line). (b) Missing RO * 2 production rates calculated as described in the text. Also shown is the total production rate for all modeled RO 2 * (thick black line). Model results are taken from the base simulation. of peroxy radical decomposition (mainly via reaction with NO). While known photolytic sources for such molecules are thought to be minor, it is possible that photolysis of yetunidentified VOC could simultaneously stimulate production of both HCHO and HO 2 , which in turn would accelerate HO x -driven photochemistry within and above the forest canopy.

Missing RO * 2 : evidence for unidentified VOC?
In contrast to HO * 2 , missing RO * 2 -mixing ratios exhibit a relatively smooth diel cycle (Fig. 5). This signal comprises most of the total missing peroxy radicals in the morning and afternoon, with a maximum at ∼16:00. The shape of the diel profile is notably similar to that of several oxidation products, including HCHO, glyoxal and PAN (Fig. 2c). Many modeled oxidation products, including first-generation peroxides and organic nitrates, also peak at this time (not shown). RO 2 sinks are dominated by reaction with NO (Fig. 9), and underprediction of HO 2 in the base simulation likely results in underestimation of RO 2 loss via reaction with HO 2 . Thus, missing RO * 2 reflects an issue with RO 2 sources.
We can estimate the magnitude of the missing RO * 2 source using an approach similar to that described for HO * 2 (Eq. 3). This method requires calculation of the missing RO * 2 lifetime; however, this value depends on the assumed structure of these peroxy radicals. Figure 9a illustrates this point for three representative peroxy radicals. The lifetime of CH 3 O 2 and MBOAO2 (the primary RO 2 from MBO oxidation) ranges from 30 to 60 s throughout the day, except in the morning when NO concentrations spike. In contrast, the lifetime of the acetyl peroxy radical, CH 3 CO 3 , is typically < 20 s. These differences arise mainly from NO reaction rate constants, which are 7.7, 9.0 and 20 × 10 −12 cm 3 molec −1 s −1 at 298 K for CH 3 O 2 , MBOAO2 and CH 3 CO 3 , respectively. Figure 9a also shows the concentration-weighted average RO * 2 lifetime for all model species in the RO * 2 group. Coincidentally, this lifetime is nearly identical to that of MBOAO2 even though this radical is not included in RO * 2 . Figure 9b compares the total production rate of modeled RO * 2 with missing RO * 2 production rates as calculated via the RO 2 * equivalent of Eq. (3). The magnitude of the missing production rate is similar to that of the "known" production rate except when RO * 2 is assumed to have a lifetime comparable to CH 3 CO 3 . For the other three cases, missing RO * 2 production follows a diurnal pattern similar to its concentration profile (Fig. 5c) except in the morning, where the steady-state assumption may be invalid due to rapidly changing NO concentrations. Recall that the absolute magnitude of this source is dependent on our estimate of missing RO * 2 and thus is highly uncertain.
The contrast between diurnal cycles of production rates for "known" and missing RO * 2 (Fig. 9b) demonstrates that the processes driving missing RO * 2 are not solely sun-driven (i.e., OH reaction with VOC). Figure 10 shows modeled tendencies for the same three representative RO 2 species (MBOAO2, CH 3 O 2 and CH 3 CO 3 ). Overall radical production tracks with solar radiation, but some processes exhibit minor diurnal asymmetry. In particular, production of CH 3 O 2 and CH 3 CO 3 radicals from reactions of other RO 2 with NO maximizes slightly after midday due to both an increase in NO (Fig. 2a) and the buildup of RO 2 . Despite these features, OH is still the main driver for classical RO 2 production, and none of the 347 modeled-RO 2 species exhibit a profile similar to that of the missing RO * 2 . While amplification of a purely OH or light-dependent process may be able to explain missing RO * 2 in the morning, such a source cannot explain the afternoon maximum. These results lead us to consider other non-OH RO 2 sources that may be underrepresented in the model mechanism.
One potential candidate for missing RO * 2 is the acyl peroxy (AP) radical family. AP radicals are a special class of peroxy radical that can react with NO 2 to form a metastable acyl peroxy nitrate (APN): Temperature controls the lifetime of APNs; for the conditions of BEACHON-ROCS (4-29 • C), the lifetime of PAN ranges from 23 h to 24 min. APNs can act as a source or sink of AP radicals depending on equilibrium conditions and the strength of primary AP sources (LaFranchi et al., 2009). Our model predicts that PAN, the most abundant APN, is a net sink for acetyl peroxy radicals (Fig. 10c); however, the net rate of CH 3 CO 3 loss via PAN formation is small compared to primary CH 3 CO 3 production and reaction with NO, indicating near-steady state conditions (Cleary et al., 2007;LaFranchi et al., 2009). Moreover, the model predicts a number of additional APNs, with PAN (which is constrained by observations) comprising only 19-39 % of the total budget. The next most abundant APN is the MCM species C4PAN5, a byproduct of MBO oxidation, at 17-22 %. The shape of the missing RO * 2 profile may imply an underestimation in the source of AP radicals from decomposition of such APNs. Assuming steady-state conditions for APNs, the AP radical concentration from this source alone is given by where k 16r and k 16f are the reverse and forward rate constants for Reaction (R16). Based on this equation, errors in APN concentrations and/or reaction rates could lead to underprediction of AP radicals. Errors in APN concentrations are likely not a viable explanation. Evidence from other investigations suggests that > 90 % of the total peroxy nitrate budget is comprised of only a handful of APNs, mostly PAN (Wooldridge et al., 2010); thus, it is likely that modeled APN concentrations are already overestimated. Errors in rate constants are more probable, as MCM rate constants for APN formation and loss are assumed equal to that of PAN, except for decomposition of PPN and MPAN, which follow IUPAC recommendations (Jenkin et al., 1997;Saunders et al., 2003). Laboratory data from Kirchner et al. (1999) indicate that decomposition rates generally decrease with increasing size and decreasing electronegativity of the organic functional group. In contrast, IUPAC (International Union of Pure and Applied Chemistry) recommends decomposition rate constants for PPN and MPAN that are ∼10 % faster than that of PAN. Formation rate constants have not been measured for species other than PAN, though one study has suggested that PPN formation may be 11 % slower than that of PAN (Sommariva et al., 2011). To completely explain missing RO * 2 , we estimate that the equilibrium constant (k 18f /k 18r ) for model APNs would need to decrease by more than a factor of 10, well beyond the likely uncertainty in this value. Moreover, model simulations show that most of the growth in the RO 2 pool from such a change is due not to AP radicals themselves but rather to the RO 2 products of the reaction of AP with NO. Thus, we conclude that AP radicals are not a major component of missing RO * 2 . Other RO * 2 generation mechanisms to consider include reaction of VOC with ozone or nitrate radical (NO 3 ). In the base model simulation, ozone chemistry contributes 10-20 % to the daytime peroxy radical budget, while NO 3 chemistry is only significant at night. Ozonolysis of unidentified VOC has been invoked previously to explain anomalously high ozone fluxes (Goldstein et al., 2004;Hogg et al., 2007;Kurpius and Goldstein, 2003), oxidation product concentrations (Holzinger et al., 2005) and sulfuric acid levels (Mauldin et al., 2012) in other forests. Decomposition of Criegee intermediates can simultaneously generate OH and RO 2 radicals, with measured yields ranging from 0.06 to near-unity (Aschmann et al., 2002;Atkinson and Arey, 2003;Shu and Atkinson, 1994). In a detailed modeling study, Wolfe et al. (2011b) established an upper limit for RO 2 production from ozonolysis of "very reactive" VOC of 60 pptv min −1 , similar to both our missing RO * 2 source and the missing HCHO source inferred by DiGangi et al. (2011). The latter study also determined that any missing VOC should exhibit a light-dependent emission profile similar to that of MBO.
To test this hypothesis, we implement an additional set of reactions following the very reactive VOC mechanism described by Wolfe et al. (2011b). Specific reactions are listed in Table S3. Rate constants for initial oxidation of this hypothetical VOC are assumed equal to those of βcaryophyllene, while reactions of the peroxy radical products are assumed to be similar to those of the β-pinene-derived Atmos. Chem. Phys., 14, 4715-4732, 2014 www.atmos-chem-phys.net/14/4715/2014/  Fig. 11. Comparison of RO x observations with 0-D model results using the very reactive VOC mechanism. Solid black lines with shaded gray areas represent observations and their associated uncertainties. Model simulations include the base (solid blue line) and ModOH (dashed magenta line) scenarios augmented with very reactive VOC chemistry; in the latter case, OH concentrations are not constrained to observations. For comparison with HO * 2 and RO * 2 in (c) and (d), the total modeled RO 2 is subdivided into two groups as described in the text. Model uncertainties are excluded for clarity. radical BPINAO2. As a modification to the original mechanism, we discriminate between RO 2 made by OH, O 3 and NO 3 chemistry, since we anticipate that OH-derived RO 2 would be detected as HO * 2 in the PeRCIMS inlet. The yield of OH and RO 2 from ozonolysis is set to the upper limit of 0.1 recommended by Wolfe et al. (2011b). Very reactive VOC mixing ratios, shown in Fig. S8 in the Supplement, are fixed to a diurnal cycle that scales with the observed flux of the sum of MBO and isoprene (Kaser et al., 2013a). The scaling factor of 0.23 is chosen to optimize model-measurement agreement for total peroxy radicals in the base scenario. We caution that inferred VRVOC (very reactive VOC) mixing ratios depend directly on the assumed reaction rate constants and product yields -the product of which determines the RO 2 production rate. In other words, this calculation effectively constrains the VRVOC reactivity, as discussed in Wolfe et al. (2011b). For these conditions, modeled OH reactivity increases by as much as 30 %.
We implement this mechanism for both the base and ModOH scenarios; results are shown in Fig. 11. Modelmeasurement agreement for all radicals improves markedly on incorporating very reactive VOC chemistry (compare to Fig. 3), though the unique diurnal patterns of HO * 2 and RO * 2 are not captured. With OH constrained to observations, model agreement with HO * 2 improves due to (1) RO 2 from reaction of very reactive VOC with OH and (2) increased HO 2 from reaction of new RO 2 radicals with NO. With OH determined by the model, however, HO * 2 is overpredicted in the afternoon. This is mainly driven by excess OH co-produced with RO 2 during ozonolysis of very reactive VOC. On the other hand, the latter fraction of RO 2 also improves agreement with RO * 2 . Thus, the ozonolysis of unidentified hydrocarbons alone cannot provide closure of the RO x budget unless (1) OH is not produced with the same yield as RO 2 , contrary to canonical mechanisms, or (2) additional processes are invoked that affect the diel OH profile. Nonetheless, these results strongly support an intimate chemical link between missing peroxy radicals and other yet-unexplained observations (i.e., HCHO and ozone fluxes) within the canopy airspace.

Conclusions
Using the comprehensive suite of observations from the 2010 BEACHON-ROCS field campaign, we have explored the detailed chemistry of peroxy radicals in a rural environment dominated by biogenic hydrocarbons. Total peroxy radical concentrations are among the highest yet reported, exceeding 100 pptv on every day of observations and reaching as high as 180 pptv. Box model calculations underpredict total peroxy radicals by as much as a factor of 3, indicative of missing sources. Though the PeRCIMS instrument does not provide a strict segregation between HO 2 and RO 2 radicals, the data alludes to several distinct sources. High levels of HO * 2 at midday, combined with a clear dependence on radiation and a lack of similar behavior in RO * 2 , suggest a missing photolytic source of HO 2 . The magnitude of this missing source is highly uncertain, and its exact nature remains a mystery. RO * 2 is also underpredicted, but the diel profile of missing RO * 2 more closely resembles that of oxidation products and light-dependent VOC emissions. Implementing an additional chemical mechanism involving ozonolysis of putative "very reactive" VOC greatly improves model-measurement agreement with all radicals. With OH determined by the model, this same mechanism degrades agreement with measured OH and HO * 2 unless the OH yield is assumed to be lower than the RO 2 yield, contrary to classical chemistry. While these results do not provide closure of the radical budget, they do imply that similar mechanisms may underlie both missing peroxy radicals and other indicators of faster-than-expected chemistry at this and other forests.
Failure to accurately represent peroxy radical chemistry in biogenic regimes will limit the reliability of model results and predictions. For example, the impact of peroxy radicals on ozone production is a complex balance between the NO turnover rate and the production of radical reservoirs, such as alkyl nitrates and peroxy acyl nitrates. Modeled production of such compounds depends directly on the calculated RO 2 distribution, and changing the assumed yield of NO x reservoir species can alter ozone production on regional and larger scales (Farmer et al., 2011;Paulot et al., 2012). Interconversion of reactive nitrogen also affects estimates of nitrogen deposition, which represents a critical link between atmospheric composition and ecosystem health (Sparks et al., 2008). Peroxy radical chemistry is also intimately tied to atmospheric oxidizing capacity. In environments where HO 2 is a dominant precursor of OH, missing peroxy radical sources will lead to underestimation of the overall rate of oxidation. Understanding this chemistry is particularly critical in high-VOC, low-NO x regimes where radical cycling is controlled by RO x + HO 2 reactions and, potentially, unimolecular decomposition of larger RO 2 (Archibald et al., 2010;Stavrakou et al., 2010). For all these reasons, it is imperative that we identify and eliminate remaining gaps in our understanding of chemistry in such environments -especially if mechanistic shortcomings are as substantial as implied by this study.
Closure of the RO x budget will require a comprehensive understanding of both primary radical sources and the processes that control radical cycling in the continental boundary layer. Of critical importance is continued measurement of organic peroxy radicals, as these are too often a key missing constraint in modeling studies. Further speciation of these compounds would also greatly aid identification of errors in current chemical mechanisms. While it may not yet be technically feasible to segregate all individual RO 2 , progress could be made by determining the distribution of different RO 2 families (i.e., acyl peroxy, β-hydroxyalkylperoxy, etc.; Whalley et al., 2013b). Measurements of radical reservoirs and termination products, such as organic peroxides (ROOH) and alkyl nitrates (RONO 2 ), will also play a vital role in this regard. Observations of total ozone production (Cazorla and Brune, 2010) may also provide a useful check on observations of total peroxy radical concentrations while offering an integrated perspective on pertinent chemistry in biogenic en-vironments. Finally, detailed comparisons of regional-scale chemical transport models with in situ chemical observations and derived properties (e.g., ozone production) are needed for full evaluation of the impact of mechanistic uncertainties on current and future predictions of atmospheric composition and air quality.