Observation and modelling of HOx radicals in a boreal forest

Measurements of OH and HO2 radicals were conducted in a pine-dominated forest in southern Finland during the HUMPPA-COPEC-2010 (Hyytiälä United Measurements of Photochemistry and Particles in Air Comprehensive Organic Precursor Emission and Concentration study) field campaign in summer 2010. Simultaneous side-by-side measurements of hydroxyl radicals were conducted with two instruments using chemical ionization mass spectrometry (CIMS) and laser-induced fluorescence (LIF), indicating small systematic disagreement, OHLIF / OHCIMS = (1.31 ± 0.14). Subsequently, the LIF instrument was moved to the top of a 20 m tower, just above the canopy, to investigate the radical chemistry at the ecosystem-atmosphere interface. Comprehensive measurements including observations of many volatile organic compounds (VOCs) and the total OH reactivity were conducted and analysed using steady-state calculations as well as an observationally constrained box model. Production rates of OH calculated from measured OH precursors are consistent with those derived from the steady-state assumption and measured total OH loss under conditions of moderate OH reactivity. The primary photolytic sources of OH contribute up to one-third to the total OH production. OH recycling, which occurs mainly by HO2 reacting with NO and O3, dominates the total hydroxyl radical production in this boreal forest. Box model simulations agree with measurements for hydroxyl radicals (OHmod. / OHobs. = 1.00 ± 0.16), while HO2 mixing ratios are significantly under-predicted (HO2mod. / HO2obs. = 0.3 ± 0.2), and simulated OH reactivity does not match the observed OH reactivity. The simultaneous under-prediction of HO2 and OH reactivity in periods in which OH concentrations were simulated realistically suggests that the missing OH reactivity is an unaccounted-for source of HO2. Detailed analysis of the HOx production, loss, and recycling pathways suggests that in periods of high total OH reactivity there are additional recycling processes forming OH directly, not via reaction of HO2 with NO or O3, or unaccounted-for primary HOx sources. Under conditions of moderate observed OH reactivity and high actinic flux, an additional RO2 Abstract. Measurements of OH and HO 2 radicals were conducted in a pine-dominated forest in southern Finland

conditions is the hydroxyl radical (OH), which dominates the cleansing capacity of the atmosphere.
In the lower troposphere, the main primary source of OH on a global scale is the photolysis of ozone (O 3 ) and the subsequent reaction of the excited oxygen atom with water vapour (Levy II, 1971): Other sources of OH, e.g. photolysis of peroxides or ozonolysis of alkenes, are known as well Ren et al., 2008, and references therein). Once formed, OH reacts rapidly -with a lifetime of typically less than 1 s. Many atmospheric compounds (e.g. CO or O 3 ) subsequently produce hydroperoxyl radicals (HO 2 ). The oxidation of hydrocarbons by OH leads to the formation of peroxy radicals (RO x = RO 2 + HO 2 ). In the presence of nitric oxide (NO), RO 2 is converted to HO 2 , which reacts with O 3 or NO forming OH (e.g. Martinez et al., 2003). The main HO x (= OH + HO 2 ) sinks are self-reactions with radicals and the formation of acids and peroxides. Forests cover about one-third of the earth's total land surface (FAO, 2010). They are known to be an important global source of biogenic volatile organic compounds (BVOCs) that are very reactive towards OH (Williams, 2004). A major contribution to BVOCs, such as isoprene, monoterpenes (e.g. αpinene), and sesquiterpenes (e.g. farnesene), is emission by vegetation (Fehsenfeld et al., 1992;Guenther et al., 1995;Kesselmeier and Staudt, 1999). Therefore, forest emissions are expected to significantly influence the oxidation capacity of the atmosphere. Measurement of OH and HO 2 radicals in forest environments, however, unveils serious deficits in our understanding of the underlying processes. Different types of forests are known to emit various characteristic BVOCs influencing oxidation chemistry. Several studies in and above isoprene-rich forest environments have been conducted in the past. Aircraft measurements of atmospheric trace gases performed over the pristine Amazon rainforest during the GABRIEL campaign Kubistin et al., 2010) unexpectedly showed higher OH concentrations than predicted from box and global models. Furthermore, Kubistin et al. (2010) reported that isoprene acts as a buffer for the hydroxyl radicals, which is related to an increased HO x recycling in the oxidation mechanism of BVOCs at low NO x conditions, when the OH recycling reaction of HO 2 with NO is inhibited. Similarly, other studies from tropical forest regions Whalley et al., 2011) reported that increased recycling would be necessary to explain the measured HO x values. An increase of the deposition rates for OH reactants, like methyl vinyl ketone (MVK) and methacrolein (MACR), can help explain the measurement results for ground-based observations during the OP3 campaign in Borneo (Pugh et al., 2010). The authors conclude that the inadequacies apparent in the model are related to the representation of detailed physical and micrometeo-rological effects rather than errors in the chemical scheme.
The observations were not compared to measurements of the total OH reactivity in the latter study. However, comparison of measured and calculated total OH reactivity in this environment showed that the physical loss of species that react with OH plays a significant role in the calculated OH reactivity. Furthermore, a significant sink of OH appears to be missing (Edwards, 2013). Measurements conducted in a deciduous forest (Tan et al., 2001) also showed an underestimation of observed OH concentrations from box model simulations. Although two follow up studies at the same site showed model-predicted OH generally being in reasonable agreement with the measured OH, the model did underestimate the OH concentrations observed by a factor of about 3, in the afternoon during warmer periods (Griffith et al., 2013). It remains unclear if higher biogenic VOC emissions within those periods caused instrumental interference as recently suggested by Mao et al. (2012).
Further, Kim et al. (2013) reported reasonable agreement between observed OH and that from steady-state calculations using observations, particularly measurements of HO 2 , in a monoterpene-dominated forest environment. Constrained box model calculations under-predicted the observed HO 2 by as much as a factor of 8 and underestimated the observed OH by a factor of 4. The authors concluded that OH recycling occurs mainly via the reaction of HO 2 with NO in this forest, which is characterized by high 2-methyl-3-buten-2ol (MBO) and monoterpene emissions (Kim et al., 2013). Wolfe et al. (2013) found evidence for missing peroxy radical sources within a rural forest canopy.
Studies on oxidation processes in monoterpene-dominated environments are rare. Direct OH reactivity measurements in a boreal forest, conducted by Sinha et al. (2008), and a box model study investigating the OH reactivity budget ) revealed a significant fraction of "unknown OH reactivity". Comprehensive measurements in the same boreal forest were conducted during the HUMPPA-COPEC-2010 campaign (Hyytiälä United Measurements of Photochemistry and Particles in Air -Comprehensive Organic Precursor Emission and Concentration study) . The aim of the HUMPPA-COPEC-2010 campaign, by performing observations of VOCs, HO x , and total OH reactivity, is to increase our understanding of atmospheric oxidation processes in coniferous forests. The HO x budget is explored using direct calculations from measured species as well as an observationally constrained chemical box model in steady-state conditions.

Methodology
In order to measure OH and HO 2 radicals, the HORUS instrument (HydrOxyl Radical Measurement Unit based on fluorescence Spectroscopy; Martinez et al., 2010) was deployed during the intensive HUMPPA-COPEC field experiment in summer 2010. HORUS uses the laser-induced fluorescence based on fluorescent assay by gas expansion (LIF-FAGE; Hard et al., 1984) technique to detect OH and HO 2 . A comprehensive data set including measurements of the main oxidants OH, O 3 , and NO 3 ; important trace gases such as CO, NO x , H 2 O 2 , HCHO, and HONO; anthropogenic and biogenic VOCs, and their oxidation products, inorganic chemical constituents, aerosol properties, aerosol size distributions, as well as photolysis frequencies and other meteorological parameters was compiled at the field station SMEAR II (Station for Measuring Forest Ecosystem-Atmosphere Relations). An overview can be found in Williams et al. (2011).

Measurement site and instrumentation
The field site is located in a boreal forest in Hyytiälä, southern Finland (61 • 51 N, 24 • 17 E, 181 m a.s.l.) (Vesala et al., 1998). The largest city near the station is Tampere (about 200 000 inhabitants), located about 60 km S-SW of the measurement site (Hari and . The SMEAR II station is equipped with several masts and towers surrounded by a more than 40 yr old pine-dominated forest (Pinus sylvestris L.). The canopy height during the measurement period was about 20 m (for more detailed information about the continuous measurements and the infrastructure see Vesala et al., 1998;Kulmala et al., 2001;. During the intensive measurement period of the HUMPPA-COPEC-2010 campaign an additional scaffolding tower was set up at the field site within a clearing (the HUMPPA tower, ∼ 24 m high), reaching about 2-3 m above the canopy top. The sensors and inlet lines of instruments measuring reactive species like OH, HO 2 , O 3 , NO x , NO 3 , HONO, H 2 O 2 , organic peroxides, HCHO, monoterpenes, the associated photolysis frequencies (J NO 2 , J O( 1 D) ), and the total OH reactivity were collocated on the top platform of the HUMPPA tower to investigate the photochemistry at the ecosystem-atmosphere interface.
Since the campaign instrumentation and meteorological conditions are described elsewhere , only a brief description of the instruments used in this study is given here; time resolution and lower limits of detection are summarized in Table 1.
Ozone was measured using a UV instrument, sharing an inlet line with the chemiluminescence detection system (CLD) for measuring NO and NO 2 . NO 2 was measured in-directly by conversion to NO using a blue light converter. The instrument set-up was similar to the well-established one described by Hosaynali  for a different field campaign. NO 3 and N 2 O 5 were measured by cavity ring-down spectroscopy (Schuster et al., 2009;Crowley et al., 2010). Two long-path absorption photometer systems (LOPAP) were set up within the forest and on the HUMPPA tower, measuring HONO (Kleffmann et al., 2002). Hydrogen peroxide and the sum of organic peroxides were observed with a wet chemical system based on derivatization and fluorescence enzyme (DEF) described by Klippel et al. (2011). Measurements of carbon monoxide were conducted using a commercial vacuum UV resonance fluorescence CO instrument (AeroLaser GmbH, Garmisch-Partenkirchen, Germany). Detection of HCHO was based on the Hantzsch reaction and subsequent quantification of the reaction product via fluorescence detection. BVOCs were measured by several mass spectrometers. Gas chromatograph mass spectrometry (GC-MS) was used to investigate alkanes, alkenes, and particularly, isoprene and monoterpenes (Yassaa et al., 2012). Proton transfer reaction mass spectrometry (PTR-MS) was applied for the observation of methanol, acetone, acetonitrile, isoprene, total monoterpenes, benzene, and toluene. In the case of isoprene measurements, GC-MS data is used for further analysis, due to higher uncertainty in PTR-MS measurements, probably caused by a cold-trap connected to the PTR-MS instrument. Furthermore, the PTR-MS isoprene data might be affected by an interference due to 2-methyl-3-buten-2-ol (MBO) which is detected at the same mass-tocharge ratio as isoprene using conventional H 3 O + ionization (Goldan et al., 1997;Williams et al., 2001;Karl et al., 2012). Thus, PTR-MS data was only used for estimating MBO (see Sect. 2.4.1).
Photolysis frequencies (J NO 2 , J O( 1 D) ) were measured within the clearing at the forest ground as well as on top of the HUMPPA tower with a set of filter radiometers at each position (Junkermann et al., 1989). Water vapour, temperature, and pressure were recorded at several different levels on a 75 m meteorological mast located about 100 m away. The total reactivity towards OH was determined by a comparative reactivity method (Sinha et al., 2008;Nölscher et al., 2012). This method is based on the competitive scavenging of OH by a reference gas (pyrrole) and atmospheric trace gases. A chemical ionization mass spectrometry (CIMS) instrument was deployed to measure OH on the ground (Petäjä et al., 2009).

OH and HO 2 measurements by LIF-FAGE
Observations of OH and HO 2 concentrations were conducted using the Max Planck Institute for Chemistry (Mainz, Germany) HO x instrument based on laser-induced fluorescence, HORUS. The HORUS instrument is originally based on the design of GTHOS (Ground-based Tropospheric Hydrogen Oxides Sensor), the Penn State HO x instrument described by Faloona et al. (2004). HORUS is described in detail by Martinez et al. (2010). OH molecules are selectively excited at low pressure by pulsed UV light at around 308 nm on resonance with the Q 1 (2) transition line (A 2 + − X 2 , v = 0 ← v = 0), and fluorescence is detected using gated microchannel plate (MCP) detectors. The UV light for excitation of the hydroxyl radicals is provided by an Nd : YAG pumped, pulsed, tunable dye laser system Wennberg et al., 1994) operated at a pulse repetition frequency of 3 kHz. During the HUMPPA-COPEC-2010 campaign, optical fibers about 4 m in length were used to transfer the 308 nm light to the detection system. The sample air is drawn through a critical orifice (pinhole size of about 1.2 mm) into the low pressure detection cell, achieving a constant volume flow of about 10 slm (p = 1013 hPa, T = 273.15 K). In order to maximize the fluorescence signal at a certain radical level (i.e. maximizing the instrument sensitivity) a White cell set-up crossing the detection volume with 32 light paths is used (White, 1942). The detection of HO 2 is achieved via chemical conversion to OH by adding NO in excess to the total flow of sample air downstream of the OH detection. The sum of remaining ambient OH plus OH originating from HO 2 conversion is measured in a second detection axis. Calibrations of the HORUS instrument were performed on a regular basis.

Calibration
For calibration of the HORUS instrument, OH and HO 2 radicals were produced by photolysis of water vapour in hu-midified zero air, as described by Martinez et al. (2010). Exact knowledge of the actinic flux density of the mercury lamp (Pen-ray line source, LOT-Oriel, Germany) used for the photolytic radical production is crucial for this method. The mercury lamp used for calibration of the HORUS instrument during HUMPPA-COPEC-2010 was calibrated using the actinometry method by N 2 O photolysis, as described by Martinez et al. (2010), immediately before and after the campaign. The resulting actinic flux density at different N 2 O mixing ratios is shown in Fig. 1. Both measurements of the actinic flux density, 0 , before and after the field campaign, agree within uncertainties and indicate the long-term stability of the actinic flux density. From the overall fit we obtained 0 = (3.68 ± 0.04)×10 12 cm −2 s −1 with a reasonable quality of fit. The systematic error of the actinic flux density measurement is calculated to be about 17 %, similar to the value by Martinez et al. (2010).
Furthermore, the instrument sensitivity is limited by quenching of the fluorescence, which happens mainly due to water vapour. To minimize this effect, the HORUS system is operated at low pressure (∼ 3-5 hPa), but still high enough to keep wall losses small and provide enough molecules for excitation. The quenching effect by H 2 O is considered in our calculations. An additional dependency on water vapour of about 12 % and 17 % decrease in sensitivity per 1 % increase in water vapour mixing ratio was observed for the first and second fluorescence cell, respectively. This additional water effect indicates further losses at higher water mixing ratios possibly due to formation of OH-water clusters during the cold-adiabatic expansion of the sample air while entering the The data shown here is the combined data set from two actinometric measurements, one of them conducted before (red dots) and a second one after (black triangles) the field campaign, indicating a good long-term stability of the lamp actinic flux density. Error bars indicate the propagated statistical variability of the calculated flux density. low pressure detection system. Changing instrument sensitivity with respect to water vapour mixing ratio, which is not caused by quenching, was reported by others (Hofzumahaus et al., 1996;Holland et al., 1998), hypothesizing that condensation processes during the cold-adiabatic expansion are causing this effect. The above-mentioned quantification of the additional water dependency during instrument calibration was used to correct for this dependency. Unfortunately, the MCP detectors changed in sensitivity during the HUMPPA-COPEC-2010 field campaign, decreasing over time. Calibrations of the instrument were conducted about every second day to keep track of changes in sensitivity. Within some limitations, the behaviour of the laser scattering inside the system is an indicator of the sensitivity changes in the system. For the laser scatter signal, factors such as laser power fluctuations and background reflections have to be taken into account. Differing amounts and composition of aerosols in the sample air therefore might cause variable scatter. However, comparison of the laser scatter signal in ambient air with the quasi-simultaneous measured sensitivity during calibration shows a linear dependency. This functional dependency was applied to correct for the changing sensitivity (see Fig. 2).

Interferences
Knowledge about possible interferences and avoiding those when indicated is required in order to measure OH and HO 2 reliably. Interferences can be caused by the instrument itself (e.g. laser generated OH) and/or atmospheric substances = 0.96) with a non-significant offset was observed during instrument calibrations. The functional dependency obtained from linear regression was applied to the measurements to correct for changes in instrument sensitivity by continuously monitoring the laser scatter signal. A similar correction function for the change in sensitivity of the first detection cell has also been determined empirically and applied.
which fluoresce at wavelengths similar to the hydroxyl radical. Laboratory studies testing the effect of sulfur dioxide, formaldehyde, nitrous acid, nitric acid, acetone, hydrogen peroxide, and various hydrocarbons on the OH signal did not show any significant interference for measurements in the atmosphere (Kubistin, 2009;Faloona et al., 2000;Ren et al., 2004). A negative interference on the OH signal by naphthalene was observed in polluted urban environments and can be used for the specific measurement of this compound . It was recently reported that LIF measurements can be affected from internally generated OH (Mao et al., 2012;Novelli et al., 2012). Therefore, a chemical modulation method to determine the background signal for the measurement of atmospheric OH (as proposed by Brune et al., 2010) was applied to the HORUS-LIF for the first time during HUMPPA-COPEC-2010. A new injection unit IPI (inlet pre-injector) was developed and optimized to scavenge more than 95 % of atmospheric OH by periodic injection of a chemical reactant in front of the standard inlet. In order to minimize wall losses, a bypass flow, large compared to the sample flow into the detection system, was established. Optimization of this injection system included tests using different reactants, e.g. propylene and hexafluoropropene, and varying injection and bypass flow conditions to determine the best parameters for continuous operation. There is a different publication on the characterization of the OH scavenger injection device IPI by Novelli et al. (2014).
Measurements of HO 2 concentrations are conducted using chemical conversion into OH via the reaction with NO and the subsequent detection of the hydroxyl radicals by the LIF In the atmosphere, RO 2 is also converted into HO 2 in the presence of NO (Reactions R4 and R5). These reactions were expected to be negligible in low pressure FAGE detection systems due to the reduced oxygen concentration and the short reaction time between injection of NO and detection of OH within the system (Heard and Pilling, 2003).
We observed a small interference (< 10 %) originating from methyl peroxy radicals in the laboratory for the HO-RUS instrument, comparable to the results reported by Holland et al. (2003), who detected an interference smaller than 5 % for their LIF-FAGE instrument. Significant interferences in HO 2 measurements by the LIF-FAGE technique from RO 2 species resulting from the OH-initiated oxidation of alkenes and aromatics have been observed by  and others (Dillon, 2011). In contrast to alkane-based peroxy radicals, which are formed in the reaction of VOC + OH via H-atom abstraction, alkene-based peroxy radicals (mainly produced via OH-addition on the C=C double bond) form specific RO by reaction with NO that can decompose under the low pressure conditions forming HO 2 rapidly. The conversion efficiency is limited by Reaction R4. Thus, any potential interference from alkenes on HO 2 can be reduced by reducing reaction time and/or NO concentration. During HUMPPA-COPEC-2010, NO mixing ratios of about 400 ppmV occurred inside the detection system of HORUS by injection of pure NO into sample air, yielding a conversion efficiency from HO 2 to OH of more than 95 % at the operating conditions of the instrument. This could also have caused a high conversion efficiency for interfering RO 2 species as described above. Nevertheless, this interference depends strongly on the available VOCs for the production of RO 2 in the investigated environment as well as on the instrumental set-up which defines the conversion efficiency of those peroxy radicals inside the instrument. The observed HO 2 can be interpreted as the sum of atmospheric HO 2 and a contribution from the effective interference by specific RO 2 : The relative detection sensitivities α i RO 2 for the specific RO 2 compared to HO 2 are strongly dependent on the instrumental set-up and can reach values up to about 0.9 Lu et al., 2012;Whalley et al., 2013). However, the speciation of RO 2 in the observed environment strongly influences the effective interference. Based on model simulations, the magnitude of the RO 2 interference on HO 2 measurements by some LIF instruments in different environments was estimated between 10 % up to about 30 % (Lu We observed a small interference (< 10 %) originating 75 from methyl peroxy radicals in the laboratory for the HO-RUS instrument, comparable to the results reported by Holland et al. (2003), who detected an interference smaller than 5 % for their LIF-FAGE instrument. Significant interferences in HO 2 measurements by the LIF-FAGE technique 80 from RO 2 species resulting from the OH-initiated oxidation of alkenes and aromatics have been observed by  and others (Dillon, 2011). In contrast to alkanebased peroxy radicals, which are formed in the reaction of VOC + OH via H-atom abstraction, alkene-based peroxy 85 radicals (mainly produced via OH-addition on the C-C double bond) form specific RO by reaction with NO, that can decompose under the low pressure conditions forming HO 2 rapidly. The conversion efficiency is limited by Reaction R4. Thus, any potential interference from alkenes on HO 2 can 90 be reduced by reducing reaction time and/or NO concentration. During HUMPPA-COPEC-2010 NO mixing ratios of about 400 ppmV occurred inside the detection system of HORUS by injection of pure NO into sample air, yielding a conversion efficiency from HO 2 to OH of more than 95 % 95 at the operating conditions of the instrument. This could also have caused a high conversion efficiency for interfering RO 2 species as described above. Nevertheless, this interference depends strongly on the available VOCs for the production of RO 2 in the investigated environment as well as on the instru-00 mental setup which defines the conversion efficiency of those peroxy radicals inside the instrument. The observed HO 2 can be interpreted as the sum of atmospheric HO 2 and a contribution from the effective interference by specific RO 2 : The relative detection sensitivities α i RO 2 for the specific RO 2 compared to HO 2 are strongly dependend on the instrumental setup and can reach values up to about 0.9 Lu et al., 2012;Whalley et al., 2013). However, the speciation of RO 2 in the observed environment strongly influences the effective interference. Based on model simulations the magnitude of the RO 2 interference on HO 2 measurements by some LIF instruments in different environments was estimated between 10 % up to about 30 % (Lu et al., 2012;Mao et al., 2012;Whalley 15 et al., 2013;Griffith et al., 2013).
NO titration experiments during calibration and ambient air measurements in two forest environments in Germany have been conducted after HUMPPA-COPEC-2010, quantifying the maximum overall interference by RO 2 in the HO-20 RUS instrument in these environments to be less than 20 % (Tatum . To further investigate the effect of RO 2 interference on the measurement of HO 2 concentrations during HUMPPA-COPEC-2010, we calculated the H 2 O 2 budget taking the 25 production by the self-reaction of HO 2 and the loss via pho- tolysis and deposition into account and compared the calculated hydrogen peroxide with the measurements (Fig. 3).
The production of H 2 O 2 by HO 2 shows a quadratic dependency and is therefore highly sensitive to the HO 2 con-430 centration. The decay of hydrogen peroxide in the afternoon is mainly determined by the deposition process. A reasonable deposition rate of 4 × 10 −5 s −1 , corresponding to a deposition velocity of 4 cm s −1 in a 1 km high boundary layer, was used for the calculation. This is comparable to values re-435 ported by Hall and Claiborn (1997) for a boreal forest, ranging from 1 to 5 cm s −1 . Removal of hydrogen peroxide by photolysis makes up to 10 % of the total H 2 O 2 loss. Hydrogen peroxide concentrations derived using the measured HO 2 is in reasonable agreement with the measured H 2 O 2 .

440
In some cases measured HO 2 is still not sufficient to explain the measured H 2 O 2 concentrations. Even though under some conditions hydrogen peroxide measurements might be affected by mixing with different air masses, e.g. from the residual layer, which have not been considered in our cal-445 culation, the comparison gives confidence that the measurements of HO 2 are not subject to a major interference. Assuming an interference on the observed HO 2 concentration of 50 % and 30 %, i.e. using 50 % and 70 % of the observed HO 2 , respectively, for simulation does not suffice to repro-450 duce the observed H 2 O 2 concentration, while using twice the observed amount of HO 2 results in a significant overestima-  Griffith et al., 2013).
NO titration experiments during calibration and ambient air measurements in two forest environments in Germany have been conducted after HUMPPA-COPEC-2010, quantifying the maximum overall interference by RO 2 in the HO-RUS instrument in these environments to be less than 20 % (Tatum . To further investigate the effect of RO 2 interference on the measurement of HO 2 concentrations during HUMPPA-COPEC-2010, we calculated the H 2 O 2 budget taking the production by the self-reaction of HO 2 and the loss via photolysis and deposition into account and compared the calculated hydrogen peroxide with the measurements (Fig. 3).
The production of H 2 O 2 by HO 2 shows a quadratic dependency and is therefore highly sensitive to the HO 2 concentration. The decay of hydrogen peroxide in the afternoon is mainly determined by the deposition process. A reasonable deposition rate of 4 × 10 −5 s −1 , corresponding to a deposition velocity of 4 cm s −1 in a 1 km high boundary layer, was used for the calculation. This is comparable to values reported by Hall and Claiborn (1997) for a boreal forest, ranging from 1 to 5 cm s −1 . Removal of hydrogen peroxide by photolysis makes up to 10 % of the total H 2 O 2 loss. Hydrogen peroxide concentrations derived using the measured HO 2 are in reasonable agreement with the measured Atmos. Chem. Phys., 14, 8723-8747, 2014 www.atmos-chem-phys.net/14/8723/2014/ H 2 O 2 . In some cases measured HO 2 is still not sufficient to explain the measured H 2 O 2 concentrations. Even though under some conditions hydrogen peroxide measurements might be affected by mixing with different air masses, e.g. from the residual layer, which have not been considered in our calculation, the comparison gives confidence that the measurements of HO 2 are not subject to a major interference. Assuming an interference on the observed HO 2 concentration of 50 % and 30 %, i.e. using 50 % and 70 % of the observed HO 2 , respectively, for simulation it does not suffice to reproduce the observed H 2 O 2 concentration, while using twice the observed amount of HO 2 results in a significant overestimation (Fig. 3). This result is independent of deposition velocities ranging between 1 and 5 cm s −1 . Although the magnitude of the RO 2 interference during HUMPPA-COPEC-2010 cannot be conclusively derived there is no evidence for an extraordinary large interference in the HORUS instrument compared to other LIF systems. Therefore, a contribution of 30 % to the observed HO 2 signal (presently the maximum value observed in LIF instruments) is considered as an upper limit estimate of the RO 2 interference for further analysis.

OH budget -calculated based on observations
Subsequent to the ground-level comparison with the OH-CIMS (1-8 August), the HORUS instrument was moved to the top of the HUMPPA tower to investigate the photochemistry at the ecosystem-atmosphere interface. Making use of the comprehensive measurements of atmospherically relevant species, described in Sect. 2.1, the known daytime production rates of OH can be calculated. Besides the photolytic sources (Reactions R6 and R7), ozonolysis of different biogenic VOCs (Reaction R8) contributes to the primary production of OH. Secondary sources, e.g. the recycling of HO 2 by NO and O 3 (Reactions R11 and R12) and peroxide photolysis (Reactions R9 and R10), additionally play an important role in OH radical production. We define primary production as follows: Secondary production/recycling : Under steady-state conditions, which can be assumed for short-lived compounds like OH, the sum of these production rates should equal the total loss of hydroxyl radicals, which can be derived from the product of the measured OH concen-tration and the total OH reactivity measurements (k OH ).
The comparison of the sum of production rates from known pathways with the measured total loss of hydroxyl radicals can be used as a tool to unveil gaps in our understanding of the underlying processes. Furthermore, the importance of the different contributions to the total production of OH can be determined. A more detailed analysis was also conducted taking into account unmeasured BVOCs and higherorder oxidation products using a measurement constrained box model (see Sect. 2.4).

Box model
To investigate the influence of unmeasured intermediate reaction products, the observations were compared to simulations with version 3.0 of the chemical box model CAABA/MECCA (Chemistry As A Boxmodel Application/Module Efficiently Calculating the Chemistry of the Atmosphere) by Sander et al. (2011). We constrained the model with observed, complete 5 min data sets. The simulations were conducted in steady-state mode, i.e. until the relative change in number concentration of OH and HO 2 was less than 10 −6 within 1 s, when steady-state conditions for both species were reached. This corresponds to a spin-up period of typically around 48 h. MECCA contains a comprehensive atmospheric chemistry reaction scheme. However, since we focus on organics, we switched off halogen and sulfur chemistry, as well as heterogeneous and aqueous phase reactions. A list of the chemical reactions used in this study, including rate coefficients and references, is available in the Supplement. In the base configuration, version 2 of the isoprene chemistry from the Mainz Isoprene Mechanism (MIM2) was used, considering 68 species and 195 reactions (Taraborrelli et al., 2009based on Pöschl et al., 2000. For sensitivity studies, the recently developed isoprene mechanism MIM3 (Taraborrelli et al., 2012), which includes additions to the isoprene chemistry such as the photo-oxidation of unsaturated hydroperoxy aldehydes, and a preliminary version of the monoterpene mechanism (MTM) (Taraborrelli et al., in preparation) were also used. The latter is based on a MIM2-like version of MIM3 (MIM3*), i.e. hydroperoxy aldehyde chemistry, H-shifts, and RO 2 +HO 2 reactions are considered in the manner of MIM2, while updated estimates of rate constants from MIM3 were retained. Furthermore, it has a representation for the oxidation of the major terpenes during HUMPPA-COPEC-2010, those being α-pinene, β-pinene, β-myrcene, 2 -carene, 3carene, and α-farnesene chemistry. The oxidation of the first two monoterpenes is taken from the Master Chemical Mechanism (MCM v3.2, Jenkin et al., 1997;Saunders et al., 2003, via the website: http://mcm.leeds.ac.uk/MCM), with updates based on recent literature. Carene is assumed to yield the same products as α-pinene. Finally, the oxidation of βmyrcene and α-farnesene is simplified and partially follows an isoprene-like oxidation. An overview of the key differences between the chemical mechanisms used for box model simulations in this study is given in Table 2.
Deposition was included for the species listed in Table 3, according to values given in the literature or derived from measurements.
For the numerical integration of the resulting set of ordinary differential equations (ODEs), the KPP software (Sandu and Sander, 2006) with a positive definite Rosenbrock solver and automatic time-step control was used. For the photolysis frequencies of NO 2 and O 3 to form O( 1 D), measured values (as described above) were used. Photolysis frequencies for other observed photolabile species are calculated based on measured J O( 1 D) and J NO 2 using parameterizations from spectroradiometer measurements at other field sites (Bohn et al., 2008). Therefore effects of cloud coverage and aerosols are considered implicitly. In the case of missing methane (CH 4 ) data, box model simulations were conducted using the median value of CH 4 observed, 1.79 ppmV. The 1σ variability in CH 4 measurements during the entire campaign was 0.03 ppmV. Measured photolysis frequencies below the lower limit of detection were set to zero, which affects the derived photolysis frequencies from other species accordingly. The uncertainty of simulated HO x caused by these assumptions is less than 10 %. Data sets missing any other of the species used as model input were omitted from this study.

Significance and sensitivity
The significance of the discrepancies between observed and simulated HO x concentrations depends on the uncertainty of the HO x measurements as well as on the uncertainty of the model simulation. The latter includes the uncertainties of all measured input variables such as trace gas concentrations and meteorological parameters as well as the uncertainties due to the chemical mechanism applied. Sensitivity analysis on measured species by varying constraint parameters within their uncertainty can be used to identify the impact on the model result. In the case of unmeasured species, e.g. oxidation products and intermediates, the uncertainties depend on the uncertainty of their production and destruction pathways as well as the uncertainty of the chemical mechanism. The uncertainties of the simulated OH and HO 2 concentrations related to the reaction rate coefficients were estimated by Monte Carlo simulations as described by Sander et al. (2011). The result of one Monte Carlo simulation in the case of OH applying the isoprene mechanism (MIM2) is shown as an example in Fig. 4. The histogram shows a binned frequency distribution of the for the reaction rate coefficients on a single above-canopy data set in the early afternoon. The resulting distribution shows a slight skewness with a median of 6.15 × 10 6 molec cm −3 and a mean of (6.21 ± 0.86)×10 6 molec cm −3 . The dark blue bars indicate the ±1σ range.
model OH resulting from the individual Monte Carlo runs, simulated with diverse sets of rate coefficients. This yields a 1σ uncertainty of the simulated OH resulting from the uncertainty of rate coefficients of 15 %.

Observations
During the HUMPPA-COPEC-2010 field experiment gradients of the hydroxyl radical were measured using the CIMS instrument from University of Helsinki on the ground (Petäjä et al., 2009) and the HORUS-LIF instrument above the canopy. To assure the comparability of both instruments and techniques, they had to be compared side-by-side under ambient conditions. The HORUS-LIF instrument already participated in the international HO x Comp 2005 project, a groundbased intercomparison of six OH instruments (4 LIF, 1 CIMS, 1 DOAS 1 ) performing measurements in SAPHIR (Simulation of Atmospheric PHotochemistry In a large Reaction Chamber) as well as in ambient air. The HORUS-LIF showed good agreement with CIMS measurements (R 2 = 0.96) during daytime. However, we were expecting from that study to see a night-time signal in ambient air whereas the CIMS method usually does not detect night-time OH (Schlosser et al., 2009).
During HUMPPA-COPEC-2010 the two OH instruments were compared at the beginning of the field experiment (27-30 July) before starting the gradient measurement of OH concentrations. The inlet system of the LIF was placed next to the CIMS inlet on the ground. The meteorological conditions at the field site within the four days of instrument comparison  Groß, 2013) Ozonolysis reaction yields MCM v3.1 updated yields updated yields (Taraborrelli et al., 2012) (Taraborrelli et al., 2012) 1,5-/1,6-H-shifts none included none (Peeters et al., 2009) HPALD photo-oxidation none included none (Taraborrelli et al., 2012) monoterpene chemistry none none included Table 3. Deposition rates/velocities used in the model. were dominated by above average temperatures, exceeding 25 • C during noon, and mainly south-easterly winds, without rainy periods. The result of these four days of instrument comparison is shown in Fig. 5. OH radical concentrations observed by both techniques are generally in agreement. The LIF data shows stronger fluctuations compared to the measurements by CIMS on the same averaging interval timescale of 30 min. Daytime maximum OH concentrations reached about 1.5-2 ×10 6 molec cm −3 . During night-time, both instruments observed OH concentrations below 5×10 5 molec cm −3 , but still well above the lower limit of detection of the CIMS, which was 5 × 10 4 molec cm −3 at a time resolution of 30 s. For the LIF, the detection limit was 4.8 × 10 5 molec cm −3 at a time resolution of 60 min. The larger limit of detection as well as the increased fluctuations in the LIF observation compared to CIMS is mainly caused by application of the chemical modulation method. To determine atmospheric OH, the in- terference signal is subtracted from the total observed signal. The detection limit is therefore not only given by instrument properties, like laser power, optical properties, and detector efficiency; it is also prone to the atmospheric variability both of OH and of the species causing the interference inside the detection unit. Nevertheless, the good agreement during the comparison (Fig. 6) provides confidence in the chemical modulation method and that the LIF measurements are not affected by additional unknown OH interferences resulting from the laser fluorescence technique .
Finally, the LIF instrument was moved to the top of the HUMPPA tower to investigate the radical chemistry at the interface between atmosphere and ecosystem while the CIMS stayed on the ground. The resulting hydroxyl radical measurements are presented in Fig. 7. Simultaneous on-ground and above-canopy OH measurements revealed a factor of 2-3 difference in observed concentration maxima, reaching values up to about 3 × 10 6 molec cm −3 on the tower. During night-time, the above-canopy and on-ground hydroxyl radical observations showed similar values. Linear correlation of OH and photolysis frequency J O( 1 D) was previously found during groundbased campaigns (Rohrer and Berresheim, 2006). However, the slope varies with location, depending on the abundance of VOCs and NO x . The comparison of the on-ground and above-canopy correlations shows that higher OH values on the tower are driven by higher radiation but does not suggest differences in the chemical regime (Fig. 8).
The night-time OH, which was occasionally well above the detection limit of both instruments, cannot be explained by supporting observations, with known sources accounting for about 20 % of the total OH production necessary to explain the measured concentrations. NO 3 was always below the lower limit of detection of about 1 pptV (at 5 min time resolution) for the CRD instrument . However, the production rate for NO 3 from the reaction of NO 2 and O 3 is of the order of 1 × 10 6 molec cm −3 s −1 . Assuming this would directly cause an equally high OH production in the oxidation process of VOCs, which is an unlikely high upper estimate, this would still only explain an additional 10 to 15 % of the missing production during nighttime. The ozonolysis of unmeasured VOCs and their oxidation products and enhanced HO x recycling are potential candidates for explaining the missing OH production. A more detailed analysis of the most relevant OH production terms is given in Sect. 3.1.1. Time series of supporting observations are shown in Fig. 9

Radical budget based on observations
Hydroxyl radical production was calculated taking the primary production by O 3 photolysis as well as ozonolysis of observed hydrocarbons and recycling via HO 2 by reaction with NO and O 3 into account (see Sects. 2.3, Reactions R6-R12). It reaches an on-ground maximum of about 1 × 10 7 molec cm −3 s −1 and an above-canopy maximum of 1.4 × 10 7 molec cm −3 s −1 around local solar noon as shown in Fig. 10.   Atmos. Chem. Phys., 14, 8723-8747, 2014 www.atmos-chem-phys.net/14/8723/2014/ Since NO and NO 2 were not measured at ground level, they were derived from the above-canopy measurements assuming constant NO x with height, using the filter radiometer measurements of the on-ground and above-canopy photolysis frequency J NO 2 to calculate partitioning between NO 2 and NO. The above-canopy enhanced radical production is caused by higher J O( 1 D) observed on the tower and by enhanced recycling (Fig. 10) of hydroxyl radicals (60-80 %). The ozonolysis of measured biogenic VOCs plays a minor role as a source of OH during daytime but becomes more important during night-time.
The budget of steady-state OH was calculated as described in Sect. 2.3, using the available measurements of the chemical species contributing to the production rates and the total above-canopy OH reactivity observed as constraints. Mean OH reactivity measured during this period was 11.5 s −1 , varying typically between the lower limit of detection (3-4 s −1 ) and about 30 s −1 , with some peaks reaching above 70 s −1 (Nölscher et al., 2012).
The known OH sources are almost sufficient to close the budget above the canopy. At maximum, isoprene contributes less than 10 % to the total OH reactivity measured during HUMPPA-COPEC-2010 due to its low mixing ratio (typically below 200 pptV). Observed terpenes, being similar or Atmos. Chem. Phys., 14, 8723-8747, 2014 www.atmos-chem-phys.net/14/8723/2014/ Figure 10. Average on-ground (left panel, 27-30 July) and abovecanopy (right panel, 1-8 August) diurnal OH production. The whiskers indicate the variability of the total loss rate, calculated from total OH reactivity and hydroxyl radical measurement using LIF data.
In addition, HCHO photolysis is a direct radical source, also contributing to HO 2 production during daytime. The cycling reactions of RO 2 species with NO can become the largest contributor to hydroperoxyl radical production, depending on the abundance of organic peroxy radicals and NO.
The loss of HO 2 is dominated by reactions with NO and O 3 , which are at the same time the predominant source of OH. Radical-radical termination reactions (Reactions R22-R24) yielding peroxides and water act as a sink for HO x radicals.
Unfortunately, RO 2 was not measured during HUMPPA-COPEC-2010 prohibiting the calculation of a HO 2 budget from observations in a similar manner as the OH budget. Nevertheless, assuming steady-state conditions for HO 2 and taking the hydroperoxyl radical measurements into account, the RO 2 concentration can be estimated.
where b is a branching ratio and k i denotes the reaction rate constant associated with Reaction Ri. From this it follows For calculation, the lumped rate constant k R19 = 2.7 × 10 −12 × e (360/T ) cm 3 molec −1 s −1 from the Master Chemical Mechanism (MCM v3.2, Jenkin et al. (1997); Saunders et al. (2003), via the website: http://mcm.leeds.ac.uk/MCM) and a branching ratio b = 0.9 was used. Typically, there was about twice as much RO 2 as HO 2 (RO 2 / HO 2 : median 1.8; min. 0.5; max. 4.5). This estimate of RO 2 is calculated neglecting isomerization reactions forming HO 2 , thus providing an upper estimate of the actual RO 2 concentration. Nonetheless, the significance of these additional reactions strongly depends on the speciation of RO 2 , which is unknown. The RO 2 estimate was confirmed by calculation of the modified Leighton ratio (RO 2 / HO 2 : median 1.8; min. 0.1; max. 6.3) (Leighton, 1961).
Thus, organic peroxy radicals seem to play a major role in the radical photochemistry in the observed boreal forest environment, providing a strong link between OH and HO 2 .
The importance of radical cycling via HO 2 shown in the budget demands careful examination of the reliability of HO 2 measurements. The measurement of HO 2 by conversion to OH using NO can be severely affected by RO 2 reacting with NO also producing hydroxyl radicals . However, this interference depends on the instrumental setup . It is to be quantified specifically for each instrument and depends on the composition of RO 2 in ambient air. The HO 2 observed during HUMPPA-COPEC-2010 therefore gives an upper limit for the atmospheric HO 2 including an unknown contribution from RO 2 . To put limits on the atmospheric HO 2 , the budget of H 2 O 2 was analysed. The H 2 O 2 comparison (Fig. 3)

HO x recycling pathways
Production, loss, and recycling pathways of HO x above the canopy under various conditions are shown in Tables 4 and S1 in the Supplement. Average rates calculated from observations highlight once more the importance of OH recycling via HO 2 . Ozonolysis of BVOCs is an important primary radical source at low radiation levels. Photolytic OH production from O 3 and HONO photolysis are more relevant primary radical sources during daytime. The contribution of acetone photolysis to HO 2 formation is small at the surface, typically less than 1 % of the total direct radical sources.
The main sink terms are radical self-reactions, particularly HO 2 + RO 2 . Since RO 2 was not measured during HUMPPA-COPEC-2010, it is estimated from HO 2 steady-state conditions (see Sect. 3.1.1). The total OH production, P total OH , is calculated from total OH reactivity data and the hydroxyl radical measurements assuming steady-state conditions (Eqs. 3 and 4). The loss rate of OH via reaction with VOCs and unmeasured oxygenated VOCs (OVOCs) is derived by subtracting the known OH loss rates caused by CO, O 3 , HCHO, H 2 O 2 , NO 2 , NO, SO 2 , and HO 2 from the total loss observed (k OH [OH]).
The data set can be grouped into conditions of low and high radiation, and moderate and enhanced observed total OH reactivity. The resulting cases during daytime at moderate and enhanced OH reactivity conditions are presented in Table 4. In the Supplement of this publication, a similar HO x recycling figure under conditions of moderate and high observed total OH reactivity at low radiation (J O( 1 D) ≤ 3×10 −6 s −1 ) can be found (see Table S1 in the Supplement).
Under daylight conditions (J O( 1 D) > 3 × 10 −6 s −1 ), at moderate observed total OH reactivity (k OH ≤ 15 s −1 ) a total loss rate of L total OH ≈ 3.3 × 10 6 molec cm −3 s −1 is calculated from observations. The known OH sources, i.e. the primary production of OH plus the reactions of HO 2 with NO (χ (NO) ≈ 46 pptV) and O 3 , close the OH budget. An OH loss rate of about 1.2 × 10 6 molec cm −3 s −1 is missing under these conditions, which is close to the combined uncertainty of 40 %. The secondary OH production due to the reactions of HO 2 with NO and O 3 account for 73 % of the known OH production.
OH is directly lost by radical termination reactions with NO 2 ,NO, and HO 2 . Furthermore, OH reacts with CO, O 3 , HCHO, SO 2 , and H 2 O 2 , yielding HO 2 radicals. The remaining observed total OH reactivity constrains the maximum production of RO 2 from the reaction of OH with VOCs. However, this production rate is only 75 % of the RO 2 loss rate via reaction with NO, which is the dominant source of HO 2 . This indicates a missing RO 2 source. Besides the ozonolysis of unmeasured BVOCs, reservoir species such as PAN, which thermally decompose (showing a strong temperature dependency), could contribute to the missing RO 2 production rate which is of the order of 1.2 × 10 6 molec cm −3 s −1 .
Under daylight conditions (J O( 1 D) > 3 × 10 −6 s −1 ), with high observed total OH reactivity (k OH > 15 s −1 ) the largest known source of OH is again HO 2 recycling by reaction with NO and O 3 . However, about 70 % of the total OH production is unknown. The excess OH reactivity (equivalent to an average OH loss rate of 27 × 10 6 molec cm −3 s −1 ) is available for production of RO 2 radicals. RO 2 radicals react with NO forming HO 2 with a yield of about 90 %, representing the main source of HO 2 . In comparison, the direct production of HO 2 radicals from HCHO photolysis is negligible. The RO 2 loss by reaction with NO forming HO 2 or nitrates compensates approximately 50 % of the potential RO 2 production rate calculated from unaccounted-for observed total OH reactivity. An RO 2 loss rate of 14 × 10 6 molec cm −3 s −1 , is due to other processes, e.g. direct OH recycling, not via the reaction of HO 2 + NO/O 3 . Furthermore, an OH production rate of the order of 20 × 10 6 molec cm −3 s −1 is missing from a closed OH budget under these conditions. The reaction of specific alkylperoxy radicals with HO 2 could represent an additional RO 2 sink and OH source as indicated by pathway A in Table 4. Studies investigating the OH production from the reactions of ethyl peroxy, acetyl peroxy, and acetonyl peroxy radicals with HO 2 revealed OH yields up to 70 % (Thornton et al., 2002;Hasson et al., 2004;Jenkin et al., 2007;Dillon and Crowley, 2008). Assuming that all RO 2 species, as an upper limit, would react with HO 2 forming OH, this would contribute an additional 10 × 10 6 molec cm −3 s −1 to OH production. However, this would still not be sufficient to close the OH budget under these conditions of high actinic flux and high OH reactivity. A direct recycling mechanism, returning OH from RO 2 in the absence of NO, by a 1,5-H-shift, was proposed by Peeters et al. (2009) for the OH-initiated degradation of isoprene. A recent chamber study suggests a reduced reaction rate constant by about a factor of 2 with respect to the value given by Peeters et al. (2009) for this 1,5-H-shift (Fuchs et al., 2013;Crounse et al., 2011;da Silva et al., 2010). Nevertheless, at low NO levels (χ (NO) ≈ 100 pptV) these isomerization reactions for isoprene-related RO 2 radicals forming OH can become competitive with the traditional reactions of RO 2 with HO 2 , RO 2 , and NO. Similarly, other RO 2 radicals, such as conjugated alkadienes, e.g. myrcene, many monocyclic monoterpenes, and also sesquiterpenes, can undergo a fast H-shift isomerization (Peeters et al., 2001;Vereecken et al., 2007Vereecken et al., , 2012. Monoterpenes and sesquiterpenes have many more pathways to oxidize than isoprene. However, not all channels lead to OH production. Still, an OH production Atmos. Chem. Phys., 14, 8723-8747, 2014 www.atmos-chem-phys.net/14/8723/2014/ • about 25 % of P total OH by HO 2 + NO/O 3 (P OH, HO 2 +NO/O 3 ≈ 3 × 10 6 molec cm −3 s −1 ) (P OH, HO 2 + NO/O 3 ≈ 6 × 10 6 molec cm −3 s −1 ) • additional RO 2 source is required, • up to 50 % of the RO 2 production rate imply unknown e.g. by VOC + O 3 and/or loss processes that could lead to direct OH recycling decomposition of nitrates (PAN) • RO 2 sink is missing → potential OH source ⇒ alkylperoxy radicals +HO 2 (Thornton et al., 2002;Hasson et al., 2004;Jenkin et al., 2007;Dillon and Crowley, 2008) and H-shifts (Peeters et al., 2009) are likely rate of the order of 6 × 10 6 molec cm −3 s −1 is missing. A direct OH source or additional RO 2 production, balanced by an equivalent loss of RO 2 forming OH via mechanisms like the ones mentioned above could close the OH budget. Under conditions of low radiation (J O( 1 D) ≤ 3×10 −6 s −1 ) the classical recycling of HO 2 to OH by reaction with NO and O 3 is inhibited, mainly due to the low NO concentration. Comparison of the known production rates with the derived total production rate (Eq. 4) reveals a missing fraction of 40 % under conditions of moderate total OH reactivity and 80 % under conditions of enhanced total OH reactivity. When the total OH reactivity is moderate, the missing OH source is in the same order of magnitude of the observed NO 3 production rate of 1 × 10 6 molec cm −3 s −1 . Since the NO 3 concentration always remained below the lower limit of detection, its reactivity had to be fast, potentially to some extent producing OH. Also, ozonolysis of unmeasured BVOCs could directly produce OH and possibly close the OH budget. Under conditions of high total OH reactivity RO 2 loss processes not yet considered are implied that could lead to direct OH recycling as proposed above for daytime, high total OH reactivity conditions.

Box model simulations
The previous discussion indicates that under conditions of low reactivity and high radiation, recycling of OH in this environment occurs mainly by HO 2 . In order to identify the influence of unmeasured oxidation products on the HO x budget and to examine the current understanding of the underlying processes, box model simulations have been conducted.
The CAABA/MECCA box model was applied in steadystate mode with concentrations of NO, CO, O 3 , H 2 O 2 , organic peroxides, isoprene (C 5 H 8 ), terpenes, HONO, HCHO, and H 2 O as well as photolysis frequencies constrained to measured values. Simulations were only done when data of all key constraints were available (see Sect. 2.4).
Applying the chemistry scheme from the Mainz Isoprene Mechanism (MIM2), OH concentrations are overestimated by the model by about 40 % on average (Fig. 11). cates that the missing reactivity is an unaccounted source of HO 2 . Furthermore, the recycling reaction of NO and O 3 with 935 the missing HO 2 has the potential under most conditions to compensate for the additional OH loss, which has not been considered yet. This preserves the good agreement of simulated and observed OH even when the model accounts for all the observed total OH reactivity. However, when high to-940 tal OH reactivities (> 20 s −1 ) were observed, the enhanced HO 2 recycling is not sufficient to reproduce the observed OH. Additional mechanisms, e.g. direct OH recycling as discussed for isoprene-dominated, low-NO x environments, or an additional primary source of OH is required to explain the 945 observed hydroxyl radical concentrations under these conditions.
To investigate the hypothesis that the OH reactivity missing in the model using the terpene mechanism acts as a source of HO 2 , additional reactivity towards OH was in-950 troduced by adding the surrogate molecule X which behaves chemically identically to α-pinene and its concentration was iteratively tuned for each dataset in such a way that the total OH reactivity in the model agreed with the observed OH reactivity. To match the observed total OH reactivity, up to 955 3 ppbV of X (typically a factor of 5 to 10 times the measured α-pinene) had to be added. Tuning the simulation to the observed total OH reactivity does not improve the HO 2 agreement (HO 2 mod. /HO 2 obs. =0.4), as shown in Fig. 13. The production of OH from HO 2 is not sufficient to compensate 960 for the enhanced reactivity in the model, leading to underprediction of hydroxyl radicals in the simulation. The unaccounted OH reactivity in the simulation might still be linked to the missing source of HO 2 , nevertheless, this shows that α-pinene-like chemistry for the compound X does not suffice 965 to provide this connection. Figure 11. Simulated vs. observed OH concentrations and HO 2 mixing ratios, applying the MIM2 chemistry scheme and the recently proposed MIM3 including new additions in the isoprene chemistry.
HO 2 is underestimated in the simulation. The modelmeasurement discrepancy can be divided into two groups: The ratio of simulated-to-observed HO 2 concentrations is about 0.7 for a minor part of the data set, while the simulation significantly underestimates the observed total OH reactivity for the rest of the data set. The ratio between simulated and observed HO 2 concentrations in this case is only about 0.3. A recently published chemical reaction scheme (MIM3) including new additions to the isoprene chemistry, such as the photo-oxidation of unsaturated hydroperoxy aldehydes (Taraborrelli et al., 2012), produces even more hydroxyl radicals leading to an overestimation by a factor of up to 3 in the simulation as compared to the observations. The ratio between simulated and observed hydroperoxyl radicals increases to HO 2 mod. / HO 2 obs.
= 0.6 on average, still separating into two regimes with a minor part of the data set showing a ratio of HO 2 mod. / HO 2 obs.
= 1.5, while most of the data shows a ratio of HO 2 mod. / HO 2 obs.
= 0.6. Since isoprene was not the predominant biogenic VOC during HUMPPA-COPEC-2010, contributing less than 10 % to the total OH reactivity measured, the terpene mechanism MTM was added to the chemistry mechanism. Inclusion of terpene chemistry helped to reproduce the OH reasonably in the simulation (Fig. 12).
The ratio between simulated and observed HO 2 slightly changed compared to the isoprene chemistry reference run, but still separates into two regimes. At lower observed HO 2 levels, a ratio of up to HO 2 mod. / HO 2 obs.
= 0.8 is reached for a minor fraction of the data set. Only about 25 % of the observed HO 2 concentration can be reproduced by the model on average, when the simulated total OH reactivity does not match the observed reactivity.
The under-prediction of HO 2 , when OH reactivity is missing in the model while OH is reproduced accurately, indicates that the missing reactivity is an unaccounted-for source of HO 2 . Furthermore, the recycling reaction of NO and O 3 with the missing HO 2 has the potential under most conditions to compensate for the additional OH loss, which has not been considered yet. This preserves the good agreement of sim-ulated and observed OH even when the model accounts for all the observed total OH reactivity. However, when high total OH reactivities (> 20 s −1 ) were observed, the enhanced HO 2 recycling is not sufficient to reproduce the observed OH. Additional mechanisms, e.g. direct OH recycling as discussed for isoprene-dominated, low-NO x environments, or an additional primary source of OH is required to explain the observed hydroxyl radical concentrations under these conditions.
To investigate the hypothesis that the OH reactivity missing in the model using the terpene mechanism acts as a source of HO 2 , additional reactivity towards OH was introduced by adding the surrogate molecule X which behaves chemically identically to α-pinene and its concentration was iteratively tuned for each data set in such a way that the total OH reactivity in the model agreed with the observed OH reactivity. To match the observed total OH reactivity, up to 3 ppbV of X (typically a factor of 5 to 10 times the measured α-pinene) had to be added. Tuning the simulation to the observed total OH reactivity does not improve the HO 2 agreement (HO 2 mod. / HO 2 obs. = 0.4), as shown in Fig. 13. The production of OH from HO 2 is not sufficient to compensate for the enhanced reactivity in the model, leading to under-prediction of hydroxyl radicals in the simulation. The unaccounted-for OH reactivity in the simulation might still be linked to the missing source of HO 2 ; nevertheless, this shows that α-pinene-like chemistry for the compound X does not suffice to provide this connection.

Significance and uncertainties
To quantify the significance of model-measurement discrepancies, the uncertainties in both observations and simulations have to be considered. The uncertainties of HO x measurements are listed in Table 1. The uncertainties of simulated HO x related to the reaction rate coefficients were determined via Monte Carlo simulations as described in Sect. 2.4.1. Monte Carlo analysis using the terpene mechanism yields uncertainties of 14 % and 11 % for the simulated OH and HO 2 , Atmos. Chem. Phys., 14, 8723-8747, 2014 www.atmos-chem-phys.net/14/8723/2014/  Fig. 11. Simulated vs. observed OH concentrations and HO2 mixing ratios, applying the MIM2 chemistry scheme and the recently proposed MIM3 including new additions in the isoprene chemistry. cates that the missing reactivity is an unaccounted source of HO 2 . Furthermore, the recycling reaction of NO and O 3 with 935 the missing HO 2 has the potential under most conditions to compensate for the additional OH loss, which has not been considered yet. This preserves the good agreement of simulated and observed OH even when the model accounts for all the observed total OH reactivity. However, when high to-940 tal OH reactivities (> 20 s −1 ) were observed, the enhanced HO 2 recycling is not sufficient to reproduce the observed OH. Additional mechanisms, e.g. direct OH recycling as discussed for isoprene-dominated, low-NO x environments, or an additional primary source of OH is required to explain the 945 observed hydroxyl radical concentrations under these conditions.
To investigate the hypothesis that the OH reactivity missing in the model using the terpene mechanism acts as a source of HO 2 , additional reactivity towards OH was in-950 troduced by adding the surrogate molecule X which behaves chemically identically to α-pinene and its concentration was iteratively tuned for each dataset in such a way that the total OH reactivity in the model agreed with the observed OH reactivity. To match the observed total OH reactivity, up to 955 3 ppbV of X (typically a factor of 5 to 10 times the measured α-pinene) had to be added. Tuning the simulation to the observed total OH reactivity does not improve the HO 2 agreement (HO 2 mod. /HO 2 obs. =0.4), as shown in Fig. 13. The production of OH from HO 2 is not sufficient to compensate 960 for the enhanced reactivity in the model, leading to underprediction of hydroxyl radicals in the simulation. The unaccounted OH reactivity in the simulation might still be linked to the missing source of HO 2 , nevertheless, this shows that α-pinene-like chemistry for the compound X does not suffice 965 to provide this connection.  Fig. 13. Simulated vs. observed OH concentrations and HO2 mixing ratios, applying the MIM3*+MTM chemistry scheme and tuning the model to the observed total OH reactivity by introducing an additional α-pinene equivalent.

Significance and uncertainties
To quantify the significance of model-measurement discrepancies the uncertainties in both, observations and simulations, have to be considered. The uncertainties of HO x mea-970 surements are listed in Table 1. The uncertainties of simulated HO x related to the reaction rate coefficients were determined via Monte-Carlo simulations as described in Sect. 2.4.1. Monte-Carlo-Analysis using the terpene mechanism yields uncertainties of 14 % and 11 % for the simulated 975 OH and HO 2 , respectively. Measured and simulated HO x concentrations including their 1σ-uncertainties are compared in Fig. 14. Modelled and observed OH within their uncertainties partly agree within a factor of two. Significant discrepancies by more than a factor of two are present for a minor part 980 of the dataset (22 %). For HO 2 the underestimation by more than a factor of two in the model is significant for a large part of the dataset (52 %).
The uncertainties in observed trace gas species and photolysis frequencies used to constrain the model can also have 985 a direct impact on the uncertainty of simulated HO x . Even less obviously these parameters which are kept fixed at the observed values within the simulation, influence the abundance of unmeasured trace gases and intermediate products, therefore also showing an indirect effect on the resulting 990 calculation of HO x concentrations. Sensitivity studies were conducted by varying the observed trace gas concentrations separately by a factor between 0.5-10 to investigate the overall influence on the simulated HO x (Table 5) with the terpene mechanism.

995
Substances relevant for primary production of HO x show the highest impact in this sensitivity study. Doubling the O 3 or water vapour concentration leads to about 25 % increase in simulated OH and about 14 % enhanced HO 2 . An even larger effect (OH: +36 %, HO 2 : +25 %) is found for pho-1000 tolysis elevated by a factor of two. Constraining the model with twice the amount of NO observed causes 27 % more hydroxyl radicals in the simulation. The simulated HO 2 decreases by 21 %. Reducing the CO concentration by a factor of two leads to a 8 % increase in simulated OH and a 8 % 1005 decrease in simulated HO 2 while doubling the observed CO yields a 13 % decrease in OH and a 13 % increase in HO 2 . Doubling formaldehyde or H 2 O 2 only has a minor influence on the simulated HO x . Changing the deposition rates by a factor of two causes no significant change in HO x . None of 1010 these input parameter variations yield a change in OH and/or HO 2 of a factor of two or more. All photolysis frequencies multiplied by a factor of ten could cause such changes in simulated HO x . Unless the uncertainties of the species used to constrain the simulation (Table 1) exceed the variation inter-1015 vals, these uncertainties individually can neither explain the discrepancies between observation and modelled HO 2 , nor cause a significant disagreement for OH.
Free running HCHO, i.e. not constraining the simulation to the HCHO observations, leads to reduced OH and HO 2 1020 concentrations. Formaldehyde concentrations in the free running model are up to 15 times higher than observed levels. This suggests a formaldehyde sink not represented in the box model, possibly due to transport processes or uptake by plants (Lowe and Schmidt, 1983;Benning and Wahner, 1025 1998). However, it was shown that changes in the deposition velocity have only a small impact on the simulated HO x . The overprediction of formaldehyde in this case is more likely due to inadequate representation of the HCHO sources in the simulation. Another sensitivity simulation by also constrain-1030 ing the model to estimated 2-methyl-3-buten-2-ol (MBO) concentrations was conducted. Unfortunately, MBO, which is often referred to as "the isoprene of coniferous forests" was not directly measured during HUMPPA-COPEC-2010, though PTR-MS measurements of isoprene were affected 1035 by an MBO interference, as described in Sect. 2.1. Previous measurements in this forest showed that MBO emission Figure 13. Simulated vs. observed OH concentrations and HO 2 mixing ratios, applying the MIM3*+MTM chemistry scheme and tuning the model to the observed total OH reactivity by introducing an additional α-pinene equivalent.
respectively. Measured and simulated HO x concentrations including their 1σ uncertainties are compared in Fig. 14. Modelled and observed OH within their uncertainties partly agree within a factor of 2. Significant discrepancies by more than a factor of 2 are present for a minor part of the data set (22 %). For HO 2 the underestimation by more than a factor of 2 in the model is significant for a large part of the data set (52 %).
The uncertainties in observed trace gas species and photolysis frequencies used to constrain the model can also have a direct impact on the uncertainty of simulated HO x . Even less obviously, these parameters which are kept fixed at the observed values within the simulation influence the abundance of unmeasured trace gases and intermediate products, therefore also showing an indirect effect on the resulting calculation of HO x concentrations. Sensitivity studies were conducted by varying the observed trace gas concentrations separately by a factor between 0.5-10 to investigate the overall influence on the simulated HO x (Table 5) with the terpene mechanism.
Substances relevant for primary production of HO x show the highest impact in this sensitivity study. Doubling the O 3 or water vapour concentration leads to about a 25 % increase in simulated OH and about for 14 % enhanced HO 2 . An even larger effect (OH: +36 %, HO 2 : +25 %) is found for photolysis elevated by a factor of 2. Constraining the model with twice the amount of NO observed causes 27 % more hydroxyl radicals in the simulation. The simulated HO 2 decreases by 21 %. Reducing the CO concentration by a factor of 2 leads to a 8 % increase in simulated OH and a 8 % decrease in simulated HO 2 while doubling the observed CO yields a 13 % decrease in OH and a 13 % increase in HO 2 . Doubling formaldehyde or H 2 O 2 only has a minor influence on the simulated HO x . Changing the deposition rates by a factor of 2 causes no significant change in HO x . None of these input parameter variations yield a change in OH and/or HO 2 of a factor of 2 or more. All photolysis frequencies multiplied by a factor of 10 could cause such changes in simulated HO x . Unless the uncertainties of the species used to constrain the simulation (Table 1) exceed the variation intervals, these uncertainties individually can neither explain the discrepancies between observation and modelled HO 2 , nor cause a significant disagreement for OH.
Free running HCHO, i.e. not constraining the simulation to the HCHO observations, leads to reduced OH and HO 2 concentrations. Formaldehyde concentrations in the free running model are up to 15 times higher than observed levels. This suggests a formaldehyde sink not represented in the   Figure 14. Uncertainties of HO x simulations (using the terpene mechanism) with respect to measurement uncertainties. The solid line represents the 1 : 1 ratio of observed and simulated HO x . Data points showing significant discrepancies within their uncertainties by more than a factor of 2 ±1 (dashed lines) are shown in red. box model, possibly due to transport processes or uptake by plants (Lowe and Schmidt, 1983;Benning and Wahner, 1998). However, it was shown that changes in the deposition velocity have only a small impact on the simulated HO x . The overprediction of formaldehyde in this case is more likely due to inadequate representation of the HCHO sources in the simulation. Another sensitivity simulation, by also constraining the model to estimated 2-methyl-3-buten-2-ol (MBO) concentrations, was conducted. Unfortunately, MBO, which is often referred to as "the isoprene of coniferous forests" was not directly measured during HUMPPA-COPEC-2010, though PTR-MS measurements of isoprene were affected by an MBO interference, as described in Sect. 2.1. Previous measurements in this forest showed that MBO emission rates were only 1-3 % of the total monoterpene emission rate (Tarvainen et al., 2005). In American pine forests it can be more important. However, the PTR-MS signal was used as an upper limit estimate of MBO for this simulation to investigate the influence on the simulated HO x . This MBO causes, on average, about 0.5 s −1 additional reactivity towards OH, similar to the reactivity by methane. Only about 20 % of the observed HO 2 can be explained by the model and the observed OH is under-predicted by 40-60 % in this simulation. Considering the results from the above-mentioned sensitivity tests and significance analysis, the discrepancy between measured and simulated HO 2 is significant; thus, the chemical mechanism applied in the simulations needs further improvement to reproduce the HO 2 for the observed boreal forest environment.

Sources and sinks of HO x
The production rates of the hydroxyl radical in the box model using the terpene mechanism are shown in Fig. 15 with respect to their contribution to the total OH production derived from measurements. The data set is again grouped by conditions of different radiation and total OH reactivity. The total production rate of hydroxyl radicals in the simulation accounts only for about 10 % of the observed production when the total OH reactivity is high (k OH > 15 s −1 ). This is mainly due to the strong underestimation of HO 2 by the model under these conditions (see Sects. 3.1.1 and 3.1.2). Photolytic production accounts for one-third of the OH formation in the simulation. A similar contribution is found by HO 2 recycling via reaction with NO and O 3 . Ozonolysis of BVOCs and a minor contribution by other species produced in the model account for the remaining OH production. At lower radiation (J O( 1 D) ≤ 3 × 10 −6 s −1 ), the ozonolysis of BVOCs becomes relatively more important and the recycling reaction of HO 2 with O 3 becomes more relevant in comparison to the reaction with NO. Under conditions of moderate total OH reactivity (k OH ≤ 15 s −1 ) and low radiation (J O( 1 D) ≤ 3×10 −6 s −1 ), about one-third of the observed total OH production is represented by the box model simulation. Ozonolysis of observed biogenic VOCs accounts for, on average, 15 % of the total production rate. Even though the hydroxyl radical production rates due to recycling of HO 2 via the reactions with NO and O 3 are based on the underestimated HO 2 by the box model, they are relevant source terms contributing 13 % to the total OH production. Photolytic sources and production due to  other species play a minor role. By assuming observed rather than model calculated HO 2 , this contribution would increase to 30 %.
Almost one-quarter of the OH during periods characterized by high radiation values (J O( 1 D) > 3 × 10 −6 s −1 ) and moderate total OH reactivity (k OH ≤ 15 s −1 ) is produced by photolytic sources. Ozonolysis of BVOCs and other sources in the simulation contribute 4 and 2 % to the total observed OH production. HO 2 recycling via NO dominates the OH production (37 %). Nonetheless, about 30 % of the total OH observed production is not represented in the simulation due to the underestimation of HO 2 and the associated recycling pathways.
OH reactivity contributions calculated from individually measured compounds compared to directly measured OH reactivity revealed 58 % missing OH reactivity under "normal" boreal conditions and up to about 90 % under "stressed" boreal conditions (i.e. prolonged high temperature) during HUMPPA-COPEC-2010 (Nölscher et al., 2012). About 50 % of missing OH reactivity was reported from the same site during summer in 2008 by Sinha et al. (2010) when more typical boreal conditions at temperatures lower than in 2010 prevailed. The loss of OH in terms of OH reactivity in the simulation is presented in Fig. 16.
Similar to the findings of another modelling study based on different measurements conducted at the field station SMEAR II , the inorganic contribution (CO, O 3 , H 2 , H 2 O 2 , NO, NO 2 , HO 2 , and HONO) is significant (12 %). A large sink for OH is due to organic compounds, most importantly monoterpenes (4 %), methane (3 %), and isoprene (2 %). In contrast to the study by Mogensen et al. (2011), second-order and higher organic reaction products by the model contribute an additional 34 % to the total OH reactivity. On average, almost one-third of the observed total OH reactivity remains unexplained. The large contribution by second-order and higher organic reaction products in the model is mainly caused by aldehydes coming from monoterpene oxidation (∼ 30 %) and secondary products of isoprene oxidation (MVK + MACR ∼ 3 %). Primary biogenic emissions, e.g. unmeasured monoterpenes and sesquiterpenes, photo-oxidation products, as well as anthropogenic pollutants occasionally transported to the measurement site are likely candidates causing the "missing" fraction of total OH reactivity. HO 2 radicals are lost by reaction with other radicals as well as recycled towards OH by reaction with NO and O 3 (Fig. 17). The biggest contribution to the loss of HO 2 under all conditions is the yielding peroxynitric acid (HNO 4 ). However, under standard conditions, HNO 4 is rather unstable and decomposes typically within a few seconds. This creates an equilibrium between HNO 4 and HO 2 +NO 2 which is strongly temperature dependent (Dentener et al., 2002). In the simulation, the production and decomposition of HNO 4 occur at the same rate, thus the net effect is zero for the hydroperoxyl radicals.
At conditions of low radiation (J O( 1 D) ≤ 3×10 −6 s −1 ), hydroperoxyl radicals reacting with RO 2 and HO x contribute to the total loss of HO 2 in a similar amount as the recycling reactions with NO and O 3 . This is still the case at conditions of enhanced observed OH reactivity and low radiation. The higher absolute HO 2 loss is caused by enhanced recycling as well as enhanced loss by radical-radical reactions.
At high photolysis frequencies (J O( 1 D) > 3 × 10 −6 s −1 ), both the total HO 2 production and loss increase by about a factor of 2 compared to night-time. The HO 2 loss contribution from the recycling reaction with NO is predominant (41 %) at a total OH reactivity k OH ≤ 15 s −1 . This is followed by the reaction with RO 2 , contributing about 12 %. Other reactions, e.g. HO 2 +NO 3 or HO 2 +HO x , contribute only a few percent. When the total OH reactivity is high, k OH > 15 s −1 , the OH recycling reaction of HO 2 with NO becomes less important (23 %) in the simulation, whereas the reaction with RO 2 contributes the most to the total HO 2 loss. In addition to HNO 4 decomposition, the reaction of OH with carbon monoxide and the reaction of RO 2 with NO are the main sources of HO 2 in the model. The photolysis of formaldehyde contributes less than 1 % to the total HO 2 production under all conditions. Ozonolysis of monoterpenes yields up to 8 % when photolysis is low but is rather unimportant at higher J values. OH reactions with O 3 , HCHO, and H 2 O 2 are minor contributors (typically in total about 4 %) in HO 2 production. Differences between conditions of moderate and high total OH reactivity show up in the reactions of OH+CO and RO 2 + NO, forming HO 2 . The absolute HO 2 production under conditions of enhanced radiation and high observed total OH reactivity is smaller compared to the case with enhanced radiation and moderate total OH reactivity. Under these conditions, the simulation underestimates the observed HO 2 most, indicating that important species and HO 2 production pathways are missing in the chemical mechanism, very likely including unmeasured BVOCs and their oxidation products. Unfortunately, there were no direct observations of RO 2 during HUMPPA-COPEC-2010. Thus, we are lacking the possibility to constrain the model by RO 2 observations, which would improve our understanding of the underlying processes.

Summary and conclusions
The HORUS instrument was operated for the first time using the IPI-LIF-FAGE technique to measure OH during HUMPPA-COPEC-2010. To identify possible measurement interferences, an instrument comparison between the CIMS and HORUS-LIF instruments was conducted at the beginning of the campaign on the ground. The hydroxyl radical measurements by the two independent techniques show generally good agreement. During daytime, above-canopy OH concentrations up to 3.5 × 10 6 molec cm −3 were observed, a factor of 2 to 3 times higher than the on-ground measurements. During night-time these differences in OH vanished, but still, during some nights both instruments, (onground) CIMS and (above-canopy) IPI-LIF-FAGE, detected significant amounts of OH that cannot be fully explained by known production rates calculated from observations. The missing OH source during night-time of the order of 1 × 10 6 molec cm −3 s −1 could partly be related to NO 3 . The NO 3 mixing ratio during HUMPPA-COPEC-2010 was always below the lower limit of detection of the CRD instrument . However, it could nevertheless be produced at a rate of up to 1 × 10 6 molec cm −3 s −1 from the reaction of NO 2 with O 3 . RO 2 radicals from NO 3 -initiated VOC oxidation potentially undergo isomerization reactions or react with HO 2 as well, forming OH. Thus, NO 3 might, at least under some conditions, contribute to the missing OH source during night-time. Ozonolysis of unmeasured VOCs is also likely to contribute.
Comparison of the calculated total OH production rate with the total loss rate calculated from OH concentrations Atmos. Chem. Phys., 14, 8723-8747, 2014 www.atmos-chem-phys.net/14/8723/2014/ and total OH reactivity measurements showed that the known OH sources are almost sufficient to close the budget above the canopy. Detailed analysis of the radical production, loss, and recycling pathways revealed that OH recycling in the observed boreal forest environment occurs mainly by recycling via HO 2 under conditions of high radiation and moderate observed total OH reactivity. In addition, recycling mechanisms of OH, not via reaction of HO 2 with NO/O 3 , are likely under conditions of enhanced total OH reactivity. The role of potentially undetected VOCs and oxidation products in missing OH reactivity cannot be understood from observations alone. Hence, box model simulations have been considered for further investigation. The chemistry of the condensed isoprene mechanism (Mainz Isoprene Mechanism) is shown to be deficient for the observed monoterpene-dominated boreal forest environment. Isoprene levels of typically less than 200 pptV contributed at most 10 % to the total observed OH reactivity, thus leading to an overprediction of OH due to the missing sinks by a factor of up to 3 and HO 2 being significantly under-predicted (HO 2 mod. / HO 2 obs. = 0.3). Inclusion of the Mainz Terpene Mechanism (MTM) to account for the reactivity towards OH due to observed terpene species and their oxidation products leads to much better agreement between observed and simulated OH concentrations. However, this is due to two compensating effects. On average, about one-third of the observed total OH reactivity is not reproduced in the simulation, thus leading to underestimation of the total sink of OH. On the other hand, the production of OH is significantly underestimated due to the under-predicted HO 2 available for recycling by reaction with NO and O 3 , which was shown to be an important source of OH by direct calculation from observations. Tuning the simulation of the observed total OH reactivity by adding an unknown compound which behaves like α-pinene did not improve the HO 2 mod. / HO 2 obs.
agreement. HO 2 , being recycled towards OH, was not sufficient to compensate for the resulting enhanced OH loss. The biggest model-observation discrepancies for HO 2 occurred when missing OH reactivity was highest. Therefore, this provides evidence that the missing OH reactivity in the simulation is a source of HO 2 . However, a single terpene following the chemical mechanism of α-pinene as described in the preliminary terpene mechanism cannot account for this alone. The OH production in this forest environment seems to be understood reasonably well under moderate total OH reactivity conditions (Table 4). The loss of OH cannot be fully explained, even through inclusion of higher-order oxidation products by application of a box model. Similarly to the study of Kim et al. (2013), HO 2 is underestimated in simulations, indicating missing BVOCs, which could also account for the missing reactivity in the model. Additional recycling processes of OH, not via HO 2 + NO/O 3 , are indicated under conditions of high observed total OH reactivity by detailed analysis of the HO x budget calculated from observations. RO 2 measurements could help constrain simulations better and further improve the understanding of the radical cycling processes in this boreal forest (and elsewhere). The oxidation capacity in the observed forest environment is mainly defined by high recycling probabilities at moderate NO x levels. Chemical mechanisms need to include a comprehensive representation of the most abundant biogenic VOCs, e.g. in this case monoterpenes and their oxidation chemistry, in order to reproduce the radical photochemistry when isoprene is not the predominant BVOC.
The Supplement related to this article is available online at doi:10.5194/acp-14-8723-2014-supplement.