Annual cycle of volatile organic compound exchange between a boreal pine forest and the atmosphere

Long-term flux measurements of volatile organic compounds (VOC) over boreal forests are rare, although the forests are known to emit considerable amounts of VOCs into the atmosphere. Thus, we measured fluxes of several VOCs and oxygenated VOCs over a Scots-pine-dominated boreal forest semi-continuously between May 2010 and December 2013. The VOC profiles were obtained with a proton transfer reaction mass spectrometry, and the fluxes were calculated using vertical concentration profiles and the surface layer profile method connected to the Monin-Obukhov similarity theory. In total fluxes that differed significantly from zero on a monthly basis were observed for 13 out of 27 measured masses. Monoterpenes had the highest net emission in all seasons and statistically significant positive fluxes were detected from March until October. Other important compounds emitted were methanol, ethanol+formic acid, acetone and isoprene+methylbutenol. Oxygenated VOCs showed also deposition fluxes that were statistically different from zero. Isoprene+methylbutenol and monoterpene fluxes followed well the traditional isoprene algorithm and the hybrid algorithm, respectively. Emission potentials of monoterpenes were largest in late spring and autumn which was possibly driven by growth processes and decaying of soil litter, respectively. Conversely, largest emission potentials of isoprene+methylbutenol were found in July. Thus, we concluded that most of the emissions of m/z 69 at the site consisted of isoprene that originated from broadleaved trees. Methanol had deposition fluxes especially before sunrise. This can be connected to water films on surfaces. Based on this assumption, we were able to build an empirical algorithm for bi-directional methanol exchange that described both emission term and deposition term. Methanol emissions were highest in May and June and deposition level increased towards autumn, probably as a result of increasing relative humidity levels leading to predominance of deposition.


Introduction
Knowledge on biogenic emissions of volatile organic compounds (VOCs) has been continuously increased as a result of a development of modelling methods and extended measurement network community (Guenther et al., 1995(Guenther et al., , 2006(Guenther et al., , 2012Sindelarova et al., 2014). VOCs, such as monoterpenes 25 and isoprene, make a major contribution to the atmospheric chemistry, including tropospheric ozone formation, control of atmospheric radical levels, and aerosol particle formation and growth. Therefore, these compounds affect both local and regional air quality and the global climate (Atkinson and Arey, 2003;Kulmala et al., 2004;Spracklen et al., 2008;Kazil et al., 2010).
In addition to terpenoids, vegetation also emits copious amounts of oxygenated volatile organic 30 compounds (OVOCs). Their contribution to the total biogenic VOC budget has been estimated to be ca. 10-20 % in carbon basis (Guenther et al., 2012;Sindelarova et al., 2014). Due to their lower reactivity, OVOCs have only a minor role in the boundary layer chemistry but they can be transported to the upper troposphere where for example methanol can possibly have a major effect on oxidant formation (Tie et al., 2003;Jacob et al., 2005). Methanol emissions have been widely stud- 35 ied in recent years (e.g. Guenther et al., 2012 and references therein). However, it has been recently observed that methanol has also significant deposition at some ecosystems. This deposition could be related to the night-time dew on surfaces (Holzinger et al., 2001;Seco et al., 2007;Wohlfahrt et al., 2015) but removal mechanisms of methanol from the surfaces are still unknown (e.g. Laffineur et al., 2012). In global estimates, methanol deposition is usually determined with a deposition velocity that 40 is tuned to fit concentration observations, leading possibly to uncertainties in methanol budget estimates (Wohlfahrt et al., 2015). Other OVOCs than methanol are even more poorly described in the global scale (Karl et al., 2010).
Generally, boreal forests are important emitters of for example monoterpenes, even though their contribution to global total VOC emission is surpassed by isoprene emission from tropical rainforest 45 (e.g. Guenther et al., 2012). However, the negative temperature-monoterpene emission-aerosol feedback on the regional climate is estimated to be significant (up to −0.6 Wm −2 K −1 , see Paasonen et al., 2013, and also Spracklen et al., 2008).
In order to describe the VOC exchange processes in models, continuous long-term ecosystem, or canopy, scale flux measurements play an important role (Guenther et al., 2006). They can be 50 used to study the dependencies of these fluxes on environmental variables. Also, even when the process understanding has been obtained by for example laboratory experiments, the evaluation of model in ecosystem scale is a crucial step towards reliable global exchange estimates. Unfortunately, the ecosystem scale flux measurements are rare. As an example, even though branch scale monoterpene emissions from Scots pine are well-studied (Ruuskanen et al., 2005;Tarvainen et al., 55 2005; Hakola et al., 2006;Aalto et al., 2014Aalto et al., , 2015, ecosystem scale emissions from Scots pine dominated forests have been mainly explored in short campaigns (Rinne et al., 2000bGhirardo et al., 2010). Longer time series have also consisted of measurements from May to Septem-ber only (Räisänen et al., 2009;Taipale et al., 2011). This has had a direct effect on the capability of models to predict monoterpene concentrations (Smolander et al., 2014). 60 Thus, we have measured ecosystem scale fluxes of VOCs using the proton transfer reaction quadrupole mass spectrometer (PTR-MS, Lindinger et al., 1998) above a Scots pine dominated forest in Hyytiälä at SMEAR II (Station for Measuring Forest Ecosystem-Atmosphere Relations) since 2010. In this study, we quantify the ecosystem scale VOC emissions and deposition at a boreal forest site throughout the seasonal cycle. The most important ecosystem scale VOCs emitted at the 65 site are monoterpenes and methanol , thus we concentrate on these compounds separately. Isoprene is also analysed more precisely because despite to its importance in the global scale, ecosystem scale emissions have remained unstudied in Scots pine dominated forests.
In the case of monoterpenes and isoprene, we will examine emissions with algorithms suggested by Guenther et al. (1993) and Ghirardo et al. (2010). Our purpose is to study how well the algorithms 70 are able to predict ecosystem scale fluxes, and how much there is seasonal variation in emission potentials. As the last aim, we examine the importance of the methanol deposition, and develop a simple empirical algorithm describing the bi-directional exchange needed to achieve more precise methanol flux budgets. This algorithm is evaluated against the measurements.

Measurement site and VOC concentration calculations
All measurements were conducted in Hyytiälä, Finland, at SMEAR II (Station for Measuring Forest Ecosystem-Atmosphere Relations, 61 • 51 ′ N, 24 • 17 ′ E, 180 m a.m.s.l., UTC+2). Hyytiälä is located in the boreal region and the dominant tree species in the flux footprint is Scots pine (Pinus sylvestris). In addition to Scots pine, there are some Norway spruce (Picea abies) and broadleaved 80 trees such as European aspen (Populus tremula) and birch (Betula sp.). The forest is about 50 years old and the canopy height is currently ca. 18 m. Hari and Kulmala (2005), Haapanala et al. (2007) and Ilvesniemi et al. (2009) have given a detailed description about the station infrastructure and surrounding nature.
The proton transfer reaction quadrupole mass spectrometer (PTR-MS, manufactured by Ionicon 85 Analytik GmbH, Innsbruck, Austria) was measuring 27 different masses (see Table 1) using a 2.0 s sampling time from six different measurement levels at a tower which was mounted on a protruding bedrock, ca. 2 m above the average forest floor. Two of the measurement levels (4.2 and 8.4 m) were below the canopy and four of them (16.8, 33.6, 50.4 and 67.2 m) above it. VOC fluxes were derived from the profile measurements with the surface layer profile method. The temperature was 90 also measured at the VOC sampling levels with ventilated and shielded Pt-100 sensors. A 3-D acoustic anemometer (Gill Instruments Ltd., Solent 1012R2) was installed at height of 23 m and it was used for determining turbulence parameters, including turbulent exchange coefficients. The photo-synthetic photon flux density (PPFD, Sunshine sensor BF3, Delta-T Devices Ltd., Cambridge, UK) was measured at the height of 18 m. The relative humidity (Rotronic AG, MP102H RH sensor) was 95 measured at the height of 16 m.
The PTR-MS was located inside the measurement cabin and samples were drawn down to the instrument using heated 14 mm i.d. PTFE tubing of equal length at all levels. The sample lines were 100 m long until the end of 2013 and 157 m from 2013 onwards. The change was due to the extension of the tower from 73 to 127 m length. A continuous air-flow was maintained in the tubes 100 (43 L min −1 ). From these lines a side flow of 0.1 L min −1 was transferred to PTR-MS via a 4 m PTFE tube with 1.6 mm i.d. During the measurements, the instrument was calibrated roughly every second week using two VOC standards (Apel-Riemer). The calibrations were performed with manually operated flow measurements until 7 July 2011 (Taipale et al., 2008). From that date onwards, the flow levels were obtained with a mass flow controller (Kajos et al., 2015). The volume mixing ratios 105 were calculated using the procedure described in detail by Taipale et al. (2008). The primary ion signal m/z 19 (measured at m/z 21) had some variations over the years being approximately around 10 − 30 × 10 6 cps. SEM was always optimized before a calibration, and we used same SEM-model (MasCom MC-217) over all years.
The instrumental background was determined every third hour by measuring VOC free air, pro-110 duced with a zero air generator (Parker ChromGas, model 3501). In addition, the estimated oxygen isotope O 17 O was subtracted from m/z 33 to avoid contamination of methanol signal. The isotope signal was estimated by multiplying the measured signal of m/z 32 by a constant O 17 O/O 2 ratio (0.00076, see Taipale et al., 2008). Samples for the zero air generator were taken outside of the measurement cabin close to the ground, and the stability of the zero air generator was followed con-115 tinuously. We found that the generator had some problems at m/z 93 but this did not affect on the flux calculations as the same zero air signal was subtracted from each concentration level.

Flux calculation procedure
The flux of a compound can be written as In this study, fluxes were quantified using the surface layer profile method. Detailed description of the flux calculation is given by Rantala et al. (2014), who use the term profile method of this variant of gradient method. Below we give only a brief outline of the method.
According to the Monin-Obukhov theory, a concentrationc(z j ) can be calculated at any height z j 125 in the surface layer using the formulā where In here, k = 0.4 is the von Kármán constant (e.g. Kaimal and Finnigan, 1994), Ψ h (ζ) is the integral form of the dimensionless universal stability function for heat, z 0 is the roughness length, and ζ = (z − d)/L is the dimensionless stability parameter where L is the Obukhov length (Obukhov, 1971) and d the zero displacement height. L has been derived using dimensional analysis and it has the 135 following form whereθ v is the potential virtual temperature, g the acceleration caused by gravity (g ≈ 9.81 m s −2 ) and (w ′ θ ′ v ) s the turbulent heat transfer above the surface (in our case at 23 m). z 0 is the surface roughness length, z M the highest measurement level, and variables x i+1 i refer to the average values 140 between heights z i and z i+1 . Using the equations above, the surface layer parameter c * , and the flux, can be derived using the least square estimate (a linear fit).
For the flux calculation procedure, we selected d = 13 m and γ = 1.5 between the two lowest levels (Rantala et al., 2014). Between other measurement levels, the roughness sublayer correction factor γ was assumed to be 1, i.e. no corrections were applied. Our lowest and highest measurement  Based on those results, we decided to use the profile method for long-term measurements at the site as the DEC-method was often found to have problems in determining low VOC fluxes. For example, the lag-time determination was turned out to be difficult in conditions where values are usually close to flux detection limit. Moreover, the high frequency losses are currently unknown for many VOCs as the response time of the PTR-MS has been studied for water vapour only (Rantala et al., 2014). On from further analysis. Finally, we disregarded 2.5 % of the lowest and highest values from every month as outliers.
The filtering criteria applied were strongly turbulence dependent, which implies that night-time values had higher probability to be rejected. Therefore, monthly means, later introduced, were derived from gap-filled fluxes. In the gap-filling procedure, the missing flux values were replaced by a 165 corresponding value from median diurnal cycle, calculated from the measurements made within 16day-window around a missing value (Bamberger et al., 2014). However, there had to be at least one measured value on both sides of a missing value in the gap filling window, otherwise that missing value was not gap-filled In this study, we have often used a relative error, ∆R, that is defined as where h corresponds to measured flux values and q to values given by an algorithm. Pearson's correlation coefficient, r, was used widely through the study as well, and it is hereafter referred as correlation.
Algorithm optimization is applied many times, and all fits were based on, if not stated other-175 wise, least squares minimization and trust-region-reflective method that is provided as an option in MATLAB (function fit).

Emission algorithms of isoprene and monoterpenes
The well-known algorithm for isoprene emissions (E iso ) is written as 180 where E 0,synth , C T and C L are same as in the traditional isoprene algorithm (Guenther et al., 1991(Guenther et al., , 1993. The shape of this algorithm is based on the light response curve of electron transport activity and the temperature dependence of the protein activity. Similar behaviour for methylbutenol (MBO) emissions from Ponderosa pine has been suggested by for example Gray et al. (2005).
The algorithm we used for monoterpene emissions is the hybrid algorithm where f synth ∈ [0 1] is the ratio E 0,synth /E 0,hybrid (Ghirardo et al., 2010;Taipale et al., 2011). E pool is the traditional monoterpene algorithm by Guenther et al. (1991) and Guenther et al. (1993) and Γ = e β(T −T0) the temperature activity factor, where β = 0.09 K −1 and T 0 = 303.15 K. The hybrid algorithm is based on the observation that part of the monoterpene emission even from conifer-190 ous trees originates directly from synthesis. Therefore, it can be calculated using algorithm similar to isoprene emission algorithm while the rest originates as evaporation from large storage pools (Ghirardo et al., 2010). The latter can be calculated using exponentially temperature dependent al-gorithm, as the temperature dependence of the monoterpene saturation vapour pressure is approximately exponential (Guenther et al., 1991(Guenther et al., , 1993. The formula, is hereafter referred as the pool algorithm.

Net exchange algorithm of methanol
The total exchange of methanol consists of both emission term, E meth , and deposition term, D meth .
Therefore, an algorithm for the methanol flux, F meth , has the form of According to observations, biogenic methanol production is mainly temperature dependent, with photosynthesis having no direct role (Oikawa et al., 2011). Instead of that, the emissions are potentially controlled by stomatal opening, as methanol has high water solubility, i.e. low Henry's constant (e.g. Niinemets and Reichstein, 2003;Filella et al., 2009). Therefore, we assumed that a part 205 of the emissions could be represented by the traditional temperature activity factor Γ multiplied by a light dependent scaling factor of stomatal conductance. In addition, methanol is also produced by non-stomatal sources, such as decaying plant matter (Schade and Custer, 2004;Harley et al., 2007;Seco et al., 2007). Moreover, Aalto et al. (2014) observed with chamber studies that at least part of the methanol emissions is independent of light during springtime. Hence, we estimated that the total 210 methanol emission, E meth , is determined as where E 0,meth and f stomata ∈ [0 1] are an emission potential and a fraction of stomatal controlled emissions, respectively. The light dependent scaling factor of stomatal conductance, G light , was estimated as where α = 0.005 µmol −1 m 2 s is the same as used by Altimir et al. (2004) for pine needles. The stomatal conductance is also dependent on for example the temperature and vapour pressure deficit but their effect is much weaker than the effect of light at the site (Altimir et al., 2004). For the temperature activity factor, we used a parameter β = 0.09. In principle, β should be determined from 220 measurements but we wanted to have as few experimental parameters as possible. Therefore, we used the same value as for monoterpenes.
We assumed that methanol is deposited on wet surfaces, such as on dew, in a way that the methanol concentration at the absorbing surface is zero. Thus, a deposition term, D meth , was estimated to be

225
where ρ methanol is a mass mixing ratio measured at z = 33.6 m and V d a deposition velocity. The function f (RH) defines a filter of relative humidity (RH) in a such way that where RH 0 was determined from the measurements. The deposition velocity V d was determined by a resistance analogy: where R a is the aerodynamic resistance, R b the laminar boundary-layer resistance, and R w a surface resistance. The aerodynamic resistance is written as: where the correction factor γ(z 1 , z 2 ) = 1.5 as with the flux calculations. R b was determined by a 235 commonly used formula (Wesely and Hicks, 1977) where η is a diffusivity of methanol and κ a thermal diffusivity of air. The factor R w was assumed to be constant and it was determined from the measurements. In reality, R w might be also consisting of stomatal uptake due to oxidation of methanol into formaldehyde on leaves (Gout et al., 2000).

240
Consequently, the assumption of a constant value is a very rough estimate. However, in order to simplify the algorithm as much as possible, the parameterized deposition velocity consisted only of the factors R a , R b and a constant R w . We used the constant values of 1 m and 13 · 10 −6 m 2 s −1 for the surface roughness length (z 0 ) and for the diffusivity of methanol (η), respectively. The diffusivity of methanol was approximated at 273.15 K using Chapman-Enskog theory (e.g. Cussler, 1997).

245
Generally, the diffusion coefficient, and thus the deposition velocity, would be larger at higher temperatures. However, using the constant value causes only a minor error. We assumed also a constant value for the thermal diffusivity of air (κ = 19 · 10 −6 m 2 s −1 ).

Statistical significance of fluxes 250
For the analysis of seasonal cycle the fluxes were divided into twelve monthly bins, each containing data from a specific month of all years. To study whether the measured fluxes from each month differed significantly from zero or not, numbers of positive and negative fluxes were counted. The null hypothesis was that there is no flux, thus the counts of positive and negative values are equal.
Finally, it was determined from the binomial distribution with a confidence level of 99.9937 % ("4σ",

255
Clopper-Pearson method) whether a fraction of positive and negative values could be generated by a random process (the null hypothesis), or if there was a real positive or negative flux, i.e. the null hypothesis was rejected. We made the test for both night-(2-8 a.m.) and day-time (11 a.m.-5 p.m.) fluxes separately. Measurements from January and February were excluded from the analysis due to the lack of data points. Measurements at higher mass-to-charge ratio (m/z) than 137 were also 260 left out from the analysis due to a very low sensitivity of the PTR-MS at those masses. In addition, identification of the heavier masses was proven to be extremely difficult.
Altogether, 13 masses (excluding monoterpene fragments at m/z 81) had statistically significant fluxes on a monthly scale (Table 2). One should note that the masses for which no significant flux ( Table 2) Surprisingly, statistically significant formaldehyde fluxes were also observed. However, formaldehyde is poorly detected and quantified with the PTR-MS due to its low proton affinity. Thus, the observed fluxes may be related for example to the behaviour of water vapour (de Gouw and Warneke, . We tried to minimize the interference of water vapour using a normalization method which takes into account changes in water cluster ions (Taipale et al., 2008). There were also other controversial discoveries such as net emissions of m/z 93. A compound at m/z 93 is usually connected with toluene but it might be a fragmentation product of p-cymene as well (Ciccioli et al., 1999;Heiden et al., 1999;White et al., 2009;Ambrose et al., 2010;Park et al., 2013). We found a depen-  (Table 2). Therefore, we concluded that acetonitrile had a major contribution to the observed depo-295 sition of m/z 42.
The measured fluxes do have significant uncertainties. Some of these are random in nature and will thus cancel out with data analysis of sufficiently large data set. Some of the uncertainties are more systematic and may bias average flux values presented. The surface layer profile method itself may have a systematic error of about 10 % (Rantala et al., 2014). In addition, monoterpene fluxes 300 are underestimated up to few percent by the chemical degradation (Spanke et al., 2001;Rinne et al., 2012;Rantala et al., 2014). Our calibration procedure may also contain systematic error sources.
This concerns especially the indirect calibration if molecules are fragmented, such as in the case of methylbutenol at m/z 87 (Taipale et al., 2008). In addition to systematic errors, random flux uncertainties are also several hundreds of percent for many compounds (Rantala et al., 2014). On the other 305 hand, when averaging over a sample size of ca. a hundred data points, a random uncertainty of the average is decreased to the scale of 10 %.
After the addition of a mass flow controller to the calibration system in 7 July 2011, the sensitivities of methanol were observed to be highly underestimated. The reason was unknown but the biased sensitivities were probably caused by an absorption of methanol on metal surfaces of the mass 310 flow controller (Kajos et al., 2015). Therefore, methanol concentrations were derived from general transmission curves (indirect calibration) after that date ( Table 2). The indirect calibration might potentially lead to large systematic errors. However, no rapid changes in the methanol concentrations were observed after 7 July 2011. is more likely to be isoprene or MBO. MBO is produced by conifers (Harley et al., 1998) whereas many broad-leaved trees are high isoprene emitters (Sharkey and Yeh, 2001;Rinne et al., 2009).

Monoterpene and isoprene fluxes
In order to quantify the emission potentials for isoprene+MBO, measured flux values were fitted against the isoprene algorithm (Eq. 7) for each month separately. We found a significant corre-325 lation between the measurements and the calculated emissions from May, June, July and August (Table 3). Here we defined that the measurements and the calculated values correlated significantly if the p value (p) of the correlation (r) was smaller than 0.0027 (3σ-criteria). In June, July, and August, the measured fluxes were also clearly light dependent (Fig. 3). Shapes of the curves in the Fig. 3 go near to zero when PPFD is zero and the normalized values have also their saturation point 330 around PPFD = 500 µmol m −2 s −1 where C L is also already larger than 0.8 (Fig. 3). In May, the dependency between the measured fluxes and light was, however, unclear. However, the calculated values corresponded well with the measured ones as is seen in Fig 4. Previous emission studies with chamber method with gas chromatography have shown that Scots pines emit MBO much more than isoprene (Tarvainen et al., 2005;Hakola et al., 2006). However, May and June, and Aalto et al. (2014) showed that the increased MBO emissions during early summer were related to new biomass growth. In the case of isoprene emissions from aspen, the maximum should come later in July (Fuentes et al., 1999). In this study, the maximum emission potential of 345 m/z 69 was observed in July, indicating that most of the emissions of m/z 69 might actually consist of isoprene. Maximum net emissions of m/z 87 were also detected in July (Table 2)

Monoterpenes, their emission potentials and differences to branch scale studies
Monoterpenes are emitted by Scots pine (Hakola et al., 2006), Birch (Hakola et al., 2001) and forest floor (Hellén et al., 2006;Aaltonen et al., 2011Aaltonen et al., , 2013 at the site. According to Taipale et al., 2011, Scots pine is the most important monoterpene source in summer but its fraction of the total emission in spring and fall have remained unstudied. Therefore, monoterpene fluxes from spring-and autumn-355 time will be analysed more carefully in this chapter. Unsurprisingly, a seasonal cycle of monoterpene fluxes correlated roughly with the temperature (Fig. 2). To examine a response of monoterpene fluxes to the temperature and light in more detail, the fluxes were fitted against the hybrid algorithm, and the pool algorithm (Eqs. 8 and 9) for each month separately (Fig. 5). We found a correlation (p value was smaller than 0.0027) between the 360 measurements and the hybrid algorithm from April to October (Table 4).
Significant monoterpene fluxes were also observed in March but no dependence with the temperature was found. This is most probably due to the low temperatures and its diurnal variation, letting the random variation in the flux data to dominate. In addition, Aalto et al. (2015) observed that freezing-thawing cycles may increase the monoterpene emission capacity of Scots pine shoots; 365 in late autumn and early spring such cycles are frequent and potentially hide the relation between temperature and emissions at least partially. Nevertheless, monoterpene fluxes in March were in a reasonable range being lower than in April (Table 2, Fig 6).
Correlations between measured fluxes and the hybrid emission algorithm were better than those between measured fluxes and the pool algorithm in every month analysed (Table 4). In addition, 370 relative errors (Eq. 6) between the measured fluxes and the hybrid algorithm were also smaller than the relative errors between the measured fluxes and the pool algorithm. Thus, the hybrid algorithm worked better than the pool algorithm in every month. The result was expected as Taipale et al.
(2011) showed that ecosystem scale monoterpene emission from Scots pine forest, measured by the disjunct eddy covariance method, has a light dependent part. In addition, Ghirardo et al. (2010) 375 has shown by stable isotope labeling that a major part of the monoterpene emissions from conifers originates directly from synthesis (de novo). In this study, the ratios f synth = E synth /E pool varied between 0.36 (July) and 0.79 (October) whereas Ghirardo et al. (2010) estimated that the fraction of the de novo emissions from Scots pine seedlings to be around 58 %, and Taipale et al. (2011) estimated the fraction to be around 40% for the Scots pine ecosystem. Generally, these estimates fit 380 well our results considering the relatively large uncertainties (Table 4).
In the case of the hybrid algorithm, the largest emission potentials were found in May and in October ( (2005) found that the emission potential of monoterpenes was five times higher in early summer than in late summer. In that study, however, the parameter β was ca. 0.18 in the early summer and only ca. 0.08 in the late summer which makes the direct comparison of the emission potentials between the seasons difficult.

395
The hybrid algorithm matched with measurements especially well from May until July when ∆R < 50 % and r > 0.6. Conversely to those months, the measurements from October were noisy leading to somewhat unreliable fitting parameters (Table 4 and Fig. 5). Compared to earlier estimates on autumn monoterpene emissions based on extrapolation of short measurement campaigns (e.g. Rinne et al., 2000a), the autumnal monoterpene emissions were larger than expected. Although one 400 should keep in mind that the data set of this study from October was relatively small, and the results are therefore less representative than from other months. Nevertheless, increased microbiological activity in the fall has been observed to have an effect on the monoterpene emissions ( 100 ng m −2 s −1 ).
In addition to the temperature and light intensity, monoterpene emissions have been also connected to other abiotic stresses, such as mechanical damage, high relative humidity, drought, and 410 increased ozone level (e.g. Loreto and Schnitzler, 2009 and references therein). At the ecosystem level, such stress related emissions could often increase monoterpene fluxes. Thus, they will be incorporated into emission potentials even though the pool algorithm or the hybrid algorithm cannot describe those stress emissions at a process level. We found for example a weak dependency between relative humidity and monoterpene fluxes in low (PPFD < 50 µmol m −2 s −1 ) light conditions 415 (Fig. 7). Nevertheless, the measured mean fluxes differed from the predicted mean emissions only a few percent in monthly basis, i.e. in our dataset clear signals of stress related emissions in a temporal scale of one month were not found (see also Fig 4).
Overall, there were some results that were not totally corresponding with previous monoterpene studies. According to Hakola et al. (2006), monoterpene emissions from two Scots pine branches We found periods of net deposition for various OVOCs: methanol, acetaldehyde, acetone and acetic acid. Although for acetic acid, the observed deposition was weak. In the fall, methanol and acetone fluxes were even dominated by deposition (Table 2). Methanol, acetone and acetaldehyde fluxes had also a negative correlation with the relative humidity (RH) which might indicate the deposition is connected with moisture, such as water films on plant surfaces. However, after normalizing fluxes 440 with the temperature and light, only methanol had a statistically significant relationship with RH (95 % confidence level). Figure 9 shows how both temperature and light classified methanol fluxes behave as a function of relative humidity. The deposition starts at around RH = 75 %, therefore that value was selected as the threshold value RH 0 (Eq. 14). Although, the method of selecting the threshold value RH 0 is somewhat subjective, the value RH 0 = 75 % is well in line with earlier 445 observations by Altimir et al. (2006) who found the surface water film starting to occur when RH 60...70 %. The surface resistance R w (Eq. 15) was determined by minimizing the relative error between the calculated and measured methanol fluxes in Jul-Aug when the fluxes were the largest.
On average, the smallest relative error was obtained with a value of R w = 120 s m −1 , thus it was selected to be the constant resistance. Methanol could also deposit to the stomata. However, at least 450 part of the deposition should happen on the non-stomatal surface as the lowest mean concentrations were measured close to the ground during night time (Fig 8).
Measured methanol fluxes were fitted against the exchange algorithm (Eq. 10) for each month. The seasonal behaviour of the emission potentials was found to be similar to monoterpenes: both compounds have the maximum emission potentials in late spring and in autumn, and the lowest emission 455 potential in late summer (Table 5). The high emission potential in May (and June) is probably partly related to growth processes as methanol emissions correlate with leaf growth (e.g. Hüve et al., 2007).
The ratio f stomata (Eq. 11) had somewhat opposite cycle with the maximum values recorded in summer and the lowest values in spring. This could be related to non-stomatal emissions in springtime, most probably from decaying litter that is re-exposed after snowmelt. The behaviour is visible in 460 Fig. 3 where normalized methanol emissions are presented as a function of PPFD from each month. Generally, the algorithm was able to represent the measured values well (Figs. 10 and 4). An exception is May when the measured median day-time values were much lower than calculated values.
The relative errors were larger compared with the corresponding results of monoterpenes in every month. This indicates that the measured methanol fluxes were either noisier than measured monoter-465 pene fluxes, or our exchange algorithm could not describe methanol fluxes as well as the hybrid or the pool algorithm describes monoterpene emissions. For example, the parameterization of the RHfilter (Eq. 14) might bring a considerable uncertainty because as there may be deposition already at lower relative humidities than RH = 75 %. Moreover, the shape of the RH response curve f (RH) is probably smoother than a step function (Eq. 14). Nevertheless, the deposition seems to have an 470 important role in a methanol cycle between a surface and the atmosphere. Based on our calculations, the total deposition from April to September was slightly lower than 40 % compared with the total emissions within the same period (Fig. 11). However, it is impossible to distinguish which part of the deposited methanol evaporates back into the atmosphere again. Part of the deposited methanol is removed irreversibly from the atmosphere, as the mean methanol flux is negative in October (Ta-475 ble 2) but the removal processes of methanol from surfaces are generally unknown. Laffineur et al.
(2012) estimated that a half lifetime for methanol in water films is 57.4 h due to chemical degradation but the origin of the process was unidentified. The methanol sink has been also connected to consumption by methylotrophic bacteria (Duine and Frank, 1980;Laffineur et al., 2012).

Conclusions
Using VOC data set from four years, we were able to detect monthly mean fluxes for 13 out of 20 masses (excluding masses heavier than m/z 137) that were statistically different from zero. The largest positive fluxes were those of monoterpenes through almost the whole year, whereas different 495 oxygenated VOCs showed the highest negative fluxes, i.e. deposition. Oxygenated VOCs had also considerable net emission in May and early summer.
The hybrid algorithm described monoterpene fluxes better than the pool algorithm as expected.
However, the differences in correlations and relative errors between the pool and the hybrid algorithm were rather small. In the case of the hybrid algorithm, the highest emission potentials of 500 monoterpenes were recorded in May, and on the other hand in October, probably due to different growing and decaying processes. One should still keep in mind that interannual variations of the emission potentials were considerable in May. This indicates that a one year data set might be too short for determining representative estimates for emission potentials.
Most of the flux observed at m/z 69 was estimated to be isoprene, likely emitted by the nondom- A considerable amount of OVOCs was found to be deposited into the forest, especially in the 510 fall. We observed that the methanol deposition is probably related to water films on surfaces, which can be parameterized. Deposition mechanisms for other measured OVOCs were left unknown as no significant relationship between the fluxes and the relative humidity or other environmental parameters was found. Nevertheless, mean acetone and also methanol fluxes were negative in autumn, which indicates that after depositing, those compounds were not fully re-evaporated back into the 515 atmosphere. Hence, a sink mechanism for some OVOCs should exist. Overall, we estimated that the cumulative deposition of methanol (April-September) is slightly less 40 % compared with the corresponding cumulative methanol emissions. In reality, the fraction is even larger as methanol has probably net deposition in October-December.
Constructing a simple mechanistic algorithm to describe a methanol exchange between the sur- in August, which indicates that the largest emissions originate from growth processes. It was also observed that summertime emissions are strongly light dependent whereas springtime emissions are more driven by the temperature. One possible explanation is that methanol emissions are controlled by stomatal opening during summer, while in spring time the methanol might be produced partly by decaying litter.
methanol flux: synthesis of micrometeorological flux measurements, Atmos. Chem. Phys., 15, 7413-7427, doi:10.5194/acp-15-7413-2015Phys., 15, 7413-7427, doi:10.5194/acp-15-7413- , 2015. Table 1. The compound names and the formulas listed below in third and fourth column, respectively, are educated estimates for the measured masses (see e.g. de Gouw and Warneke, 2007). However, also other compounds might have a contribution at the measured masses (e.g. m/z 85, see Park et al., 2013). The second column shows whether a sensitivity was determined directly from the calibration or not (derived from a transmission curve, i.e. calculated), and which compounds were used in the calibrations.        September has the value of 100 %. One should note that due to uncertainties in the calculations, substraction between the cumulative emission and the cumulative deposition is unequal to the cumulative flux (Table 5).