The eddy covariance measurements were conducted at the Siikaneva 1 site of the Station for Measuring Ecosystem-Atmosphere Relations II (SMEAR II). Siikaneva is an oligotrophic fen located 5 km west of the Hyytiälä Forestry Field Station in southern Finland (61∘49′60′′ N 24∘11′32′′ E; 162 m a.s.l.). The vegetation in the ground layer is dominated by peat mosses (Sphagnum balticum, S. papillosum, S. magellanicum, and S. majus), while the dominating vascular plant species are sedges Carex rostrata, C. limosa, C. lasiocarpa, and Eriophorum vaginatum. The microtopography consists of different plant community types that vary from wet hollows to intermediated sedge-dominated lawns to dry hummocks dominated by dwarf shrubs (Andromeda polifolia, Betula nana, Rubus chamaemorus, and Vaccinium oxycoccos) for 200 to 400 m in all directions, and the fen is surrounded by Scots pine forest. A comprehensive vegetation inventory of the site was conducted during the peak growing season in 2017. The annual mean temperature from 1981–2010 was 4.2 ∘C, and the annual total precipitation was 711 mm (Pirinen et al., 2012). Siikaneva has been a well-established site for eddy covariance measurements of greenhouse gases since 2005 (Aurela et al., 2007; Rinne et al., 2007) and has also been part of the integrated carbon observation system (ICOS) network since 2017 (Rinne et al., 2018). Recently, Junninen et al. (2022) performed a comprehensive observation in Siikaneva 1 to investigate oxidation products of monoterpene, clustering, and NPF.

The EC measurements took place in the growing season from late spring to early summer (from 19 May to 28 June 2021), 2.4 m above the fen. Figure S1 shows a satellite view of the sampling site, overlaid with the average flux footprint calculated using a two-dimensional model (Kljun et al., 2015). Over 90 % of the campaign-average flux footprint is within the wetland area and ∼150 m from the sampling site. Therefore, the flux results have a negligible contribution from the surrounding Scots pine forest. Meteorological parameters such as air temperature, photosynthetically active radiation (PAR), soil moisture, and humidity were obtained from the SmartSMEAR website (https://smear.avaa.csc.fi/; last access: 1 February 2022; Junninen et al., 2009). Air temperature and relative humidity (RH) were measured with an HC2 sensor (Rotronic AG, Switzerland) at a 2 m height in Siikaneva, and PAR was measured by a Li-190SZ quantum sensor (LI-COR, Inc., USA). Leaf area index (LAI) was modeled and calculated for the 90 % contribution of flux footprint area during the measurement campaign separately for shrubs (LAIshrub), other vascular plants (LAIother), and all vascular plants (LAItotal). LAI was first calculated for each plant community type of the site (see Table S1). Leaf count and average leaf size for each vascular plant species were conducted every third week from permanent ICOS study plots. The area-based LAI of each vascular plant species per plot was then calculated by multiplying the leaf count by the average leaf size. Finally, area-based LAI was weighted with their portion in the flux footprint (Supplement Table S1) to obtain LAIshrub, LAIother, and LAItotal of the footprint area. Table S1 also shows the mean moss cover as % in the footprint.

The Vocus-PTR was placed on a wooden platform at the Siikaneva 1 site. We adapted the inlet design from Fischer et al. (2021), who used a similar setup and essentially the same sampling strategy for EC flux measurements. A sonic anemometer (METEK USA-1) for measuring the vertical and horizontal components of the wind vector was mounted 0.4 m above the main inlet to minimize the disturbance from the high inlet flow. A horizontal core sub-sampling (Teflon tubing, 10 mm i.d.) of 5 L min-1 perpendicular to the main inlet was drawn from within the entrance cone of the main flow to the Vocus-PTR, which minimized sample interactions with the main inlet walls and also substantially reduced subsequent wall losses of less volatile compounds such as SQT, DTs, and oxygenated VOCs, as only 100 sccm of the sample flow would enter the instrument reaction region. Figure 1 shows the setup used here, including a schematic of the flows involved in sampling, calibrating, and zeroing.

The details about the Vocus-PTR calibrations, the data pre-processing, and the EC flux calculations with the innFLUX code package (Striednig et al., 2020) are provided in Supplement Sect. S1.1 to S1.5.

BVOC measurements at a high time resolution of 10 Hz were performed using a Vocus-PTR, described in detail in Krechmer et al. (2018). This novel instrument uses PTR as an ionization technique but differs from earlier-generation PTR in many ways. The key differences are described briefly.

A low-pressure (1.5 mbar) focusing ion molecular reactor (FIMR) uses a radiofrequency (RF) field to focus the ions on the central axis. This drastically improves the detection efficiency of product ions. Furthermore, the sensitivity of Vocus-PTR is independent of ambient humidity due to the high water mixing ratio in the FIMR (up to 15 % v/v). Of the 5 L min-1 sampled from the core of the main inlet, only 100 sccm are subsampled (again from the core of the flow) using a capillary inlet, while the rest is directed to exhaust (Fig. 1). This further reduces wall losses and improves the sensitivity of less volatile compounds. The time-of-flight (ToF) mass analyzer had a mass resolving power of > 10 000 m dm-1 throughout the campaign, enabling high-resolution peak identification and typically unambiguous attributions of elemental formulas. The Vocus-PTR also has an inbuilt dynamic dilution setup for automated calibration. We operated the FIMR at a pressure of 1.5 mbar, a temperature of 50 ∘C, and an electric field of 40 V cm-1, resulting in field strength (E/N) of 119 Townsend (10-17 V cm2) (Krechmer et al., 2018).

We compared the average BVOC concentration from Vocus-PTR measurements with offline GC–MS samples collected on 28 June 2021, the last day of the campaign. Ambient air was sampled through three cleaned stainless steel tubes (Markes Intl., UK) filled with adsorbents (Tenax TA and Carbograph 5, 100 mg of each, mesh 60/80, Supelco, Bellefonte, PA, USA) at different locations: close to the wetland surface (∼ 5 cm height), on top of the Vocus-PTR box (2 m height), and near the surface of the wooden platform (∼ 10 cm height). Specifically, using handheld pumps, 160 sccm of flow was directly drawn through the adsorbent tubes for 30 min without pre-treatment, such as an ozone trap. After collection, the tubes were analyzed by GC–MS (GCMS-QP2020, SHIMADZU, Kyoto, Japan) with a split mode (1:20).

Global BVOC emission models, like the model for emissions of gases and aerosols from nature (MEGAN), parametrize the emission rate of a terpene, i (Ei), by the emission factor at standardized conditions (EFi at T=30 ∘C, PAR = 1000 µmol m-2 s-1) and is driven by light and temperature. In a pre-MEGAN version (hereafter referred to as “G93”), the isoprene emission is driven by instantaneous leaf temperature and PAR, and MT emissions are driven only by leaf temperature. The G93 model was developed for individual leaves but can be used for a whole canopy by treating vegetation as one big leaf and so does not consider LAI (Guenther et al., 1993). In the canopy-scale MEGANv2.1 model, the emissions of isoprene, MT, and SQT are driven by instantaneous and past 1 d and 10 d leaf temperature and PAR. However, DTs are not included in the model. MEGANv2.1 takes LAI into account, using a canopy environmental model, and the emissions of MT and SQT have both light-dependent and light-independent fractions. The standard conditions for the MEGANv2.1 are an LAI of 5 and average canopy environmental conditions of the past 24 to 240 h, leaf temperature of 24 ∘C, and PAR of 200 µmol m-2 s-1 for sunlit leaves. We use an Excel-based version of MEGANv2.1 for a single location to model the terpene emissions at our Siikaneva site, hereafter referred to as “MEGANv2.1”. The parameters and EF values for Arctic C3 grass were used to compare our measured emissions to the model predictions. We did not consider the variations in soil moisture and CO2 for these calculations. The details about the MEGANv2.1 framework are given in Guenther et al. (2012).

To calculate the EF from our dataset, we used G93 for isoprene and MT and a simplified version of MEGANv2.1, hereafter referred to as G2012, without considering any canopy environment model and assuming all leaves to be sunlit for isoprene, MT, and SQT. Briefly, the emission rate of a terpene, i (Ei), is characterized by its EFi, LAI, and the light (CL) and temperature (CT) activity factors, as shown in Eq. (1). 1Ei=EFi⋅LAI⋅CL⋅CT The explicit parametrizations for the light and temperature activity factors of G93 and G2012 are given below. 2CL(G93)=1-LDF+LDF⋅0.0027⋅1.066⋅PAR1+0.00272⋅PAR2,3CLG2012=1-LDF+LDF⋅CP⋅α⋅PAR1+α2⋅PAR2,4CTG93=1-LDF⋅expβ⋅T-303+LDF⋅expCT1⋅T-303R⋅303⋅T1+expCT2⋅T-313R⋅303⋅T,5CT(G2012)=1-LDF⋅expβ⋅T-303+LDF⋅EoptCT2⋅expCT1⋅XCT2-CT1⋅1-exp230⋅X, where R is the gas constant, CT2 is 231 kJ mol-1, and LDF is the light-dependent fraction (between 0 and 1). For G93, the LDF is set to 1 for isoprene and 0 for MT and omits SQT. The LDF for G2012 is set to 1 for isoprene, 0.525 for MT, and 0.6 for SQT. Since the LDF is 0 for MT in G93, CT is an exponential function of T, and CL is 1. For isoprene in the G93 algorithm and all compound groups in the G2012 algorithm, the CL is a sigmoidal function of PAR, and CT is an exponential response function of temperature with a maximum at around 35–40 ∘C.

The light-dependent terms of CL and CT in G93 are based on the instantaneous leaf temperature (T in Kelvin) and PAR, whereas the G2012 version of the CL and CT emission activity algorithm includes the instantaneous conditions along with their 1- and 10 d histories as the average values for those periods (P24, P240, T24, and T240). The parameters describing the impact of past temperature and light conditions in the G2012 algorithm are calculated with the following equations: 6α=0.004-0.0005ln⁡P240,7CP=0.0468⋅exp⁡0.0005⋅P24-200⋅P2400.6,8Eopt=Ceo⋅exp0.05⋅T24-297⋅exp⁡0.05⋅T240-297,9x=1Topt-1TR,10Topt=313+0.6⋅T240-297. β, CT1, and Ceo are empirically determined coefficients for each emission chemical class. β is 0.13 for isoprene, 0.1 for MTs, and 0.17 for SQTs. CT1 is 95 kJ mol-1 for isoprene, 80 kJ mol-1 for MTs, and 130 kJ mol-1 for SQTs. Ceo is 2 for isoprene, 1.83 for MTs, and 2.37 for SQTs. Equation (1), including and excluding LAI, was used to determine the EF of different terpenes from the wetland.

We performed the eddy covariance flux measurements in Siikaneva during the growing season, from 19 May to 28 June 2021. The minimum air temperature during the campaign was -1.3 ∘C, measured in the early morning of 25 May 2021, and the maximum air temperature was 31.7 ∘C, observed in the afternoon on 22 June 2021. The average air temperature during the campaign was 15.2 ± 6.5 ∘C. The PAR flux varied from 0–1888 µmol m-2 s-1,, and the campaign average was 463 ± 493 µmol m-2 s-1. The soil moisture was very high and only varied between 0.76–0.87 m3 of water per m3 of soil. The water table level varied from -12 to 1.3 cm. Thunderstorms accompanied occasional rainfall, and the measurements were interrupted due to power failures on three occasions. There was another interruption from 4th June afternoon to 7th June midnight due to a blockage of the Vocus-PTR capillary. Due to these interruptions, we obtained Vocus-PTR data for 60 % of the 41 d of the campaign. We calculated the leaf temperature for the fen using a canopy environment model (Goudriaan and Van Laar, 1994; Leuning, 1997), using grass as the vegetation type with five canopy layers. The minimum leaf temperature during the campaign was -3.1 ∘C, calculated for the early morning of 25 May 2021, and the maximum leaf temperature was 31.9 ∘C, calculated for the afternoon of 22 June 2021. The average leaf temperature during the campaign was 15.3 ± 6.8 ∘C. The estimated leaf temperature is typically within < 1 ∘C of the air temperature throughout the campaign, and hereafter, we refer to both temperatures as the same. The meteorological parameters measured in the SMEAR II Siikaneva 1 station for PAR, relative humidity, air temperature, the estimated leaf temperature, and the concentrations of the four measured terpene classes (isoprene, MTs, SQTs, DTs) are given in Fig. S5 in the Supplement.

Isoprene showed the highest ambient mixing ratio among all the terpenes. We observed a relatively high ambient isoprene mixing ratio with a mean of 0.62 ± 0.8 ppbv during the campaign. A previous study at the same site during the same season reported a mean of 0.23 ± 0.009 ppbv (Haapanala et al., 2006). We observed a clear diurnal pattern in the ambient isoprene mixing ratio, reaching as high as 4.6 ppbv at 11:30 (UTC +2) on 22 June 2021. We observed a substantial accumulation of MTs, SQTs, and DTs during the night at Siikaneva, with MT concentrations reaching above 12 ppbv in the early morning on 11 June 2021. The highest SQT and DT mixing ratios were observed at 22:30 on 25 June 2021, at 1.22 ppbv and 23 pptv (parts per trillion by volume). We assume that these nighttime peak mixing ratios were due to the formation of a very shallow boundary layer in atmospherically stable conditions over the wetland at nighttime. A recent study also observed this phenomenon at the same site, and they have connected it to stronger NPF in the wetlands than in the surrounding boreal forests (Junninen et al., 2022). MT, SQT, and DT mixing ratios generally dropped during the daytime, presumably due to an increasing boundary layer height, ensuing dilution, and higher oxidant levels. The campaign-average mixing ratios were 1.1 ppbv for MT, 100 pptv for SQT, and 3 pptv for DTs. DTs have very low volatility and were thought not to be emitted by terrestrial vegetation for a long time. However, Li et al. (2020) reported average mixing ratios of around 2 pptv in the French Landes forest during summertime, and Fischer et al. (2021) measured 0.1 pptv of DT above a boreal forest during spring.

Figure S6 compares the terpene mixing ratios measured by the GC–MS at different locations and the terpene mixing ratios measured using the Vocus-PTR at 12:00 on 28 June 2021. We observed a steep concentration gradient in isoprene concentration with vertical height, indicating very high nearby emissions. Near the wooden platform and close to the wetland, the isoprene mixing ratio was about 4–5 ppbv (parts per billion by volume) and was reduced to half on top of the Vocus (2 m above the wetland). The mixing ratios for isoprene obtained from the adsorbent tube on top of the Vocus (40 cm below the high flow inlet) were comparable to those of the Vocus-PTR (1.7 vs. 1.4 ppbv).

We detected α-pinene, 3-carene, β-pinene, camphene, and limonene in the GC–MS analysis. Table S2 shows the different monoterpenes and their amount in the three adsorbent tubes (in ng). α-pinene was the most abundant monoterpene, followed by 3-carene. We compared the sum of speciated MTs measured by GC–MS with the sum of all MTs measured by Vocus-PTR (from the parent ion signal (C10H17+) using the α-pinene sensitivity). The GC–MS obtained a mixing ratio of 300 pptv, whereas Vocus measured 580 pptv for MTs. The GC–MS MTs mixing ratio may be underestimated due to not detecting monoterpenes other than in Table S2 and the oxidation of sampled MT, e.g., by O3 (Helin et al., 2020). In addition, the Vocus-PTR might overestimate the mixing ratios due to the possibility of other terpenes fragmenting to C10H17+. The GC–MS could not reliably detect any SQT, although we detected 100 pptv of sesquiterpenes in the VOCUS-PTR. Since the mixing ratios of isoprene near the inlet agreed between both methods within the different analytical capabilities of the two instruments, we believe the difference in mixing ratios of the larger terpenes could be due to the unintended loss of these compounds with low volatility and high reactivity.

As part of the EC flux analysis with the innFLUX code (Striednig et al. 2020), the data quality was investigated and grouped into nine quality classes (see Sect. S1.3). For the following flux analysis, we have only used data that fulfills the criteria for quality classes 1–3 (Table S3) and with a minimum friction velocity of 0.1 m s-1. Most of these data points were measured during the daytime. About 70 % of the available daytime data passed the filtering criteria. We had 496, 425, 483, and 449 data points of 30 min flux time intervals for isoprene, MT, SQT, and DTs that passed the quality control.

We observed vertical fluxes for more than 250 out of 1072 peaks in our mass spectrometric dataset. Besides terpenes, we also detected oxygenated VOCs, including smaller ones such as methyl ethyl ketone (C4H8O) and methyl vinyl ketone (C4H6O), and compositions consistent with the first oxidation products of sesquiterpenes (e.g., C15H24O3, C15H24O4). Earlier BVOC emission studies at Siikaneva were carried out by gas chromatography using chamber and relaxed eddy accumulation (REA) techniques. They reported 9 and 29 compounds each (Haapanala et al., 2006; Hellén et al., 2006), mostly C2- to C10-hydrocarbons. Many of these smaller organics have lower proton affinities than water and could not be detected using Vocus-PTR. Other studies using PTR-based measurement techniques at other locations have similarly detected vertical fluxes for 200 or more compounds (Fischer et al., 2021; Millet et al., 2018; Park et al., 2013). In the following, we focus our analysis on the emissions of the terpene compound groups.

Several studies have shown that isoprene is the main BVOC emitted by boreal wetlands (Klinger et al., 1994; Janson and De Serves, 1998). In this study, the highest emission fluxes were observed for isoprene: more than a factor of six times higher than for MTs (Fig. 2b), on average. Overall, we observed a similar emission pattern for all terpenes. We also report the first-ever emission of DTs from wetlands (Fig. 2d). We observed relatively low emissions for all the terpenes from the start of the campaign till the end of May. During that period, the average temperature was below 10 ∘C, and the LAI (∼ 0.2) was very low. We observed downward flux for MTs during nighttime from 1 to 12 June. However, most nighttime data did not pass the EC quality controls and could thus not be evaluated. DT fluxes were erratic at the beginning of the campaign until the end of May since the measurements were close to the limit of detection (LoD).

Overall, the terpene emissions increased substantially over the course of the campaign due to the increase in temperature and the progression of the growing season of the vegetation in the fen. LAI varied from 0.2 to 0.54, increasing continuously throughout the campaign from the beginning to the peak of the growing season (Fig. 2). The maximum flux for isoprene of 19.5 nmol m-2 s-1 was observed at noon (12:00–12:30) on 22 June 2021, and the maximum flux for MT of 2.2 nmol m-2 s-1 in the afternoon (14:30–15:00) of the same day, which was the day with the highest temperature during the campaign (31.7 ∘C at 15:00). During these maximum flux time intervals, the temperature was 30 and 31.5 ∘C, and PAR was 1533 and 1409 µmol m-2 s-1, respectively. The maximum emission for SQTs of 1.25 nmol m-2 s-1 was observed already on 21 June (10:00–10:30) at 27.9 ∘C and 1470 µmol m-2 s-1 of PAR, and the maximum emission for DTs of 0.018 nmol m-2 s-1 was observed on 22 June (10:30–11:00) at 29.5 ∘C and 1435 µmol m-2 s-1 of PAR. Figure S7 shows the diurnal patterns of the terpene fluxes with the campaign-wide median values and the interquartile ranges (IQR). The average fluxes observed during the day (PAR > 50 µmol m-2 s-1) were 1.96 nmol m-2 s-1 for isoprene, 0.33 nmol m-2 s-1 for MTs, 0.2 nmol m-2 s-1 for SQTs, and 0.003 nmol m-2 s-1 for DTs. The maximum nighttime deposition flux observed for MTs was -0.11 nmol m-2 s-1. All other terpenes had one order of magnitude lower deposition fluxes.

Previous studies at the same site have measured isoprene along with other small hydrocarbons, namely C2–C5 alkanes and halocarbons (Haapanala et al., 2006; Hellén et al., 2006). The previous REA study at the site reported a maximum of 3 nmol m-2 s-1 for isoprene on 4 August 2004 at 25 ∘C and 1500 µmol m-2 s-1 PAR (Fig. 3 in Haapanala et al., 2006). The other previous study used enclosure measurements and reported a much lower isoprene emission flux of 0.4 nmol m-2 s-1 on 18 August 2005 at 25 ∘C temperature and 1135 µmol m-2 s-1 PAR. That enclosure study suggested that Sphagnum moss could be a source of the isoprene emitted by the fen (Hellén et al., 2006). Both studies have not reported emission fluxes of any terpenes other than isoprene from this site.

In our study, SQTs emitted from the boreal fen were almost of the same magnitude as the emitted MTs. In contrast, a recent study in a subarctic fen in northern Finland (Hellén et al., 2020) reported that SQT emissions were significantly higher than MT emissions. Their highest SQT emission was 0.012 nmol m-2 s-1, and MTs had a flux of 0.004 nmol m-2 s-1 at 20 ∘C temperature and 560 µmol m-2 s-1 PAR, which is much lower than our observations.

Only a few ecosystem-scale EC/disjunct EC studies have been conducted in similar high latitude environments, three in Scandinavia (Seco et al., 2022, 2020; Holst et al., 2010) and one in an Alaskan tundra ecosystem (Potosnak et al., 2013). Seco et al. (2022) reports only isoprene emission from Abisko, Sweden (68∘20′ N, 19∘03′ E), and Finse, Norway. They report comparable emissions from both sites of 2–6 nmol m-2 s-1, with peaking values of up to 11 nmol m-2 s-1. Another recent study by Seco et al. (2020) at a subarctic mire in Abisko reports maximum isoprene emission of 11.3 nmol m-2 s-1 at 21.2 ∘C air temperature and 1339 µmol m-2 s-1 of PAR on 16 July 2018, which is about the same magnitude we observed in our study. They also report a maximum MT emission flux of 0.2 nmol m-2 s-1 at 19.6 ∘C air temperature and 1145 µmol m-2 s-1 of PAR, which is more than an order of magnitude less than the highest monoterpene emission observed in our study. During their study, they had leaf temperature measurements and had much higher leaf temperatures than air temperatures. The setup we used for our canopy environment model was able to predict their leaf temperature from their air temperatures. We believe the higher leaf temperatures for their campaign could be due to high PAR, which heats the leaf surface, but the air temperature remains lower due to large-scale meteorological reasons. Seco et al. (2020) could not detect any ecosystem-scale SQT or DT emissions due to the limitations of their measurement setup. Another disjunct EC study (Holst et al., 2010) at the same Abisko mire site as Seco et al. (2020) reports a maximum isoprene emission flux of 6.4 nmol m-2 s-1 at 22 ∘C air temperature and 1200 µmol m-2 s-1 of PAR on 3 August 2006. The Alaskan study was conducted in a moist tundra ecosystem (68∘38′ N, 149∘38′ W) (Potosnak et al., 2013) and reports a maximum hourly isoprene emission flux of 5.5 nmol m-2 s-1, observed at an air temperature of 22 ∘C and a PAR level over 1500 µmol m-2 s-1. These isoprene emissions are 3–4 times lower than the maximum isoprene emission flux observed in our study but have at least 10 ∘C lower temperatures than observed in our campaign. We observed isoprene emissions of 2–10 nmol m-2 s-1 when temperatures and PAR were comparable. Moreover, even though these studies were conducted in similar high latitude environments with low species diversity, they have different plant species mixes, and we might expect different emissions. The Swedish mire has vegetation similar to our site in its wet areas, whereas the vegetation in the drier area is different. In the Alaskan study, the vegetation is dominated by different species such as Salix spp. This also points to the need for more than one flux study in each ecosystem.

Several chamber studies report isoprene fluxes from boreal fens, starting with Janson and De Serves (1998) in central Sweden (Stormossen, 60∘10′ N, 17∘05′ E). Using a static chamber technique, they report isoprene fluxes up to 9.3 nmol m-2 s-1 during August. A study by Bäckstrand et al. (2008) reported an average flux of non-speciated total non-methane volatile organic compound (NMVOC) flux of 140 mg C m-2 d-1, which would convert to 27 nmol m-2 s-1 assuming all the NMVOC was isoprene. This is about the same magnitude as our observed maximum fluxes.

No other study has ever reported ecosystem-scale DT emissions, except for Fischer et al. (2021). They used a similar setup as ours in Hyytiälä, above a boreal forest canopy close to our site. They found an average emission of 0.15 pmol m-2 s-1, which is more than an order of magnitude lower than the DT emissions we measured here. Measuring accurate emissions of the larger terpenes, such as SQT and DTs, is very important since very little of their emission potential is known. Therefore, their impact on CCN growth is uncertain. A recent study has found boreal wetlands to initiate stronger NPF than boreal forests (Junninen et al., 2022). We speculate that the unexpectedly high emissions of SQT and DTs could contribute to the strong NPF observed there.