Leaf BVOC emissions fluxes, above-canopy monoterpene concentration, and aerosol particle size and number concentration were measured during the growing season 2009 at the research station Spasskaya Pad, located ca. 40 km to the northeast of Yakutsk (62∘15′18.4′′ N, 129∘37′07.9′′ E) and centred in the Eastern Siberian larch biome (Kobak et al., 1996; Tchebakova et al., 2006). In the northern direction, no major pollution sources exist within hundreds of kilometres, the nearest mining areas are concentrated to the south and west of Yakutsk. The predominant air flow to the site is either from southern (via Yakutsk) or northern locations. Forest fires contribute to aerosol load in summer.

An eddy covariance tower for measurements of forest–atmosphere exchange of CO2, water vapour, and sensible heat was established at Spasskaya Pad in the late 1990s (Ohta et al., 2001; Dolman et al., 2004) in a L. cajanderii forest growing on permafrost soil with an understory vegetation consisting of ericaceous shrubs. The forest has an average age of ca. 185 years and canopy height is little less than 20 m. Maximum one-sided larch leaf area index (LAI) in summer is around 2 (Ohta et al., 2001; Takeshi et al., 2008). In 2009, leaf samples for BVOC analyses were taken, accessing the upper part of the canopy from a scaffolding tower located within a few hundred metres of the eddy flux tower (Kajos et al., 2013). Using a custom-made Teflon branch chamber, air filtered of O3 was sampled onto Tenax-TA/Carbopack-B cartridges with a flow rate of 220 mL min-1. A total of 5–12 samples were taken during the day from two trees on south-facing branches approximately 2 m below the tree top. The cartridge samples were stored at 5∘ C during the campaigns, transported afterwards to Helsinki and thermally desorbed and analysed using a thermal desorption instrument (Perkin-Elmer TurboMatrix 650, Waltham, USA) attached to a gas chromatograph (Perkin-Elmer Clarus 600, Waltham, USA). For details on chamber, adsorbents, and laboratory measurements, see Haapanala et al. (2009), Ruuskanen et al. (2007), and Hakola et al. (2006).

Monoterpene concentrations and forest–atmosphere exchange fluxes were measured with a high-sensitivity quadrupole proton-transfer-reaction mass spectrometry (PTR-MS; Ionicon, Innsbruck, Austria) located in a hut at the foot of the eddy covariance tower. Sample air was drawn through a heated PFA tube using a 20 L min-1 flow from the inlet located at 30.3 m above ground. While reporting here on monoterpenes only, a range of masses, corresponding to BVOCs (e.g. isoprene, methanol, acetaldehyde), were sampled sequentially, with typical dwell times of 0.5 s and scanning sequences of around 4 s. Measurement set-up, disjunct eddy covariance flux calculations, and quality control followed Holst et al. (2010). It was not possible to import a gas calibration standard to Spasskaya Pad due to security and customs restrictions, and thus the PTR-MS could not be calibrated on-site. However, the instrument had been calibrated before and after the field campaign using a gas standard mixture from Ionimed (Innsbruck, Austria) using the same detector and instrument settings as during the field campaigns.

Aerosol particles were continuously monitored with a scanning mobility particle sizer (SMPS) located at the foot of the eddy covariance tower, connected to a differential mobility analyser (DMA, medium Hauke type, custom built; for size segregation of aerosol particles) in front of a condensation particle counter (CPC, model 3010, TSI Inc., USA; for determining the number of the size segregated particles). The system was identical to the one described and evaluated in Svenningsson et al. (2008). Scans across the size range of 6–600 nm were completed every 5 min. The SMPS data were used to determine occasions of aerosol particle nucleation. The growth rates (GRs) were calculated from log-normal modes fitted to the measured particle size distribution following Hussein et al. (2005). The time evolution of the diameters at which the fitted modes peaked was inspected visually, and the GR was determined with linear least squares fitting to these peak diameters whenever a continuous increase in diameter was observed. In this analysis we calculated GRs for particles from 25 to 160 nm.

The source rate for condensing vapour (Q, molecules cm3 s-1) was determined by calculating the concentration of condensable vapour needed to produce the observed GR (CGR, cm-3; Nieminen et al., 2010, 2014) and the condensation sink from the particle size distribution (CS, s-1; Kulmala et al., 2001). In steady state the sources and sinks for the condensing vapour are equal, and thus we determined the source rate as Q=CGR⋅CS.

We applied the dynamic global vegetation model (DGVM) LPJ-GUESS (Smith et al., 2001; Sitch et al., 2003), including algorithms to compute canopy BVOC emission following Niinemets et al. (1999), Arneth et al. (2007b), and Schurgers et al. (2009a) and permafrost as adopted from Wania et al. (2009) (Table 1). LPJ-GUESS simulates global and regional dynamics and composition of vegetation in response to changes in climate and atmospheric CO2 concentration. Physiological processes like photosynthesis, autotrophic, and heterotrophic respiration are calculated explicitly; a set of carbon allocation rules determines plant growth. Plant establishment, growth, mortality, and decomposition, as well as their response to resource availability (light, water), modulate seasonal and successional population dynamics arising from a carbon allocation trade-off (Smith et al., 2001). Fire disturbance is included in the model (Thonicke et al., 2001). Similar to other DGVMs, a number of plant functional types (PFTs) are specified to represent the larger global vegetation units (Sitch et al., 2003).

BVOC emissions models, whether these are linked to DGVMs or to a prescribed vegetation map, all rely on using emission potentials (E*, leaf emissions at standardized environmental conditions) or some derivatives in their algorithms. In LPJ-GUESS, production and emissions of leaf and canopy isoprene and monoterpenes are linked to their photosynthetic production, specifically the electron transport rate, and the requirements for energy and redox equivalents to produce a unit of isoprene from triose phosphates (Niinemets et al., 1999; Arneth et al., 2007b; Schurgers et al., 2009a). A specified fraction of absorbed electrons used for isoprene (monoterpene) production (ε) provides the link to PFT-specific E* (Arneth et al., 2007a); in case of monoterpenes emitted from storage, an additional correction is applied to account for their light-dependent production (taking place over parts of the day) and temperature-driven emissions (taking place the entire day) (Schurgers et al., 2009a). Half of the produced monoterpenes were stored, whereas the other half was emitted directly (Schurgers et al., 2009a). Values for E* were similar to the global parameterization for most of the model's PFTs (Schurgers et al., 2009a), with the exception of boreal needle-leaf summergreen (BNS) “larch” PFT (see results).

Contrasting the stimulation of BVOC emissions in a warmer and more productive environment, higher CO2 concentrations have been shown to inhibit leaf isoprene production. Even though the underlying metabolic mechanism is not yet fully understood, this effect has been observed in a number of studies (for an overview see Fig. 6 in Arneth et al., 2011). Due to limited experimental evidence, whether or not a similar response occurs in monoterpene producing species cannot yet be confirmed, especially in species that emit from storage. The model is set up to test this hypothesis. Arneth et al. (2007a) proposed an empirical function for CO2 inhibition, based on the ratio of leaf internal CO2 concentration at a standard atmospheric CO2 level (taken as 370 ppm) and at the given atmospheric CO2 levels of the simulation year (both calculated for non-water-stressed conditions); the relationships have been shown since to fit an updated compilation of observations well (Arneth et al., 2011). The algorithm that describes the CO2 inhibition of BVOC emission can either be enabled or disabled in the model (Arneth et al., 2007a) and simulations results thus compared (see Fig. A1, Appendix; Table 2).

LPJ-GUESS was recently expanded with a permafrost module following Wania et al. (2009; Miller and Smith, 2012) in which a numerical solution of the heat diffusion equation was introduced. The soil column in LPJ-GUESS now consists of a snow layer of variable thickness, a litter layer of fixed thickness (5 cm), and a soil column of depth 2 m (with sublayers of thickness 0.1 m) from which plants can extract non-frozen water above the wilting point. A “padding” column of depth 48 m (with thicker sublayers) is also present beneath these three layers to aid in the accurate simulation of temperatures in the overlying compartments (Wania et al., 2009). Soil temperatures throughout the soil column are calculated daily, in addition to change in response to changing surface air temperature and precipitation input and the insulating effects of the snow layer and phase changes in the soil's water.

Here we run the model with 0.5∘ spatial resolution, using climate and atmospheric CO2 as driving variables as described in the literature (Smith et al., 2001). Values describing growth and survival of the BNS (larch) PFT were adopted from previous studies (Sitch et al., 2003; Hickler et al., 2012; Miller and Smith, 2012), but with the degree-day cumulative temperature requirements on a 5∘ basis (GDD5) to attain full leaf cover reduced from 200 to 100 (Moser et al., 2012). Minimum GDD5 to allow establishment was set to 350 resulting in establishment of seedlings in very cold locations. Soil thermal conductivity was 2 W m-1 K-1. The model was spun up for 500 years to 1900 values using CO2 concentration from the year 1900 and repeating de-trended climate from 1901 to 1930 from CRU (Mitchell and Jones, 2005). Historical (20th century) simulations used observed CO2 concentrations and based on variable CRU climate. Simulations for the 21st century were based on ECHAM climate, using RCP 8.5 emissions (Riahi et al., 2007). The model requires daily radiation, precipitation, and maximum and minimum air temperatures as input (Arneth et al., 2007b). Climate output from ECHAM was interpolated to the CRU half-degree grid, and monthly values interpolated to daily ones (see Ahlström et al., 2012, and references therein). These daily fields were then bias corrected using the years 1961–1990 as reference period, as in Ahlström et al. (2012). CO2 inhibition of BVOC emissions were switched on and off in separate simulations to assess the sensitivity of our results to this process. Totals across Siberia were calculated for a grid box that ranged from 46 to 71∘ N and 76 to 164∘ E. Simulated changes in total carbon uptake or losses were translated into radiative forcing following (IPCC, 2007), assuming a 50 % uptake in oceans in case of a net loss to the atmosphere (Sitch et al., 2007).

To model the effect of BVOCs on CCN concentrations, we use the global aerosol–climate model ECHAM5.5-HAM2 (Zhang et al., 2012). ECHAM5.5-HAM2 includes the aerosol components of black carbon, organic carbon, dust, sea salt, and sulfate (Table 1) and describes the aerosol size distribution with seven log-normal modes. The microphysics module M7 (Vignati et al., 2004) includes nucleation, coagulation, and condensation. In this study, we use the ECHAM5.5-HAM2 version with activation type as described in Makkonen et al. (2012a, b). For simulating secondary organic aerosol, we use the recently developed SOA module (Jokinen et al., 2015). The SOA module explicitly accounts for gas-phase formation of extremely low-volatility organic compounds (ELVOCs) from monoterpene oxidation. The module implements a hybrid mechanism for SOA formation: ELVOCs are assumed to condense to the aerosol population according to the Fuchs-corrected condensation sink, while semi-volatile organic compounds (SVOCs) are partitioned according to organic aerosol mass. While simulated ELVOCs are able to partition more effectively to nucleation and Aitken mode, hence providing growth for nucleated particles to CCN size, SVOCs primarily add organic mass to accumulation and coarse aerosol modes. A total SOA yield of 15 % from monoterpenes is assumed (Dentener et al., 2006). While similar assumption on total SOA yield is applied by most aerosol–climate models, the simulated SOA is likely to be underestimated (e.g. Tsigaridis et al., 2014).

ECHAM5.5-HAM2 was run with different BVOC emission scenarios that were simulated for the years 2000 and 2100 off line with LPJ-GUES (see previous section). The model uses T63 spectral resolution with 31 vertical hybrid sigma levels. The simulations apply present-day oxidant fields as in Stier et al. (2005). All simulations are initiated with a 6-month spin-up, followed by 5 years of simulation for analysis. The model climate is nudged towards ERA-40 reanalysis year 2000 meteorology, an approach that is widely used in aerosol–climate assessments (K. Zhang et al., 2014). Present-day wildfire and anthropogenic aerosol and precursor emissions are applied for all simulations (Dentener et al., 2006). One of the foci here is a BVOC, comparing present-day and future BVOC emissions (choosing a conservative estimate of E*=1.9 µg C m-2(leaf) h-1; see results section for further details on E*) but keeping other emissions constant (oxidant fields and nudging meteorology are same for both 2000 and 2100). The emissions of dust and sea salt are modelled interactively (Zhang et al., 2012).

The analysis of model results includes total particle number concentrations and CCN at 1 % supersaturation (CCN(1 %)), since it reflects the changes in Aitken mode concentrations and local changes in precursor emissions. While “realistic” supersaturations are generally lower, choosing CCN(1.0 %) concentration provides the upper limit for CCN concentrations. The simulations are also used to assess the radiative effects of SOA. In the simulations, the aerosol concentrations are interactively coupled to the cloud-microphysics scheme (Lohmann et al., 2007) and to the direct aerosol radiative calculation. The aerosol indirect effect is evaluated as a change in cloud radiative forcing (ΔCRF). The direct aerosol effect accounts only for clear-sky short-wave forcing (ΔCSDRF). The radiative effects are calculated as differences from two time-averaged 5-year simulations as ΔCRF=CRF(BVOC2100)-CRF(BVOC2000)ΔCSDRF=CSDRF(BVOC2100)-CSDRF(BVOC2000). Subscripts “2000” and “2100” denote that BVOC emissions from this year were used, while other model conditions were based on present-day values.

LPJ-GUESS reproduces the present-day circumpolar permafrost distribution (Fig. 1; shown as circumpolar map for comparison with Tarnocai et al., 2009) and, with the exception of the Kamchatka peninsula, simulates also the expanse of the larch-dominated forests in Eastern Siberia (Fig. 1; Miller and Smith, 2012; Wagner, 1997). Maximum LAI calculated by the model for the Spasskaya Pad forest (62∘15′18.4′′ N, 129∘37′07.9′′ E; 220 m a.s.l.), where the BVOC measurements were obtained, was 2.0 (averaged over years 1981–2000; not shown) and is in good agreement with the measured values during that period (1.6; Takeshi et al., 2008). Total present-day modelled soil C pools over the top 2 m in Eastern Siberia are 216 Gt C and, for circumpolar soils summed for latitudes above 40∘ N, 454 Gt C (Table 2). Recent estimates of C stored in northern latitude soils affected by permafrost were 191, 495, and 1024 Gt C in the 0–30, 0–100, and 0–300 cm soil layer respectively, based on extrapolating observations stored in the Northern Circumpolar Soil Carbon Database (Tarnocai et al., 2009). These numbers indicate that the values calculated with LPJ-GUESS are lower than observation-based ones, most likely underestimating C density in particular in the soil layers below a few tenths of centimetres.

Kajos et al. (2013) measured for the first time MT E* from L. cajanderii. Their measurements, taken over an entire growing season at Spasskaya Pad, suggested values of E* ranging from 1.9 µg C m-2(leaf) h-1 at the lower end to 9.6 µg C m-2(leaf) h-1 at the upper. Applying a weighted measured-average E* of 6.2 µg C m-2(leaf) h-1 in LPJ-GUESS led to average summer daily monoterpene emissions of 2.9 (± 0.8, 1 standard deviation, June) mg C m-2 d-1, and 2.2 (± 0.8, July) mg C m-2 d-1 for the grid location representing the Spasskaya Pad site, for the same year of measurements as reported in Kajos et al. (2013). These values are within 25 and 10 % of measured values (3.3 ± 2.9 mg C m-2 d-1, June; 2.4 ± 1.6 mg C m-2 d-1, July), even though the modelled day-to-day variation was smaller, which is expected when applying grid-averaged climate as model input. By comparison, for a boreal Scots pine forest in southern Finland the average June and July monoterpene emissions were somewhat larger than the values for larch (3.8 and 5.1 mg C m-2 d-1 respectively; Rantala et al., 2015). For a Larix kaempferi-dominated temperate forest in Japan, Mochizuki et al. (2014) extrapolated, based on their measurements, summertime maxima of ca. 10 mg C m-2 d-1. Across the Siberian larch biome, applying E* of 6.2 µg C m-2 increased simulated total present-day MT emissions from 0.11 TgCa-1 (as in Schurgers et al., 2009a) to 0.21 TgCa-1, or to 0.42 TgCa-1 when the maximum E* was used (Table 2). The observed range in E*, and the calculated range in total emissions across Siberia, might reflect variability in tree microclimate or genetic variability or might have been induced by (undetected) mechanic or biotic stress during the time of measurements (Staudt et al., 2001; Bäck et al., 2012; Kajos et al., 2013). While our data are insufficient to make a finite suggestion of L. cajanderi E*, the measurements provide evidence for potentially substantially higher MT emissions from Siberian larch than previous estimates.

New particle formation events (Fig. 2a) were observed regularly at Spasskaya Pad. The calculated volumetric source rates of condensing vapours (Q), the product of vapour concentration required for the observed particle growth rate and particle loss rate (Kulmala et al., 2005), increased exponentially with temperature (Fig. 2b). MT concentrations increased with temperature as well, with a slope relatively similar to that found for the Q vs. T relationship (Fig. 2c). Consequently, a positive relationship emerged between Q and MT concentration (Fig. 2d), which supports previous field and laboratory evidence that MTs and their oxidation products are a main precursor to the observed particle formation and growth.

Figure 2d shows the connection between the BVOC concentration and the formation rate of vapours causing the growth of the aerosol particles. Even though the monoterpene concentrations were measured above and the aerosol growth rates below the canopy, the observed correlation indicates that BVOC concentration is an important contributor to the regional aerosol growth and supports the theory that the condensation of organic vapour is largely responsible for the formation of secondary organic aerosol (Hallquist et al., 2009; Carslaw et al., 2010). Substantial within-canopy chemical reactions would be expected to worsen the relationship. The correlation depicted in Fig. 2d is determined in particular by the formation of secondary organic aerosol on pre-existing aerosol particles, whereas the nucleation rate of new aerosol particles seems not to be dominated by the landscape-scale emissions and surface concentrations of BVOCs. For instance, most nucleation events in a Scots pine dominated landscape in Finland have been found in spring, when measured monoterpene concentrations in the near-surface were about one-tenth of the summertime maximum (ca. 60 ppt, vs. up to 500 ppt; Haapanala et al., 2007; Lappalainen et al., 2009). In our study, we found MT concentrations of similar magnitude to these.

In contrast to temperature and BVOC concentrations, levels of radiation, which can be considered a surrogate for the concentration of the OH radical (OH⚫), did not affect Q (Fig. 2b), even though OH⚫ has been considered an important player in aerosol formation. Rohrer and Berresheim (2006) showed a strong correlation between solar ultraviolet radiation and OH⚫ concentration at the Hohenpeissenberg site in Germany. Furthermore, Hens et al. (2014) demonstrated that the daytime OH⚫ concentrations in (especially) boreal forest depend on solar radiation, both above and below the canopy. Hence, the poor relation between the source rate of condensing vapour and orders-of-magnitude variation in levels of radiation (Fig. 2b) indicates that OH-radical concentration did not have a major impact on Q. This agrees with the findings by Ehn et al. (2014) that ozone instead of OH⚫ is an important, if not the main, atmospheric agent oxidizing organic vapours into a chemical form that condenses on particle surfaces. Since the relative variation in ozone concentrations is much smaller than in BVOC (or OH) concentrations (Hens et al., 2014), the similarity in the dependencies of Q and MT concentration and temperature (Fig. 2b and c) is in favour of a more significant role of ozone than of OH in the formation of condensable vapours. In general, our results indicate that factors and processes besides the concentrations of SO2 and OH⚫ seem to limit aerosol production in unpolluted environments (Kulmala et al., 2005).

In a warmer environment with higher atmospheric CO2 levels, the simulations indicated a drastically reduced area of permafrost in Siberia (Fig. 1). Total net primary productivity in the simulated domain increased from an annual average of 3.5 to 5.9 PgC a-1 at the end of the 21st century. An overall C loss of 100 PgC assumed to be in the form of CO2 (since the model does not yet include a dynamic surface hydrology which would be necessary to assess changing methane emissions) at the end of the 21st century, compared to present-day conditions, was calculated from the shrinking Siberian areas of permafrost (Table 2). However, warming and higher levels of atmospheric CO2 led also to increasing LAI and to larch-dominated areas showing the expected north and north-eastwards shift (Fig. 1) compared to present-day climate (Miller and Smith, 2012). The carbon uptake in expanding vegetation into permafrost-free areas, combined with enhanced productivity across the simulation domain overcompensates for the losses from C pools in permafrost areas (Table 2).

Without CO2 inhibition of BVOC emissions future MT emissions were, as expected, notably enhanced: directly as a result of warmer leaves, but augmented by the future higher LAI of larch and evergreen conifers (Fig. 1d; Table 2, Fig. A1). Since the emissions scale with the emission factors applied, the proportional increase between present-day and future climate conditions is independent of the value of E*. Whether or not leaf MT emissions are inhibited by increasing atmospheric CO2 levels to a similar degree to what was found for isoprene is difficult to assess from today's limited number of studies (e.g. Niinemets et al., 2010, and references therein). We included both simulation results in Table 2 since similarities in the leaf metabolic pathways of isoprene and MT production suggest such an inhibition, but possibly this effect does not become apparent in plant species where produced MTs are stored unless the storage pools become measurably depleted by the reduced production. By contrast, species emitting MTs in an “isoprene-like” fashion immediately after production should more directly reflect CO2 inhibition. Evergreen conifers typically emit most MTs from storage pools, although recent experiments have shown that some light-dependent emissions also contribute to total emission fluxes. Accordingly, based on the leaf-level measurements, larch could follow a hybrid pattern between emission after production and from storage (Kajos et al., 2013). Without accounting for CO2 inhibition, MT emissions across the model domain more than doubled (Fig. 1; Table 2) by 2100, as a consequence of higher emissions per leaf area due to warmer temperatures and of the larger emitting leaf area in response to higher photosynthesis. With CO2 inhibition included, simulated changes were negligible, similar to what was shown in previous BVOC simulations with the model (Arneth et al., 2007a, 2008).