Field measurements were made at the SMEAR II site (Station for Measuring Forest Ecosystem–Atmosphere Relations) at Hyytiälä Forestry Field Station of the University of Helsinki (61.845∘ N, 24.288∘ E; 181 ma.s.l.). The station features a largely homogeneous stand of Scots pine (Pinus sylvestris) planted in 1962 . The forest floor is covered by mosses (Dicranum polysetum, Hylocomium splendens, and Pleurozium schreberi) and understory herbs including bilberry (Vaccinium myrtillus) and lingonberry (Vaccinium vitis-idaea) . The climate is boreal, with 30-year-average January and July temperatures of -7.2 and 16.0 ∘C, respectively . The average annual precipitation is 711 mm, with summer and fall receiving somewhat more than winter and spring . Meteorological and ancillary data such as surface pressure, air temperature, relative humidity, radiation, precipitation, and soil temperature and moisture are continuously monitored at the SMEAR II site (see , for description of the site infrastructure). These data are available online at http://avaa.tdata.fi/web/smart/smear/.

Soils at the site are podzols of depths varying from 0.5 to 1.6 m, developed from glacial deposits. The O horizon is a porous mor-humus layer laden with fine roots and mycorrhizae, distinct from the mineral soil underneath. The thickness of the O horizon varies from 1 to 5 cm , the bulk density is 0.10 gcm-3, and the porosity is 0.67 m3m-3 . The O horizon is highly acidic (pH = 2.9 to 3.6) and rich in carbon content (31–45 wt%). The mineral soil underneath is of sandy loam texture but also has a high fraction of gravels and stones . The A horizon is 4–8 cm thick, and it has a porosity of 0.61 m3m-3 and a carbon content of 3–6 wt%. Beneath the A horizon, porosity and carbon content decrease with depth. The mineral soil is less acidic than the humus layer, with pH around 4 to 5.

A quantum cascade laser spectrometer (QCLS, Aerodyne Research Inc., Billerica, MA, USA) was used to measure concentrations of COS, CO, CO2, and H2O at 1 Hz. The instrument had overall uncertainty (1 SD) of 7.5 ppt (parts per trillion) for COS, 3.3 ppb for CO, and 0.23 ppm for CO2 . An oil-free dry scroll pump (Varian TriScroll) was connected to the QCLS to pull the sampling air through the analyzer. The QCLS was housed inside a small cabin that was not air-conditioned. An automatic background correction was performed every 6 h with an ultrahigh-purity (> 99.999 %) nitrogen cylinder to remove the curvature effect in the baseline spectra. An air purifier (Gatekeeper CE-500K-I-4R) was used to scrub trace amounts of CO from the cylinder air. The instrument was calibrated against three working standards that had been calibrated to the NOAA or WMO scale in the laboratory .

Soil fluxes were measured in two automated soil chambers (LI-8100A-104, LI-COR Biosciences, Lincoln, NE, USA) modified to avoid COS emission artifacts from chamber materials and operated in a flow-through configuration . These modifications included replacing the chamber bowl and soil collar with stainless steel components, and removing or replacing other small COS-producing parts. Dark chambers were selected to prevent photochemical production of COS e.g., or CO e.g., at the soil surface during chamber measurements. The two chambers were placed in similar environments, about 10 m apart. The moss layer or any other vegetation was removed to expose the humus layer inside the chambers.

The sampling system used a multi-position valve (Valco Instruments Co., Inc.) to sample each soil chamber once per hour. Air was sampled from the open chamber for 3 min; then the chamber was closed and the headspace air was sampled for 9–10 min, followed by sampling from the open chamber for 2 min. Soil chamber 2 was added to the sampling system on 30 July 2015. Prior to this date, soil chamber 1 was measured twice per hour.

To ensure that the chamber materials do not show apparent fluxes that bias the measurements, we conducted blank-chamber tests with soil chamber 1 at the start of the campaign. The chamber footprint was sealed off with a Teflon FEP film to exclude soil fluxes and measure only the fluxes from chamber materials. We found that the apparent fluxes of CO and CO2 from the blank chamber were 0.00 ± 0.10 and -0.05 ± 0.15 µmolm-2s-1, respectively (Fig. S1 in the Supplement), and hence were not statistically different from zero. However, we found a small positive COS flux (0.66 ± 0.48 pmolm-2s-1) that was statistically different from zero in a one-sample t test (t = 3.40 and p = 0.02). We did not observe the blank-chamber COS emissions to depend on temperature as in , since daily temperature changes were small at the site. Because the same chamber was previously used at a site with strong diurnal temperature variations but did not show temperature-dependent COS emissions , we assumed the blank-chamber COS flux to be constant throughout the campaign period. We therefore subtracted the mean blank-chamber COS flux from the measured COS fluxes in both chambers and included the uncertainty term from the blank-chamber COS flux into the calculation of the overall flux uncertainty. In doing so, we also assumed soil chamber 2 to have the same blank-chamber COS flux as soil chamber 1, since the chamber materials were the same.

The tubing connecting the chambers to the QCLS (Synflex 1/4 in.) was flushed continuously to minimize wall effects for the sampled gases. The segment of inlet tubing inside the chambers was perforated to enhance the mixing of chamber air. The outlet tubing was pushed into the center of the chamber bowl, with a filter attached to it. Airflow into the chambers was provided by a diaphragm pump (KNF N811) with inlet at 0.5 m height in the vicinity of the chambers. The air flowing through the pump did not show enhanced COS or CO concentrations. The flow rates into the chambers were set to 1.5 slpm (standard liter per minute) before and 2.1 slpm after 19 August 2015. Flow rates at the chambers, and pressure and flow rate at the pump inlet were checked during regular site visits. To correct for drifts, we interpolated the time series of chamber flow rate linearly from a set of discrete field measurements, including the measured flow rate values and the estimated values derived from linear correlations with pump flow rates and inlet pressure. The residence time of the air in the chamber was 3–4 min, as calculated from the chamber effective volume (6.1 L) and the flow rate.

In flow-through chambers, any imbalance between inlet and outlet flows creates pressurization in the chamber headspace that drives vertical advection in the soil column, leading to biases in measured fluxes . For soil fluxes, underpressure seems more problematic than overpressure, because it would siphon up the soil pore air that is usually enriched in CO2 by a few orders of magnitude. To prevent pressure-related flux biases, we set the inlet flow slightly higher than the outlet flow, with the small residual flow (approximately 0.1 slpm) equilibrated through a vent at the top of the chamber . The residual flow that was dissipated would not affect the mass balance calculation, because fluxes were always calculated using the flow rates measured at the inlet.

Fluxes were calculated from the mass balance equation of chamber headspace concentrations during chamber closure. Assuming the chamber air is well mixed, the rate of change of headspace concentration of a gas species is the balance of the inlet flux, the outlet flux, and the soil flux. The inlet concentration is assumed to be the ambient concentration measured before chamber closure, and the outlet concentration is what the analyzer measures during chamber closure. We therefore obtain an equation of mass balance in the chamber, VdCdt=q(Ca-C)+FA, where C (molm-3) is the chamber headspace concentration; Ca (molm-3) is the ambient concentration; q (m3s-1) is the inlet flow rate; V (m3) and A (m2) are the chamber volume and footprint area, respectively; and F (molm-2s-1) is the flux rate to be determined. By solving the differential equation of mass balance, the soil flux rate F is then obtained from least square fit of the chamber headspace concentration vs. time.

We implemented a baseline correction to account for changes in ambient concentrations during chamber closure and instrument drift. The inlet concentration was interpolated between the two opening periods before and after chamber closure. This zero-flux baseline was subtracted from chamber concentrations before calculating the flux from least square fitting. Some measurement periods had wavelike noise in all measured gas concentrations, likely due to instrument instability, which prevented the calculation of reliable fluxes. Causes of the instrument instability were unclear, and we did not find it to be related to condensation in the chamber. The affected flux data points were filtered out by diagnosing the concentration vs. time plots, and conspicuous outliers were also removed (Table S1 in the Supplement).

Soil moisture data were measured with the Campbell TDR100 time-domain reflector (Campbell Scientific, Logan, UT, USA) and provided by the SMEAR II database. Sensors were in close proximity to the chambers (∼ 5 m). Since soil moisture measurements were associated with high-frequency random noise, we ran a Savitzky–Golay filter with a 1-day window to smooth the data while retaining the daily trends. After early September 2015, soil moisture measurements had frequent gaps. A more complete time series was available from soil profile measurements about 30 m north from the chamber site. Soil A horizon moisture at this site in August was highly correlated with both the A horizon (r2=0.88) and humus layer (r2 = 0.93) moisture measurements near the chambers. We reconstructed the missing measurements at our soil plots from the linear regressions using August data. The gap-filled soil moisture data of both layers generally agreed well with the intermittent measurements during that period (root mean square error (RMSE) = 0.042 m3m-3 for the humus layer and 0.015 m3m-3 for the A horizon, respectively).

To extract smooth patterns of temperature and moisture dependence of soil fluxes, we ran a 2-D local regression (LOESS) on COS, CO, and CO2 fluxes against humus layer temperature and moisture (predictors). Unlike linear regression, LOESS is a non-parametric method that does not require any analytical expression of the underlying relationships. At each data point, a low-degree polynomial is fitted to all its neighboring points, weighted by distances, to give a smoothed estimate at the current point .

Soils in both chambers behaved as COS sinks, with average fluxes of -2.8 (±1.0) and -2.5 (±1.2) pmolm-2s-1 for SC1 and SC2, respectively (Fig. a; Table ). The two chambers exhibited broadly similar patterns of COS fluxes (Figs. a and a–d). COS emissions were rare, accounting for only 0.1 % cases of SC1 and 1.5 % cases of SC2. Most emission cases were not statistically different from zero, and the few large emissions appeared to be transient and isolated cases unrelated to temperature or moisture change. Overall, soil COS fluxes at this site were comparable to reported values in similar ecosystems, for example, -2.5 pmolm-2s-1 from a Swedish boreal forest soil in .

There was a weak increasing trend in soil COS uptake (or decreasing net COS flux) throughout the campaign (Table ). This increasing trend was not significant during the peak growing months (July and August) but became stronger towards the end of the growing season (September and October). This trend in COS uptake appeared to coincide with the decreasing trends in soil temperature and moisture (Fig. d and e).

No significant diurnal trend was observed in COS fluxes (Fig. a–d), although surface (0.5 m) COS concentration often changed from around 300 pmolmol-1 at night to 400 pmolmol-1 at midday. Surprisingly, we found no correlation between COS fluxes and ambient COS concentrations (r = 0.005; Fig. S2 in the Supplement), indicating that COS uptake was not concentration-limited at the daily timescale. The deposition velocity of COS (uptake normalized by concentration) showed a weak diurnal variability (Fig. a and b); however, this seemed to be an apparent effect of the lower ambient COS concentration at night .

Smoothed 2-D patterns of soil COS uptake as a function of soil temperature and moisture were constructed with the LOESS method (Sect. ). Soil moisture rather than temperature was the dominant physical driver of soil COS uptake, with uptake rates decreasing with increasing moisture (Fig. a). There was only a weak tendency of increasing COS uptake with decreasing temperature. Soil COS uptake was positively correlated with soil respiration (Fig. ), consistent with previous observations . However, the relationship between soil COS uptake and respiration seemed to divide into different branches delimited by soil temperature bins (Fig. ).

CO was also taken up by soils in both chambers, with average fluxes of -1.00 (±0.43) and -0.76 (±0.43) nmolm-2s-1 in SC1 and SC2, respectively (Table ). Although the two chambers were placed in similar conditions, SC1 always had slightly stronger uptake than SC2, indicating small-scale heterogeneity. Both chambers had only a few and very weak CO emission cases (0.1 % of SC1 and 0.5 % of SC2).

In contrast to soil COS flux, there was clear diurnal variability in CO flux, consistent across all months (Fig. ). CO uptake was significantly lower during the daytime than at night, with up to 0.5 nmolm-2s-1 difference (30–50 % of nighttime CO uptake). We found weak correlations between soil CO flux and ambient CO concentration (r = -0.590 and -0.317 for SC1 and SC2, respectively, Table ; Fig. S2). However, the relative diurnal amplitudes of CO concentration were too small (less than 10 % in monthly-mean diurnal trends; not shown) to explain the diurnal variability in CO uptake. Significant diurnal variability was also present in CO deposition velocity (Fig. c and d), with the midday values about 40 % smaller than those around midnight. The diurnal variability in CO uptake was also unrelated to soil temperature, since it was out of phase with soil temperature variations (Fig. ). Instead, CO uptake was weakly correlated with the below-canopy radiation (rank correlation = 0.51 and 0.35 for SC1 and SC2, respectively; Fig. S6 in the Supplement), which varied diurnally, suggesting that the midday reduction in CO uptake might be partially due to photochemical CO production at the soil surface.

During the campaign period, the month with the highest soil CO uptake was September (Table ; Fig. ). The increase of CO uptake in September appeared to result from the decrease of soil moisture rather than changes in soil temperature (Fig. ). CO uptake was also weakly correlated with CO2 flux (Fig. ; Table ), branching into different clusters depending on temperature bins. This pattern resembled the relationship between COS and CO2 fluxes (Fig. ).

The average CO2 fluxes during the campaign period were 3.2 (±1.3) and 3.8 (±1.9) µmolm-2s-1 for SC1 and SC2, respectively. Soil respiration showed strong seasonal variations that correlated with soil temperature changes (Table  and Fig. ; Figs. S3 and S4 in the Supplement). Monthly mean soil respiration in SC1 increased from 3.0 µmolm-2s-1 in July to 4.1 µmolm-2s-1 in August, the warmest month. As soil temperature began to decrease in September, soil respiration dropped to 3.2 µmolm-2s-1 in SC1 but increased in SC2, indicating possible small-scale differences between the two chamber locations. There was no well-defined relationship between soil moisture and respiration (Fig. ).

We did not see strong temperature-driven diurnal trends in soil respiration (Fig. ), mainly because daily temperature variations were small (2–3 ∘C diurnal amplitude in the humus layer). Temperature dependence of soil respiration at the daily timescale was generally weak since most of the days had low correlations between CO2 flux and temperature (see Fig. S8 in the Supplement for a histogram of the corresponding daily r2 values), despite the overall higher correlation between soil respiration and temperature over the campaign period (Table  and Fig. ). Interestingly, a small decrease in respiration at midday was found in the July diurnal trend of CO2 flux in SC1 (Fig. i). Daytime CO2 flux in July was on average 0.44 µmolm-2s-1 lower than nighttime CO2 flux, and this difference was statistically significant (p = 1 × 10-8 in a two-sample t test). The slightly higher nighttime respiration in July might be related to A horizon temperature that peaked at midnight. Other months did not show such a pattern.

The net soil flux of COS or CO is generally the balance of concurrent sink and source activities. Here we explore how physical and biological factors drive the variability in the net fluxes of COS and CO, and infer their effects on the gross uptake (i.e., actual microbial uptake without accounting for the concurrent production) of COS and CO.

Soil moisture is the key determinant of the net soil COS flux (Fig. a), reflecting the variability of the gross COS uptake, since COS production is controlled mainly by temperature . Due to the fact that net uptake dominates soil COS flux, and that soil temperature does not show a strong variability (Fig. a–d), COS production, if present at all, is likely a minor and constant component. The moisture dependence of COS uptake activity typically manifests as a bell-shaped curve with a moisture optimum . Below the moisture optimum, microbial uptake of COS is limited by water availability, whereas above it, COS uptake is limited by the diffusional supply of COS from the atmosphere because soil gas diffusivity decreases with moisture content . In this study, the decrease of COS uptake with increasing soil moisture (Fig. a) is characteristic of the diffusion-limited regime in COS uptake. The diffusion-limited regime indicates that the moisture optimum for COS uptake was likely below the observed soil moisture range and that most COS uptake happened beneath the surface humus layer.

There is only a weak tendency of decreasing net COS uptake with increasing temperature (Fig. a). The lack of strong temperature dependence is not surprising given that 90 % of the data were measured in the narrow temperature range of 8.3–16.4 ∘C (humus layer). A temperature optimum for COS uptake cannot be identified within the observed temperature range but likely exists below this range since COS uptake here tends to increase slightly with decreasing temperature (Fig. a). The low temperature optimum would differ from previous laboratory studies that report temperature optimum values of around 20∘ , suggesting a need to constrain this parameter under field conditions.

Soil CO flux variability is also dominated by moisture dependence and did not show any significant temperature dependence (Fig. b). Previous studies show that the moisture dependence of CO uptake also follows a bell-shaped curve with a moisture optimum, which qualitatively resembles that of soil COS uptake . Because CO uptake in the soil is subject to the same reactive transport mechanism as in COS uptake, the decrease of CO uptake with increasing soil moisture indicates that CO uptake is also diffusion-limited and the majority of uptake activity should happen below the surface humus layer. The absence of temperature-driven variability in net CO uptake suggests the lack of significant abiotic thermal production of CO, yet other production activity may still exist.

A unique feature of soil CO flux is the diurnal cycle showing reduced daytime net uptake (Fig. e–h), contrasting with soil COS flux that shows no significant diurnal variability. The correlation between CO uptake and below-canopy radiation (Sect. ; Fig. S2) implies that there might be photochemical production at the surface humus layer during the campaign. Although opaque chambers were used to prevent photochemical production of CO during measurements (Sect. ), when the chamber was open and not being measured, the soil surface was exposed in the sun and photochemical production might have happened. CO production at the surface, if present, may alter the vertical CO profile and consequently affects surface flux measurements, because gas transport in the soil column is a slow process. Strong diurnal variability in soil CO flux due to photochemical production has previously been reported in a boreal forest , a temperate mixed-wood plain , and a reed canary grass cropland . Hence, photochemical production is a possible daytime source of CO that can be responsible for the diurnal cycle of net CO flux. Future studies on soil CO flux will need to operate the chamber in constant darkness to confirm or falsify the existence of photochemical CO production.

Both COS and CO fluxes appear to correlate well with soil respiration when divided into specific temperature bins (Fig. ). To characterize the sensitivities of COS and CO fluxes to respiration, mean flux ratios (FCOS:FCO2 and FCO:FCO2) are calculated in each temperature bin, approximated by the slope of the zero-intercept linear regression (Fig. ). The sensitivity of COS or CO uptake to respiration is stronger at lower temperature, as indicated by the more negative FCOS:FCO2 or FCO:FCO2 ratio (Fig. ). The temperature dependence of the FCOS:FCO2 ratio shows an asymptotic feature at higher temperature, which corresponds well with the temperature dependence of the concentration-normalized COS-to-CO2 flux ratio (also known as soil relative uptake, SRU) reported from temperate forest soils . Note that the normalization by ambient concentrations usually shrinks SRU by a factor of 1 to 1.15 with respect to the flux ratio but does not change its temperature dependence feature. The range of SRU values (-1.8 to -0.5) in is therefore comparable to the range of FCOS:FCO2 ratio in SC2 but significantly smaller than that in SC1 (Fig. ). The qualitative similarity in the temperature dependence of the FCOS:FCO2 ratio from different sites suggests that this relationship can be a generalizable feature for soils. Interestingly, the transition from linear increase to constant ratio occurs around 10 ∘C at our site (Fig. ), compared to 30 ∘C in the temperate sites , indicating a soil- or site-specific feature. In addition, the newly discovered relationship between the FCO:FCO2 ratio and temperature can be used to simulate soil CO uptake empirically.

There remains the question of why the relationship between COS or CO uptake and respiration divides into different branches defined by soil temperature (Fig. ), despite the observation that COS or CO uptake does not show strong temperature dependence (Fig. ). Since microbial activity underpins COS or CO uptake, each branch in the uptake–respiration pattern may represent the behavior of a distinct microbial group. The temperature dependence of flux ratios hence may indicate shifts in active COS- and CO-consuming microbial groups caused by temperature. For example, microbial groups that have optimal uptake at a lower temperature may have a stronger sensitivity of COS (or CO) uptake to respiration than groups active at higher temperature. The asymptote values of FCOS:FCO2 and FCO:FCO2 (Fig. ) hence would reflect the uptake to respiration sensitivity behavior of microbial groups active at higher temperature. Collectively, the sum of COS uptake (or CO uptake) contributions from the ensemble of microbial groups may not show a well-defined temperature response like that in respiration. Although we cannot rule out the possibility that the flux ratio vs. temperature pattern is influenced by divergent temperature responses of microbial uptake and abiotic production e.g.,, abiotic production is unlikely to correlate with respiration. A mechanistic explanation for the temperature dependence of flux ratios requires a clear-cut separation of uptake and production processes and further understanding of the microbial communities involved in COS uptake or CO uptake.

Neither COS flux nor CO flux exhibits strong seasonality, mainly because soil temperature variation was not strong and soil moisture was not low enough to severely limit microbial activity. For example, have shown that COS uptake is inhibited when soil moisture is below around 0.10 m3m-3 in incubation experiments, yet in this campaign soil moisture was above this threshold most of the time. From August to October, monthly mean net soil uptake of COS increased slightly from 2.7 to 3.8 pmolm-2s-1, and that of CO increased from 1.0 to 1.2 nmolm-2s-1, despite a significant drop in humus layer temperature from 13.5 to 5.5 ∘C. As discussed previously, the increase was due to increased gas diffusivity caused by declining soil moisture. Soil microbial activities for COS and CO uptake at this site appeared tolerant of low temperature at the end of the growing season (October/November). Given the lack of temperature-related seasonality in COS or CO uptake, it is possible that a shift in active microbial groups might have acted to stabilize the overall uptake against changes in soil physical variables.

Soil respiration in SC1 increased significantly from July to August (3.0 to 4.1 µmolm-2s-1), concurrent with a slight increase in the mean soil temperature (12.8 to 13.5 ∘C). As shown in , soil respiration at the site is controlled not only by temperature but also by gas diffusivity and photosynthate input from the vegetation. Since soil moisture dropped greatly after July (Fig. ), the increase in respiration was more likely driven by the aeration of soil than by just a slight increase in soil temperature. In early October, soil respiration was suppressed by the abrupt decrease in soil temperature and gradually dropped to below 1 µmolm-2s-1 (Fig. ). In addition, the declining ecosystem photosynthetic activity after August would reduce photosynthate input to the soil and therefore might also contribute to the decrease in respiration. However, since the soil plots in both chambers were cleared of understory vegetation before the campaign, autotrophic respiration would rely on photosynthate supply from trees at a distance and should therefore make up a much smaller fraction in the total soil respiration than at a vegetated soil plot. The smaller contribution of autotrophic respiration is also supported by the lower August soil respiration of around 4 µmolm-2s-1 in our campaign compared with the August soil respiration of around 6 µmolm-2s-1 in from a vegetated area at the same site. Overall, the seasonal pattern of CO2 flux is mainly driven by soil temperature and moisture, and to a lesser extent by photosynthate input.

The soil at this boreal forest site (podzol) was consistently a weak sink of COS during the late growing season. Soil uptake of COS was around 10–20 % of the daytime mean ecosystem uptake of 20.8 pmolm-2s-1 with daytime defined as the period with solar elevation angle > 20∘ therein. Soil COS uptake did not show strong diurnal and seasonal variations, nor did it show any abrupt increase induced by rain events cf.. Moreover, soil COS uptake variability was well explained by changes in soil moisture and respiration, and it would be possible to construct an empirical model for soil COS flux based on its relationship with soil moisture, temperature, and CO2 flux (Figs.  and ). Hence, we expect that soil COS uptake can be reliably accounted for when using COS as a photosynthetic tracer at this site.

Since soil COS uptake did not show a well-defined response to soil temperature or moisture but correlates well with respiration when divided into specific temperature bins (Figs.  and ), simulating soil COS uptake in this boreal forest will rely on using soil respiration as an important statistical predictor. In this case, the parameterization scheme used in and the empirical relationship based on the soil relative uptake ratio (COS uptake to CO2 emission ratio normalized by their concentrations) as in will be useful in predicting COS uptake, provided that diffusion in the soil column is properly resolved e.g.,.

The global soil CO budget remains uncertain due to limited field observations and the lack of modeling studies. Our results from a boreal forest site help bridge the gap in the understanding of soil CO exchange in this important biome. CO uptake by soil is usually characterized by the deposition velocity of CO, because ambient CO concentration varies spatially. At this site, CO deposition velocity lies in the range of 0.1–0.35 mms-1 (Fig. ), similar to previous studies in boreal forests and slightly less than those in temperate forests . CO deposition velocity shows a weak decreasing trend with increasing soil moisture (Fig. S7 in the Supplement), which is broadly similar to the negative correlation found in , indicating a prominent diffusional control on CO uptake.

Globally, soils contribute to a significant but poorly constrained CO sink. The current estimate of global soil CO uptake (∼ 300 Tgyr-1 in ) is equivalent to a global mean uptake of 2.3 nmolm-2s-1, averaged over the total land area (1.49 × 108 km2). This global mean value is significantly higher than the mean CO uptake (∼ 1 nmolm-2s-1) observed at this boreal forest in the late growing season. Recent field observations of soil CO uptake in temperate ecosystems are also smaller than this global mean, for example, less than 1 nmolm-2s-1 in a grassland in Italy and 0.78 nmolm-2s-1 in a grassland in Denmark . If we assume the soil from this site is representative of boreal forest soils, it follows that temperate and tropical soils must have higher CO uptake capacity to compensate for the relatively low soil CO uptake in the boreal forest compared with the global mean, or the current estimate of global soil CO uptake needs to be revisited.