The timing of snow melt controls the annual CO₂ balance in a subarctic fen

1. Introduction

[2] Despite the short growing season and the relatively low gross primary production rates, northern peatlands have accumulated a large fraction (about one-third) of the soil carbon in terrestrial ecosystems [Gorham, 1991], mostly during the Holocene [Turunen et al., 2002]. This has been enabled by the limited decomposition of the organic matter in the typically wet and cool soils. The increase of about 0.05°C yr⁻¹ in summer temperatures during the last four decades, and the consequent drying of the soil have turned many ecosystems in Alaska into sources of CO₂ [Oechel et al., 1993, 2000; Welker et al., 2000]. Net carbon loss has also been found in other northern areas [Heikkinen et al., 2004; Griffis and Rouse, 2001], although many arctic and subarctic ecosystems continue to act as CO₂ sinks [Soegaard and Nordstroem, 1999; Harazono et al., 2003]. Climate warming has also been detected in northern Europe [Tuomenvirta, 2004]. At Sodankylä (67°N, 26°E) in northern Finland, an increase of 0.02°C yr⁻¹ on average has been observed in the mean annual temperatures over the period 1961–2000, while in spring (March–May) the increase has been 0.04°C yr⁻¹ [Tuomenvirta et al., 2001].

[3] Using correlation analysis, we studied which hydrometeorological factors are the most important determinants of the carbon balance of a fen on a monthly and annual scale. We utilize here the first continuous multi-year data-set of CO₂ fluxes from the subarctic region. The measurements were conducted on a natural wetland at Kaamanen in northernmost Finland in 1997–2002 using the micrometeorological eddy covariance technique.

2. Data

[4] The measurement site is situated on a mesotrophic fen within the aapa mire region at Kaamanen in northern Finland (69°08′N, 27°17′E, 155 m above sea level). The fen is about 7000 years old and has a long-term apparent rate of carbon accumulation of 11 g C m⁻² yr⁻¹ (Turunen, unpublished data), which is somewhat lower than the average (16 g C m⁻² yr⁻¹) determined for similar mires [Turunen et al., 2002]. The surface pattern of these northern fens (aapa mires) consists of wet hollows and drier strings (hummocks). The annual precipitation sum (470 mm) markedly exceeds the annual evapotranspiration (260 mm). The hollows are most of the time inundated, but the strings are constantly above the water table, receiving nutrients mainly from precipitation. The height of the strings is 0.3–0.8 m, and they cover 40% of the surface area of the fen. The peat depth is about 1 m [Heikkinen et al., 2002]. The site has no permafrost, but thin lenses of ice may remain in the well-insulated strings until late summer. The hollows are covered by different sedges (Carex spp.) and some moss species, while the higher strings are dominated by various shrubs, such as Ledum palustre, Empetrum nigrum, Rubus chamaemorus and Betula nana. The maximum single-sided LAI is estimated to be 0.7 [Aurela et al., 2001].

[5] The CO₂ flux measurements were conducted at a height of 5 m using the eddy covariance technique. The instrumentation included an SWS-211 (Applied Technologies, Inc.) three-axis sonic anemometer and a LI-6262 (Li-Cor, Inc.) CO₂/H₂O analyzer. A series of calculations and corrections involving the collected data were performed off-line [Aurela et al., 2001]. The gaps in the time series were filled by parameterisation of the net ecosystem CO₂ exchange and the ecosystem respiration [Aurela et al., 2002].

3. Results

[6] During the six measurement years, the Kaamanen fen acted as a net sink of atmospheric CO₂. The annual balance ranged from −4 to −53 g C m⁻² yr⁻¹, with a mean of −22 g C m⁻² yr⁻¹ (Figure 1a). The interannual differences originate almost completely from the variations during the snow-free period. Even though the efflux (23–26 g C m⁻²) in the wintertime (November–April) constitutes an important part of the annual balance [Aurela et al., 2002], its contribution to the interannual variation is minor.

[7] We found no significant correlation between the annual CO₂ balance and the annual mean of key hydrometeorological quantities (Table 1). The correlations were also poor for the summer (June–August) mean of most variables. However, two hydrological variables, viz. river discharge and the water table depth (WTD), correlated significantly with the CO₂ balance. These correlations originate in the timing of the spring flood. Excluding the flood period, the WTD data for July–August do not explain the interannual variations. Furthermore, the annual balance correlates with various other descriptors of spring conditions, including the beginning of the growing season, the start of the sink period and the length of the sink period. The best correlations were found with the snow melt date and the associated mean temperature (Figure 2). In 1997–2002, the snow melt date varied from April 30 to May 23 and was accompanied by a large variation in the onset of the growing season and the CO₂ fluxes (Figure 1a).

[8] The monthly CO₂ balances demonstrate the significance of the spring phenology (Figure 1b). The highest uptake rates are regularly observed in July, but the variation in the monthly balance at this time is very low. In June, the mean net uptake is 28% of that in July, but the range of variation is over four times as high as in July. The variation in August is also higher than in July, mostly due to the senescence timing, but is considerably lower than in June. The onset of senescence varies less than the onset of the growing season, as senescence is partly governed by the day-length. The shorter day and consequently smaller daily uptake also reduce the variation in autumn.

[9] The monthly correlations show that during the non-growing season there is a consistent pattern of temperatures controlling the monthly balances (Table 2). In autumn (September–December), with ceasing photosynthesis and limited or absent snow cover, the air temperature is the dominant factor. As the snow pack thickens, the soil respiration processes become detached from the ambient air, and in spring (February–May) the soil temperature constitutes a more important control. As expected, in June the balance depends on the same factors that dominate on an annual scale, i.e., the snowmelt date (r = 0.88), spring temperatures (r = −0.84) and the onset of the growing season (r = 0.90). The good correlations with WTD and river discharge have the same origin; the earlier the snow melts, the less water there is in pools in June. In July and August, the CO₂ balance does not correlate significantly with any single variable.

[10] Increasing temperatures accelerate both growth and decomposition processes, thus affecting the opposite components of the net ecosystem exchange (NEE) of CO₂. Our data support the conclusion that during the growing season an enhanced gross photosynthesis (GP) tends to be balanced by a higher ecosystem respiration (ER) [Bubier et al., 2003] and that the remaining variations in NEE mostly result from those in GP [Griffis and Rouse, 2001; Arneth et al., 2002]. In June, however, we observe no such balance: a change in conditions that enhance GP induces a much lesser change in ER (Figure 3). This in accordance with the hypothesis that advancing the beginning of the growing season effectively increases the annual net carbon gain in northern high latitudes, as has been previously suggested, based on atmospheric CO₂ data [Keeling et al., 1996], satellite measurements [Myneni et al., 1997], and ecosystem-scale flux measurements [e.g., Black et al., 2000].

[11] By taking into account the efflux of methane (4 g C m⁻²) [Hargreaves et al., 2001] and leaching of dissolved organic carbon (8 g C m⁻²) [Kortelainen et al., 1997; Aurela et al., 2002], we estimate that the fen acts as a significant sink of carbon only during the two years with the earliest snow melt. It is important to emphasize that the six years considered in this study contain the extremely dry summer of 1997 (lowest precipitation in 1961–2003) and the early spring of 2002 (second earliest snow melt date in 1961–2003). On the other hand, the monthly mean temperatures in July were relatively stable, which may partly explain the low variation in the CO₂ balances. The early spring had a significant effect on the annual CO₂ balance, whereas the low precipitation had only a limited influence on WTD and the CO₂ fluxes.

4. Discussion

[12] Long-term data from arctic and subarctic wetlands are scarce and predominantly collected using the chamber technique. However, the conclusion has been drawn that many of these ecosystems have turned into sources of CO₂ owing to the warming and drying conditions, and thus the future of this large carbon reservoir is considered threatened [Oechel et al., 1993, 2000; Welker et al., 2000; Griffis and Rouse, 2001; Heikkinen et al., 2004]. Our results, which are based on continuous ecosystem-scale measurements of the CO₂ flux, disagree with these conclusions. We suggest that those non-permafrost wetlands that are characterised by an annual spring flood and hydrological buffers (surface adjustment of floating peat [Nykänen et al., 2003] and a dynamic connection to a catchment-scale water system) will predominantly benefit from a warmer climate, especially from a warmer spring and earlier snow melt. The idea that humidity conditions and the changes in WTD overshadow the influence of rising temperature and prolonged growing season [McCarthy et al., 2001; Gorham, 1991] may be correct for more southern bogs, where even the present WTD fluctuations are markedly greater [Lafleur et al., 2003; Arneth et al., 2002].

[13] Our results have significant implications for the estimation of the climatic responses of the carbon balance of northern wetlands. A change in the snow melt timing is likely to increase the annual net CO₂ uptake of a typical fen in the climatically-sensitive area between the boreal and permafrost regions. Our data suggest a change of 2.0 g C m⁻² yr⁻¹ in the annual CO₂ balance per one-day change in the snow melt date (Figure 2a). With the mean shift of 0.14 d yr⁻¹ for the last 40 years (based on the linear regression of data from Sodankylä), this corresponds to an overall increase in the sink term of 11 g C m⁻² yr⁻¹. This sensitivity estimate is based on the simple assumption that the annual variation in NEE can be totally explained by the snow melt date. In reality, the environmental conditions during summer and autumn have their effect on the annual balance, but the present data shows that the timing of the snow melt is by far the most important factor. On a decadal timescale, the nutritional status and plant composition of the ecosystem will also play a role, which calls for controlled field experiments and modelling studies.

5. Conclusions

[14] The eddy covariance measurements conducted on a subarctic fen at Kaamanen in 1997–2002 indicate that the fen is presently a small sink of atmospheric CO₂ (−22 g C m⁻² yr⁻¹). The interannual variation of the CO₂ balances originates mainly from the variations during the snow-free period, even though the wintertime efflux (23–26 g C m⁻²) is an essential part of the annual balance. The snow melt timing was the most important single factor determining the annual carbon balance, whereas the hydrometeorological conditions during the growing season had only a minor effect on the annual balance.