The Stordalen catchment, located in the discontinuous permafrost region, is situated around 9.5 km east of the Abisko Naturvetenskapliga Station (ANS, Abisko Research Station; 68∘20′ N, 19∘03′ E). The catchment, covering around 16 km2, has a steep topography on the southern part and turns into the lower flat peatland region to the north (Fig. 2b). The mean air temperature obtained from ANS records was -0.7 ∘C for the period 1913–2002 (Christensen et al., 2004) and 0.49 ∘C for the period 2002–2011 (Callaghan et al., 2013). Climate details of the two warming periods in 20th century are given in Callaghan et al. (2010). The Stordalen catchment is composed of birch forest, tundra heath, peatland, lakes and rivers. Permafrost is only found in the palsa, and recent permafrost loss and conversion of palsa into non-permafrost Eriophorum sites have been related to the observed warming trends in the region (Åkerman and Johansson, 2008). The Sphagnum sites have a variable active layer depth and are dominated by Sphagnum spp. (PFT: pmoss), while the Eriophorum sites are dominated by flood-tolerant Eriophorum spp. (PFT: WetGRS). Birch forest (PFT: IBS, Betula pubescens ssp. czerepanovii), which forms the treeline in this region, is mainly distributed around and to the south of the peatlands and extends upwards to mountain slope areas (Fig. 2). Above the birch treeline, there are extensive areas of subarctic ericaceous dwarf-shrub heath dominated by the evergreen species Vaccinium vitis-idaea and the deciduous species V. uliginosum and V. myrtillus. Willow shrubs (Salix spp.) and dwarf birch shrubs (Betula nana) occur at the fringes of the forest and in wetter areas above treeline. The heath shrub species also occur as a ground layer in the forest (see Callaghan et al. , 2013, for details of vegetation distribution and changes). Around 5 % of the catchment is covered by lakes or small rivers (Lundin et al., 2013). A detailed description of catchment hydrological conditions can be found in Olefeldt et al. (2013). The map in Fig. 2a is based on an object-based vegetation analysis in this area (Lundin et al., 2015). The classification is based on an aerial imagery data set collected in August 2008, with a spatial resolution of 0.08 by 0.08 m.

The observations used to evaluate the model include (1) EC-tower-measured CO2 NEE covering the palsa and Sphagnum sites during 2008 and 2009 (Olefeldt et al., 2012); (2) EC-tower-measured CO2 NEE at the Eriophorum site from 2006 to 2008 (Christensen et al., 2012); (3) 2 years (2006–2007) of EC-tower-measured CH4 fluxes at the Eriophorum site (Jackowicz-Korczyński et al., 2010); (4) analyzed DOC concentrations at the palsa/Sphagnum site and Eriophorum site from 2008 (Olefeldt and Roulet, 2012); (5) measured DIC export and DOC concentration at six sampling points from 2007 to 2009 (Olefeldt et al., 2013); (6) EC-tower-measured birch forest CO2 NEE for four continuous years, 2007–2010 (Heliasz, 2012); and (7) catchment-level annual CO2 and CH4 emission from lakes and streams from 2008 to 2011 (Lundin et al., 2013).

The catchment level C budget was calculated using estimated C fluxes (CFi; after first accounting for observed DOC or DIC fluxes, where available) weighted by land-cover fractions (fi in Eq. 1). The symbol i in Eq. (1) represents different land-cover types – in our case, birch forest, peatland (Sphagnum and Eriophorum sites), tundra heath and lake/river. To identify the temporal variations in C fluxes, the identification of different land cover (except peatland regions) is based on the dominant PFTs for each grid cell during the reference period of 2001–2012. Additionally, the aerial photo-based classification of different peatland types is also used when estimating peatland fluxes as a whole (Olefeldt et al., 2012). Call=∑i=1i=5CFi⋅fi

The modeled DOC exports based on runoff estimations are compared with the measured DOC fluxes. The model generally underestimated annual runoff (the measured and modeled mean annual runoff for six points are 279.65 and 207.75 mm, respectively), but the modeled accuracy varies from point to point as well as for different years (Tang et al., 2014a). The average DOC export of the birch forest over 3 years across sampling sites from 2007 to 2009 are 3.65, 2.77 and 2.33 g C m-2 yr-1 from the observations and 3.09, 2.03 and 1.69 g C m-2 yr-1 from the model estimations, respectively. The downward trend over the 3 years is captured and is related to lower precipitation during the latter years. The observed DOC export rates at the palsa/Sphagnum and Eriophorum sites in 2008 are directly used to represent different types of peatland export level and the values are 3.35 and 7.55 g C m-2 yr-1 for the palsa/Sphagnum and Eriophorum sites, respectively (Olefeldt and Roulet, 2012). DIC export is currently beyond the scope of our model but nonetheless contributes to the whole C budget. We used an averaged DIC export value of 1.22 g C m-2 yr-1 based on the published data in Olefeldt et al. (2013) and DIC export is included in the estimation of the catchment C budget below (see Table 2).

Investigations of aquatic system C emission in the Stordalen catchment were conducted by Lundin et al. (2013) during the years 2008–2011. Around 5 % of the total catchment area (0.75 km2) is classified as aquatic systems, with lakes accounting for 96 % of the aquatic area. Both lakes and streams contribute to the emissions of CH4 and CO2, but the streams dominate CO2 emission while the lakes dominate CH4 emissions. Averaged across the catchment area, the measured annual CO2 and CH4 emissions from streams are 10.1 and 0.06 g C m-2 yr-1, respectively, while the lake emitted 0.5 g C m-2 yr-1 as CO2 and 0.1 g C m-2 yr-1 as CH4 (Lundin et al., 2015). Since river and lake processes are at present beyond the scope of the model, the contribution of aquatic systems to the catchment C fluxes is purely based on the observed data (see Table 2).

One of the benefits of using a dynamic vegetation model is that it allows us to investigate the vegetation and C-budget responses to climate change. In this section, annual variations of both climate variables (temperature and precipitation) and different ecosystem C fluxes are presented for the whole simulated period. In addition, the normalized catchment C budget is also estimated.

Our simulation results suggest that the temperature increase (around 2 ∘C; see Fig. 6a–c) together with a CO2 increase to 639 ppm by 2080 could greatly increase the productivity of the birch forest (Figs. 6, 2a–c and 5) as well as peatland CH4 emissions (see Fig. 6a–c). The DOC export from birch sites also increases slightly over the study period (Fig. 6, 3a–c). Besides leading to an increase in ecosystem biomass in grid cells currently occupied by birch forest (Fig. 5c, d), warming will also result in the current tundra heath close to the birch treeline being replaced by an upward expansion of the birch forest (Fig. 5a, b). This is indicated by the changes to birch treeline's uppermost elevation in Fig. 6, 8a–c. The dramatic increase in C uptake in the tundra heath region (Fig. 6, 5a–c) is largely a result of this vegetation succession. Meanwhile, the successive degradation of permafrost and slightly higher annual precipitation may result in more anaerobic conditions for the modeled peatland. Combined with warmer soil conditions, these result in both higher decomposition rates of soil organic matter and greatly increased CH4 production. Moreover, a stronger response of respiration than NPP to the temperature increase reduces the net C uptake in the peatland (see Fig. 6, 4a–c). The catchment, as a whole, shows an increased C uptake (see the trend line in Fig. 6, 7a) over the 1913–2080 period. The averaged C budgets for the two selected periods (1961–1990 and 2051–2080) are -1.47 and -11.23 g C m-2 yr-1, respectively, with the increase being dominated by the large increase seen in the uptake rates at the birch forest and tundra heath sites.

During the 2051–2080 period, both boreal needleleaf shade-tolerant spruce (PFT: BNE, e.g., Picea abies) and boreal needleleaf (but less shade-tolerant) pine (PFT: BINE, e.g., Pinus sylvestris) start to appear in the birch-dominated regions (Fig. 5). In the current tundra regions, the coverage of HSS greatly increases at the expense of LSE. For the northern parts of the catchment, birch forest densification is observed with future warming, while the greatest relative changes in biomass occur near the treeline. The increased temperature together with increased CO2 concentration by 2080 are very likely to increase CO2 uptake in both tundra heath and forested areas, though the nutrient limitations are not included in this version of the model (but see Smith et al., 2014). A summary of the modeled catchment C components and C fluxes from different sources is given in Table 2. Additionally, the modeled estimations during the warm period of 2000–2005 are also presented in Table 2, and the positive mean value of catchment C budget is seen to be an exception with reference to the main reference period. Furthermore, the annual variations of the modeled C fluxes in different periods are also presented in Table 2. We find that different land-cover types could shift rapidly from being a C sink to a C source in the future (see the mean plus 1 SD value).

To illustrate the impact of CO2 increases alone on the modeled C budget, the differences between simulations (ΔC fluxes) with and without a CO2 increase since 1960 are shown in Fig. 7. The original outputs of those two simulations are plotted separately in Fig. S1 in the Supplement, together with the statistical significance values (p). Interestingly, the simulation with constant CO2 forcing after 1960 significantly reduces the birch and tundra uptake (positive values in Fig. 7), whereas the peatland NEE and CH4 are not strongly influenced by the CO2 increase. The catchment C-budget dynamics are consistent with the changes seen in the birch and tundra heath regions. Furthermore, the magnitudes of ΔC fluxes for the birch and tundra NEE show an increasing trend after 1971, which is also seen in the ΔC fluxes for the annual GPP and respiration (see birch forest site in Fig. 7a). Since tundra heath shows a similar trend to birch forest, it is not presented in the figure. However, the relative differences of ΔC fluxes for the annual GPP and ecosystem respiration widen over time, which indicates a stronger response of GPP to the increased CO2 concentration than ecosystem respiration.

To account for the fact that CH4 is a more potent greenhouse gas than CO2, an estimation of the global warming potential (GWP) of the two simulations can be made. Assuming the relative climate impact of CH4 is 28 times greater than CO2 over a 100-year period (IPCC, 2013), the calculated GWP for the simulation with atmospheric CO2 increase is 3 times larger in the period 1961–1990 (27.3 g CO2-eq) than the period 2051–2080 (8.8 g CO2-eq). However, in the simulation without CO2 increase, the GWP in 1961–1990 (40.7 g CO2-eq) is approximately half of the GWP value for the 2051–2080 period (93.3 g CO2-eq). This shows that the change in global warming potential found in the model simulations is strongly influenced by the CO2 concentration used to force the model, as the CO2 trajectory can alter the balance between the GWP changes resulting from C uptake in the birch forest and tundra and the peatland CH4 emissions.

A temperature increase of 2 ∘C (Fig. 6) and elevated CO2 concentrations could greatly increase vegetation growth and thus the C sink of the whole catchment. However, the densification and expansion of birch forest as well as the increased presence of boreal spruce and pine PFTs in our projected period (Fig. 5) could be strongly influenced by reindeer grazing and herbivore outbreaks (Hedenås et al., 2011; Callaghan et al., 2013), even though those detected changes are consistent with other models simulations (Wolf et al., 2008a; Miller and Smith, 2012) and general historical trends (Barnekow, 1999). Furthermore, climate warming may favor the spread of insect herbivores, so an assessment of ecosystem responses to future climate change cannot ignore these disturbances (Wolf et al., 2008b) or other factors such as winter warming events (Bjerke et al., 2014).

Temperature increase results in a larger extent of permafrost degradation in the future. Meanwhile, the increased amount of available water from precipitation and lateral inflow may increase the degree of anoxia and further favors the flood-tolerant WetGRS PFT growth as well as CH4 production. However, the exact extent of wetting or drying of peatland in the future is still highly uncertain, and the model prediction depends strongly on the climate scenario, permafrost thawing, and the resulting balance between increased water availability and increased evapotranspiration. If the peatland drying is large enough, the reduced degree of anoxia could reduce CH4 emissions in the future (Wrona et al., 2006; Riley et al., 2011). In our case, the peatland, located at the lower part of the catchment, receives water from the southern mountain region, and is more likely to become wetter in the near future in response to the increases in water availability (Wrona et al., 2006). Additionally, the determination of grid cell peat fractions in our simulation is dependent on the current (historical) soil map; therefore, the projections of peatland expansion to non-peatland cells cannot be reflected in the current model predictions. This could bring some additional uncertainties into the catchment C-budget estimations (Malmer et al., 2005; Marushchak et al., 2013).

To cover all the major C components in this catchment, the current estimations of the C budget used available, observed aquatic emissions and DOC concentration and DIC export values, all of which were assumed constant for the whole simulation period, 1913–2080. However, the direction and magnitude of changes to aquatic C fluxes are hard to quantify without modeling additional processes in these systems. Previous studies have found that the substantial thawing of permafrost as well as increased precipitation in recent decades has significantly increased total organic carbon (TOC) concentrations in lakes (Kokfelt et al., 2009; Karlsson et al., 2010). The increased loadings of nutrients and sediments in lakes are very likely to increase productivity of aquatic vegetation, but the effects may be offset by the increased inputs from terrestrial organic C and increases in respiration (Wrona et al., 2006; Karlsson et al., 2010). Emissions of CH4 from aquatic systems are more likely to increase due to the longer ice-free season (Callaghan et al., 2010), increased methanogenesis in sediments and also increased CH4 transport by vascular plants (Wrona et al., 2006). Furthermore, a recent study by Wik et al. (2014) found that CH4 ebullition from lakes is strongly related to heat fluxes into the lakes. Therefore, future changes to energy fluxes together with lateral transports of dissolved C from terrestrial ecosystems to the aquatic ecosystems are especially important for predicting C emissions from aquatic systems.

As we have discussed, the dynamics of birch forest, and to a lesser extent tundra heath C assimilation, largely determine the catchment's C budget (Figs. 5 and 6), whereas the dramatic increase in CH4 can slightly offset the net climate impact of the projected C uptake. Furthermore, both modeled C budget and GWP values are very sensitive to the atmospheric CO2 levels. However, to date, there is no clear evidence showing significant long-term CO2 fertilization effects in the arctic region (Oechel et al., 1994; Gwynn-Jones et al., 1997; Olsrud et al., 2010). Two possible reasons for this lack of CO2 fertilization response might be that CO2 levels close to the moss surface in birch forest can reach as high as 1140 ppm and are normally within the range of 400–450 ppm (Sonesson et al., 1992), while tall vegetation such as shrubs and birch trees as well as peatland species have not been manipulated by CO2 fertilization. Over the long term, vegetation growth is likely a result of complex interactions between nutrient supplies (McNown and Sullivan, 2013), UV-B exposure (Schipperges and Gehrke, 1996; Johnson et al., 2002), temperature and growing season length (Heath et al., 2005) and forest longevity (Bugmann and Bigler, 2011). Therefore, more field experiments are urgently needed in order to quantify and understand the CO2 fertilization effects on the various vegetation microtypes of the subarctic environment, and particularly tall vegetation types.

Overall, the current model application has been valuable in pointing to these gaps in process understanding and meanwhile shows the importance of including vegetation dynamics in studies of C balance. Furthermore, a current inability to include the potential impacts of peatland expansion, potential increases of emissions from aquatic systems and the potential nutrient limitations on plants (but see Smith et al., 2014) and disturbances (Bjerke et al., 2014) makes it likely that our projections of the catchment C budget and CO2 GWP will vary from those that may be observed in the future. However, our high-spatial-resolution, process-based modeling in the subarctic catchment provides an insight into the complexity of responses to climate change of a subarctic ecosystem while simultaneously revealing some key uncertainties that ought to be dealt with in future model development. These developments would be aided by certain new observations and environmental manipulations, particularly of CO2 with FACE experiments of shrubs and trees, in order to improve our understanding and quantification of complex subarctic processes.