Multi-year data-model evaluation reveals the importance of nutrient availability over climate in arctic ecosystem C dynamics

Terrestrial CO₂ exchange measurements have been conducted in Zackenberg fen (Northeast Greenland, 74°N; figure 1(a)) and in Kobbefjord fen (Southwest Greenland, 64°N; figure 1(b)) since 2008 under the auspices of the cross-disciplinary GEM programme. These locations characterize high and low arctic sites (Christensen et al 2017) and are both surrounded by >1000 m.a.s.l mountains and a fjord. Annual mean temperature and precipitation are −8.6 °C and 253 mm in Zackenberg and 0.3 °C and 1081 mm in Kobbefjord during the 2008–2018 period. According to the Circumpolar Arctic Vegetation Map (Walker et al 2005) Zackenberg belongs to the subzone C bioclimate zonation with an average temperature in July of 7 °C–9 °C, while Kobbefjord fits the subzone E with an average July temperature of 11 °C–13 °C. Both fen sites have water saturated organic soils with an abundant snowmelt water supply. The precipitation falls largely as snow during the shoulder seasons and on average Zackenberg accumulates a slighter thicker (maximum) snowpack compared to Kobbefjord (1 and 0.9 m, respectively). The Zackenberg area has continuous permafrost, with maximum thaw depth variability of 0.5–1 m (Lund et al 2014), whereas no permafrost has been found in Kobbefjord (López-Blanco et al 2017). The two sites are sedge dominated fens commonly populated with Eriophorum scheuchzeri and Dupontia psilosantha (Zackenberg) and Eriophorum angustifolium and Scirpus caespitosus (Kobbefjord) (Bay 1998, Bay et al 2008) and an abundant moss layer characterized by the presence of Sanionia uncinata and Sphagnum lindbergii at each site, respectively (Hassel et al 2012). The sunlight hours from May to September differ substantially between the two sites, ranging from 14 to 21 h in Kobbefjord and from 17 to 24 h in Zackenberg.

Zoom In Zoom Out Reset image size

The EC flux data consist of high-temporal-resolution measurements for the 2008–2018 period. In Zackenberg the systems consisted of a closed-path infrared gas analyser LI-6262 (LI-COR Inc, USA) and 3-D sonic anemometer Gill R2 (Gill Instruments Ltd, UK) until August 2012, when it was upgraded to an enclosed-path LI-7200 (LI-COR Inc, USA) and Gill HS (Gill Instruments Ltd, UK). In Kobbefjord, the systems have been equipped with a closed-path infrared gas analyser LI-7000 (LI-COR Inc, USA) and a 3-D sonic anemometer Gill R3-50 (Gill Instruments Ltd, UK) until August 2018, when it was upgraded to an enclosed-path LI-7200 (LI-COR Inc, USA). The sonic anemometer in Zackenberg was installed at a height of 3 m (and the air intake was attached at the same level) while in Kobbefjord it was at 2.2 m (air intake at 2 m). In both stations we processed the high-frequency CO₂ concentration and wind components data according to standard flux community techniques (i.e. FLUXNET and ICOS), including de-spiking (Højstrup 1993), 2D coordinate rotation, time lag removal by covariance optimization, block averaging, frequency response correction (Moore 1986) and Webb-Pearman-Leuning correction (Webb et al 1980). More information on the EC system setup, flux computation, and quality checks in Lund et al (2010). We post-processed the quality-checked NEE data using gap-filling and partitioning approaches. On one hand, data-gaps have been filled with a marginal distribution sampling technique (Moffat et al 2007). On the other hand, the separation of NEE into its two modulating fluxes, GPP and R_eco, was achieved via traditional separation algorithms utilized in the FLUXNET community (Reichstein et al 2005, 2016). More information on flux gap-filling using marginal distribution sampling and flux partitioning using ReddyProc (Reichstein et al 2016) in López-Blanco et al (2017). Due to the absence of true night-time during the growing season in Zackenberg, the data have been processed using the daytime method (Lasslop et al 2010). The reporting of fluxes in this paper follows the standard micrometeorological sign of convection, i.e. the uptake of carbon (sink) is a negative flux while the release of carbon (source) is a positive flux. Moreover, this study defines the beginning of the growing season as three-consecutives days with negative fluxes (i.e. net C uptake) after the winter period, while the end of the growing season is characterized as three-consecutive days with positive fluxes (i.e. net C release).

For each site a comprehensive suite of meteorological measurements, phenology related observations, biomass and soil core samples, physical soil parameters, and soil water chemistry have been collected, processed and quality-checked from the GeoBasis and ClimateBasis subprogrammes, all freely accessible from the GEM database (data.g-e-m.dk). The meteorological datasets from nearby climate stations (<2km distance) includes data on air temperature (°C), total precipitation (mm), relative humidity (%), shortwave radiation (W m⁻²), photosynthetic active radiation (W m⁻²), and snow depth (m). The phenology related variables integrate leaf area index (m² m⁻²) at the peak of the growing season and end of the snowmelt period (Day of Year, DOY). Direct harvest measures of leaf area index (LAI) has been calculated using Image J (Schneider et al 2012) for July 2015 data in Kobbefjord (López-Blanco et al 2018) and July 2019 data in Zackenberg. The snowmelt period was classified at a pixel level (<20% snow cover) from a time-lapse camera (HP e427) following the procedures described by Westergaard-Nielsen et al (2017). C and N stocks (leaf, litter, stems, roots, mosses, and soil organic matter; g C m⁻² and g N m⁻²) were collected from 5 plots of 100 cm² area at each fen site following the procedure described in López-Blanco et al (2018). The physical soil parameters integrated in this study contains soil temperature (°C) at different soil depths and snow coverage (%) derived from the time-lapse camera. Soil water chemistry observations include dissolved organic carbon (DOC; ppm), dissolved organic nitrogen (DON; ppm), ammonium (NH₄⁺; ppm), nitrate (NO₃^-; ppm) and specific conductivity (EC; μS cm⁻¹). Further details on ancillary measurements from Zackenberg and Kobbefjord can be found in Lund et al (2012) and López-Blanco et al (2017).

We run the soil-plant-atmosphere (SPA) model (Williams et al 1996, 2000) with 11 years (2008–2018) of meteorological forcing from Zackenberg, to support previous simulations at Kobbefjord (López-Blanco et al 2018). SPA is a mechanistic point model that simulates C, water and energy cycles through eco-physiological principles in a vertically resolved canopy and soil profile. SPA models (1) a radiative transfer scheme differentiating between sunlit and shaded leaf area, (2) photosynthesis based on the classic representation of carboxylation from Farquhar and von Caemmerer (1982) model plus a stomatal conductance model that balances vapour phase losses with hydraulic supply to maximise C uptake, (3) surface energy balance and evaporation based on the Penman-Monteith method, and 4) detailed distribution of water and heat transfer through the soil profile. Furthermore, the model version used here has been refined, calibrated and validated with observational data from Kobbefjord (López-Blanco et al 2018), including implementation of important arctic related processes. In this version we (1) independently calculated maintenance respiration losses considering nitrogen (N) interactions based on formulations described by Reich et al (2008) and not as a fixed ratio, and (2) improved net C uptake timing at the beginning of the growing season by restarting growing degree day summation right after the snowmelt period, using data derived from the in-situ cameras (Westergaard-Nielsen et al 2017). In López-Blanco et al (2018) we found that the model's most sensitive parameters to NEE, GPP and R_eco were those related to leaf N traits and initial C stocks. Here we followed the approach applied to Kobbefjord data with separate calibration and validation years. Therefore, we manually calibrated the first 5 years of the time series (2008–2012) using the timing of specific snowmelt in Zackenberg to define the start of plant flush, average leaf nitrogen, leaf mass per area (LMA), the maximum foliar C stock (at the peak of the growing season), and the C and N stocks (litter, stem and roots). Finally, the Q₁₀ of foliar and root respiration rates has been increased from 2 to 3 at Zackenberg to account for plant thermal acclimation to colder temperatures (Tjoelker et al 2001, Atkin and Tjoelker 2003). The rest of parameters have been kept the same as at Kobbefjord to facilitate a model performance comparison only impacted by the environmental forcing. For validation we calculated linear goodness-of-fit (R² and RMSE) of the last 6 years (2013–2018) to evaluate the level of statistical agreement between C flux data (NEE, GPP and R_eco) and model simulations.

Overall, Zackenberg was colder (a difference of −8.9 °C) and drier (a difference of 828 mm in precipitation) compared to Kobbefjord. Zackenberg fen, located 10° north of Kobbefjord fen, had lower interannual and interseasonal temperature and precipitation variability between 2008 and 2018 (figure 1). During this period, annual mean temperatures and total precipitation ranged from −2.4 °C to 3.1 °C and from 559 mm to 1179 mm in Kobbefjord, but only between −9.7 °C and −6.7 °C and 93 mm and 436 mm in Zackenberg. Kobbefjord featured stronger interannual oscillations (greater than the −0.3 ± 2.5 °C and 1081 ± 200 mm anomalies delimited by the dotted box in figure 1(b.1) between specific series of years such as 2010 (warmer and wetter), 2011 (colder and drier) and 2012 (warmer and wetter); and 2016 (warmer and drier), 2017 (warmer and wetter) and 2018 (colder and drier). Zackenberg had smaller temperature and precipitation anomalies (not exceeding ±2.5 °C and ±200 mm figure 1(b.2).

These climatic conditions shaped both the snow regimes and the seasonality of the growing season. First, Zackenberg had 10% greater maximum snow depths in the cold season (October to May) and 23 days delay to the end-of-snowmelt periods than Kobbefjord (table 1). Second, Zackenberg fen switched from being a source to a sink of CO₂ on July 9th (17 days later than Kobbefjord) and continued with net uptake until August 23rd (5 days before Kobbefjord). Zackenberg fen had an average growing season length of 46 days compared to 66 days in Kobbefjord (table 1).

Zackenberg fen had a higher C sink strength (>170%) compared to Kobbefjord fen (figures 2 and 3) despite its higher latitude and markedly shorter growing season (table 1 and figure 2). Zackenberg fen generally acted as a sink of CO₂ over the study period, with an average NEE of −50 g C m⁻² yr⁻¹ (range +21 to −90 g C m⁻² yr⁻¹), more than twice as strong as Kobbefjord (−18 g C m⁻² yr⁻¹ with range of +41 to −41 g C m⁻² yr⁻¹)(figure 3). There were two anomalous C source years, with positive NEE; 2018 in Zackenberg and 2011 in Kobbefjord. Zackenberg 2011 has been associated with an extreme melt season (i.e. one month delay compared to the 2008–2017 period) (Christensen et al 2020). Kobbefjord featured exceptionally variable meteorology between 2010 and 2011 (López-Blanco et al 2017) facilitating optimal conditions for a biological outbreak of the noctuid moth Eurois occulta larvae (Lund et al 2017) and minimal for plant growth. The NEE of 2011 and 2018 have been ∼70 and ∼83 g C m⁻² yr⁻¹ less productive than the rest of years on average, respectively, and therefore acted as a net source of CO₂.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In general, the higher C sink strength observed in Zackenberg is linked to larger photosynthesis (i.e. more negative GPP) rather than reduced respiratory losses (i.e. more positive R_eco) (figure 3). Specifically, Zackenberg GPP was on average −252 g C m⁻² yr⁻¹ (range of −130 to −317 g C m⁻² yr⁻¹), 18% higher than Kobbefjord GPP (−213 g C m⁻² yr⁻¹, ranging from +131 to −316 g C m⁻² yr⁻¹). The respiration released from R_eco was 7% larger in Kobbefjord (195 g C m⁻²yr⁻¹, range of 145 to 280 g m⁻² yr⁻¹) than in Zackenberg (181 g C m⁻² yr⁻¹, range of 112 to 237 g C m⁻² yr⁻¹).

We found a lower sensitivity to annual air temperature from gross fluxes in Zackenberg (GPP slope = 6.1 g C m⁻² yr^{−1 °}C and R_eco slope = 1.1 g C m⁻² yr^{−1 °}C; figure 4(a)) compared to Kobbefjord (slope = 32.7 and 25 g C m⁻² yr^{−1 °}C). This finding is consistent with the lower climate interannual variability revealed in figure 1 and the smaller sensitivity in growing season length at Zackenberg (table 1). We found a similar compensatory effect (slope, intercept, R²; figure 4(b)) between photosynthesis and respiration both in Zackenberg and Kobbefjord.

Zoom In Zoom Out Reset image size

The enhanced photosynthetic activity (figures 3 and 4(b)) has been associated with higher C and N stocks and leaf traits in the aboveground domain and larger concentration levels of nutrients and minerals in soils (figure 5).

Zoom In Zoom Out Reset image size

The new in-situ information retrieved from the 2019 sampling campaign shows systematic larger C stocks, leaf mass per area (LMA), leaf N, leaf area index (LAI), and plant quality (C:N ratio) in the Zackenberg fen (figure 5(a)). The C stocks averaged at the peak of the growing season 74.6 g C m⁻² in leaves, 106.7 g C m⁻² in litter, 89.3 g C m⁻² in stems, and 405.2 g C m⁻² in mosses. This is 44.7, 40.4, 20.1, and 9.1 more g C m⁻² in leaves, litter, stems and mosses compared to Kobbefjord C stocks, respectively. Likewise, the leaf trait data pointed to consistently higher leaf N (2.3 vs 1.62 g N m⁻² leaf area), LMA (58.83 vs 56.29 g m⁻²), and LAI (1.28 vs 0.52 m² m⁻²) in Zackenberg compared to Kobbefjord (figure 5(a)). Moreover, we found that the plant quality in Zackenberg fen is 36% higher (i.e. lower C:N ratio) than Kobbefjord fen (figure 5(a)).

The water chemistry data from the first 50 cm of Zackenberg fen topsoil show consistent higher levels of Dissolved Organic C (DOC), Dissolved Organic N (DON), nitrates (NO₃⁻), ammonium (NH₄⁺), and electroconductivity (EC) during the 2015–2017 period (figure 5(b)). Overall, the belowground domain had 3 times more DOC (ppm), 5 times more nutrients such as DON (ppm) and NO₃^- (ppm), 2 times more K⁺ (ppm), and 5 times higher EC (μS cm⁻²), and slightly more acidic pH (figure 5(b)) in Zackenberg. Likewise, alkaline cations (Ca²⁺ and Mg²⁺) are 13 and 5 times higher while the acidic cations (Mn²⁺ and Fe²⁺) are 21 and 7 times higher in Zackenberg fen (figure S1 stacks.iop.org/ERL/15/094007/mmedia).

The SPA model can realistically characterise 11 years of data from the Zackenberg fen (figure 6) and 8 years of data from Kobbefjord (see López-Blanco et al (2018)), with model setup varying according to in-situ biomass and tissue N data. At Zackenberg the daily aggregated NEE (R² = 0.69; RMSE = 0.4 g C m⁻²d⁻¹), GPP (R² = 0.80; RMSE = 0.5 g C m⁻²d⁻¹) and R_eco (R² = 0.71; RMSE = 0.6 g C m⁻²d⁻¹) matched the independent summertime field observations for the validation period (2013–2018). The snowmelt period information retrieved from the photo monitoring, has been used to restart the growing degree day summation, modelling the NPP allocation into the different C pools. This implementation improved significantly the beginning of the growing season timing (R² = 0.92) and only resulted in an average 4-day shift compared to the SPA calculation.

Zoom In Zoom Out Reset image size

The SPA modelling showed that the observed difference in plant tissue N concentrations at Zackenberg relative to Kobbefjord could explain the identified increase the C uptake at the high Arctic site indicated by the EC observations (figure S2). SPA can generate a system consistent with Zackenberg fluxes only by parameterisation based on the biomass sampling data from the field campaign (average leaf N, LMA, maximum foliar C stock, initial C stocks of litter, roots and stems and C:N ratio in roots; see section 3.4). Using the initial calibration from Kobbefjord under Zackenberg climate fails to simulate annually aggregated NEE within the observation's range (figure S2). However, once the new information on C and N is updated in SPA, together with the updated Q₁₀ of foliar and root respiration rates, the yearly aggregated NEE mean for the Zackenberg site between 2008 and 2018 is −54.9 ± 50.6 g C m⁻² yr⁻¹ while the NEE value extracted from the EC tower is −56.3 ± 20.9 g C m⁻² yr⁻¹. This finding shows the importance of the C and N changes, and emphasises the importance of N to C fluxes.

Zackenberg fen features lower interannual climate variability with systematically colder and drier conditions (figure 1), thicker snowpack, later snowmelt period (table 1), and shorter periods with net photosynthetic uptake during growing seasons compared to Kobbefjord fen (table 1, figure 2). These meteorology- and phenology-related conditions have not been an obstacle to greater C uptake than Kobbefjord, on average 32 g C m⁻² more (figure 3), despite the shorter growing seasons. In this study longer growing seasons are not necessarily translated into larger net C uptake, similar to previous findings from Lund et al (2010) and Parmentier et al (2011). On average Zackenberg sequesters more C (NEE = − 70.8 g C m⁻² yr⁻¹) than other arctic wetlands at lower latitudes reported by Coffer and Hestir (2019) (table S1; mean NEE = −47.3 g C m⁻² yr⁻¹ and range of −119 to +79.3 g C m⁻² yr⁻¹) during the same period (June 1st to August 31st). Moreover, we found at least three specific extreme events that have notably disturbed the growing season C budget at both Zackenberg and Kobbefjord (figure 3). On one hand, a natural larvae outbreak had a significant impact on vegetation productivity in 2011 in Kobbefjord (Dahl et al 2017, Lund et al 2017). López-Blanco et al (2017) estimated a shift from source to sink of −30 g C m⁻² yr⁻¹ as average for the 2008–15 period to a source of 41 g C m⁻² yr⁻¹ while Lund et al (2017) reported a counterbalanced increase of C sink strength through the following 3 years. On the other hand, Christensen et al (2020) have reported multiple ecosystem effects triggered by extreme meteorological conditions indicating (1) a decrease of 18–23 g C m⁻² yr⁻¹ (close-to-zero NEE) during a 9-day rain event in 2015 and (2) a 314% weaker C sink strength in 2018 compared to the 2008–17 period forced by an extraordinary late snowmelt (i.e. 1 month delay in maximum daily CO₂ uptake and 20 days shorter growing season). Reductions of 20–40 g C m⁻² yr⁻¹ are not trivial as these are similar to typical arctic fen ecosystem annual C budgets (Parmentier et al 2011).

Zackenberg fluxes were less sensitive to temperature compared to Kobbefjord (slopes of the regressions from figure 4(a)), perhaps linked to reduced phenological variability at this site (table 1). This study also found a compensatory effect between photosynthesis (GPP) and respiratory losses (R_eco) for both sites (figure 4(b)) similar to previous findings reported in López-Blanco et al (2017) and previously described by Richardson et al (2007) and Wohlfahrt et al (2008). We noted however that the overall contribution to NEE in Zackenberg was dominated by photosynthesis (39 g C m⁻² yr⁻¹ more than Kobbefjord) compared to respiration (14 g C m⁻² yr⁻¹ less) (figures 3 and 4(b)). Consequently, controls of photosynthesis will be given a higher priority in the following sections.

Our results suggest that the limitations of plant phenology timing and colder temperatures in Zackenberg regarding net C uptake have been more than counterbalanced by the increased content of plant tissue N, linked to richer soil nutrients. Here, it is the difference in nutrient availability, not the difference in climate, that explains the divergence in net C uptake between the two sites (figures 3, 4, 5, and S2). But at each site climate governs the interannual variability. We found higher N concentration levels in above-and below-ground plant tissues and soil water. It is well-known that arctic tundra ecosystems are generally nutrient limited (Chapin III and Shaver 1985) and that soil nutrient availability shape the patterns of plant abundance (Shaver and Chapin III 1980). Yet, site-specific differences such as geology, climate boundary conditions, flora, and fauna will control differences in nutrient availability between Zackenberg and Kobbefjord, which again contributes to differences in net C uptake.

4.2.1. C and N content in the aboveground domain.

4.2.2. C and N content in the belowground domain.

Our CO₂ exchange estimations for the 2008–2018 snow-free period using local meteorological forcing at Zackenberg and Kobbefjord can explain a significant part of the high temporal variability in NEE, GPP and R_eco (figure 6, López-Blanco et al (2018)). In order to achieve the model calibration we relied primarily on in-situ data collected in 2019 during the monitoring field campaign. This aboveground biomass and soil core sampling effort was purposely designed to fill C cycle model calibration knowledge gaps. In López-Blanco et al (2018) we highlighted two important messages from Kobbefjord—N-related plant traits are the most sensitive parameters in the SPA model and therefore field data on C-N ratios decrease the model uncertainty. At Zackenberg, datasets on snowmelt period, leaf N, leaf mass per area, C:N ratio of roots, and C stocks of leaf, litter, stem, roots were critical to match model simulations with local observations of NEE (figure S2). The SPA model confirms that the differences in C cycling between the two sites are best explained when information about site-specific leaf nutrient parameters is included in the model. There is ample evidence from field manipulations of the sensitivity of primary production to nutrient additions at high latitudes, mediated by changes in plant traits (Shaver et al 2001). Our results support other Arctic biogeochemical modelling studies, e.g. TEM (McGuire et al 1992, Zhuang et al 2003) and MEL (Rastetter and Shaver 1992, Rastetter et al 2013), which concluded that regulation of arctic C cycling at the landscape and regional scales was linked to nutrient controls via C/N/P stoichiometry of plant tissues.

We also increased Q₁₀ of foliar and root respiration rates to improve the plant respiration sensitivity to temperature since Zackenberg is ∼8.3 °C colder on annual basis (figure 1). Atkin and Tjoelker (2003) have shown that Q₁₀ is not constant as it increases near-linearly with decreasing temperatures; short-term increases in temperature can have a greater potential impact on plant respiration in plants growing in cold climates (with an average leaf respiration Q₁₀ ∼2.5–3). Likewise, Heskel et al (2014) revealed how in Toolik lake Q₁₀ values also decreased with temperatures from ∼3.0 at 5 °C to ∼1.5 at 35 °C. In this version of SPA (López-Blanco et al 2018), maintenance respiration is calculated based on a modified version of the Reich et al (2008) equation built from on a strong respiration-nitrogen relationship. The fact that SPA shows a better agreement with EC observations with higher Q₁₀ (figure S2) suggests that plants are thermally acclimated to colder temperatures and that respiration triples (i.e. Q₁₀ = 3.0), and not doubles, per 10 °C rise in temperature (Atkin and Tjoelker 2003) at this high Arctic site.

In this study we show how to parameterize the SPA model with in-situ data from a single year peak season (figures 5 and S2). However, higher temporal information on C and N pool variability, similar to Arndal et al (2009) or Mosbacher et al (2019), may help understand the underlying processes and responses in shoulder season dynamics such as the snowmelt period, the rapid green-up and green-down phases and even extreme events such as a moth outbreak. Additionally, single year data may introduce bias. For example, 2019 (when the in-situ samples were collected) was an unusual warmer summer with thinner snow coverage compared to the 2008–2018 trend, and thus this year's meteorology may have enriched the allocation of biomass more than previous years. However, as noted earlier, our N samples in foliage in 2019 were close to values measured in other years. Finally, in relation with the model performance, the effect of changes in precipitation may have a role of interannual variation in fluxes. We found that the coefficient of determination for modelled vs measured NEE during anomalously dry summers was reduced (0.55) compared to wet summers (0.78) at Zackenberg. Further investigations of carbon-water interactions are required.

Using a simple set of parameters we can model high resolution temporal CO₂ exchange with a good degree of agreement in complex arctic tundra ecosystems of varying fertility and climate. We believe this modelling framework forms an ideal framework for analysing new sites. Model simulations compared with results from multiyear CO₂ exchange measurements can identify key process uncertainties, feedbacks between structure and function, and the sensitivity to extreme conditions. Two components are required to establish a description of the basic ecosystem-atmosphere interactions in the absence of direct flux measurements: (1) a quantification of the basic biomass and soil core sampling considering C and N status of the ecosystem, and (2) an ecosystem modelling component, e.g. SPA, describing the basic C dynamics based on key driving parameters, and linking C and N in plant tissues, independent of measured fluxes. Ecosystem modelling frameworks can fill process-based knowledge gaps, investigate climate feedbacks, and generate prognostic scenarios exploring the likely future implications of climate change on arctic tundra C cycle dynamics.