Integrating peatlands and permafrost into a dynamic global vegetation model: 2. Evaluation and sensitivity of vegetation and carbon cycle processes

Authors


  • This is a commentary on DOI:

Abstract

[1] Peatlands and permafrost are important components of the carbon cycle in the northern high latitudes. The inclusion of these components into a dynamic global vegetation model required changes to physical land surface routines, the addition of two new peatland-specific plant functional types, incorporation of an inundation stress mechanism, and deceleration of decomposition under inundation. The new model, LPJ-WHy v1.2, was used to simulate net ecosystem production (NEP), net primary production (NPP), heterotrophic respiration (HR), and soil carbon content. Annual peatland NEP matches observations even though the seasonal amplitude is overestimated. This overestimation is caused by excessive NPP values, probably due to the lack of nitrogen or phosphorus limitation in LPJ-WHy. Introduction of permafrost reduces circumpolar (45–90°N) NEP from 1.65 to 0.96 Pg C a−1 and leads to an increase in soil carbon content of almost 40 Pg C; adding peatlands doubles this soil carbon increase. Peatland soil carbon content and hence HR depend on model spin-up duration and are crucial for simulating NEP. These results highlight the need for a regional peatland age map to help determine spin-up times. A sensitivity experiment revealed that under future climate conditions, NPP may rise more rapidly than HR resulting in increases in NEP.

1. Introduction

[2] Peatlands represent an important ecosystem that has so far not been considered in dynamic global vegetation models (DGVMs). Peatlands are an important component of the Earth system because they accumulate carbon over long timescales [Gorham, 1991; Maltby and Immirzi, 1993; Davidson and Janssens, 2006] and they release methane, which contributes significantly to the global greenhouse effect [Denman et al., 2007]. Peatlands have been omitted from DGVMs, not because of a lack of importance, but because they constitute a very different ecosystem to all others: transitional between terrestrial and aquatic. Plants growing in permanently inundated conditions have to cope with different stress factors compared to terrestrial plants [Cronk and Fennessy, 2001; Pezeshki, 2001; Baird and Wilby, 1999]. Permanent inundation slows decomposition processes [Freeman et al., 2001] and promotes the accumulation of carbon in the soil, i.e., peat growth [Clymo et al., 1998].

[3] DGVMs are used not only for present-day carbon cycle studies, but also to study future ecosystem changes and climate feedbacks [Sitch et al., 2008]. It is likely that northern regions, where most peatlands are found at present, will experience the fastest rise in temperature over the course of this century [Christensen et al., 2007; Meehl et al., 2007b]. This region coincides with the region where almost all low-altitude permafrost is found. Freezing of soil is an important feature that should be included in DGVMs, as it determines the availability of liquid water and therefore productivity [Beer et al., 2007]. This is true for peatlands and nonpeatlands alike. Permafrost areas show slower rates of peatland carbon accumulation than nonpermafrost areas [Bridgham et al., 2006; Robinson and Moore, 2000], although this difference could be caused by soil temperature and not by the presence of permafrost per se.

[4] This paper evaluates the vegetation dynamics simulated by a widely used vegetation model, the Lund-Potsdam-Jena DGVM (LPJ) [Sitch et al., 2003; Gerten et al., 2004], modified to include peatlands and permafrost dynamics as new components [Wania et al., 2009]. The main changes to the vegetation dynamics are the introduction of Sphagnum mosses and flood-tolerant C3 graminoids as new plant functional types (PFTs) and the representation of inundation stress, root exudation, and reduced decomposition under anaerobic conditions. Physical land surface processes, governing soil temperature, water table position, active layer depth, and permafrost distribution were described and evaluated in part 1 [Wania et al., 2009]. Here we examine the effect of the new components on net primary production (NPP), heterotrophic respiration (HR), net ecosystem production NEP = NPP − HR [Chapin et al., 2006], and soil carbon dynamics.

2. Model Description

[5] The model we use here is based on the Lund-Potsdam-Jena dynamic global vegetation model version 1.2 [Sitch et al., 2003; Gerten et al., 2004]. A summary of modifications and a detailed description of changes we have made to the physical land surface processes in the model can be found in part 1 [Wania et al., 2009].

2.1. Peatland Plant Functional Types

[6] We added two new PFTs to LPJ, flood-tolerant C3 graminoids (e.g., Carex spp., Eriophorum spp., Juncus spp., Typha spp.), and Sphagnum (Table S1). In the model, these new PFTs occur only on peatlands; we therefore refer to them as “peatland” PFTs. Flood-tolerant C3 graminoids are adapted to inundation through the presence of aerenchyma, gas-filled root, stem and leaf tissue through which oxygen is transported downward to the root system. In this way the plants avoid anoxia, which would otherwise lead to accumulation of toxic substances in the roots [Wheeler, 1999]. Aerenchymatous plants are important for peatland greenhouse gas emissions because a large proportion of the methane emitted to the atmosphere in wetlands is transported via aerenchyma [Frenzel and Rudolph, 1998; Colmer, 2003; Evans, 2003].

[7] Our parameterization of Sphagnum mosses follows Yurova et al. [2007] with one exception: at high Sphagnum water content, moss photosynthetic activity is not limited by moss water content. Instead, photosynthesis can operate at full capacity at high water contents. The reason for this difference is as follows. At least some Sphagnum spp. require high pore water CO2 concentrations for growth; if such high pore water CO2 concentrations are provided, the plants grow well in water [Smolders et al., 2001]. Most studies showing an apparent decline of photosynthesis at high water content have been conducted using deionized water [Silvola and Aaltonen, 1984; Silvola, 1991; Williams and Flanagan, 1996; Titus et al., 1983], which either has very low concentrations of dissolved CO2 or is CO2-free. However, the natural environment for Sphagnum is highly acidic [Clymo, 1984a] with dissolved CO2 concentrations of up to a few mmol L−1 [Roelofs et al., 1984; Smolders et al., 2001], favorable to Sphagnum growth.

[8] Under the assumption that Sphagnum mosses can assimilate dissolved CO2 from pore water, we adjust the CO2 concentration available to the Sphagnum PFT. We use a mean pore water CO2 concentration, χp, of 934 μmol L−1 [Smolders et al., 2001] and assume that the CO2 concentration near the surface of the peat, χacro, is either the pore water concentration, if the water table is at the surface, or a linear mixture between the pore water concentration and the atmospheric CO2 concentration, χa:

equation image

where z is depth in the soil column and 〈WTP〉 is the annual mean water table position. (Note that rainwater was not considered to be an important source of CO2 for Sphagnum mosses.) This approach captures the observation that CO2 concentrations above peat hummocks are elevated (personal observations by L.P.M. Lamers cited by Lamers et al. [1999]) and hence are likely to represent a mixture of atmospheric CO2 and CO2 released from peat pore spaces. In our model photosynthesis calculations, χacro replaces the atmospheric CO2 concentration for Sphagnum.

[9] Sphagnum photosynthesis has been shown to decrease when moss water content decreases [Silvola, 1991; Williams and Flanagan, 1996, 1998]. Here we use water table position as a surrogate for moss water content [Clymo and Hayward, 1982], and we reduce moss gross primary production as a function of water table position. A daily moss photosynthesis capacity, capmoss, is set by

equation image

where acap = (1 − 0.3)/(−150 + 280). The numerical values in equation (2) were chosen so that full growth is maintained down to a water table level of −150 mm (loosely on the basis of Clymo and Hayward [1982]), while a linear decrease in moss productivity is prescribed for WTP < −150 mm. (Different species of Sphagnum show different WTP thresholds below which desiccation sets in; that is, hummock species tolerate low water contents better than lawn species, so that only a coarse approximation is possible in a global modeling context.) The lower water table position limit in LPJ-WHy of −300 mm does not lead to complete desiccation of mosses. A complete cessation of moss photosynthetic activity therefore does not occur.

[10] A growth restriction was also applied to flood-tolerant C3 graminoids. This PFT is typical in peatlands with a constantly high water table. Field experiments have shown that graminoid production increases when WTP rises and decreases when WTP falls [Thormann et al., 1998]. When the water table position falls below −100 mm, we therefore reduce this PFT's monthly gross primary production, GPPm (g C m−2) depending on mean monthly water table position (WTPm):

equation image
equation image

[11] The water table position, WTPm (mm) is negative when below the surface so that the term (WTPm + 100)/200 will be negative and result in a reduction of GPPm.

2.2. Inundation Stress

[12] Flooding of vegetated areas leads to anoxic conditions in the rooting zone, causing accumulation of ethylene within plants and lowered redox potentials which lead to the accumulation of phytotoxins, impeding plant growth [Wheeler, 1999; Pezeshki, 2001]. If not adapted to inundation stress, plants die within a few days under such conditions [Wheeler, 1999; Larcher, 2003]. Each plant functional type in LPJ-WHy is assigned a water table threshold above which they experience inundation stress. Monthly gross primary production is reduced by a monthly inundation stress factor which depends on the number of days the plant functional type has experienced stress and a parameter measuring the maximal survivable duration of inundation, which counts how many days per month a plant functional type can survive under inundation (Table S1).

2.3. Decomposition Rates Under Inundation

[13] Decomposition rates in LPJ-WHy are estimated by

equation image

where k is the actual decomposition rate and k10 is the decomposition rate at 10°C (for exudates, litter, and the slow and intermediate carbon pool). RT is the temperature response function according to Lloyd and Taylor [1994] and Sitch et al. [2003]. For nonpeatland grid points, Rmoist increases with increasing soil moisture w

equation image

reaching its maximum at the water holding capacity. In peatlands, however, the water holding capacity is exceeded as the soil floods. Decomposition rates under anaerobic conditions are slow because anoxia limits phenol oxidase activity and causes phenolic compounds to accumulate [Freeman et al., 2001]. Phenolic compounds suppress the activity of hydrolase, a pivotal enzyme for the degradation of organic material [Freeman et al., 2001]. Ratios of observed aerobic to anaerobic CO2 production, raer:anaer, range from 1.4–2.7 [Segers, 1998]. We therefore calculate the moisture response as

equation image

giving values for Rmoist of 0.37–0.71. After some experimentation and examination of the resulting decomposition rates, we chose a value of Rmoist of 0.35 to be used in equation (4) for peatland grid cells.

2.4. Litter Accumulation

[14] Litter in LPJ is accumulated at the end of the year and added to the litter pool all at once in January. This scheme can result in decreasing decomposition over the year as less litter becomes available in later months. To avoid this effect, the annual freshly accumulated litter in LPJ-WHy is split evenly between all months, and after each month, one twelfth of the newly accumulated litter is added to the existing litter pool. This modification leads to a more homogeneous decomposition rate over the year.

2.5. Soil Carbon Storage

[15] In LPJ, decomposed litter is split into a fraction that goes directly to the atmosphere as heterotrophic respiration and a fraction that goes into soil carbon storage. The soil carbon pool consists of an intermediate and a slow carbon pool with turnover times of 33.3 and 1000 years, respectively, at 10°C [Sitch et al., 2003]. These turnover times are weighted by functions of soil temperature and soil moisture (equation (4)).

[16] The peat column may be separated into the acrotelm and the catotelm [Ingram, 1983]. The main distinction between the two is that the catotelm has constant water content whereas the acrotelm has a fluctuating water content. One consequence of this difference is that the catotelm experiences continuously anaerobic conditions while the acrotelm has periodically aerobic conditions. Decomposition is thus much slower in the catotelm. Peat is partially decomposed organic material produced by the vegetation on top of the acrotelm and slowly pushed downward as more material accumulates. In steady state, the acrotelm maintains its depth [Clymo et al., 1998]. Once peat is below the minimum water table position and is therefore continuously inundated, it becomes part of the catotelm. In order to mimic the carbon transfer from the acrotelm to the catotelm, we assume that the intermediate carbon pool in LPJ-WHy represents the acrotelm and the slow pool the catotelm. In LPJ-WHy, carbon is transferred from the acrotelm (intermediate carbon pool) to the catotelm (slow carbon pool) at a rate of 12 g m−2 a−1 [Clymo, 1984b]. This transfer begins only once the acrotelm has reached a carbon content of 7.5 kg C m−2, which constitutes a fully developed acrotelm. The threshold of 7.5 kg C m−2 is based on the assumptions that the acrotelm is 0.3 m thick and the bulk density of the acrotelm is 50 kg m−3 [Clymo and Hayward, 1982; Clymo, 1984b], corresponding to about 25 kg C m−3 if we assume that 50% of dry matter is carbon (approximately as given by Gorham [1991] and Clymo et al. [1998]).

[17] In addition to this new carbon transfer from the acrotelm to the catotelm, we maintain the carbon input to the slow carbon pool from litter accumulation in LPJ, which in “reality” may correspond to decaying roots. The majority of accumulating, nondecomposing litter goes into the intermediate soil carbon pool; only 1.5% goes into the slow carbon pool.

2.6. Root Exudates

[18] Root exudates are critical for methane production [Chanton et al., 1995; Ström et al., 2005; Christensen et al., 2003] and also important for the terrestrial carbon cycle as, on average, 15–20% of photosynthesized carbon is exuded by plant roots [Whipps and Lynch, 1983; Helal and Sauerbeck, 1984; Hütsch et al., 2002]. We set the fraction of NPP produced as root exudates to 17.5% and subtract this from monthly NPP. The exudates pool has a fast turnover rate, k10(exu), in the range of 1–2 weeks at 10°C. Its overall decomposition rate follows equation (4). Even though methane emissions are not considered in this current manuscript, we wanted to mention the inclusion of root exudates in LPJ-WHy since carbon allocated to root exudates will cause a minor temporal shift of heterotrophic respiration.

2.7. Leaf Dark Respiration

[19] We use a single calculation method for the leaf dark respiration Rleaf throughout the model:

equation image

where Vmax is the maximum rate of CO2 uptake and b is a scaling factor set to 0.015, 0.02, and 0.03 for C3 plants, C4 plants [Haxeltine and Prentice, 1996; Sitch et al., 2003], and mosses [Yurova et al., 2007], respectively.

3. Data and Experiments

[20] Here we describe data and experiments briefly and refer the reader to part 1 [Wania et al., 2009] for a detailed description. LPJ-WHy reads in a soil type classification for each grid cell based on FAO data [Zobler, 1986; FAO, 1991; Sitch et al., 2003] overlain by organic soil carbon data from the IGBP-DIS map [Global Soil Data Task Group, 2000]. Climate data to drive LPJ-WHy were taken from the Climate Research Unit time series data CRU TS 2.1 [New et al., 1999; Mitchell and Jones, 2005]. This data set provides monthly air temperature, cloud cover, total precipitation, and number of wet days for the time period 1901–2002. Atmospheric CO2 concentrations for 1901–2002 were taken from Etheridge et al. [1996] and Keeling and Whorf [2005]. The first 10 years of the CRU data were repeated 100 times to give 1000 years of model spin-up. This ensures that carbon stocks for nonpeatlands are in equilibrium before the model is driven with the data from 1901–2002 (transient run). Carbon stocks in peatlands do not reach equilibrium in 1000 years of spin-up time, as discussed in section 4.4. During spin-up only, we include an extra 0.5 mm d−1 run-on forcing for peatland sites to avoid large vegetation fluctuations caused by climatic oscillations.

[21] Observations of net ecosystem production (NEP) were collected from sites where monthly data were available for a period greater than 1 year before 2003. We chose Degerö in Sweden, Kaamanen in Finland, Mer Bleue in Canada, and Zotino in Russia (Table 123). The free software package R version 2.5.1 was used for statistical analyses [R Development Core Team, 2008]. Future projections by the ECHAM5 GCM were obtained from the World Climate Research Programme's (WCRP's) Coupled Model Intercomparison Project phase 3 (CMIP3) multimodel ensemble [Meehl et al., 2007a]. The sensitivity experiment was run under CO2 concentrations of 380 ppmv and 780 ppmv.

Table 1. Net Ecosystem Exchange Values (Eddy Covariance Technique) Reported in the Literaturea
SiteCoordinatesTypeTime PeriodMeanRangeReference
  • a

    Values in parentheses indicate the length of measurement in days. All data are given as g C m−2 a−1.

  • b

    Site used for seasonal NEP comparison.

  • c

    Chamber measurements only.

Bakchar bog, WSL, Russia56.85°N82.97°Ebog1999−108 Friborg et al. [2003]
Churchill, Canada58.75°N94.07°Wfen1997–1999−28−55/−10Lafleur et al. [2001]
Degerö, Swedenb64.18°N19.55°Emixed acid mire2001–2002−55−61/−48Sagerfors et al. [2008]
Faejemyr, Sweden56.25°N13.55°Etemperate, eccentric bog2005–2006−22±5.4Lund et al. [2007]
FCRN tower, Alberta, Canada54.82°N113.52°Wrich fen2004−144±32Syed et al. [2006]
Kaamanen, Finlandb69.13°N27.28°Esubarctic fen1997–2002−22−56/−1Aurela et al. [2004]
Mer Bleue, Canadac45.4°N75.5°Wraised ombrotrophic bog1999−60 Moore et al. [2002]
Mer Bleue, Canadab45.4°N75.5°Wraised ombrotrophic bog1998–2004−40−110/0Roulet et al. [2007]
SW Ireland51.92°N9.92°Wblanket bog2003–2004−55−61/−49Sottocornola and Kiely [2005]
    Average ± S.D. 59.3 ± 41.4 
Table 2. Summary of Simulated Net Primary Production, Heterotrophic Respiration, and Net Ecosystem Production for the Gerten Version of LPJ, LPJ-WHy Without Peatlands, and LPJ-WHy With Peatlands Weighted by Fractional Peatland Cover in Each Grid Cella
Model VersionNPPHRNEPSoil C
  • a

    Flux values are mean values over 1991–2000 measured in Pg C a−1. Soil carbon content (Soil C) of the intermediate and slow carbon pools is given as Pg C.

Gerten17.0615.411.65891
LPJ-WHy16.7615.800.96930
LPJ-WHy peat16.2215.181.04968
Table 3. Correlation of Monthly Modeled Versus Observed NEPa
SiteEquationPearson's rp
  • a

    M, monthly modeled NEP; O, observed NEP.

DegeröO = 5.06264 + 0.25882 M0.78<0.001
KaamanenO = −0.0855 + 0.27403 M0.77<0.05
ZotinoO = 0.79806 + 0.29014 M0.85<0.001
Mer BleueO = 5.3029 + 0.1789 M0.41<0.01

4. Results and Discussion

[22] As DGVMs are principally designed for application at the regional to global scale [Smith et al., 2001; Thonicke et al., 2001; Sitch et al., 2003; Gerten et al., 2004], we focus on testing the regional performance of LPJ-WHy. In particular, we concentrate on the evaluation, seasonality, and sensitivity of NEP. We note that the only way to determine if an ecosystem has a positive or a negative land-to-atmosphere carbon exchange is by including all of the carbon species involved in the exchange, i.e., carbon monoxide, methane, dissolved organic and inorganic carbon, volatile organic compounds, and lateral transfer of particulate carbon [Chapin et al., 2006]. Chapin et al. refer to the sum of all of these fluxes as Net Ecosystem Carbon Balance (NECB). Ideally, to simulate the terrestrial carbon balance a model would take account of all of these fluxes. Methane flux has been included into a later version of LPJ-WHy [Wania, 2007], and work to include isoprene into the model is underway (P. N. Foster, personal communication, 2008), but we are still far from being able to simulate NECB comprehensively. Here we focus exclusively on CO2 exchange.

4.1. Effects of Spin-up Procedures

[23] Peatland formation requires inundation, either caused by stagnation of water, flooding rivers, or terrestrialization of shallow water bodies and lakes [Wieder, 2006]. To reflect these conditions, in peatland grid cells, LPJ-WHy considers the catotelm to be continuously inundated from the beginning. For the acrotelm, however, there are two possibilities. One can begin the simulation with either a water-filled acrotelm or an empty acrotelm. We use an initially empty acrotelm. This slightly unrealistic initial condition and the necessary, but again unrealistic, model spin-up procedure causes problems at some sites, problems that are alleviated by the application of an extra, artificial source of water during the spin-up period only. This additional source of water can be thought of as a run-on term. The difference in water table positions due to this artificial run-on can be seen in Figure S1. (Results from the transient run are not affected.) Figure S2 shows how the added run-on affects NPP of the peatland-specific PFTs at two sites described in part 1 [Wania et al., 2009]. At Salmisuo, run-on during spin-up affects NPP during spin-up, but not later. At BOREAS, on the other hand, flood-tolerant C3 graminoids do not grow when no run-on is added during spin-up, and this problem seems to persist into the transient period (Figure S2), although total NPP is not significantly affected. NEP shows no significant effect of spin-up method on the transient period (not shown).

4.2. Net Primary Production

[24] Distributions of simulated NPP values for flood-tolerant C3 graminoids and Sphagnum mosses in circumpolar regions are highly skewed. The flood-tolerant C3 graminoids have a modeled median NPP value close to zero, but a mean value of 97 g C m−2 a−1. Modeled Sphagnum NPP values are similarly skewed, but with a mean of 159 g C m−2 a−1. Nungesser [2003] summarized NPP values for ten Sphagnum species. Our mean value of 159 g C m−2 a−1 (i.e., 378 g dry matter m−2 a−1) lies at the upper end of the range of these values, but is not outside the observed range. The majority of peatlands in our study show a modeled NPP of <600 g C m−2 a−1, but higher values are found in Great Britain and Ireland.

[25] A histogram of modeled NPP values shows one peak at 0–20 g C m−2 a−1 and a second peak at 200–220 g C m−2 a−1 for both PFTs (not shown). The peak close to zero can be explained by the absence of vegetation in many grid cells in the far north characterized by constantly low water table positions and harsh climatic conditions. Combined NPP for flood-tolerant C3 graminoids and Sphagnum shows a peak at zero, a second peak at 180–200 g C m−2 a−1 and a plateau from 320–460 g C m−2 a−1. Observations of NPP for three microsites at Mer Bleue, Canada, gave average aboveground NPP values of 140–170 g dry matter m−2 a−1 for vascular plants in a poor fen and a bog [Moore et al., 2002]. Moore et al. estimated Sphagnum NPP as between 140 and 225 g dry matter m−2 a−1 and total aboveground NPP as between 290 and 360 g dry matter m−2 a−1. Assuming a carbon content of 50%, we obtain a range of 145–180 g C m−2 a−1 for total aboveground NPP. Wieder [2006] indicated that total aboveground NPP for North American boreal and sub-boreal bogs and fens were 462 and 319 g dry matter m−2 a−1, respectively, i.e., approximately 231 and 160 g C m−2 a−1, only slightly higher than Moore et al.'s estimates. LPJ-WHy produces larger numbers but includes belowground NPP; as Moore et al. [2002] found, belowground biomass in peatlands is typically 4 times as large as aboveground biomass. How this ratio translates to an aboveground-belowground NPP ratio depends on the relative turnover times of aboveground and belowground biomass.

[26] Despite this observation, it remains possible that the model overestimates NPP. Photosynthesis rates in peatlands typically do not reach their theoretical optima for the prevailing light intensity and water availability [Arneth et al., 2006], perhaps because nitrogen or phosphorus limitation or accumulation of phytotoxins [Pezeshki, 2001; Cronk and Fennessy, 2001] may limit photosynthetic capacity. Bog and fen species from a study site in Maryland, United States, showed lower leaf nitrogen concentrations than terrestrial plants of the same group [Aerts et al., 1999], pointing to a potentially decreased Rubisco concentration, which could be parameterized in the model following the approach by St-Hilaire et al. [2008]. Xu-Ri and Prentice [2008] showed that consideration of the effect of nitrogen uptake on the carbon cycle helped to bring modeled forest NPP in boreal regions closer to observations. In peatlands, where water is abundant but nitrogen is often in short supply [e.g., Malmer and Wallen, 2005; Bragazza et al., 2003, 2005], it may similarly be worthwhile to include nitrogen limitation in order to better simulate NPP (for comparison, see Stöckli et al. [2008]).

4.3. Net Ecosystem Production

4.3.1. Regional NEP Estimates

[27] Figure 1 shows NEP for the circumpolar region north of 45°N for four different model configurations. “LPJ Gerten” (Figure 1a) refers to the Gerten et al. [2004] version of LPJ on which we based our model development. The Gerten et al. model does not include peatlands or permafrost. Figure 1b shows the effect of including permafrost. Decreases in NEP are observed throughout Siberia and in northern Canada and Alaska. The decreases in central and eastern Siberia are mainly due to a reduction in NPP of boreal summergreen needleleaf trees (i.e., Larix spp.). This change can be explained by reductions in soil water availability when soil water is allowed to freeze [Beer et al., 2007].

Figure 1.

Annual simulated net ecosystem production (NEP) averaged over 1991–2000. (a) Gerten et al. [2004] version of LPJ, (b) all grid cells without peatland hydrology switched on, (c) peatland grid cells only, unweighted, and (d) overall NEP, combining generic and peatland values by weighting by fractional cover of peatlands. Regional summed values of NEP, where NEP = NPP − HR, for Figures 1a, 1b, and 1d are listed in Table 2.

[28] Figure 1c shows NEP for grid cells with organic soil. The highest peatland NEP was modeled in eastern Europe, the southern fringe of the West Siberian Lowlands, central Canada, and eastern Alaska. However, not all peatlands showed positive NEP values. Figure 1d shows the combined effect of including permafrost and peatlands. To produce this map, the model was run twice, once with and once without peatland hydrology. Overall results were calculated as a weighted average of the peatland and nonpeatland values, weighted by the fractional peatland cover for each grid cell (see part 1 [Wania et al., 2009, Figure S2]). The overall effect of adding peatlands to the permafrost is an increase in NEP compared to permafrost alone (Figure 1b versus Figure 1d) and a decrease in NEP compared to peatlands alone (Figure 1c versus Figure 1d).

[29] Figure 2 shows the frequency distribution of modeled and observed annual NEP for peatlands (observations as listed in Table 1). The simulated values peak at an NEP of −5–0 g C m−2 a−1, but then extend towards higher values of 20–50 g C m−2 a−1. The observations (based on CO2 flux measurements) have a median of 40 g C m−2 a−1. This comparison suggests that LPJ-WHy is capable of representing reasonable annual NEP values for peatlands.

Figure 2.

Histogram of annual NEP for peatland grid cells. References for observations (values shown as ticks below main histogram plot) are listed in Table 1.

4.3.2. Interannual Variability and Trends

[30] Figure 3 illustrates the sensitivity of modeled carbon fluxes to climate variability. The mean annual NPP of all grid cells for the 45–90°N region over the 1901–2002 time period is 14.9 Pg C, with a range from 13.6 to 17.0 Pg C. A clear positive trend over the century is visible

equation image

where NPP is given in Pg C a−1 and y is the calendar year. This trend is significant at the 99% level: significance levels are 99% or greater throughout the paper, unless otherwise stated. The positive trend is at least partially caused by the rise in atmospheric CO2 concentration over the course of the simulation. However, NPP is correlated not only with atmospheric CO2 concentration, but also with soil temperature (correlation not shown). We conducted an experiment in which atmospheric CO2 concentrations were held constant at 321 ppmv, i.e., the mean over 1901–2002. This resulted in a weaker NPP trend:

equation image

with NPP again in Pg C a−1. These results indicate that temperature plays only a minor role in the modeled NPP trend.

Figure 3.

Interannual variability of net primary production (NPP), heterotrophic respiration (HR) and net ecosystem production (NEP) from 45–90°N, and soil temperature at 25 cm depth for the time period 1901–2002.

[31] Between 1901 and 2002, NEP fluctuated between around 0 and 1.7 Pg C a−1, with a mean of 0.88 Pg C a−1. Values for the global biosphere are estimated to be approximately 0.3 ± 0.9 Pg C a−1 for the 1980s [Prentice et al., 2001] and 1.0 ± 0.6 Pg C a−1 for the 1990s [Prentice et al., 2001; Denman et al., 2007]. Sitch et al. [2008] reported results from five DGVMs, finding a mean global NEP minus fire carbon flux over the 1990s of 1.5–2.8 Pg C a−1. LPJ-WHy's circumpolar mean NEP for the 1980s is 1.04 Pg C a−1, of which 0.8 Pg C a−1 is lost because of fire. Peatlands contribute 0.14 Pg C a−1 to the circumpolar NEP, but fire consumes 0.058 Pg C a−1, and the balance of accumulation and fire is 0.082 Pg C a−1. Clymo et al. [1998] estimated the current carbon sequestration (i.e., NEP minus fire flux) by northern peatlands to be 0.067 Pg C a−1.

[32] In contrast to NPP, NEP does not show a significant trend (Figure 3). As NPP rises, heterotrophic respiration also rises, although with a time lag, keeping NEP more or less stable. Heterotrophic respiration is correlated to CO2 concentration, soil temperature, and NPP, whereas NEP is not correlated to any of these factors (not shown). However, when keeping atmospheric CO2 constant over the course of the simulation, we observe a decline in NEP from about 1970 onwards (Figure S4), brought about by a stronger increase in HR compared to NPP. If these findings are realistic, they could have profound implications for the carbon balance in northern regions. In the future, photosynthesis will reach a saturation point at which additional CO2 will not further enhance NPP, but as temperatures continue to rise, HR will also rise. When HR reaches a point where it starts to exceed NPP, NEP will become negative. Although this study does not consider the full NECB [Chapin et al., 2006], it is clear that once NEP becomes negative, the total carbon balance will certainly be negative.

4.3.3. Seasonality

[33] Observed and modeled monthly NEP from four peatland sites are compared in Figure 4 (see Table 3 for the relevant statistics). There is a general overestimation of the amplitude of seasonal NEP variation in the model. The NEP observations range from −20 to 50 g C m−2 month−1, whereas the modeled NEP ranges from −90 to 120 g C m−2 month−1. The overestimation of the NEP range would be expected if NPP is overestimated. On the one hand, higher NPP values would contribute to greater positive NEP values in summer by sequestering more CO2. On the other hand, higher NPP would also lead to greater negative NEP in winter as more carbon is available for heterotrophic respiration. However, NEP is underestimated in August and September 1998 and 2001 and September 2002, when the modeled water table position was at its lower boundary of −300 mm for the Mer Bleue site. The simulated water table positions show a difference of 400 mm between their maximum value in spring and their lowest position in autumn, and this change is comparable to the observed drop of water table position [Lafleur et al., 2005]. LPJ-WHy may thus exaggerate the drop in NPP because of a lower water table in a dry summer. The same pattern can be found at the Degerö site for September 2002, when the water table dropped leading to a reduction in NPP, while HR stayed high, causing modeled NEP to plummet. Even though studies have examined the effects of flooding on the physiology of peatland plants [Pezeshki, 2001; Baird and Wilby, 1999, and references therein], few have looked at the effect of intermittent drought on peatland plants [Li et al., 2004]. Sphagnum mosses are likely to experience drought stress relatively quickly [Clymo and Hayward, 1982], but graminoids have roots that enable them to access water from deeper soil layers [Bernard and Fiala, 1986; Kohzu et al., 2003] and may therefore cope better with lower water tables. More experimental work is needed to relate photosynthesis rates to fluctuations in water table position.

Figure 4.

Observed versus modeled monthly NEP for four sites: Degerö, Sweden (2001–2002), Kaamanen, Finland (average over 1997–2002), Zotino, Russia (1998–2000), and Mer Bleue, Canada (1998–2002).

4.4. Soil Carbon

[34] Figure 5 shows differences in soil carbon content between the same four model configurations as used in Figure 1. In North America, inclusion of permafrost causes a slight southward shift of high-carbon soils in arctic Canada and of soils with medium carbon content in southern central Canada. In Eurasia, the changes are larger, with higher carbon content in northern Scandinavia and the West Siberian Lowlands. The areas of Russia with low soil carbon become more extensive because of the effect of soil freezing on water availability [Beer et al., 2007]. However, with the model peatland hydrology enabled, we find high soil carbon contents in most peatlands (Figures 5c and 5d) because of the deceleration of decomposition. The modeled soil carbon content of 40–50 kg C m−2 is lower than estimates of 96–133 kg C m−2 reported in the literature [Gorham, 1991; Turunen et al., 2002]. The explanation for the lower soil carbon contents lies in the relatively short spin-up time of LPJ-WHy.

Figure 5.

Annual simulated soil carbon averaged over 1991–2000. (a) Gerten et al. [2004] version of LPJ (891 Pg C), (b) all grid cells without peatland hydrology switched on (930 Pg C), (c) peatland grid cells only, unweighted (427 Pg C), and (d) overall soil carbon, combining generic and peatland values by weighting by fractional cover of peatlands (968 Pg C). Values in parentheses are the total carbon stored in soils from 45–90°N.

[35] Soil carbon in nonpeatland grid cells reaches equilibrium after a few hundred years, so that a spin-up time of 1000 years is appropriate. However, peatlands accumulate carbon over long timescales and reach equilibrium only when carbon gained equals carbon lost [Clymo, 1984b]. The critical factor in this balance is the catotelm. As carbon accumulates in peatlands, the catotelm deepens. Decomposition in the catotelm is slow, but does occur, and once the catotelm is thick enough, it contributes enough respiration to counterbalance soil carbon gains. LPJ-WHy was run for a spin-up time of 90,000 years, plus 102 years of transient run, for five sites differing in mean annual temperature. Figure 6 shows the evolution of soil carbon over 90,102 years at these sites. Michigan, the warmest site, reaches 95% of its equilibrium value (40 kg C m−2) after less than 10,000 years. Ayach-Yakha, the coldest site, takes almost 50,000 years to reach 95% of its equilibrium value of >200 kg C m−2. The correlation between mean annual air temperature, Ta (°C), and the final soil carbon content, Csoil (kg C m−2), is

equation image
Figure 6.

Soil carbon content for five sites through 90,102 years of simulation. Values in parentheses are mean annual air temperature in °C.

[36] The correlation between mean annual air temperature and time until 95% of the equilibrium soil carbon content is reached τ95 (years) is

equation image

[37] These correlations indicate a very close relationship between air temperature and soil carbon accumulation rate, in agreement with Clymo et al. [1998], who analyzed data from almost 800 sites in Finland and found a clear decrease in accumulation rates from south to north.

[38] Carbon accumulation in LPJ-WHy is influenced by the rate of flux of carbon from the acrotelm to the catotelm (see section 2.5), currently set to 12 g C m−2 a−1 [Clymo, 1984b]. In addition, a small fraction of freshly accumulated litter (1.5%) is added to the slow carbon pool. A commonly used term in the peatland research community is the “long-term apparent rate of carbon accumulation” (LORCA). Reported values of LORCA (in g C m−2 a−1) are 14.0 ± 3.7 to 21.9 ± 2.8 for the interval from 400 to 3000 years ago for two cores at Mer Bleue, Canada [Roulet et al., 2007], and 18.5 for undrained Finnish mires [Turunen et al., 2002], where the lowest values were found in fens (16.9), and the highest values were found in raised bogs (26.1). The LORCA values for Mer Bleue compared well to a 6-year-long mass balance study resulting in a carbon sink of 21.5 ± 39.0 g C m−2 a−1 [Roulet et al., 2007]. However, Tolonen and Turunen [1996] found that the true rate of carbon accumulation is only about 60% of LORCA, which would bring the rate of carbon accumulation down to 8.4–13.1 g C m−2 a−1, similar to the value of 12 g C m−2 a−1 used in this work.

[39] In summary, LPJ-WHy can accumulate peat realistically in the form of carbon storage in the slow carbon pool. The accumulation continues until a stable state is reached, at which point the contribution of heterotrophic respiration is large enough that the NEP is close to zero. The time taken for a system to approach this steady state depends on the mean annual air temperature. None of the five sites tested above were in soil carbon equilibrium after the standard spin-up time of 1000 years. One key unknown prevents us from modeling total column carbon content more realistically, namely the time since peat initialization. What we need is a complete gridded data set of peat initialization similar to that shown by MacDonald et al. [2006]. Further work in this direction is underway, and spatial data may be available to the modeling community soon (P. Oksanen, personal communication, 2008).

[40] The estimate of carbon stored in peatland soil is 270–500 Pg C [Gorham, 1991; Maltby and Immirzi, 1993; Davidson and Janssens, 2006; Turunen et al., 2002], while LPJ-WHy simulates 427 Pg C. However, the peatland map used for our study overestimates peatland area by a factor of 1.67 relative to Aselman and Crutzen [1989] and 2.61 relative to the map of Prigent et al. [2007]. If we correct the peatland area downwards, we obtain a corresponding circumpolar peatland soil carbon content of 164–256 Pg C. These values are at the lower limit of current estimates, but are derived from simulations where spin-up time was limited to 1000 years. We expect soil carbon content to rise as the spin-up time lengthens until we reach equilibrium conditions at all points. We thus expect our estimate of total carbon in peatlands to be somewhat low.

4.5. Effect of Spin-up Duration on NEP

[41] The length of the spin-up procedure influences NEP by increasing heterotrophic respiration from the growing slow carbon pool. The length of the spin-up period matters only until the soil carbon has reached equilibrium. For nonpeatland sites, an equilibrium is established after 1000 years by the use of an analytical solution for the exponential carbon accumulation. In order to test the potential bias of peatland NEP when limiting spin-up to 1000 years, the same spin-up comparison simulations used to produce Figure 6 were used to calculate NPP, HR, and NEP values for comparison between the 1000 year and 90,000 year spin-up approaches (Table S3). NEP did not differ significantly between the 1 k and 90 k runs at the three warmer sites (Michigan, Salmisuo, and Degerö), but showed greater differences at the two colder sites (BOREAS and Ayach-Yakha). For the two latter sites, a significant difference was found in HR but not in NPP. This is as expected since HR depends on carbon pool size while NPP does not.

[42] For sites in colder areas, it will therefore be important to consider the site's history when assessing simulated NEP. The BOREAS site has positive NEP after the 1k spin-up, but is close to zero after the 90 k spin-up. The Ayach-Yakha site is more extreme, switching from a strong positive NEP to a negative NEP from the 1 k to the 90 k spin-up.

4.6. Sensitivity Experiment

[43] In order to test the sensitivity of vegetation dynamics to changes in climate, six sites were chosen, and LPJ-WHy simulations were performed for different climate conditions, independently increasing the monthly mean air temperature and monthly precipitation across ranges selected to correspond to changes projected by an ECHAM5 GCM simulation forced by the IPCC SRES A2 emissions scenario. The results of experiments using an atmospheric CO2 concentration of 780 ppmv are shown in Figure 7, and for comparison, results of experiments using CO2 concentration of 380 ppmv are shown in Figure S5. The results may be examined in conjunction with the land surface process sensitivity experiments in part 1 [Wania et al., 2009].

Figure 7.

Sensitivity of net primary production (NPP), heterotrophic respiration (HR), and net ecosystem production (NEP) at six sites (Table 6 in part 1 [Wania et al., 2009]) in different zones of projected temperature increase under an atmospheric CO2 mixing ratio of 780 ppmv. (a) Temperature change projected by the ECHAM5 GCM under the IPCC SRES A2 emissions scenario: this was derived by averaging the 30-year mean for 2071–2100 over the three ECHAM5 ensemble runs and subtracting the 30-year mean over the three ensembles for 1961–1990. (b–d) Absolute values of (Figure 7b) net primary production, (Figure 7c) heterotrophic respiration, and (Figure 7d) net ecosystem production as air temperature and precipitation change. The dots on Figures 7b–7d indicate climate conditions at each site for the year 2100 projected by ECHAM5 GCM simulations driven by the IPCC SRES A2 scenario. The dashed line shows a possible trajectory from present-day to projected 2100 conditions, assuming that temperature and precipitation change in concert.

[44] Figure 7a shows projected changes of mean annual air temperature by the end of this century under the SRES A2 scenario. This was used to guide the setup of the sensitivity experiment. The six sites chosen are located in areas that experience temperature increases of 4 to >7°C. Figure 7b shows the change in NPP under precipitation (ΔP) and temperature increases (ΔT). The starting NPP values (bottom left corner of each individual panel) are under 300 g C m−2 a−1 for sites 1, 4, and 6, 300–400 g C m−2 a−1 for sites 3 and 5, and over 600 g C m−2 a−1 for site 2. These results are obtained using an atmospheric CO2 concentration of 780 ppmv and are therefore somewhat higher than present-day values. When using an atmospheric CO2 concentration of only 380 ppmv, the starting NPP for site 2 is between 500 and 600 g C m−2 a−1 (Figure S5). Higher NPP values at 780 ppmv are mainly caused by an increase in vascular plant NPP, whereas Sphagnum mosses either show a slight increase or a decrease due to complex interactions between Sphagnum moss growth and water table position or due to competition with other PFTs.

[45] Sites 1, 3, 4, and 5 show an increase in NPP as temperature rises, but increases in precipitation have only a minor impact. In contrast, site 2 shows a decline in NPP followed by an increase as temperatures are increased under conditions of constant precipitation. However, when temperature and precipitation increase together, the precipitation changes outweigh the effects of rising temperatures and NPP remains constant (dashed lines in Figure 7, which connect starting conditions with end conditions as projected for 2100 by ECHAM5). This pattern observed for site 2 reflects the changes in water table position shown in part 1 [Wania et al., 2009]. The increase of NPP above 6°C at 0% ΔP and at 9°C with 10% ΔP is due to a change in vegetation. While ΔT is low or ΔP is high, water tables remain high and support peatland PFTs. However, when ΔT reaches 6°C without any increase in precipitation, water tables drop far enough to allow tree or C3 grass PFTs to survive. This transition in vegetation is also seen at site 6, where a diagonal zone of lower net productivity separates two zones of higher NPP, which again tallies nicely with the pattern of water table positions in part 1 [Wania et al., 2009].

[46] Heterotrophic respiration, shown in Figure 7c, behaves similarly to NPP. Variations in water table position do not influence these results, as the moisture response in peatlands in LPJ-WHy is fixed (equation (6)).

[47] The variable of most interest is NEP, shown in Figure 7d. Sites 3 and 4 increase from a “neutral” NEP to a clearly positive NEP along the trajectory of future climate conditions, as indicated by the dashed lines in each panel. Site 5 appears to be stable over almost the entire temperature and precipitation range examined, with a decrease in NEP under only the warmest and driest conditions. Site 1 shows a strong increase in NEP with temperature from 100–150 g C m−2 a−1 to over 200 g C m−2 a−1. NEP decreases with increasing precipitation, but at the greatest ΔP, values increases again. This pattern can be explained by a rising water table which enables the growth of peatland PFTs (not shown), leading to lower NEP because the NPP of peatland PFTs is lower than that of generic PFTs (indirectly seen in Figure 7b by the decrease in NPP at ΔT of 0–2°C and ΔP of 30–50%.). Sites 2 and 6 also experience a change in vegetation composition due to a large drop in water table (see Figure 7 in part 1 [Wania et al., 2009]). Under cooler and wetter conditions, when water tables are high, peatland PFTs thrive, while under warmer and drier conditions, when water tables are low, generic PFTs grow. Interestingly, NEP is higher when generic PFTs are present (top left corner of sites 2 and 6 in Figure 7d). This effect is caused by slightly higher NPP combined with lower HR. The model's behavior in simulating lower respiration under lower water table conditions is plausible, as lower water tables coincide with lower soil temperatures (not shown), reducing the rate of respiration. Finally, we would like to add a word of caution concerning the interpretation of the sensitivity results. Model PFTs experience increases in CO2, temperature, and precipitation, without any limitation in mineral nutrients. In the real world, plant growth is limited primarily by nitrogen and phosphorous availability, and it is questionable whether nutrient supply under changing environmental conditions will keep up with plant demand.

5. Conclusions

[48] The introduction of permafrost into LPJ-WHy had a strong impact on the simulation of NEP and soil carbon storage in some of the coldest areas of the circumpolar region, where the lack of freezing water permitted too much productivity in previous model versions. The addition of peatlands further reduced NPP, but had only a small positive effect on circumpolar NEP. Both permafrost and peatlands increased the soil carbon content of the circumpolar region, a crucial element for future projections of the global terrestrial carbon cycle. There is much uncertainty concerning the future of the terrestrial carbon sink [Friedlingstein et al., 2006]. Inclusion of permafrost and peatlands into models will have implications for the simulation of this carbon sink.

[49] LPJ-WHy is able to simulate the seasonality of NEP realistically, but not its amplitude. We speculate that this is due to overestimation of NPP, possibly due to lack of nutrient limitation in the model. Photosynthesis is generally lower in plants growing in peatlands compared to similar plants growing in other ecosystems [Arneth et al., 2006]. This is presumably caused by lack of nutrients, most likely nitrogen, lack of oxygen, accumulation of phytotoxins, or by a combination of these factors [Pezeshki, 2001; Cronk and Fennessy, 2001]. Long-term trends in NPP and heterotrophic respiration are visible over the last third of the 20th century. The increase in NPP is mainly driven by increasing atmospheric CO2 concentrations, while the increase of heterotrophic respiration is caused by rising temperatures.

[50] Modeled annual NEP of peatlands was shown to lie within the range of observations. This means that our simple future sensitivity experiment has some validity, despite deficiencies in the current model. The general trend projected by our experiment is towards higher NPP and heterotrophic respiration and slightly higher NEP in a warmer, high-CO2 climate. The addition of methane emissions from peatlands is the next step in LPJ-WHy development [Wania, 2007] and will provide further insight into the changing carbon and greenhouse gas balance of boreal peatlands.

Acknowledgments

[51] We thank Nigel Roulet, Peter Lafleur, and Elyn Humphreys of the NSERC-CFCAS/Fluxnet Canada (Canadian Carbon Project Mer Bleue observatory) for the provision of NEE data for the Mer Bleue site. We thank Mats Nilson and Jörgen Sagerfors for providing NEE data for the Degerö site, funded by Swedish Research Council for Environment (grant 621-2001-1911), Agricultural Sciences and Spatial Planning (grant 21.4/2003-0876), and the Kempe Foundation for grants for the micrometeorological instruments. We acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and the WCRP's Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multimodel data set. Support of this data set is provided by the Office of Science, U.S. Department of Energy. We acknowledge the valuable input by Paul Miller on code development and manuscript preparation. We thank Renato Spahni for help with the provision of the IGBP-DIS soil carbon data, Angela Gallego-Sala for discussion of the manuscript, Paul Valdes for the provision of computer facilities time, and Gethin Williams for IT support. R.W. was sponsored by a studentship of the Department of Earth Sciences, University of Bristol, and by the EU-Project HYMN (GOCE-037048). I.R. was supported by a NERC e-Science studentship (NER/S/G/2005/13913). The manuscript has greatly improved after reviews by Nigel Roulet and an anonymous reviewer.

Ancillary