Integrating peatlands and permafrost into a dynamic global vegetation model: 1. Evaluation and sensitivity of physical land surface processes


  • This is a commentary on DOI:


[1] Northern peatlands and permafrost soils are associated with large carbon stocks. Rising temperatures are likely to affect the carbon balance in high-latitude ecosystems, but to what degree is uncertain. We have enhanced the Lund-Potsdam-Jena (LPJ) dynamic global vegetation model by introducing processes necessary to simulate permafrost dynamics, peatland hydrology, and peatland vegetation. The new version, LPJ-WHy v1.2, was used to study soil temperature, active layer depth, permafrost distribution, and water table position. Modeled soil temperatures agreed well with observations, apart from a Siberian site where the soil is insulated by an extensive shrub layer. Water table positions were generally in the range of observations, with some exceptions. Simulated active layer depth showed a mean absolute error of 44 cm when compared to observations, but the error was reduced to 25 cm when the soil type for seven sites was manually corrected to mirror local conditions. A sensitivity test, in which temperature and precipitation were varied independently, showed that soil temperatures and active layer depths increased more under higher temperatures when precipitation was increased at the same time. The sensitivity experiment suggested persisting wet conditions in peatlands even under temperature increases of up to 9°C as long as annual precipitation is allowed to increase with temperature to the extent indicated by climate model experiments.

1. Introduction

[2] Dynamic global vegetation models provide a versatile platform for studying the interactions between vegetation, the carbon (C) cycle and climate [Potter and Klooster, 1999; Thonicke et al., 2001; Smith et al., 2001; Sitch et al., 2003; Gerten et al., 2004; Thonicke et al., 2005; Kucharik et al., 2006; Schaphoff et al., 2006; Sato et al., 2007; Beer et al., 2007; Sitch et al., 2008]. However, these models have not until now simulated peatland ecosystems, which cover about 3 × 106 km2 north of 40°N [Matthews and Fung, 1987; Aselman and Crutzen, 1989]. Peatlands are important for the carbon cycle because they have accumulated 270–500 Pg C in boreal and subarctic regions through the course of the Holocene [Gorham, 1991; Maltby and Immirzi, 1993; Davidson and Janssens, 2006; Turunen et al., 2002] and they contribute significantly to global methane emissions [Chen and Prinn, 2006; Bousquet et al., 2006]. The estimated overall present day radiative effect of peatlands on the climate is between −0.2 and −0.5 W m−2, i.e., a net cooling [Frolking and Roulet, 2007].

[3] Peat is partially decomposed organic matter. Its accumulation depends on water table position, soil temperature, and net primary production, with water table position being most important [Rouse et al., 1997]. Peat deposits form under inundated anaerobic conditions where decomposition is slower than accumulation [Maltby and Immirzi, 1993; Rouse et al., 1997]. The balance between peat accumulation and decomposition varies with the spatial and temporal extent of inundation. Permafrost formation can either impede drainage, leading to increased inundation and therefore to peat accumulation, or permafrost establishment can lead to drier conditions and reduced carbon accumulation rates [Robinson and Moore, 2000]. The presence or absence of permafrost is thus an important factor in peatland dynamics. In North America, 36% of the total peatland area (513,000 km2) is underlain by permafrost [Bridgham et al., 2006]. In turn, peat also affects permafrost dynamics through changes in the thermal properties of soils [Nicolsky et al., 2007]. Bridgham et al. [2006] reviewed the existing literature and found that permafrost peatlands in North America store 53.5 Pg C, and nonpermafrost peatlands store 124.6 Pg C, with current sequestration rates of 6.6 and 22.6 Tg C a−1, respectively. Using the areal information provided by Bridgham et al. [2006], we derive carbon accumulation rates of 12.9 g C m−2 a−1 for permafrost peatlands and 25.3 g C m−2 a−1 for nonpermafrost peatlands.

[4] Projections for the 21st century indicate that northern high-latitude ecosystems and carbon budgets will be disproportionately affected by temperature increases [Christensen et al., 2007; Meehl et al., 2007b]. Current climate projections suggest a mean annual temperature rise of 3.5–7.5°C and a very likely increase in winter and a possible increase in summer precipitation in northern high latitudes [Walsh et al., 2005; Meehl et al., 2007b; Christensen et al., 2007]. The increase in atmospheric CO2 concentrations by the end of the century will depend on measures taken to mitigate the current trend in emissions, and also on feedback responses in the climate system, to which peatlands and permafrost are almost certain to contribute [Friedlingstein et al., 2006; Lemke et al., 2007].

[5] In order to capture these feedback mechanisms in climate and vegetation models, the first step is to include permafrost and peatlands into the models. Some land-surface models, terrestrial ecosystem models, and global climate models have already incorporated permafrost processes and representations of organic soils and wetland hydrology [Comer et al., 2000; Zhuang et al., 2001, 2004; Nicolsky et al., 2007; Lawrence and Slater, 2007]. Dynamic global vegetation models, which fully couple ecosystems and physical processes, have so far only incorporated an empirical relationship between climate and permafrost, and no peatland hydrology [Beer et al., 2007].

[6] We present a first approach to including soil thermal dynamics, peatland hydrology, and peatland-specific plant functional types into a dynamic global vegetation model. In this first paper, we focus on the physical factors influencing the carbon cycle at high latitudes. A companion paper considers vegetation and carbon cycle processes [Wania et al., 2009]. We use the Lund-Potsdam-Jena dynamic global vegetation model [Sitch et al., 2003; Gerten et al., 2004], incorporating the necessary physical processes to model peatlands and permafrost. We evaluate modeled land surface processes against observations and test the sensitivity of the key outputs (water table position, soil temperature, and active layer depth) to changes in climate.

2. Model Description

[7] Our starting point for model development was the Fortran 77 version of the Lund-Potsdam-Jena dynamic global vegetation model (LPJ) [Sitch et al., 2003], with modifications to the hydrology scheme as reported by Gerten et al. [2004] (Version 1.2).

[8] LPJ simulates vegetation dynamics at a global, regional, or site scale by modeling atmosphere-vegetation carbon and water fluxes, plant physiology, phenology, establishment, mortality, and fire. Processes are simulated on a daily, monthly, or annual time step as appropriate. Plant functional types are defined using physiological parameters influencing growth and bioclimatic limits for establishment and mortality. As input data, the model requires monthly climate data (air temperature, total precipitation, fractional sunshine hours, and wet days), soil type information, and an annual value of the global atmospheric carbon dioxide mixing ratio.

[9] The model was modified as follows:

[10] 1. We implemented a soil temperature solver to simulate temperature changes as a function of depth. This implementation enables us to determine active layer depths and the location of surface permafrost, i.e., permafrost within the top 2 m of soil.

[11] 2. The LPJ hydrology module was split into two parts: for nonpeatland sites, we modified the scheme of Gerten et al. [2004] to enable the simulation of freeze-thaw cycles and to adapt the model to more soil layers than previously used; for organic soils, we use a new parameterization appropriate for peatlands that captures the diplotelmic nature of peat, i.e. an acrotelm and catotelm [Ingram, 1978].

[12] 3. The treatment of surface snowpack was modified to include some of the effects of snow ageing. Snow density now varies with snow age instead of being fixed, allowing for changes in thermal insulation properties.

[13] 4. Two new plant functional types, namely flood-tolerant C3 graminoids and Sphagnum mosses were added to the ten existing plant functional types (see part 2 [Wania et al., 2009]).

[14] 5. An inundation stress mechanism was introduced, reflecting the decrease in productivity experienced by non-flood-adapted plant functional types under anoxic conditions.

[15] 6. Root exudates were added to LPJ. A fraction of the net primary production is put into an exudates pool, which is then either respired as CO2 or (in a later version of the model) used to generate methane.

[16] 7. Decomposition rates are slowed under inundation.

[17] 8. For the calculation of decomposition, litter accumulation is spread equally over twelve months instead of being added to the litter pool all at once in January.

[18] 9. Leaf dark respiration is simulated in a consistent way related to carboxylation capacity (see part 2 [Wania et al., 2009]).

[19] Points 1 to 3 are covered in part 1, and points 4 to 9, referring to vegetation related changes, are described in part 2 [Wania et al., 2009]. The new model version is called LPJ-WHy v1.2. “WHy” stands for Wetland Hydrology. Model source code, documentation, and boundary condition files will be archived at the Oak Ridge National Laboratory Distributed Active Archive Center for biogeochemical dynamics at

2.1. Soil Processes

[20] The standard LPJ soil model, with two soil layers extending over 1.5 m, is insufficient for resolving permafrost soil dynamics. We expand the modeled soil column to a depth of 10 m, split into the top 2 m, which we refer to as soil, and the bottom 8 m, which we call padding. We are mainly interested in processes in the soil layers as soil temperature and water availability in these layers will affect net primary production, competition between plant functional types, and decomposition rates. We represent the upper part of the soil column using eight layers with thicknesses of 0.1, 0.1, 0.1, 0.2, 0.2, 0.3, 0.5, and 0.5 m.

[21] Three more model layers sit on top of the soil. From bottom to top, these are a mixed layer, a snow layer, and an air layer. The mixed layer sits directly on top of the soil layers and has a variable thickness, updated daily. In nonpeatland grid cells, the mixed layer is composed only of litter. Litter depth varies with the amount of aboveground litter biomass (a minimum depth of 50 mm is prescribed to avoid numerical instabilities in the soil temperature solver). Litter depth shows interannual variability, but is stable throughout the year because the litter accumulation rate is updated only once per year. To calculate litter depth, we use bulk density values dependent on plant functional type [Wania et al., 2009, Table S1].

[22] In peatland grid cells, we assume that most of the litter is incorporated into the peat and therefore does not accumulate on top of the soil. A constant litter depth of 50 mm is therefore used. In peatlands, the water table may rise above the surface. The depth of any standing water is added to the litter layer depth:

equation image

where dmixed is the depth of the mixed layer (mm), dlitter is the depth of the litter layer (mm), and WTP is the water table position (mm), which takes positive values when the water table is above the surface.

[23] The snow layer lies on top of the litter layer. It has a variable thickness, updated daily depending on snowpack and snow density. When there is no snow, the snow layer functions as a second air layer and has the same attributes as the air layer. The overlying air layer is 1 m thick and is necessary to set the boundary conditions for the Crank-Nicolson temperature solver (see section 2.1.1). The air layer and minimum litter layer thickness were chosen so that a time step of 12 hours gave numerically reasonable results.

2.1.1. Soil Temperature

[24] Soil temperatures are simulated by numerically solving the heat diffusion equation,

equation image

using a Crank-Nicolson finite difference scheme (Text S1 together with Figure S1). Here, T is the soil temperature, t is time, z is soil depth, and D(z) is the thermal diffusivity, which varies with depth (equation (11)). The boundary conditions imposed on (2) are T(z = 0, t) = Ts(t), i.e., a direct temperature forcing of the air layer by the surface temperature Ts(t), and

equation image

at the bottom of the computational domain. This bottom boundary condition represents a vanishing heat flux at infinity, meaning that, at sufficient depth, the temperature profile becomes isothermal.

[25] The top of the computational domain is the air layer and is forced by air temperature. This is necessary because LPJ does not include a radiative balance model. To deal with the lower boundary condition, we extend the 2 m soil column with 8 m of padding layers, assuming that this is sufficiently deep to allow us to set the heat flux from the bottom layer to zero. Geothermal heat flux is assumed to be negligible for the top 10 m of soil and is neglected [Oelke and Zhang, 2004]. The padding layers use the same thermal diffusivity as the bottom soil layer but their thicknesses increase smoothly (0.82, 1.35, 2.21, and 3.62 m) toward the bottom of the computational domain. The smoothly increasing layer thicknesses minimize the additional computational expense of the padding layers by reducing the required number of layers while at the same time avoiding an abrupt transition from thin soil layers to thicker padding layers (Text S1).

2.1.2. Freeze-Thaw Cycle

[26] The thermal properties of soils vary with the fraction of water, ice, air, and organic and mineral material. The organic and mineral content of the soil depends on the soil type (Table 1). The fractional water and ice content, fwateri and ficei, of each soil layer i are updated daily. The thermal properties of the litter layer are set to the values for organic material. Ideally, the model would also take into account changes in thermal characteristics in peatlands when standing water intrudes into the litter layer, but this is not implemented in the current model version. While water is freezing or ice is melting in a soil layer i, it is assumed that the temperature, Ti, remains constant at 0°C. Energy flowing out of or into the soil layer comes from the latent heat released when freezing or consumed when thawing.

Table 1. Model Soil Parameter Valuesa
Soil TextureforgfpwpbΦb
  • a

    Parameter values are as follows: forg is the volumetric fraction of organic material, fpwp is the volumetric fraction of water at the permanent wilting point, and Φ is the porosity of the soil column. All soil texture classes except “organic” are referred to as mineral soils. All parameters are unitless fractions.

  • b

    Air Force Weather Agency [2002].

  • c

    The value of 0.439 for organic soils given by Air Force Weather Agency [2002] was too low for peat soils [Boelter, 1969].

Fine, nonvertisol0.010.1390.468

[27] In the case of freezing, this energy, Hi (J m−3), is estimated as

equation image

where Cwater is the heat capacity of water, ΔTi = TiToldi is the change in temperature of the soil layer that would occur if the temperature was not held at the freezing point, Toldi is the temperature in the soil layer on the previous day, Ti < 0°C, and Ti < Toldi. On the first day of freezing, Toldi ≠ 0. As a measure to avoid numerical instabilities caused when warm air temperatures allow rain but the soil is still frozen, freezing is only permitted when temperatures are falling. Water penetrating into the soil is added to the soil water fraction but will neither freeze when temperatures are rising, i.e., Ti > Toldi, nor will it melt the ice. The lack of heat input via precipitation and snow melt may lead to a delay of thawing of the soil in spring; this input will be included in a future model version.

[28] The fraction of water that freezes each day in soil layer i, ffreezei, can be calculated by dividing the energy released, Hi, by the latent heat of fusion, L (J m−3),

equation image

[29] Once ffreezei is determined, the water content is reduced and the ice content increased by this amount. The thawing process works similarly to freezing and occurs when Ti ≥ 0 and Ti > Toldi. When the temperature falls enough to freeze more than the remaining water, i.e., when ffreezei > fwateri, Ti is set following

equation image

[30] Similarly, when the temperature rises more than enough to thaw the remaining ice,

equation image

where fthawi is the potential fraction of ice thawed. Water held below the wilting point remains unfrozen [Granberg et al., 1999].

2.1.3. Soil Thermal Properties

[31] The soil heat capacity, Ci (J m−3 K−1), is updated daily for each layer by weighting the heat capacities of the individual components of the soil by their volumetric fraction ratios:

equation image

[32] Here, fi and C refer to the volumetric fraction and heat capacity for different components of the soil column, and the labels “air,” “water,” “ice,” “org,” “min,” and “peat” refer respectively to air, water, ice, organic soil, mineral soil, and peat (Table 2). Finally, fpwpi is the volumetric fraction of water held below the permanent wilting point.

Table 2. Thermal Properties of Air, Water, Ice, Organic Material, Mineral Material, and Peata
Air0.025b1.25 × 106c
Water0.570b4.18 × 106c
Ice2.200d3.12 × 106d
Organic material0.250e2.50 × 106c
Mineral material2.000f2.00 × 106c
Peat0.060c0.58 × 106f

[33] The thermal conductivity of soil layer i, Ki (W m−1 K−1), is given by

equation image

where fairi is the volumetric fraction of air, Kair is the thermal conductivity of air, and Kf represents the thermal conductivities of water, ice, organic soil, mineral soil, and peat, exponentially weighted by their respective fractions as

equation image

where K are the thermal conductivities of the individual soil components (Table 2), f are the volumetric fractions of the soil components, and ftotal is the total nongaseous volumetric fraction, i.e., 1 minus the fraction of air. This approach closely follows that of Granberg et al. [1999]. Given the heat capacity and thermal conductivity, the thermal diffusivity, Di (m2 s−1), of each layer can then be calculated as

equation image

2.1.4. Permafrost Location and Active Layer Depth

[34] A useful indicator for the presence or absence of permanently frozen soil is the active layer depth, defined as the 0°C isotherm, the depth to which liquid water exists continuously from the surface down [Muller, 1947; Burn, 1998]. In this study we focus on surface permafrost only. If the active layer depth is 2 m or deeper, we consider the area to be nonpermafrost. This assumption may lead to underestimation of permafrost area, but since most regions have an active layer depth of less than 2 m, we reason that the error will be small [Brown et al., 2000].

2.2. Snow

[35] Fresh snow insulates underlying soil layers from changes in surface air temperature, as its thermal conductivity is low. Later in the snow season, as snow ages, its density increases, the snow contains less air and more water, and its thermal diffusivity increases. Snow is assigned an initial density of 150 kg m−3, and the density increases linearly to 500 kg m−3 during the last quarter of the snow season [Oelke et al., 2003; Ling and Zhang, 2004]. For this calculation of snow density, snow season length is needed. Snow season length is estimated each year by counting the days when a snow layer exists. Then, assuming that snow season length does not vary significantly from one year to the next, we use the snow season length of the previous year to calculate the temporal evolution of snow density in the current year. This does not mean that we use the snow season length from the previous year to decide if snow exists or not; the previous year's snow season length is merely used to set the start point for the increase of snow density of this year's snow. Variations in snow season length from one year to the next may lead to slight shifts in the timing of the start of the increase in snow density, but we do not expect any significant impact on the insulation quality of the snowpack from year-to-year changes in snow season length.

[36] Snow is simulated using a single snow layer with varying depth that depends on the snowpack msnow (kg m−2) and the snow density ρsnow (kg m−3):

equation image

where dsnow is the snow depth in mm.

[37] As modeling snow thermal properties mechanistically is difficult, we use empirical relationships to estimate snow heat capacity and thermal conductivity. The heat capacity of snow, Csnow (J m−3 K−1), is practically the same as for ice, depends on temperature and snow density, ρsnow (kg m−3), and is given by Fukusako [1990] and Ling and Zhang [2006] as

equation image

where Tsnow is the snow temperature in Kelvin. The thermal conductivity, Ksnow (W m−1 K−1), is estimated following Sturm et al. [1997]:

equation image

2.3. Hydrology

[38] A soil type map is used to determine which model grid cells are peatland and which are nonpeatland. The peatland hydrology is adopted for all grid cells with organic soil (Figure S2).

2.3.1. Nonpeatland Hydrology

[39] The hydrology scheme for nonpeatland soils is modified from Gerten et al. [2004]. Changes were necessary because of the new layering scheme.

[40] Soil water holding capacity, now including water held below the permanent wilting point, is reduced by the ice content so that the actual water holding capacity (AWHCi) in layer i is

equation image

where WHCi is the water holding capacity of layer i. The ice fraction in each layer, ficei, is updated every day within the soil temperature routine and feeds back into the hydrology module on the next day.

[41] The “old” upper layer of LPJ, which extended to a depth of 0.5 m, corresponds to the top four soil layers in the new model. Daily changes in water content are calculated differently for these upper layers than for the other, deeper, soil layers. The daily change in water content Δwi (mm) is

equation image

where P is precipitation, M is meltwater Es is evaporation from bare soil, ETup is transpiration from the upper soil layer, i.e., transpiration weighted by root fraction in the upper 0.5 m (for root fractions see Wania et al. [2009, Table S1]), Δzi is the depth of soil layer i, and Δzup is the depth of the upper soil layer (0.5 m).

[42] The calculation of bare soil evaporation, Es (mm d−1), was modified to a quadratic dependence on soil water content to improve evaporation estimates (D. Gerten, personal communication, 2005):

equation image

[43] Here, Eeq is the daily equilibrium evapotranspiration, α is the Priestley-Taylor coefficient, which we set to 1.32 following Gerten et al. [2004], wup is the water content in the upper 0.5 m of soil, and ϕtotal is the foliar projective cover summed over all plant functional types.

[44] Evapotranspiration is simulated in the same way as by Gerten et al. [2004], as the minimum of a plant- and soil-limited supply function and the atmospheric demand [see Gerten et al., 2004, equations 11 and 12].

[45] When the water content of any layer in the upper 0.5 m exceeds the actual water holding capacity (AWHCi), the surplus water, Rsurf is available for percolation, Rperc, to the lower 1.5 m of the soil:

equation image

where k is the soil-type-dependent conductivity [Gerten et al., 2004] and wup is the water content of the upper soil layer. The percolated water, Rperc, is removed from the surplus water pool represented by Rsurf, and is added equally to the lower layers. The change in water content Δwi in each of the layers between 0.5 and 2 m is given by

equation image

where ETlow is the transpiration over the entire depth from 0.5–2 m and Δzlow is therefore 1.5 m. If the water content of any of the lower layers exceeds the actual water holding capacity, the water is added to the subsurface runoff, Rsubsurf.

[46] Fractional water and ice content in each layer are calculated daily. The water content is updated in the hydrology routine and then passed on to the soil temperature routine. There, the freezing and thawing processes are calculated for each sublayer as described in section 2.1.2. The updated water and ice content are then used in the hydrology routine on the next time step. Only liquid water held above the permanent wilting point is available for plant uptake.

2.3.2. Peatland Hydrology

[47] For organic soils (Figure S2), a peatland-specific hydrology scheme is used. Peatland soils in LPJ-WHy are divided into the acrotelm, defined as the upper 0.3 m, which experiences a fluctuating water table, and the catotelm, the underlying permanently inundated layer. The catotelm is 1.7 m thick, to permit calculation of the temperature down to 2 m. The simulation of the water table position follows Granberg et al. [1999] with minor modifications; Granberg et al.'s approach has also been applied by Zhuang et al. [2004] and Weiss et al. [2006].

[48] The modeled water table position is defined so that positive values correspond to standing water above the surface and negative values mean that the water table is below the surface. Our modifications to Granberg et al.'s [1999] approach involve the introduction of standing water above the surface and the use of Simpson's rule to numerically integrate water content over the unsaturated zone of the acrotelm to improve conservation of water in the hydrology code.

[49] The change in total water volume, ΔVtot (mm), in the acrotelm is derived from

equation image

where P is precipitation, M is meltwater, ET is evapotranspiration, and Qrunoff is surface runoff, all in mm. The total water volume is consequently given by

equation image

where W is water content, I is ice content, and S is standing water. If the total volume is greater than the depth of the acrotelm (zacro) times its porosity (Φ), Vtot > zacro Φ, standing water will occur up to 0.1 m height, and excess water above this height will run off. The threshold value of 0.1 m for runoff was chosen on the basis of observations that show water table positions above the surface [e.g., Shannon and White, 1994; Booth et al., 2005]. The acrotelm porosity is set to 0.9, close to 0.92 as used by Granberg et al. [1999]; the catotelm porosity takes the value for organic soils from Table 1. The total water volume is used to estimate the water table position, WTP (mm), as

equation image

[50] The second and third cases in equation (22) are from Granberg et al. [1999], where fwatersurfmin = 0.25 is the minimum fractional water content at the surface and the gradient az represents the increase of surface water content from fwatersurfmin to the maximum water content, which is the acrotelm porosity, Φ, minus a constant gas fraction of 0.08 [Hogg and Wein, 1988; Strack et al., 2005]. The water content of the unsaturated acrotelm, i.e., between the surface and the water table, is calculated by integrating fwater over 0.01 m thick layers using Simpson's rule.

[51] Data showing the relationship between evaporation and water table position are sparse; however, the ratio of actual evapotranspiration (including evaporation and transpiration) to potential evapotranspiration (ET/ETpot) has been correlated to water table position [Kim and Verma, 1996; Lafleur et al., 2005]. We assume in LPJ-WHy that transpiration will not decrease significantly as the water table position drops from the surface to 0.3 m, because the water content in the unsaturated zone will be sufficient to meet transpirational demand at any possible water table position. A decrease in evapotranspiration thus arises from a decrease in evaporation from the moss layer. From Kim and Verma's results we derive an approximation to actual evapotranspiration ET,

equation image

where ETpot is the potential evapotranspiration, corresponding to Eeq in equation (17), and WTP is the water table position (mm). When using water table positions of −750 to −300 mm observed at the Mer Bleue Bog in southern Ontario, equation (23) gives (ET/ETpot) ratios of 0.46 to 0.8, which are slightly higher than the ones estimated by Lafleur et al. [2005] (0.44 to 0.59). It should be noted that in equation (23), ET is independent of foliar biomass, which may constitute a problem for future projections, where foliar biomass, and with it ET, is likely to increase.

[52] Vertical drainage through peat is negligible compared to losses by evapotranspiration and was not included in the peatland hydrology [Boatman and Tomlinson, 1973; Evans et al., 1999]. Surface runoff in peatlands depends on the water table position; to model this, an exponential dependency on the water table position was used, loosely based on Evans et al. [1999], giving

equation image

which gives a maximum runoff of 2.7 mm d−1 at a water table position of 100 mm and a minimum of 0.23 mm d−1 at −149 mm. Runoff is permitted only when the fractional ice content of the uppermost layer is less than 0.05 and the snow depth is less than 10 mm.

[53] During the model spin-up period, an additional water source of 0.5 mm d−1 representing run-on is added to the acrotelm. It was found that this helped to avoid large vegetation fluctuations during model spin-up due to oscillations in the climate data.

3. Data and Experiments

3.1. Soil Map

[54] One input required by LPJ-WHy is a map classifying each land point by soil type. The soil map originally used for LPJ classified soils into eight soil texture types derived from a soil type map based on FAO data [Zobler, 1986; FAO, 1991; Sitch et al., 2003] (see Table 1). For use in LPJ-WHy, this map was augmented using information on organic soil carbon from the IGBP-DIS map [Global Soil Data Task Group, 2000]. Organic soils accumulate under anaerobic conditions, hence the organic carbon content of soils can be used as an indicator of inundation and thus peatland extent. Each 5′ × 5′ grid cell in the IGBP-DIS map with a soil carbon content of more than 31 kg m−2 in the top 1 m was considered to be organic soil. From this, a 1° × 1° resolution map showing fractional peatland cover per grid cell was produced (Figure S2). The resulting peatland area between 45°N and 90°N is 5.37 × 106 km2, a value that exceeds other estimates [Matthews and Fung, 1987; Aselman and Crutzen, 1989; Prigent et al., 2007]. For grid cells with organic soil, the model was run twice (once with peatland and once with nonpeatland hydrology), and the results were weighted by the fractional cover of peatland.

3.2. Twentieth Century Climate Data

[55] Climate data to drive LPJ-WHy were taken from the Climate Research Unit climatology data set CRU CL 1.0 (Figures 1 and 2) and the time series data CRU TS 2.1 [New et al., 1999; Mitchell and Jones, 2005]. Both data sets, used at a 1° × 1° resolution, provide monthly air temperature, cloud cover, total precipitation, and number of wet days, CRU CL 1.0 as a climatology for the years 1961–1990 and CRU TS 2.1 as a time series for 1901–2002. The time series data were used to permit effective comparison of individual model years to observations (Figures 3, 4, and 6). Atmospheric carbon dioxide concentrations for 1901–2002 were taken from Etheridge et al. [1996] and Keeling and Whorf [2005].

Figure 1.

Examples of LPJ-WHy output. Monthly values for net primary productivity (NPP) per plant functional type are shown as colored bars. Snow depth is plotted as gray columns with snow flake pattern (out-of-scale values are labeled). The water table is the boundary between the brown unsaturated soil and either water (blue) or ice (light gray). Although the brown area representing the unsaturated soil will have high water content close to the water table, we do not show saturation levels here in order to highlight the water table position. Soil temperature is indicated by the vertical trace which changes color from blue (cold) to red (warm) and is shown only for the top meter of soil. The temperature line is gray where the temperature is held at 0°C during freezing or thawing.

Figure 1.


Figure 2.

Mean monthly temperature (line) and precipitation (bars) for the four sites in Figure 1 (data taken from the CRU CL 1.0 climatology [New et al., 1999]).

Figure 3.

Comparison of simulated and observed soil temperatures at different soil depths (values in the top left corner of each plot) at four sites. Air temperatures are shown to provide a comparison between meteorological data from the sites and the CRU climatology data used to drive LPJ-WHy. Standard deviations (SD) were only available for observations from (Figure 3b) the Lakkasuo, Finland, site, for which one SD is denoted by gray shading. To keep the temperature range plotted within reasonable bounds, soil surface (0 cm) instead of air temperatures were plotted for (Figure 3c) the Spasskaya, Siberia, site.

Figure 4.

Comparison of simulated and observed water table position at four sites. For the Salmisuo, Finland, site, observations were available from two locations within the site with different hydrological characteristics, a lawn and a flark.

3.3. Spin-up Procedure

[56] To spin the LPJ-WHy model up using the climatology data set CRU CL 1.0, the 12-monthly data were simply repeated for 1000 years. When using the time series data CRU TS 2.1, the first 10 years of the data were repeated 100 times to give 1000 years of spin-up time. This procedure ensures that carbon stocks in nonpeatlands and permafrost are in equilibrium before performing the transient portion of the simulation. In fact, peatland carbon stocks did not reach equilibrium after 1000 years; the consequences of this for net ecosystem production and soil carbon content are discussed in part 2 [Wania et al., 2009].

3.4. Observations

[57] Observations of meteorological, soil, and hydrological parameters from several boreal sites were used for model evaluation. Air and soil temperatures were taken from Point Barrow in Alaska, the Spasskaya forest near Yakutsk in Siberia, the Lakkasuo mire in central Finland, and the BOREAS Northern Study Area in Manitoba, Canada (Table 3). Point Barrow lies in the continuous permafrost region and is vegetated with wet sedge tundra. The Spasskaya larch forest is another example from the continuous permafrost region with an active layer depth of about 1.2 m. The Lakkasuo weather station, maintained by the Hyytiälä Forestry Field Station of Helsinki University, provided air and soil temperature data from three microsites at the Lakkasuo mire complex, which covers ombrotrophic bogs and minerotrophic fens. The soil and air temperatures at the BOREAS site are from the old black spruce forest (BOREAS-NOBS, TF-3).

Table 3. Names, Locations, and References for Data Used in the Evaluation of Soil Temperature and Water Table Position in Figures 3 and 4a
Site NameCoordinatesReference
  • a

    We refer to these sites by the names given in bold font.

  • b

    Boreal Soil Hydrology and Carbon (BOSHAC) project led by the University of Harvard.

Point Barrow, Alaska, United States71.3°N156.8°WWelker and Fahnestock [2003]
BOREAS NOBS, Manitoba, Canada55.8°N99.3°WDunn et al. [2006], BOSHAC projectb
Marcell Experimental forest, Minnesota, United States47.5°N93.5°WDise et al. [1993]
Lakkasuo mire, Finland61.8°N24.3°EWeiss et al. [2006], Hyytiälä Forestry Field Station of Helsinki University
Salmisuo mire, Finland62.8°N30.9°ESaarnio et al. [1997]
Spasskaya forest, Yakutsk, Russia62.3°N129.6°EOhta et al. [2001], Ohta et al. [2003], Suzuki and Ohata [2003]

[58] Measured water table positions were taken from the BOREAS Northern Study Area in Canada, the Lakkasuo mire and the Salmisuo mire in Finland, and a site in Minnesota, United States. The BOREAS data are again from the old black spruce forest (BOREAS-NOBS, TF-3). The Lakkasuo mire data, the Salmisuo mire data, and the Minnesota data were digitized from publications (see Table 3 for sources).

[59] Active layer depth observations were taken from the Circumpolar Active Layer Monitoring (CALM) network [Brown et al., 2000], from which we used mean values of 20 sites over the period 1991–2000 (Table 5); note that the time series for some sites were incomplete, and the average over less than 10 years was taken. For detailed information consult Brown et al. [2000] or the CALM network Web site at

3.5. Evaluation Methods

[60] In order to evaluate the quality of fit between observed and modeled soil temperatures, we use Willmott's index of agreement (W) [Willmott, 1982] and the mean absolute error (MAE), defined as

equation image


equation image

where the Pi are model predictions, the Oi are observations, equation image is the mean of the observations, and n is the number of observations. Higher values of W and lower values of MAE indicate a better match between predictions and observations.

[61] The free software package R version 2.5.1 was used for statistical analyses [R Development Core Team, 2008].

3.6. Sensitivity Experiment

[62] We chose the ECHAM5 general circulation model (GCM) as a representative of the World Climate Research Programme's (WCRP's) Coupled Model Intercomparison Project phase 3 (CMIP3) multimodel ensemble [Meehl et al., 2007a]. Under the IPCC SRES A2 emissions scenario, the ECHAM5 GCM projects temperature increases for 2071–2100 compared to 1961–1990 ranging from <4°C in western Europe, southern Scandinavia, the west coast of North America, and the southern end of Greenland to >7°C in large areas of Siberia and the north of the Canadian and Alaskan Arctic (Figure 7). Precipitation projections for the same time interval suggest much wetter conditions in most areas in North America with increases of 10–30% being widespread and increases of >30% found mainly in Alaska and parts of Canada (Figure S3). Some eastern regions of Scandinavia are projected to experience rainfall increases of 30–40%. Siberia may face precipitation increases of 30 to >50%.

[63] Loosely on the basis of this projection of future climate, a sensitivity study was performed to examine the effect of temperature and precipitation increases on peatland and permafrost dynamics. The sensitivity of active layer depth, soil temperature, and water table position to these factors was examined by imposing increases of mean monthly temperature ranging from 0–9°C in 1°C intervals and increases of monthly total precipitation ranging from 0–50% in 10% intervals (60 simulations per site). Experiments showed that changes in atmospheric carbon dioxide concentration had little effect on the physical land surface variables presented here, so that we show only results based on a carbon dioxide concentration of 780 ppmv.

4. Results and Discussion

4.1. Overview of the Effects of New Features in LPJ-WHy

[64] Figure 1 illustrates output from LPJ-WHy as annual time series of soil temperature, hydrology, and vegetation composition for four sites chosen to exemplify the new features we have added to the model, viz. fluctuating water table position, calculation of soil temperatures in 12 layers to a depth of 10 m, thawing and freezing of soil water, and two new plant functional types. Temperature and precipitation climatologies for the example sites are shown in Figure 2.

[65] The first site is the Lakkasuo site in Finland, at 62.75°N, 30.75°E (Figures 1a and 2a), which experiences mild temperatures and high precipitation. This site has higher water tables than the others examined here and little ice formation during the winter. It is the only site of the four that lacks permafrost. The reason for the absence of permafrost is not high summer temperatures, but mild winter temperatures and a thick snow blanket that prevents the underlying soil from losing energy in the winter. The peatland hydrology scheme allows the water table to rise during the melt season, while surface snow has yet to melt (Figure 1a, April). This treatment is slightly unrealistic, but is justified by the difficulty of simulating snow ageing dynamics more mechanistically. The snow model in LPJ-WHy does not incorporate a representation of melted water suspended in snow and treats snow ageing via a simple linear increase in snow density (section 2.2). The Lakkasuo site is used for the evaluation of soil temperature and water table position.

[66] In contrast to the warm Finnish site, the soil at the Siberian site at 69.75°N, 67.25°E (Figures 1b and 2b) is almost completely frozen and thaws only during 4 months of the year to a depth of approximately 0.35 m. With these colder conditions and a long snow season, the growing season lasts only for 3–4 months. Only the Sphagnum mosses thrive in this harsh environment. Their productivity peaks in July and August (Figures 1b and 2b). October (Figure 1b) and November (Figure 1d) illustrate how freezing in LPJ-WHy starts at the top of the soil, leading to liquid water being trapped in the middle soil layers before eventually turning to ice.

[67] The other two sites illustrated in Figure 1 are the BOREAS Northern Study Area site in central Canada, at 55.75°N, 98.25°W (Figures 1c and 2c) and a site in northern Canada at 61.75°N, 120.25°W (Figures 1d and 2d). These were chosen to show the potential sensitivity of the vegetation distribution to relatively small differences in climate. The temperatures for these two sites are comparable, but the precipitation is higher at the BOREAS site (Figures 2c and 2d). The higher precipitation at the BOREAS sites results in higher water table positions and a typical peatland vegetation composition of flood-tolerant C3 graminoids and Sphagnum mosses (Figure 1c). When the water table is lower, however, tree plant functional types can grow on peatlands, as seen in Figure 1d. The BOREAS site is also used for the evaluation of soil temperature and water table position in sections 4.2 and 4.3.

[68] The different behaviors illustrated in Figure 1 critically affect the vegetation composition in peatland and permafrost regions, and the balance between the factors involved is delicate enough that a detailed process-based model like LPJ-WHy is needed to capture the important interactions.

4.2. Soil Temperature Evaluation

[69] The annual cycle of soil temperature simulated by LPJ-WHy was compared to observations at a range of soil depths at four sites (Figure 3 and Table 4), two peatland sites (Lakkasuo and BOREAS) and two nonpeatland sites (Point Barrow and Spasskaya).

Table 4. Soil Temperature Evaluation Statisticsa
DepthObserved ± S.D.Modeled ± S.D.WMAE (°C)
  • a

    Mean values for observed and modeled soil temperatures (in °C) are given together with their standard deviation (S.D.). Willmott's index of agreement (W) and the mean absolute error (MAE) are explained in section 3.5.

Point Barrow, Alaska
Air−11.7 ± 10.4−13.3 ± 11.10.963.2
5 cm−10.0 ± 8.6−10.9 ± 7.00.972.3
20 cm−9.7 ± 7.7−10.6 ± 6.40.972.0
40 cm−9.3 ± 7.1−10.3 ± 5.70.962.0
Lakkasuo, Finland
Air3.0 ± 8.23.7 ± 7.00.942.7
5 cm3.7 ± 4.74.2 ± 4.70.990.7
20 cm3.9 ± 3.64.3 ± 2.80.970.9
40 cm4.4 ± 1.24.4 ± 0.60.770.8
Spasskaya, Siberia
Air0.9 ± 13.8−1.2 ± 16.40.973.8
5 cm−1.2 ± 9.6−6.9 ± 14.70.876.8
20 cm−1.3 ± 5.3−6.9 ± 8.60.738.6
40 cm−1.5 ± 1.8−6.8 ± 6.30.455.7
BOREAS, Canada
Air−0.6 ± 14.5−0.2 ± 12.50.954.5
5 cm3.1 ± 5.02.3 ± 5.10.971.3
20 cm2.6 ± 3.91.6 ± 3.20.951.1
40 cm2.0 ± 2.70.2 ± 0.50.512.0

[70] Simulated soil temperatures match observations well at Point Barrow, Alaska and Lakkasuo, Finland (Figures 3a and 3b). Willmott's index of agreement yields high scores for both sites for five of the six soil depths considered (Table 4). The bias compared to observations at Point Barrow is 2.0–2.3°C, whereas the bias for the Lakkasuo site is only 0.7–0.9°C (Table 4). The simulated propagation of temperature variations into deeper soil layers at the Lakkasuo site shows minor deviations from the observations, but considering the mismatch between the resolution of the model output (a 1° × 1° grid box) and observations (a single point with a microclimate influenced by vegetation, topography, slope aspect, and soil type), the agreement is encouraging.

[71] At the Spasskaya site in Siberia, LPJ-WHy modeled soil temperatures better in summer than in winter, when there are wide discrepancies (Figure 3c). Willmott's index of agreement is only 0.45–0.87, and the bias is 5.7–8.6°C (Table 4). It appears that some factor acted to insulate the soil from variations in air temperature, as heat loss from even shallow soil depths of 10 and 20 cm was minimal. These discrepancies are likely caused by an overlying icy snow layer and a thick dwarf shrub layer, e.g., Vaccinium spp., both of which insulate the underlying soil and dampen temperature variations (T. Ohta, personal communication, 2007). The dwarf shrub layer could in principle be represented in the model by introducing one or two further plant functional types [Wolf et al., 2008]. However, more detailed modeling of snow characteristics lies beyond the scope of LPJ-WHy.

[72] At the BOREAS site in Canada, the model underestimates temperatures in summer at 50 cm depth (Figure 3d). The deviation of simulated temperature from observed values (on average 2°C) could be caused by lower thermal diffusivities because of the presence of excess water. The modeled water table position was always above the observed position (Figure 4c). The agreement between observations and model result is W = 0.97 (5 cm layer) and W = 0.95 (20 cm layer) and shows a bias of 1.3 and 1.1°C, respectively (Table 4).

4.3. Hydrology Evaluation

[73] Figure 4 shows simulated and observed water table positions for four evaluation sites. The simulated water table position at the Lakkasuo mire follows the observations closely until July, but the observed rise in water table position starting in August was not reproduced by the model (Figure 4a). At a second site in Finland, at Salmisuo, LPJ-WHy simulated a water table position in the range of the lawn microsite type (Figure 4b), but the modeled water table shows wider fluctuations than observed. These results can be set in context by noting that peatlands are generally characterized by heterogeneous microtopography, with hummocks representing the driest sites and pools or flarks the wettest. Hollows or lawns are states of intermediate wetness. The simulated water table results for our two Finnish sites suggest that LPJ-WHy simulates conditions that lie well within this moisture regime.

[74] Model results for the BOREAS site in Canada show water table positions consistently too close to the surface compared to the observed water tables (Figure 4c), whereas the Minnesota site exhibits the opposite problem with the model simulating water tables significantly deeper than observed for the second half of the year (Figure 4d).

[75] These results highlight an important issue when modeling peatlands in LPJ-WHy, namely, that inflow and outflow are not captured. At the BOREAS site, where the observed peat layer is only about 0.6 m thick (A. Dunn, personal communication, 2007), LPJ-WHy overestimates the water table position. This could be caused by subsurface drainage from the mineral soil layers below the peat: subsurface drainage in LPJ-WHy is set to zero. In an experiment run where we set the drainage rate to 0.15 mm d−1, LPJ-WHy simulates a water table position for the BOREAS site that matches the observed data much more closely. However, at a global scale, peat deposits are estimated to be 2.3–2.5 m deep on average [Gorham, 1991; Clymo et al., 1998], and we therefore assume that the mean error in simulated water table position would be smallest when setting the drainage rate to zero. A reduction in uncertainty when modeling circumpolar peatlands will not be possible until a detailed peat depth map is available.

[76] The opposite problem to that seen at the BOREAS site was observed at the Minnesota site, where the water table position was underestimated (Figure 4d). This site is classified as a poor fen [Dise et al., 1993], and since fens by definition receive water from groundwater or overland flow [Aselman and Crutzen, 1989], it seems likely that the Minnesota site was influenced by another water source in addition to precipitation. To model fens effectively, it would be necessary to incorporate a hydrological routing scheme to represent flow of water from each grid cell to its neighbors.

[77] Maximum monthly evapotranspiration in the model reached 86, 82, 92, and 112 mm for the Lakkasuo, Salmisuo, BOREAS, and Minnesota sites, respectively. Monthly evapotranspiration rates of up to 75 mm were reported for a central Swedish mire [Kellner and Halldin, 2002], slightly lower than the rates modeled by LPJ-WHy. Annual total evapotranspiration for the four sites was 390, 393, 411, and 569 mm for the Lakkasuo, Salmisuo, BOREAS, and Minnesota site, respectively. Annual evapotranspiration for the Mer Bleue Bog in Canada was estimated to be between 422 and 520 mm [Lafleur et al., 2005], very close to the LPJ-WHy results. Figure 5 shows the annual evapotranspiration for all peatland grid cells and reveals a general trend in evapotranspiration rates from south to north, with the highest evapotranspiration rates (>500 mm) in southeastern Canada and some individual grid cells in Europe and eastern Asia and the lowest rates in the high Arctic.

Figure 5.

Annual evapotranspiration for peatland grid cells simulated with LPJ-WHy, averaged over 1991–2000 using the CRU TS 2.1 data set. Evapotranspiration values are not weighted by the fraction of peatlands; that is, the values are representative only for the peatland area in each grid cell.

4.4. Permafrost Evaluation

[78] The modeled present-day surface permafrost area between 45°N and 90°N of 20.0 × 106 km2 is comparable to the 25.5 × 106 km2 observational estimate covering the entire Northern Hemisphere, including permafrost on the Tibetan plateau and Himalayas [Anisimov and Nelson, 1997].

[79] Simulated active layer depth for the entire surface permafrost zone shows a poor match with observations from the CALM network (Figure 6, R2 = 0.01, p = 0.7). A significant divergence from the CALM observations can be seen in Canada west of the Hudson Bay at the Baker Lake site (Table 5), where exceptionally deep active layer depths were observed and could not be reproduced by LPJ-WHy. The CALM observations for Baker Lake appear to constitute an exception to the norm for this area as the active layer depths in a drainage area 80 km west of Baker Lake were recorded to be between 0.5 and 1.25 m [Roulet and Woo, 1986], which compares well to our modeled 0.76 m.

Figure 6.

Permafrost distribution and comparison of simulated and observed active layer depth for the region 45–90°N. The underlying map shows results from an LPJ-WHy simulation averaged over the years 1991–2000 while the dots are observations from the same period, where available. Nonwhite areas indicate regions where LPJ-WHy models permafrost in the top 2 m of soil.

Table 5. Sites From the Circumpolar Active Layer Monitoring Network Used in Figure 6a
CALM CodeNameCoordinatesObs (cm)Mod1 (cm)Mod2 (cm)
  • a

    Observed active layer depths (Obs), modeled active layer depths using the soil type given by the soil type map (Mod1) and modeled using a different soil type (Mod2). Data are averaged over 1991–2000 (or over as many years as were available for the observations). Soil types are either mineral (m) or organic (o). CALM, Circumpolar Active Layer Monitoring network [Brown et al., 2000].

C6Parsons Lake68.97°N133.56°W83.6105.2(m)-
C12Great Bear River64.92°N125.59°W74.561.8(o)-
C17Sheldrake River56.63°N76.1°W136.3150.1(m)-
C20Baker Lake64.17°N95.5°W163.376.8(m)-
C22Kluane Lake60.96°N138.43°W56.0134.7(m)34.0(o)
United States
U4West Dock70.37°N148.55°W31.2128.0(m)32.0(o)
U13Toolik LTER Transect68.62°N149.6°W39.972.4(m)-
U18Bonanza Creek LTER64.75°N148.0°W50.897.0(o)-
U27Council Grid65.45°N164.62°W63.0144.2(m)45.2(o)
G3Disko Island69.25°N53.18°W67.532.6(m)-
S2Abisko68.33°N18.83°E59.648.5 (o)-
R1Nadym, West Siberia65.33°N72.92°E56.553.3(o)-
R2Ayach-Yakha, Vorkuta67.59°N64.19°E65.6183.3(m)40.0(o)
R3Marre Sale, Yamal Peninsula69.72°N66.76°E108.231.3(o)104.0(m)
R8Tiksi (Game), Lena Delta71.59°N128.79°E42.393.3(m)31.0(o)
R9Cape Rogozhny, Chukotka64.79°N176.97°E43.3102.8(m)37.4(o)
R14Chukochya River, Kolyma69.49°N156.99°E40.847.6(o)-
R18Mt. Rodinka, Kolyma68.75°N161.5°E73.781.4(m)-
R25Jakutskoe Lake69.85°N159.5°E30.041.9(o)-

[80] There are several sites where LPJ-WHy overestimates the active layer depth, e.g., two sites in Alaska (West Dock and Council Grid), one in the Canadian Arctic (Kluane Lake), and three in Russia (Ayach-Yakha, Tiksi, and Cape Rogozhny). When these sites are simulated individually and the soil type is changed from mineral to organic, the results improve (compare columns for Obs to Mod2 in Table 5, R2 = 0.6, p < 0.001). The improvement is achieved because the soil type map that we use shows inorganic soil even though the site has very high organic content [Brown et al., 2000]. The opposite pattern can be found for a site in western Siberia (Marre Sale), where LPJ-WHy underestimates the active layer depth. At the Marre Sale site, the soil is classified as organic while in reality the organic layer makes up only 0.1–0.7 m of the total soil depth [Brown et al., 2000]. This misclassification of soil characteristics leads to errors in the prediction of permafrost occurrence, caused not by the model itself, but by problems in the input parameters. The active layer depth can vary by a factor of 2 between organic and mineral soils under otherwise identical conditions, a phenomenon also observed in the field [Brown et al., 2000].

4.5. Sensitivity Experiments

[81] Figure 7 shows changes in soil temperature at 25 cm depth, active layer depth, and water table position as temperature and precipitation are varied independently. Also shown are the locations of the sites on which we concentrate here (Table 6) and, for reference, a map of annual mean temperature changes projected by the ECHAM5 GCM under the IPCC SRES A2 emissions scenario.

Figure 7.

Sensitivity of annual soil temperature at 25 cm depth (Soil T), active layer depth (ALD), and mean summer (JJA) water table position (WTP) at six sites (Table 6) in different zones of projected temperature increase. (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) soil temperature, (Figure 7c) active layer depth, and (Figure 7d) water table position 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.

Table 6. Site Numbers, Names, and Locations for Sensitivity Results in Figure 7
  • a

    BOREAS NSA, Boreal Ecosystem-Atmosphere Study, Northern Study Area; CALM, Circumpolar Active Layer Monitoring network [Brown et al., 2000].

1Toolik Lake, Alaska, United States68.8°N149.8°W
2BOREAS NSA, Manitoba, Canada55.8°N99.3°W
3CALM site C17, Canada56.6°N76.1°W
4CALM site S2, Abisko, Sweden68.3°N18.8°E
5CALM site R1, western Siberia, Russia65.3°N72.9°E
6Eastern Siberia, Russia70.3°N147.8°E

[82] A reference map for annual mean precipitation changes can be found in Figure S3. The changes in temperature (ΔT) and precipitation (ΔP) as projected for 2100 are indicated by a black dot in each panel. The dashed line represents one potential course from present day to 2100 conditions.

[83] We first examine the sensitivity of soil temperature to temperature and precipitation perturbations (Figure 7b). All of the sites experience an expected soil temperature increase of about 4–8°C (Figure 7b). The soil temperature increases are mainly influenced by rising air temperature but are higher under higher ΔP. One possible explanation for this phenomenon is that monthly precipitation in this experiment was increased proportional to each month's present-day precipitation amount, i.e., the experiment does not account for seasonal changes in precipitation pattern. Therefore, increases in precipitation will lead to higher winter precipitation affecting snow depth and with it insulation of the underlying soil, leading to less energy loss during the winter and therefore warmer temperatures. An increase in winter precipitation is very likely to happen in the future [Meehl et al., 2007b] and will likely contribute indirectly to the warming of soils in high northern latitudes.

[84] Sensitivity of the active layer depth is shown in the Figure 7c. Sites 1 and 6, which lie in areas with a projected temperature increase of 6°C or more, show a doubling in active layer depth at temperature increases of 3 and 5°C, respectively. ECHAM5 projects temperature increases of 5 to >7°C for sites 2, 3, and 5. Under these conditions, the model indicates the disappearance of permafrost from the top 2 m of soil. Site 4 is the most sensitive to temperature increases as surface permafrost disappears at a warming of only 1°C or with an increase in precipitation alone. A thawing of permafrost in the Abisko area (site 4) has already been observed in recent years [Christensen et al., 2004; Åkerman and Johansson, 2008]. Active layer depth at sites 2 to 5 increased more quickly under higher precipitation perturbations.

[85] Summer (JJA) water table position was highly sensitive to temperature and precipitation at sites 2, 3, 5, and 6 (Figure 7d). Water table positions decreased under higher temperatures and increased under higher precipitation. When the combined effect of temperature and precipitation was considered, modeled water table positions were very similar to the present day (dashed lines and black dots in Figure 7). Sites 1 and 4, which lie at the two extremes of water table positions, show little sensitivity to either temperature or precipitation.

[86] The most striking result of this analysis is that summer water table positions might remain similar to present-day conditions under future climate change (Figure 7). In other words, peatlands could become much warmer but still remain wet, conditions likely to lead to enhanced release of methane [Christensen et al., 2004; Johansson et al., 2006].

[87] However, this analysis suffers from the limitation that it distributes changes in precipitation and temperature equally over the year. Projections by ECHAM5 indicate that temperatures in regions south of about 60°N will rise more in summer, while temperatures further north will increase more in winter [see also Christensen et al., 2007; Meehl et al., 2007b]. A more realistic approach to testing the sensitivity of LPJ-WHy to future climate conditions would be to use GCM data directly to drive the model. Results of such experiments will be presented in a future study.

4.6. Some Limitations of Our Approach

[88] The term “peatlands” encompasses a variety of complex ecosystems, ranging from ombrotophic bogs to minerotrophic fens, variations determined mainly by the influence of local groundwater hydrology. Peatlands show a large variety of small-scale microtopographical features that influence hydrology and biogeochemical processes at the small scale. However, LPJ-WHy, like all other DGVMs, is a global vegetation model that simulates processes in 2,500–10,000 km2 grid cells. At this scale, it is impossible to take account of small-scale variability, and we are forced to model average conditions. The field of global groundwater modeling is not yet sufficiently advanced to include the influence of groundwater on the local peatland hydrology. Groundwater influx, however, is important for fen ecosystems. What does this mean for our results? Firstly, we may simulate water table positions that are too low for fens in some areas, resulting in a biased vegetation composition. Secondly, we may simulate reasonable water table positions for some fen sites solely because of a fortuitous counterbalance between efflux and influx, neither of which we model. (Surface outflow in LPJ-WHy occurs only when standing water is above a fixed threshold.) We believe that one way forward with this problem is to take advantage of detailed peatland maps existing for at least some areas of the circumpolar region. These could be used to set fractions of each grid cell to either bog and fen. LPJ-WHy could then use different hydrology schemes for fens and bogs.

[89] Other observed problems in this study arise from the way that soils are prescribed. The soil map used gives only a description of the soil texture: this characteristic is then ascribed to the entire 2 m soil column, and soil thermal and hydrological parameters are assigned on the basis of the soil texture classification. At sites where the soil changes from highly organic to mineral within the top 2 m, or where the soil is shallow, LPJ-WHy cannot reproduce the observed thermal regime. Until the work described here, LPJ did not perform detailed simulation of soil temperatures, and a very basic soil classification was therefore sufficient. This soil type problem could be addressed by the use of a more detailed soil map [Batjes, 2005].

5. Conclusions

[90] This study demonstrates an approach to the integration of peatland and permafrost processes into a dynamic global vegetation model. A new hydrology scheme models the water table position for peatland grid cells. The simulation of permafrost is mechanistic: soil temperatures are simulated by solving the heat diffusion equation numerically, and processes involving thawing and freezing of soils are modeled in detail. This method has the great advantage over previous approaches of not requiring an empirical relationship, such as the frost index, to predict the occurrence of permafrost [Nelson and Outcalt, 1987; Beer et al., 2007]. Further, as a consequence of this mechanistic approach, LPJ-WHy is able to simulate discontinuous as well as continuous permafrost areas. The results for permafrost distribution and active layer depth achieved with LPJ-WHy are comparable to those produced by much more complex models [Oelke et al., 2003; Lawrence and Slater, 2005] and show generally good correspondence with observations.

[91] Variations in snow quality present a problem for all models simulating soil temperatures as snow cover is an important insulator and reduces the energy loss of soils during cold periods. The insulating qualities of snow depend on its water and air content, which can change rapidly, for instance through melting and refreezing of snow. Another contribution to the insulation of soils comes from dwarf shrubs, which can form thick mats. We plan to include evergreen and deciduous dwarf shrubs in a future version of LPJ-WHy.

[92] The sensitivity experiment reported in section 4.5 provides some insight into the effects of future climate change on land surface processes in the boreal region. However, this experiment uses a highly idealized setup under which temperature and precipitation were increased linearly and independently, which may not be particularly realistic. In order to project what will happen to active layer depth, soil temperatures, and water table positions under future climate change, dynamic global vegetation models need to be driven by the output of general circulation models. Only in this way can the complex interactions between the hydrology, soil thermal regime, and vegetation be captured.

[93] One major source of uncertainty unaddressed by our study is “What will happen to local hydrology when permafrost thaws?”. Permafrost degradation may enhance drainage, leading to drier conditions and increased decomposition rates, releasing stored carbon as carbon dioxide [Waelbroeck, 1993; Armentano and Menges, 1986]. However, permafrost degradation can also lead to inundation, which causes the formation of internal lawns and ponds, both of which enhance Sphagnum moss growth and carbon accumulation rates [Turetsky et al., 2000, 2002; Camill et al., 2001]. This is a difficult question and as yet we have neither the modeling tools nor the field observations to determine the regional consequences of permafrost thawing.


[94] We would like to thank Jukka Alm and the Hyytiälä Forestry Field Station of Helsinki University and Allison Dunn and the BOSHAC project for the provision of soil temperature data. Further we would like to thank Marko Scholze for help with data availability, Dieter Gerten for discussion of the model code, and Renato Spahni for reformatting of the IGBP-DIS soil carbon data. 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. Paul Miller provided very valuable feedback on our model development and on this manuscript. We thank Paul Valdes for the provision of computer facilities 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 was greatly improved after reviews by Nigel Roulet and an anonymous reviewer.