Landscape partitioning and environmental gradient analyses of soil organic carbon in a permafrost environment

Authors


Abstract

[1] This study investigates landscape allocation and environmental gradients in soil organic carbon (C) storage in northeastern European Russia. The lowlands of the investigated Usa River Basin range from taiga with isolated permafrost to tundra vegetation on continuous permafrost. We compile and analyze databases on soil properties, permafrost, vegetation, and modeled climate. Mean soil C storage is estimated at 38.3 kg C m−2, with similar amounts in taiga and tundra regions. Permafrost soils hold 42% of the total soil C in the area. Peatlands dominate soil C storage with 72% of the total pool and 98% of permafrost C. Multivariate gradient analyses show that local vegetation and permafrost are strong predictors of soil chemical properties, overshadowing the effect of climate variables. This study highlights the importance of peatlands, particularly bogs, in bulk soil C storage. Soil organic matter stored in permafrost has higher C:N ratios than unfrozen material. Permafrost bogs constitute the main vulnerable C pool in the region. Remobilization of this frozen C can occur through gradual but widespread deepening of the active layer with subsequent talik formation or through more rapid but localized thermokarst erosion.

1. Introduction

[2] High-latitude ecosystems are considered key components in the global carbon (C) cycle [White et al., 2000]. Over time, soils and peat deposits interact with the atmospheric C pool; photosynthesis sequesters CO2 into soil organic matter (SOM) where microbial decomposition releases C based trace gases (mainly CO2 and CH4). Because of the radiative greenhouse properties of these gases, soil processes also feedback on the global climate system. Throughout the Holocene, expanding northern peatlands have represented a major sink for atmospheric C [Smith et al., 2004]. Today, arctic and subarctic ecosystems harbor large reservoirs of soil C. Different assessments have yielded highly varying estimates of the total northern (tundra and taiga) soil C pool [e.g., Schlesinger, 1977, 1997; Post et al., 1982; Botch et al., 1995; Jobbagy and Jackson, 2000; Smith et al., 2004; Zimov et al., 2006; Schuur et al., 2008], but there is consensus that high-latitude northern soils represent a substantial part of the global C pool. To a large degree, this C is stored in Cryosols and cryic Histosols, where subzero temperatures and waterlogging hinder decomposition [Davidson and Janssens, 2006]. Gruber et al. [2004] identified permafrost soils and peatlands as major terrestrial C pools, vulnerable to remobilization through the processes of permafrost thaw and wetland drying. Schuur et al. [2008] estimate the soil C storage in the northern permafrost regions to be 1024 Pg, with peatlands contributing 277 Pg (to 3 m depth in mineral soils and total peat depth).

[3] The effects of global warming are more pronounced at high latitudes. In the last few decades average temperature increases in the Arctic have been near twice as high as mean global increases [Arctic Climate Impact Assessment, 2004]. This trend is likely to continue and the IPCC predicts increases above global averages in arctic mean temperature and winter precipitation [Christensen et al., 2007], both key factors regulating permafrost distribution. Periglacial environments stand to undergo massive changes [Stendel and Christensen, 2002] and some recent studies report present warming of permafrost temperatures [Lemke et al., 2007]. In Alaska and Siberia, permafrost thaw is changing the size and distribution patterns of thermokarst lakes [Yoshikawa and Hinzman, 2003; Smith et al., 2005]. As permafrost thaws, there is concern that tundra and peatland ecosystems may switch from C sinks into sources, creating a potential for positive feedbacks to global warming [Oechel et al., 1993; Heikkinen et al., 2004]. Other potential effects of permafrost thaw with feedbacks to the global climate system include increased fluvial export of C from periglacial watersheds and changes in the vegetation composition, nutrient dynamics and energy balance of northern ecosystems [Schuur et al., 2008].

[4] Many models of global C cycling do not include permafrost and peatland C pools [Gruber et al., 2004]. Complex feedback mechanisms and data scarcity make it difficult to accurately predict responses of northern ecosystems to global change. While the large size and vulnerability of permafrost and peatland C reservoirs is recognized, more detailed studies surrounding C storage in permafrost and peatlands is needed. Robust estimates of the size, landscape partitioning and quality of soil C pools are needed. We know of no previous studies investigating both regional scale landscape and permafrost partitioning of soil C in permafrost environments.

[5] In this study we describe the landscape distribution, quantity and quality of soil organic matter (including storage in upland soils, lake sediments and peat deposits) across the lowland taiga-tundra continuum of the Usa River Basin, northeastern European Russia. The study area ranges from taiga with isolated patches of permafrost in the south, through the taiga-tundra transition zone with discontinuous permafrost to tundra and continuous permafrost in the north. We compile and analyze databases describing soil chemical and physical properties, permafrost conditions and vegetation for 325 separate sites across the taiga-tundra continuum. Previous estimates of landscape soil C storage in the Usa Basin [see Kuhry et al., 2002] are revised and updated with new calculations and addition of new sites. We describe the partitioning of soil C in 24 different vegetation types, also including estimates of C storage in permafrost soils (above and below the active layer boundary). A subset of the database is analyzed in multivariate gradient analyses to determine how an array of environmental variables are interconnected and linked to site specific SOM quantity and quality. Constrained gradient analyses with Monte Carlo permutations are used to model connections between SOM, permafrost, vegetation and climate patterns.

2. Study Area

[6] The Usa River Basin straddles the Arctic Circle in northeastern European Russia, covering some 93,500 km2 (Figure 1). The Usa River, flowing in a southwesterly direction, constitutes the largest tributary of the Pechora River which in turn drains northward into the Barents Sea. The Ural Mountains with elevations up to 1900 m above sea level (asl) denote the eastern border, but a vast majority (∼85%) of the Usa Basin is lowland terrain ranging between 40 and 200 m asl. The lowlands are underlain by thick quaternary deposits, occasionally intersected by deep river valleys and ravines which constitute the main topographic relief.

Figure 1.

The inset map shows the location of the Usa River Basin and the Pechora and Ob rivers (west and east of the basin, respectively). The large map shows main vegetation patterns and boundaries of the regions defined for upscaling (1, taiga region; 2, (forest)-tundra region; 3, mountain region). Sites used for ordination and upscaling analyses are shown (an additional 23 taiga sites and 87 (forest)-tundra sites are lacking exact coordinates and are not displayed). No vegetation data is shown for the mountain region. Layout adapted from TUNDRA/PERUSA (CD-ROM, 2002). Projection UTM 40 N (WGS 84).

[7] The annual mean temperature in the lowlands ranges from −2.4°C in the southwest to −6.1°C in the northeast. Lowland mean annual precipitation is around 450 to 550 mm with most of the precipitation falling in summer. The orographic effect of the Ural Mountains increases the precipitation in the eastern part of the catchment. The Usa Basin lowlands span four ecoclimatic zones [Kozubov et al., 1999]; northern taiga in the southwest corner passes into extreme northern taiga further north, tundra vegetation patches first appear in the forest-tundra zone which gradually gives way to the tundra zone north of the tree line. For the purpose of soil C calculations and upscaling the Usa Basin is divided into a (1) taiga region including the two taiga ecoclimatic zones and a (2) (forest)-tundra region which includes the forest-tundra and tundra ecoclimatic zone (Figure 1); the (3) eastern mountain region is not included in this study.

[8] An estimated 84% of the basin is classified as permafrost regions, with 10% isolated permafrost, 55% sporadic or discontinuous permafrost and 19% continuous permafrost [Mazhitova et al., 2003]. Isolated permafrost, with temperatures close to 0°C, first appears in peatlands of the taiga zones. Sporadic and discontinuous permafrost underlies the wide forest-tundra transition zone. The northern lowland parts of the basin are dominated by tundra and peatlands underlain by continuous permafrost, with ground temperatures down to −4°C. River valleys are generally permafrost free throughout the Usa Basin. In a 1:1 M scale digital soil map, 17% of the total area is classified as Cryosols with another 20% classified as (potentially) cryic subgroups of Histosols (following World Reference Base for Soil Resources [1998] terminology (TUNDRA/PERUSA, CD-ROM, 2002)).

[9] Spruce (Picea obovata) is the dominant tree species in taiga and forest-tundra but downy birch (Betula pubescens) is also common, especially in young, developing forests (e.g., following clear cuts or fire disturbances). Scots pine (Pinus sylvestris) has a wide tolerance to moisture gradients and mainly appears on and around mires and on sandy upland soils. Subalpine forests including Siberian larch (Larix sibirica) and Siberian fir (Abies sibirica) grow in the foothills of the Ural Mountains. Spruce is the main tree line species in the lowlands, while larch generally forms the alpine tree line. Spruce does not generally grow on permafrost soils in the Usa Basin and the northern most tree stands occur in permafrost free river valleys. Dense willow (Salix spp.) stands are common along running water and in paludified lowland areas throughout the Usa Basin. Open, grass dominated, meadows mainly appear along river systems. Open fen and bog peatlands are found in the whole basin. Fen vegetation varies throughout the basin, but graminoids (e.g., Carex spp. and Eriophorum spp.) and mosses are common. Permafrost free bogs are typically Sphagnum spp. dominated, while raised permafrost (palsa and peat plateau) bogs generally have a dry surface dominated by prostrate dwarf shrubs, mosses and lichen. The northern tundra is dominated by deciduous dwarf shrubs (e.g., Vaccinium spp., Empetrum hermaphroditum) or Betula nana, with lichen or moss.

3. Methods

3.1. Soil Database Compilation and Environmental Information

[10] The estimates of soil organic C storage and partitioning in this paper are based on a soil database compiled from several different sources. The database contains soil chemical and physical descriptions from 373 different sites and includes both upland soils and peatlands. Forty-eight of these sites are from alpine vegetation types and were excluded from further analyses in this study because of the large differences and steep gradients in the factors affecting soil formation and soil organic C storage in alpine areas. A total of 325 lowland taiga and tundra sites remain for further analyses.

[11] All sites contain information on dry bulk density (BD, g/cm3) and C content (weight percentage) or loss on ignition (LOI) of soil layers. For each site, the soil database also lists vegetation cover, depth of soil genetic horizons down to C horizon, depth of top organics, and depth of active layer (if permafrost is present). For a subset of the database there is also data on nitrogen (N) content. Many of the sites are from the TUNDRA (88 sites) and RUSSIAN (220) data sets described and used for upscaling by Kuhry et al. [2002]. In this previous study of Usa Basin soil C storage, these pedons were combined with a land cover classification (LCC) data set of 20 nominal classes. The soil C database was also upscaled using a 1:1 M World Reference Base soil map of the Usa Basin [see Kuhry et al., 2002; Mazhitova et al., 2003].

[12] The TUNDRA and RUSSIAN databases have been revised and updated for this study. All TUNDRA sites are calculated as means of three subsites. Six of these subsites have been allocated to new vegetation classes in this reanalysis of the database. Carbon storage has been recalculated for all TUNDRA sites. Seventeen new sites had not been included in previous soil C estimates (mainly peatlands and thermokarst sediment P. Kuhry, unpublished material, 1995–2001). The vertical resolution of sampling differs between pedons as the sites come from many different sources. The RUSSIAN data set was collected during many years by different scientists and the vertical sampling resolution and site documentation is highly varying. The TUNDRA data set has at least one sample per soil horizon and other pedons are sampled at 5 or 10 cm intervals.

[13] We constructed a locally weighted scatterplot smoothing (LOESS [Wood, 2000]) regression model to approximate C % content for samples where only LOI was available. First, a LOESS regression was calculated based on individual soil samples where both LOI and C % was available (n = 469, span = 0.4, see Appendix B for more information and figure). The regression model was then used to predict C % for the soil samples were only LOI was available (n = 326) (this procedure was not applied to the RUSSIAN database as raw data is not available). The LOESS regression was performed in the open source software R [R Development Core Team, 2007], using the library mgcv [Wood, 2000].

[14] Through earlier research efforts, extensive descriptive and environmental information is available for the Usa Basin. We make use of a Land Cover Classification (LCC) produced by Virtanen et al. [2004a] using Landsat TM satellite data. The 30*30 m resolution LCC covers the entire Usa Basin and was partly ground truthed using the same plots sampled for the TUNDRA soil data set. Climate data for the Usa Basin is from a 30 year run of the 16 km resolution HIRHAM regional climate model. The parameterisation and validation of the model is described further by Christensen and Kuhry [2000]. The output of the HIRHAM model has been corrected against local climate station data and statistically downscaled to 1 km resolution using multiple regression models and kriging [Virtanen et al., 2004b; Van der Linden and Christensen, 2003]. The 1 km resolution data was used in a GIS to derive site climatic conditions for the multivariate ordinations.

3.2. Calculating C Storage, Landscape Partitioning, and Upscaling

[15] Carbon storage for each sampled soil horizon or depth interval is derived from multiplication of BD and C percent content. For each pedon, C storage (kg C m−2) is then calculated to three reference depths: 30 cm and 100 cm depth in mineral ground and peatlands and finally to full depth of peat in peatlands (hereafter termed C-30, C-100, and C-Total, respectively). For permafrost affected sites we also calculate how much C is stored in the active layer and how much C is stored in perennially frozen soil horizons.

[16] For pedons where we lacked C percent or BD data for the lower part of the pedon (C horizon) we applied default values down to 1 m depth when calculating pedon C storage. We applied defaults used by Kuhry et al. [2002], 9.6 kg C m−3 for C horizons in developed soils (forest) and 0.13 kg C m−3 for Cambisols and Fluvisols (tundra and river meadows). A default C horizon BD value of 1.31 g/cm3 was calculated from representative C horizon samples (n = 26, SD = ±0.245).

[17] To estimate landscape-scale storage in the Usa Basin lowlands, all sites are ordered according to a nominal vegetation class scheme selected to correspond with the Usa Basin LCC (see Table 2 for a list of vegetation classes [Virtanen et al., 2004a]). Mean C storage for each vegetation class is calculated as the arithmetic mean of C storage in sampled sites. While forest and tundra vegetation types generally have clear zonal distribution patterns, the azonal vegetation classes willow, fen, bog and lake are common throughout the Usa Basin. However, because of the regional scale gradients in climate and permafrost conditions, mean C storage for these classes were calculated separately for sites in the respective taiga and (forest)-tundra regions. For upscaling purposes, we consider the relative proportions of soil C stored in permafrost affected soils within our database to be representative of the entire ecoclimatic regions in which the sites are located. The results are upscaled to full areal coverage using information from the Usa Basin LCC. Human infrastructure covering ca. 415 km2 (≤0.5% of total area) is excluded from the analyses. Mean soil C storage for each vegetation type is multiplied by the areal coverage within the Usa Basin lowlands to calculate total storage and landscape partitioning of soil C.

3.3. Multivariate Gradient Analyses

[18] A subset of the soil database (n = 65) has been analyzed together with environmental data using the multivariate ordination techniques Principal Component Analysis (PCA) and Redundancy Analysis (RDA) (software CANOCO 4.5 [Ter Braak and Smilauer, 2002]). We selected sites that fulfilled the following criteria: availability of C:N data and exact coordinates.

[19] See Appendix A for a more detailed description of the multivariate ordination methods. Most of the sites used in gradient analyses are from the TUNDRA data set, with 12 additional peatland and thermokarst sites from P. Kuhry (unpublished material, 1995–2001). The data set, which includes an array of soil chemical/physical variables (termed response variables) and vegetation, permafrost and climate variables (termed environmental variables), is summarized in Table 1.

Table 1. Summary of Soil Response Variables and Environmental Variables Included in Gradient Analyses
Variable NameDescription
  • a

    Mean depth of O horizons in upland soils is 7 cm, variable includes top 7 cm of peatlands.

  • b

    Includes all O horizons and peat.

  • c

    C horizons are excluded; mineral contact samples in peatlands are included.

  • d

    All upland tundra classes as well as the two willow classes are amalgamated. The following vegetation classes from the original Usa Basin LCC are not represented in the ordination database (because of very low coverage): herbaceous forests, clear cuts, spruce-fir forest, lakes, river meadows, and bare ground.

  • e

    Sites lacking permafrost are set to 0.

  • f

    From HIRHAM model statistically downscale to 1 km resolution (precipitation adjusted for observations).

Soil Response Variables
C (0–30 cm)Pedon C storage from 0 to 30 cm depth (kg C m−2)
C (30–100 cm)Pedon C storage from 30 to 100 cm depth (kg C m−2)
C (>100 cm)Pedon C storage below 100 cm depth (kg C m−2)
C (% top organic)Percentage of total pedon C stored in top O horizonsa
C (% permafrost)Percentage of total pedon C stored in permafrost
Bulk densityMean bulk density of pedon (g cm−3)
C:N (organic)Mean C:N weight ratio of organic soil horizonsb
C:N (mineral)Mean C:N weight ratio of mineral soil genetic horizonsc
Environmental Variables
VegetationNominal vegetation classes following the terminology of Table 2d
PermafrostSite permafrost expressed as 1/active layer depthe
MAATMean annual air temperature (C°)f
MJATMean July air temperature (C°)f
Precipitation (summer)Sum of mean monthly precipitation (mm) from May to Octoberf
Precipitation (winter)Sum of mean monthly precipitation (mm) from November to Aprilf
Precipitation (annual)Annual precipitation

[20] The environmental variable data set is a combination of information on vegetation and permafrost collected in situ and climate data compiled in a Geographic Information System (GIS). All soil response variables are pedon specific and measured in situ. For each pedon a number of variables measure vertical C storage and partitioning in top organic horizons and permafrost. The Carbon/Nitrogen (C:N) ratio may be used as an indicator of SOM quality [e.g., Kuhry and Vitt, 1996]. The C:N ratio is expressed separately for organic and mineral soil horizons. We also constructed a spatial matrix on the basis of geographical locations of sampling sites. The spatial matrix is used to partial out spatial autocorrelation and ecoclimatic location from the data set, allowing more direct analyses of the impact of other environmental variables. See Appendix A for more details on the construction and testing of the spatial matrix.

[21] The gradient lengths of the soil chemical and environmental data sets were calculated in Detrended Correspondence Analysis (DCA) to 1.848 SD and 0.121 SD, respectively, confirming that our data sets show linear responses suitable for analyses in PCA and RDA [Jongman et al., 1995, pp. 154; Wagner, 2004]. Variables that are expressed as percentages or quotas were arcsin transformed to decrease the influence of extreme values. The data sets were standardized to zero mean and unit variance to enable comparisons between variables expressed in different units and variables on different scales of measurement. The soil and environmental variable data sets are ordinated in two separate PCAs and combined in several RDAs. In the two PCAs, the site permafrost variable and the nominal vegetation variables are included as supplementary variables, meaning that they do not affect the calculations but are projected into the ordination graph for interpretation purposes. In RDA analyses, soil response variables were log transformed to decrease the influence of potential outlier values. The evaluation of environmental variables in RDAs was done through unrestricted Monte Carlo permutation tests (1999 permutations).

4. Results

4.1. Upscaling and Landscape Partitioning

[22] Table 2 summarizes mean and total soil C storage to the three reference depths as well as areal coverage and number of collected samples for all different land cover types. Mean storage for the Usa Basin lowlands is estimated at 38.3 kg C m−2, with 11.3 kg C m−2 stored in the top 30 cm. When deep (>1 m) peat deposits are excluded the number drops to 27.3 kg C m−2. Mean C storage in the taiga and (forest)-tundra regions is quite similar at 37.4 kg C m−2 and 38.7 kg C m−2, respectively. The mean depth of soil genetic horizons in the Usa Basin Lowlands is 81 ± 59 cm, dropping to 60 ± 27 cm if peatlands are excluded (calculated from all sampled sites, excluding water bodies). Thirteen percent of all sites have permafrost above 1 m depth, mainly cryic-histosols in (forest)-tundra peat deposits. An additional 15% of sites have gelic soils (permafrost in the 2nd meter), mainly underlying upland tundra vegetation.

Table 2. Summary of the Databases Used for Upscaling, Ordered According to Vegetation/Land Covera
Vegetation/Land CoverbMean Carbon-30 ± SD (kg C m−2)Mean Carbon-100 ± SD (kg C m−2)Mean Carbon-Total ± SD (kg C m−2)Total Area (km2/%)Total Carbon-30 (Tg C)Total Carbon-100 (Tg C)Total Carbon-Total (Tg C)Mean Soil Depth ± SD (cm)Top of Permafrost 0 < 1 m/1 < 2 m (%)n
  • a

    Lists mean and total soil C storage to the three reference depths, coverage in the LCC, mean depth of soil formation (to C horizon), percentage of sites with top of permafrost table situated between 0 < 1 m and 1 < 2 m depth and total number of sites. Subtotals of total soil C storage and LCC coverage are provided for the main vegetation types: forests, azonal, and tundra.

  • b

    Terminology mainly following Virtanen et al. [2004a]. Human infrastructures are excluded.

  • c

    Including bog pools.

  • d

    Including thermokarst lakes.

  • e

    Polluted tundra vegetation around Vorkuta.

Forests (Spruce Dominated)
Spruce forest8.6 ± 3.214.9 ± 9.414.9 ± 9.410,161/12.286.9151.0151.057 ± 1917
Spruce forest (herbaceous)11.1 ± 2.220.3 ± 4.020.3 ± 4.01131/1.412.623.023.068 ± 2817/–6
Mixed forest9.4 ± 5.214.9 ± 6.814.9 ± 6.86019/7.356.589.589.550 ± 18–/427
Mixed forest (herbaceous)9.1 ± 3.113.8 ± 8.413.8 ± 8.4659/0.86.09.19.150 ± 344
Spruce-fir forest5.6 ± 1.89.7 ± 4.19.7 ± 4.1446/0.52.54.34.329 ± 139
Spruce forest (clear cuts)12.3 ± 8.820.9 ± 16.520.9 ± 16.5303/0.43.76.46.4100 ± 02
 
Forests (Other)
Pine forest4.5 ± 1.36.8 ± 1.46.8 ± 1.4563/0.72.63.83.851 ± 202
Birch forest4.2 ± 1.36.8 ± 2.06.8 ± 2.02905/3.512.219.719.734 ± 174
Subtotal   22,187/26.7183.0306.8306.8   
 
Azonal Vegetation
Bog, Taiga18.5 ± 3.162.3 ± 10.5115.7 ± 51.93039/3.756.3189.4351.8181 ± 5920/–22
Bog ((forest)-tundra)20.2 ± 5.858.6 ± 21.798.3 ± 56.413,266/16.0267.4777.81 303.7151 ± 7893/–20
Fen (taiga)18.3 ± 2.361.3 ± 7.8120.0 ± 57.01836/2.233.6112.6220.4191 ± 7740
Fen ((forest)-tundra)17.9 ± 2.452.0 ± 13.779.8 ± 46.02254/2.740.3117.1180127 ± 8330/–27
Treed peatland16.9 ± 3.542.8 ± 10.842.8 ± 10.85402/6.591.0231.1231.159 ± 186
Willow (taiga)9.9 ± 4.623.6 ± 4.423.6 ± 4.41743/2.117.341.241.288 ± 213
Willow ((forest)-tundra)10.9 ± 6.420.4 ± 11.920.4 ± 11.94744/5.751.796.896.857 ± 316/–18
River meadow5.2 ± 3.217.8 ± 2.917.8 ± 2.91583/1.98.228.128.194 ± 135
Taiga lakec10.2 ± 5.130.2 ± 10.656.7 ± 38.2554/0.75.616.731.4182 ± 1517
Tundra laked8.6 ± 8.117.5 ± 8.232.1 ± 22.52030/2.417.435.665.2163 ± 1194
Subtotal   36,451/43.9588.81646.41246.0   
 
Tundra Vegetation
Shrub moss tundra7.5 ± 3.714.8 ± 5.614.8 ± 5.614,946/18.0112.6220.6220.659 ± 209/4134
Shrub lichen tundra7.1 ± 4.811.0 ± 5.811.0 ± 5.83708/4.526.440.940.959 ± 2325/1712
Dwarfbirch tundra7.8 ± 4.115.1 ± 8.015.1 ± 8.03613/4.428.454.454.659 ± 277/4030
Impacted tundrae6.6 ± 2.211.5 ± 3.511.5 ± 3.5305/0.42.03.53.687 ± 26–/8622
Sparse tundra1.3 ± 0.21.4 ± 0.31.4 ± 0.3472/0.60.60.60.67 ± 3–/1002
Subtotal   23,044/27.8170320320.3   
 
Other
Mainly bare ground0.4 ± 0.030.4 ± 0.10.4 ± 0.11336/1.60.50.60.61 ± 02
 
Basin mean/total11.327.338.383,018942.32273.83177.48113/15325

[23] Open fens and bogs are clearly the most C rich vegetation types, with values from ∼80–120 kg C m−2. Treed peatlands cover a significant part of the Usa Basin, mainly in the taiga zone, but peat thickness is generally lower than in open peatlands. Soils of common upland forest and tundra vegetation classes store around 10–15 kg C m−2 while willows and river meadows have slightly higher values. Forest soils, tundra soils and peatlands cover roughly equal parts of the Usa Basin lowlands (28, 27, and 30%, respectively). Forest and tundra soils each store roughly a tenth of all soil C (Figure 2). Peatlands, however, stand out with an estimated 72% of all soil C, with a majority stored in bog peatlands. Willow stands and meadows (typically occurring along river valleys) account for 5% and lakes store 3% of all soil C. A majority of the soil C (58%) is found in permafrost free soils and more than a quarter of C in permafrost soils is stored above the active layer boundary. Open peatlands hold 96% of permafrost soil C and 98% of all C in perennially frozen horizons. Bogs are again dominant, accounting for 90% of permafrost soil C and 96% of perennially frozen horizon C.

Figure 2.

Overview of soil C partitioning. (left) The proportion of total Usa Basin C stored in different vegetation types. (middle) The permafrost partitioning of the Usa Basin C storage. (right) The partitioning of permafrost C according to land cover type. Some of the land cover types used for upscaling have been amalgamated into groups.

[24] Figure 3 presents estimated C storage and partitioning separately for the taiga and (forest)-tundra regions. In the taiga region, spruce forest and all three peatland classes are important soil C reservoirs. Spruce dominated forest types cover 44% of the taiga region, but account for only 18% of soil C storage. Fens and bogs occupy small areas but are generally deep and store much C. Only 8% of C in the taiga region is stored in permafrost soils and it is almost exclusively found in bog peatlands (palsas). Open bogs (palsas and peat plateaus) dominate the storage in the (forest)-tundra region with 60% of all soil C, most of it perennially frozen. Tundra soils are abundant in the (forest)-tundra region but store only 12% of soil C, little of which is in permafrost soils. While little C is perennially frozen in tundra soils, near half of the tundra soils are underlain by deeper permafrost (between 1 and 2 m, not shown). Soil C storage in lakes as well as willows and meadows is evenly spread between the regions and open sparse vegetation types store negligible amounts of soil C in both regions.

Figure 3.

Graph shows the estimated percentage of total land cover (crosses) and total soil C storage (bars) for different land cover types in the taiga and (forest)-tundra regions. The bars are subdivided to show storage in nonpermafrost soils as well as storage above and below active layers in permafrost soils. Some of the land cover types used for upscaling have been amalgamated into groups.

4.2. Climate, Permafrost, and Vegetation in PCA

[25] The first PCA shows the connections between an array of climatic variables (from the HIRHAM model), permafrost and land cover types for 65 sampling sites selected for multivariate gradient analyses (Figure 4). Principal components (PCs) 1 and 2, respectively, explain 58.5 and 40.5% of all data set variation. Site permafrost and vegetation classes measured in situ are included as supplementary variables and do not affect the ordination. The climatic variables are chosen on the basis of their perceived influence on site permafrost. There are two clear gradients in the data set: the temperature variables to the top left and precipitation variables to the top right. There is no correlation between the temperature and precipitation variables. All forest sites and the taiga region peatland sites are in the upper left “warmer” quadrant while a positioning in the lower right quadrant characterizes most (forest)-tundra vegetation types and willows (eight of the ten willow sites are from the (forest)-tundra region). (Forest)-tundra bogs are an exception and seem to be more affected by low precipitation values than the other northern vegetation types. Site permafrost appears to be equally affected by low temperatures and low precipitation.

Figure 4.

PCA ordination diagram of climatic environmental variables (solid black arrows) and site permafrost and vegetation (gray symbols with italic labels). PCs 1 and 2 explain 58.5 and 40.5% of total data set variance, respectively. The nominal vegetation and permafrost variables are included as supplementary variables and do not affect the ordination. See Table 1 for variable descriptions.

4.3. Soil Properties, Permafrost, and Vegetation in PCA

[26] The second PCA shows patterns in soil chemical/physical properties for the 65 pedons (Figure 5). PCs 1 and 2, respectively, explain 41.4 and 15.6% of all data set variation. Site permafrost and vegetation classes are included as supplementary variables and do not affect the ordination. The first PC is dominated by C storage which is negatively correlated to BD and sites with a high percentage of C in upper soil layers. Although BD directly affects C storage they are still negatively correlated because of the high C percent of the lighter organic soils. All peatland sites are in the right hand side of the PCA, corresponding with high C storage and low BD (more than 80% of the C in the ordination data set is found in organic horizons). Forest and tundra vegetation types are relatively closely spaced along the negative side of PC1 showing that they store comparable amounts of C with high BD. The thin top organic layers are relatively more important in the heavier upland soils. The second PC reflects C:N ratios and permafrost. C:N ratios of mineral soil horizons are negatively correlated to C:N ratios of organic soil horizons and percentage of C stored in permafrost. Sites with a large proportion of SOM stored in frozen soil horizons also have higher SOM quality in organic soil layers. Bog peatlands, in the upper right quadrant, are associated to high C:N ratios and permafrost storage while the position of fens and thermokarst lakes in the bottom right quadrant indicate storage of low-quality organic matter. The negative correlation between C:N ratios of mineral and the overlying organic horizons is likely caused by differences in the turnover rates of SOM. The upland sites are more separated along PC 2. In the upper left quadrant, tundra and pine forests are closely spaced with relatively high C:N ratios and a large proportion of C in top organics (and some frozen C in tundra sites). Willow sites store slightly more C than the forest and tundra sites and have low C:N ratios in organic horizons.

Figure 5.

PCA ordination diagram of soil chemical response variables (solid black arrows) and site permafrost and vegetation (gray symbols with italic labels). PCs 1 and 2 explain 41.4 and 15.6% of total data set variance, respectively. The nominal vegetation and permafrost variables are included as supplementary variables and do not affect the ordination. See Table 1 for variable descriptions.

4.4. Testing Environmental Control of Soil Properties in RDA

[27] Four sets of RDA analyses were performed to test to what extent site specific soil variables can be explained by the available environmental variables. A combination of the climatic variables mean annual air temperature (MAAT) and annual precipitation is also included in the analyses. Most soil response variables described in Table 1 are included in the analyses; only variable 1.5 (C, percent in permafrost) is excluded as it is autocorrelated to site permafrost. The four sets of analyses include testing all environmental variables separately: (1) with no covariables, (2) with vegetation as covariable, (3) with permafrost as covariable, and (4) with spatial covariables.

[28] Table 3 summarizes the environmental variables and their significance as explanatory variables with no covariables and with application of spatial and site specific covariables (ordination graphs not shown). The explained variance is a measure of how much of the total response variable data set variance can be explained by all the environmental variables combined. The p values are from global Monte Carlo permutation tests. The results of the first set of RDAs, without covariables, shows that site specific variables vegetation and permafrost are highly significant while no climatic variables are significant (MAAT and the MAAT + Prec. annual combination are near significant). Using the vegetation classes and site permafrost as covariables reduces the effect of local conditions to bring out regional climatic influences more clearly. Removing the effect of site vegetation greatly reduces the explanatory power of site permafrost but the performance of the climatic variables is generally enhanced. Using site permafrost as covariable does not decrease the high explanatory power of vegetation. The permafrost covariable improves all temperature based variables, while precipitation based variables lose influence. When ecoclimatic location is removed from the data set (through introduction of spatial covariables), vegetation and permafrost remain highly significant but the explanatory power of climatic variables decreases dramatically.

Table 3. Summary of the Environmental Variables Analyzed in RDAa
Environmental VariablesNo CovariablesVegetation CovariablesPermafrost CovariablesSpatial Covariables
  • a

    Analyses show the explanatory power of environmental variables over soil response variables. Monte Carlo permutations provide statistical significance of separate environmental variables under varying covariable setups. Near significant variables (p< 0.1) are in italic. One asterisk denotes a statistical significance of p < 0.05, two asterisks denote a statistical significance of p < 0.01, and three asterisks denote a statistical significance of p < 0.001. The explained variance entries show the sum of all canonical axes.

Vegetation0.001***<0.001***<0.001***
Permafrost0.001***0.30<0.01**
MAAT0.070.08<0.05*0.33
MJAT0.160.13<0.05*0.41
Precipitation (annual)0.140.110.310.50
Precipitation (winter)0.140.120.290.55
Precipitation (summer)0.140.120.320.44
MAAT and precipitation (annual)0.08<0.05*<0.05*0.43
Explained variance55.0%6.4%48.5%45.6%

5. Discussion

5.1. Total Storage and Landscape Partitioning

[29] Estimates of total storage and landscape distribution of C in this study are based on an upscaling methodology where an empirical connection between nominal vegetation classes and soil C storage is established through in situ sampling of soils in a wide range of vegetation types. Areal coverage is subsequently obtained through a satellite LCC.

[30] Our estimates of mean soil C storage in taiga and tundra biomes (37.3 and 38.7 kg C m−2, for the respective regions) are notably higher than previous estimates of western Eurasian reservoirs. Kolchugina et al. [1995] estimate that Russian taiga soils store 27.0 kg C m−2, while forest-tundra and tundra soils store 20.0 and 21.4 kg C m−2, respectively. Chestnyck et al. [1999] estimate storage in East European Russian tundra soils at 17.8 kg C m−2. Stolbovoi [2002, 2006] provide the estimates 26.9 kg C m−2 in soils of northern taiga and forest-tundra and 16.6 kg C m−2 in tundra soils. The current estimate of soil C storage for the Usa Basin lowlands (38.3 kg C m−2) is also slightly higher than the previous, methodologically alike, estimate by Kuhry et al. [2002] of 36.0 kg C m−2 (calculated from the 2002 database for 1m in mineral soils and full peatland depth with alpine habitats excluded). Our calculations are updated with a more accurate regression model for translating LOI to C percent, a more detailed subdivision of sites into separate vegetation classes and new sites (mostly peatland sites). While all the abovementioned changes contribute to the increased estimate, the biggest difference is due to an improvement in peatland representation. We estimate that peatlands hold 72% of all soil C in Usa Basin taiga and tundra, with only 30% of the surface coverage (calculated to 1 m depth in mineral soils and full peat depth). The integral role of northern peatlands in soil C storage has long been recognized [Gorham, 1991], but estimates of global storage differ widely [Gruber et al., 2004]. The uncertainties in these estimates often reflect the difficulties in assessing the real extent, depth and C content of remote and vast northern peatlands. Correct spatial representation of the fine vegetation mosaic in northern regions is a challenge, and the choice of land cover data set for upscaling or modeling is crucial [Virtanen and Kuhry, 2006].

5.2. Vertical Distribution of C in Pedons

[31] The vertical partitioning of C storage shows concurring patterns between the taiga and (forest)-tundra regions; ∼29% of C is in the top 30 cm, ∼42% is found between 30 and 100 cm and ∼28–29% is below 100 cm depth (where only peatland C storage is taken into account).

[32] There is a general trend that soils of azonal vegetation types (values calculated separately for the two regions) are deeper and more developed in the forested, southern parts of the catchment and become more shallow northward (see Table 2). Soil formation in taiga zone willow stands is on average 35% deeper than in their (forest)-tundra counterpart. Bog and fen peatlands are on average 17 and 27% deeper in the taiga zone than in the (forest)-tundra zone. However, mean storage in taiga is similar to (forest)-tundra because taiga peatlands occupy a smaller portion of the landscape.

[33] Near surface C storage is roughly equally divided between mineral soils and peatlands (peatlands have 53% and upland soils 47% of total terrestrial 0–30 cm C in the Usa Basin, Table 2). Peatlands account for 72% of the C stored between 30 and 100 cm depth (lake sediments excluded). In many periglacial regions, cryoturbation of mineral soils transfers SOM from the surface into deeper soil layers, leading to an accumulation of deep C over time [Bockheim, 2007]. There is little evidence of significant burial of organic C through cryoturbation in cryosols in the Usa Basin, as C storage in tundra sites is generally concentrated in the top organics.

5.3. Permafrost Storage

[34] Throughout the Usa Basin, 42% of the total soil C store is found in Cryosols or cryic-Histosols (with permafrost <1m depth), a third of which is in the active layer. In overview, 84% of the Usa Basin is classified as permafrost region. In a 1:1 M scale soil map, 37% of the terrain is classified as Cryosols or cryic subgroups of Histosols. At site level, 13% of 325 sites have permafrost <1 m depth, and an additional 15% of sites have permafrost <2 m depth. The gradual decrease of permafrost coverage with increased scale reflects the discontinuity and patchy spatial distribution of permafrost, especially in sporadic and discontinuous permafrost zones.

[35] Permafrost C storage is completely dominated by peatlands which account for 95% of organic C in permafrost soils and 98% of perennially frozen C (below active layer of permafrost soils). While a majority of the tundra vegetation sites are underlain by gelic soils, depth limitations exclude these deep permafrost soils from the study (1 m depth used for mineral soils in upscaling, 1–2 m depth of permafrost table in gelic soils). However, C content in these deeper layers is low, particularly because of the lack of significant cryoturbation in the area, and including the deeper permafrost layers in the estimates would not greatly affect the outcome.

[36] The large pool of permafrost C in peatlands is concentrated in raised bog peatlands (palsas or peat plateaus). Bogs hold some 90% of all permafrost C and >95% of perennially frozen C in the Usa Basin. The corresponding numbers for fen peatlands are 5% of C in permafrost soils and 2% of perennially frozen C. When permafrost is present in fen peatlands, the active layer boundary is often near the peat-mineral contact surface so that little peat is perennially frozen. Permafrost bogs are raised above the water table and thick layers of dry peat insulate the frozen layers in summer. The raised topography can also contribute to thinner insulating snow covers in winter, helping to maintain permafrost once it has established. Wet fen peat has high thermal conductivity during summer and low-lying microtopographic positions favor accumulation of insulating snow in early winter, creating local scale conditions unfavorable for permafrost formation. Palsas and peat plateaus in the Usa Basin formed as early as 9000–10000 14C a BP. They first developed as permafrost-free fens during the Early/Middle Holocene Hypsithermal. Permafrost aggradation has been reported between 3000 and 2000 14C a BP and as recent as the Little Ice Age. Frost heave resulted in the present dry bog surfaces [Oksanen et al., 2001, 2003; Kultti et al., 2004].

5.4. Patterns of Climate, Vegetation, and Permafrost in PCA

[37] The first PCA is driven by high-resolution (1 km) modeled climate parameters and climate indexes, with site permafrost and vegetation added as supplementary variables (Figure 4). The ordination results show that temperature is not correlated to precipitation but clearly divides taiga vegetation types from tundra vegetation. (Forest)-tundra bogs and thermokarst lakes, typically associated to permafrost, are closely related to in situ permafrost. As it is the modeled climate of the sites that determines the ordinated location, the positioning of the thermokarst and bog sites shows that the modeled climatic parameters can be used to identify conditions conducive to permafrost formation (all thermokarst sites naturally lack permafrost). The analysis shows that permafrost occurrence is not directly correlated with MAAT, mean July air temperature (MJAT), or summer/winter precipitation. It rather seems to be a function of low temperatures combined with low precipitation. It has long been recognized that the duration and thickness of winter snow cover is important for the ground thermal regime, especially in discontinuous and sporadic permafrost regions [Zhang et al., 2001]. Thick snow cover acts as an effective insulator and protects the ground from low air temperatures during winter, thus impeding frost penetration into the ground. But summer precipitation may also affect ground thermal regimes. High, recurrent summer precipitation at a site would increase the number of days when the organic soil horizons of the active layer are wet, thus increasing thermal conductivity and promoting permafrost thaw.

5.5. Soil Properties and Environmental Patterns in PCA

[38] The second PCA is driven by in situ data of soil chemical and physical properties, with site permafrost and vegetation added as supplementary variables (Figure 5). A strong storage gradient along PC1 (41.4% of total variance) emphasizes the role of all peatlands as being C rich, while both taiga and tundra upland sites have similar C storage despite vegetation, echoing the results from the upscaling. It is also evident that upland soils have high BD and store a high proportion of C in the top centimeters, with opposite trends for peatlands. Of the nonpeatland sites, Pine forest and tundra store least C and also have the highest concentration of C to top organic layers. Willow sites show an opposite trend while spruce dominated forests are intermediately positioned.

[39] The C:N ratio is a useful indicator of SOM quality. As a peat sequence develops, the decay of N seizes in the anaerobic catotelm while C continues to decay through e.g., methanogenesis and sulphate reduction [Kuhry and Vitt, 1996]. There is a gradient of SOM quality and permafrost C storage along PC2 (15.6% of total variance). High C:N values in organic layers are associated with bogs and a high percentage of frozen peat. As permafrost aggraded the slow anaerobic decay largely seized and C:N ratios remain high [Sannel and Kuhry, 2009]. Thermokarst and taiga fen peat is generally permafrost free and also appears to be more decomposed. There is no indication that sites with permafrost store more soil C than permafrost free sites.

[40] Mineral layer C:N is negatively correlated to the C:N of the overlying organic layers. This is probably related to the varying turnover times of organic material in soils. Soils with long turnover time have high C:N ratios in organic horizons (e.g., pine forest and tundra), but the low rates of soil formation means that little fresh organic matter is incorporated into lower soil horizons. Willow sites, on the other hand, have a small portion of total C in the top organics, corresponding with high turnover rates reflected in the low C:N ratios of organic soil horizons but relatively high C:N ratios of mineral soil horizons.

5.6. Testing the Predictive Power of Environmental Variables

[41] RDAs combined with Monte Carlo permutation tests provide a tool for statistically modeling connections between soil chemical response variables and environmental variables (site vegetation and permafrost as well as modeled climate). Environmental and spatial data matrixes can be introduced as covariables to remove their influence from the analyses.

[42] The percentage of response variable variance explained by the included environmental variables is notably high (55%, Table 3), but it decreases dramatically when the vegetation covariable is introduced. Site vegetation is a strong predictor of soil chemical properties, and is not affected by the introduction of covariables. Permafrost is a significant predictor of soil properties without covariables and with spatial covariables; but introducing the vegetation covariables greatly decreases the explanatory power of site permafrost. Permafrost occurrence is tightly linked to northern peatlands in the ordination data set. Introducing the vegetation covariable removes all of the response variable variation that can be attributed to vegetation, and the permafrost variable does not contribute significant predictive power beyond what the vegetation is already explaining. If the tundra peatland classes are removed from the vegetation covariable matrix, site permafrost remains a significant explanatory variable (data not shown).

[43] None of the modeled climate variables are statistically significant predictors when tested without covariables. The strong effects of local conditions override and mask the influence of climate on soil properties. Removing the effects of site vegetation slightly increases the performance of all climate variables, but only combining temperature and precipitation gives statistical significance. The permafrost covariable improves all temperature based variables, while precipitation based variables lose influence. Removing the impact of permafrost from the data set likely brings out the effect a temperature gradient on SOM quality (higher turnover rates in warmer temperatures). Precipitation mainly affects SOM properties indirectly, through its influence on permafrost patterns, and consequently loses explanatory power with the permafrost covariable.

6. Conclusions

[44] The Usa Basin lowlands span across the taiga-tundra continuum of western Eurasia, including all permafrost zones, from isolated patches in the south to continuous permafrost in the north. Mean soil C storage in taiga and (forest)-tundra regions is estimated at 37.3 and 38.7 kg C m−2, respectively. Peatlands account for 72% of soil C storage with only 30% of the surface area. Permafrost within the upper meter (signifying Cryosols or cryic Histosols) is found in 13% of investigated sites. These permafrost soils store 42% of the total soil C pool, of which nearly three quarters is perennially frozen below the active layer. Peatlands hold 95% of all C in permafrost terrain and 98% of all perennially frozen C, mainly in bog peatlands of the (forest)-tundra region. Multivariate gradient analyses further emphasize the role of bog peatlands in SOC storage. SOM stored in permafrost has higher C:N ratios than unfrozen material, indicating that this material is more labile and susceptible to decay if thawed.

[45] Gradient analyses of climatic patterns show that, at this regional scale, site permafrost is equally affected by temperature and precipitation variables. Site vegetation and permafrost are strong predictors of soil chemical properties, also when regional ecoclimatic patterns are accounted for (through introduction of a spatial covariable matrix). While permafrost loses its predictive power when the vegetation covariable is used, vegetation remains a strong predictor despite introduction of covariables and explains a significant part of the variance in soil properties. Local vegetation and permafrost conditions overshadow the effect of climate variables on soil properties. Introducing covariables to mask out the local conditions shows that a combination of MAAT and annual precipitation is a strong predictor of site specific soil properties.

[46] The results from this study highlight the importance of permafrost bogs as stores of large amounts of labile C. It is important that the representation of these C hot spots in global C estimates and models is as accurate as possible. Permafrost bogs are common throughout the subarctic and arctic, often found in areas no longer climatically optimal for permafrost formation [Christensen et al., 2004]. Permafrost thaw resulting in remobilization of this frozen C pool may occur through deepening of active layers or through thermokarst formation. Active layer deepening is a gradual process with large areal extent that may eventually lead to talik formation (a layer or body of perennially unfrozen ground occurring in permafrost terrain). Thermokarst formation is more localized but may occur rapidly, generally leading to altered surface hydrological conditions. While there is evidence for the active occurrence of both active layer deepening and thermokarst [Lemke et al., 2007], the relative and total magnitude of changes in C fluxes resulting from these processes remain uncertain.

Appendix A:: Multivariate Gradient Analyses

[47] See Kent [2006] for a comprehensive review on the use of ordination methods in biogeographical research. PCA is an unconstrained ordination technique commonly used for reducing dimensionality and extracting patterns from multivariate data sets with linear responses to gradients of change. PCA calculates theoretical variables (termed Principal Components, PCs) that minimize the residual sums of squares between all observations in a multidimensional matrix and the PC axes. The PCs are all orthogonal in multidimensional space (i.e., completely uncorrelated), maximizing the amount of information that is retained in each PC. RDA is an extension of PCA where the information contained in an explanatory data set of environmental variables is used to constrain the response variable data set. For each site, it only extracts the information in the soil response variables that can be explained through variations in the environmental variables. In RDA the ordination axes are derived from PCA and the technique assumes a linear response model (as opposed to a Gaussian response model). The intended use of the RDA analysis is to relate the abundance of a set of species (dependent variables, data set 1) to data describing a suite of environmental variables for the sampling sites (independent variables, data set 2).

[48] Monte Carlo permutations evaluate the null hypothesis that the soil response variables are unaffected by the environmental variables in RDA by randomized permutations of residual sums of squares, using the F ratio as test statistic [Ter Braak and Smilauer, 2002]. In the forward selection process, the environmental variable with the highest explanatory power is tested first (and included if the p value is accepted) after which the explanatory power of the remaining variables are recalculated, factoring in the variance already accounted for by the included variable. The process is used to avoid overfitting the RDA model with redundant environmental variables.

[49] A spatial matrix based on geographical locations of sampling sites can be used to partial out spatial autocorrelation and ecoclimatic location from the data set, allowing more direct analyses of the impact of other environmental variables. The matrix is constructed by adding all terms for a cubic trend surface regression (see Borcard et al. [1992] for further details). In this study a matrix based on latitude and longitude positions was used in the RDA analyses. To avoid over fitting of the model, stepwise Forward Selection of variables was used to identify the variables with highest explanatory power. The selection reduced the matrix to the three variables: x*y2, x2, and y3, together explaining 14.6% of the variation in the soil response variable matrix (Monte Carlo p values at time of inclusion in the model where 0.02, 0.06, and 0.04, respectively). The spatial covariables describe the following gradients in investigated soil properties: (x*y2) a steep diagonal gradient where a stepwise longitudinal increase (eastward) is matched by a squared latitudinal increase (northward), (y3) a weak residual latitudinal gradient and finally (x2) a longitudinal gradient. The first two gradients correspond to the major climatic gradients in the Usa Basin: MAAT is highest in the southwest and lowest in the northeast. The longitudinal gradient corresponds to a general increase in precipitation eastward in the Usa Basin. It is therefore unsurprising that introducing spatial covariables reduces the predictive power of all climate based variables and indexes. The spatial covariables do not affect site specific variables, confirming that local vegetation and permafrost are influencing soil properties throughout the studied area.

Appendix B:: LOESS Regression Model

[50] Figure B1 shows the locally weighted scatterplot smoothing (LOESS) regression model used to translate LOI percent into C percent for sites were only LOI was available (n = 326). The model is based on samples were both LOI percent and C percent was available (n = 469, C percentages determined using elemental analyzers). In the bottom left, a vast majority of samples from mineral soil horizons have <5% C. Peat samples show more spread but C content is typically 40–55%. Intermediate samples are typically in transitions between organic and mineral horizons (A horizons or basal samples from peat deposits) or gyttja.

Figure B1.

LOESS regression model for translating LOI into C percent. Regression is based on soil samples were both LOI and C percent data are available (n = 469). LOESS span is 0.4.

Acknowledgments

[51] Much of the material used in this study was collected and compiled in the EU 4th Framework TUNDRA project (contract ENV4-CT97-0522). Especially work by T. Virtanen (University of Helsinki, Helsinki), J. H. Christensen (Danish Meteorological Institute, Copenhagen), and G. G. Mazhitova (Komi Science Centre, Syktyvkar) has provided a basis for this work. Further compilation and analyses are funded through the EU 6th Framework CARBO-North project (contract 036993) and a grant of the Swedish Research Council to P. Kuhry. Comments from two anonymous reviewers helped to improve the manuscript.

Ancillary