Small net carbon dioxide uptake by Russian forests during 1981–1999



[1] A permafrost-enhanced biogeochemical process model using observed climate and CO2 data, and satellite-observed maps of forest composition and density predicts a moderate biomass increment and carbon sink of 74 and 131 Tg carbon per year (TgC/a) in the Russian forests during 1981–1999. The enhanced process model realistically represents ecosystem state in terms of river runoff, area burned by fires, vegetation productivity and biomass, in comparison to monitoring and inventory data. Rising atmospheric CO2 content is found to have been the main cause of the carbon sink. Amounting to 7% of carbon emissions from fossil fuel emissions in Eurasia, our results demonstrate a limited capability of the Russian boreal forest in its current state to compensate anthropogenic carbon emissions.

1. Introduction

[2] The terrestrial biosphere north of 25° latitude is believed to have sequestered 1–2 PgC/a from the atmosphere during the 1990s, offsetting anthropogenic carbon emissions by that amount [e.g., Prentice et al., 2000]. Magnitude, spatial patterns and underlying mechanisms of this sink are still under debate. Analyses of forest inventory data identify a moderate biomass increase in the forests of Russia of ∼76 TgC/a during the 1980s and 1990s [Shvidenko and Nilsson, 2003] while a much stronger increment in woody biomass of ∼284 TgC/a has been derived by remote sensing [Myneni et al., 2001]. However, changes in biomass are only one component of the carbon balance of forests. Net carbon exchange between the land surface and the atmosphere is determined by the balance of net primary production (NPP) of the vegetation, heterotrophic respiration of the soil (Rh) and disturbances (mostly fire). On a continental scale, these quantities can be quantified with a biogeochemical process model. Here we use the LPJ Dynamic Global Vegetation Model (LPJ-DGVM) [Sitch et al., 2003] to estimate carbon fluxes and pools in the forests of Russia as a function of monthly climate observations, soil texture and atmospheric CO2 concentration. The model was linked to satellite-observed data sets of vegetation distribution and fraction [Hansen et al., 2003; Bartholome and Belward, 2005] for a representation of land cover heterogeneity. In addition, freeze-thaw impacts on carbon and water fluxes, which are crucial in the boreal zone [Bonan and Shugart, 1989], are explicitly simulated. In doing so, we (i) estimate the biomass change and carbon balance of the Russian forests during 1981–1999, and (ii) determine the climatic factors that are responsible for a carbon uptake or release.

2. Methodology

2.1. Model-Based Estimation of Carbon Pools

[3] The LPJ-DGVM combines large-scale representations of terrestrial vegetation dynamics and land-atmosphere carbon and water exchanges in a modular framework. For a detailed description and evaluation of the model see Sitch et al. [2003] and Gerten et al. [2004]. To simulate the size of the carbon pools leaves, heartwood, sapwood, fine roots, litter and soil, the model explicitly considers key ecosystem processes such as photosynthesis, plant growth, mortality, resource competition, disturbances and respiration. Combined carbon and water fluxes are modeled on a pseudo-daily basis from monthly inputs but vegetation dynamics annually. To account for the variety of structure and functioning among plants, 4 Plant Functional Types (PFTs) are distinguished for the boreal zone: Evergreen needleleaf, deciduous broadleaf, deciduous needleleaf and grass vegetation.

[4] They contribute to the calculated carbon and water fluxes of the whole grid cell subject to the simulated and observed vegetation cover since the maximum cover of each PFT is constrained by observations (section 2.2). The extend of Russian forests is defined as all grid cells with a tree foliage cover greater than 15% in this study (see auxiliary material). Wetlands are not represented.

[5] The model was driven by 0.5 gridded monthly fields of air temperature, precipitation, number of rainy days of a month and cloud cover [Mitchell and Jones, 2005; Oesterle et al., 2003], and by nine soil texture classes [Food and Agriculture Organisation, 1991]. Annual CO2 concentrations were used that were derived from ice-core measurements and atmospheric observations, provided by the Carbon Cycle Model Linkage Project [McGuire et al., 2001], as a non-gridded global input. LPJ was run from 1901 until 2003 at a daily time step, preceded by a 1000-year spinup period using 1901–1930 climate data to bring the carbon pools and vegetation cover into an equilibrium with climate.

2.2. Satellite-Derived Land Cover for Use in a DGVM

[6] We combine two different types of remote sensing products, dominant land cover type and fractional vegetation coverage, to derive six maps with fractional coverage of the four boreal PFTs considered in the model, agricultural areas and water bodies. The information about the dominant land cover (1-km pixel size) originated from the Global Land Cover 2000 (GLC2000) project of the Joint Research Centre of the European Commission which is based on SPOT Vegetation data from 2000 [Bartholome and Belward, 2005]. The fractional tree and grass cover (500-m pixel size) was provided by the Vegetation Continuous Fields (VCF) map of the University of Maryland generated from MODIS data from 2001 [Hansen et al., 2003].

2.3. Permafrost Modeling Within a DGVM

[7] In the LPJ-DGVM soil temperature is assumed to follow a sinusoidal cycle of surface air temperature with a damped oscillation about a common mean, and a temporal lag [Campbell and Norman, 2000; Sitch et al., 2003]. Within permafrost regions, the null of this equation (0°C isotherm, daily thaw depth) is calculated numerically by applying Newton's algorithm. In doing so, the temperature below near-surface biomass (aboveground litter and grass biomass) is used, i.e., damping effects of snow and near-surface biomass are taken into account. Phase change is considered to directly impact soil moisture. In addition, water holding capacity is adjusted to the daily thaw depth. The extent of the permafrost zone is derived by the frost-index, with a threshold value of 0.6 representing the changeover from continuous to discontinuous permafrost [Nelson and Outcalt, 1987]. Outside of permafrost regions, the impact of exact frost depth in winter on vegetation is neglected, rather the whole soil is assumed to be frozen when temperature in 10 cm depth is below 0°C. Thus, the time lag between the start of snow melt and the ability of the soil to receive water in spring is also considered in non-permafrost regions of the boreal zone.

3. Results and Discussion

3.1. Model Evaluation

[8] Evaluation (see auxiliary material) of the twofold enhanced model was performed with respect to the interlinked boreal ecosystem components hydrology, fire disturbances, NPP, Rh and biomass. In this study, computed ecosystem parameters were chiefly evaluated against results produced during the project SIBERIA-II [Schmullius and Hese, 2002]. The study region spans a 3 Mio km2 transect swath from the Arctic Ocean to the southern steppe, and between the Yenisey river and Lake Baikal in Central Siberia. As part of this study, the International Institute for Applied Systems Analysis (IIASA) has provided new inventory-based information on NPP, biomass, Rh, and carbon emission by fires of this area for the year 2003.

[9] A comparison of discharges from large Arctic Siberian rivers with modeled runoffs in their watersheds demonstrates that both freeze-thaw processes in the soil and evapotranspiration have a significant impact on the hydrological regime in Northern Eurasia (Figure 1). Large runoff peaks in spring are fully explained by the frozen state of the ground during snow melt. Runoff values in summer and autumn, however, are mainly determined by vegetation density. This is evident from the effect of assimilating satellite-observed lower vegetation densities into the model.

Figure 1.

Comparison of modeled watershed runoff of the Lena river and observations of its discharge [Vörösmarty et al., 1996] between 1936 and 1984, (a) monthly long-term means and (b) annual runoff dynamics (see auxiliary material). The modeled monthly long-term mean values are shifted by 1 month to allow for time lags in lateral water flow. LPJ+perm and LPJ+perm+sat denote to model results with the consideration of freeze-thaw processes and the coupling to satellite-derived land cover maps, respectively, in addition to the model version as published before (LPJ original) [Sitch et al., 2003; Gerten et al., 2004].

[10] The water loss during snow melt due to surface runoff together with beginning transpiration by vegetation leads to dry conditions of the thawed upper active layer in spring. This drought period is responsible for a peak in fire frequency in Siberia in spring [Korovin, 1996]. If freeze-thaw effects are removed from the computations, fire return intervals of up to 1000 years are simulated in Northern Eurasia. With the improved representation of the hydrological regime, fire return intervals of 100–200 years are simulated, in accordance with observations [Bonan and Shugart, 1989]. The fire model embedded in the LPJ-DGVM [Thonicke et al., 2001] uses a statistical approach that does not resolve the interannual variability of fire occurrence, but the simulation of soil moisture taking into account freeze-thaw effects leads to a correct prediction of the mean fire probability (see auxiliary material). In the SIBERIA-II evaluation region, on average 1.2 Mha burned annually from 1992 to 2003 as derived from satellite data [Balzter et al., 2005]. The enhanced LPJ-DGVM calculates 1.5 Mha. The fire return interval has a substantial direct effect on biomass. In 2003, 43.3 TgC were consumed by fire in the SIBERIA-II study region based on the GIS analysis by the IIASA using burned area estimates from [Balzter et al., 2005]. The improved LPJ-DGVM independently agrees with 42.8 TgC. Except by fire, the amount of carbon stored in the terrestrial biosphere in Siberia is strongly determined by NPP. A function of the rather short summer season, NPP decreases with latitude in Central Siberia. By constraining the simulated vegetation fraction with satellite-derived data we are able to reproduce well the latitudinal variation of both NPP and biomass in Central Siberia and hence the underlying biogeochemical processes (Figure 2). NPP of the 3 Mio km2 SIBERIA-II study region is estimated to have been 255 gC/m2/a in 2003, which deviates only by 2.4% from the result of the IIASA GIS based inventory approach, 249 gC/m2/a.

Figure 2.

(a) Net Primary Production and (b) vegetation carbon density as a function of latitude in the SIBERIA-II study transect (80 to 110°E) in 2003 as modeled by LPJ-DGVM versions (with and without permafrost simulation, satellite-observed vegetation density patterns) versus inventory data of the International Institute for Applied Systems Analysis, Austria.

[11] Our NPP estimates are at the high end of independent estimates of 123–250 gC/m2/a in the Russian forests [Schulze et al., 1999; Lloyd et al., 2002; Röser et al., 2002; Shvidenko and Nilsson, 2003] which are based on eddy covariance measurements or forest inventory data (see auxiliary material). The modeled north-south gradient of biomass (Figure 2b) illustrates the crucial effects of both permafrost (through fire) and vegetation fraction distribution on biomass.

3.2. Climatic Causes for Changes in Carbon Pools

[12] The agreement of the DGVM output with observed river runoff, remotely sensed burned area and with inventory-based NPP and biomass gradients demonstrates that the LPJ-DGVM is able to simulate the correct magnitude of carbon exchange of the boreal forest biome in Central Siberia. It increases confidence also in the terrestrial carbon balance derived (see auxiliary material). For the whole of Russia, we calculate a moderate increase in the biomass of the Russian forests of 74 TgC/a between 1983 and 1998. This result agrees with the inventory-based analysis of changes, which arrives at 76 TgC/a [Shvidenko and Nilsson, 2003]. In contrast, the change analysis of remotely sensed cumulative growing season Normalised Difference Vegetation Index (NDVI) led to a much larger biomass increase of 284 TgC/a [Myneni et al., 2001]. The discrepancy with our results originates from differences in the estimates mainly for dense forests in the southern taiga. Here, cumulative NDVI changes were translated into biomass changes of up to 77 gC/m2/a [Myneni et al., 2001] while the LPJ-DGVM calculates values lower than 23 gC/m2/a.

[13] In order to determine the climatic factors that were responsible for this biomass increase, runs of the LPJ-DGVM were made for the time period 1901–2003 with selected climatic inputs allowed to vary as observed while keeping the remaining monthly inputs constant at average values (including the spinup time period). Previously it was shown [Lucht et al., 2002] that variations in temperature account for nearly all of the variability of LAI and NPP in the global boreal zone. Our results, however, show that increasing atmospheric CO2 concentrations were an additional cause for rising NPP in the Russian forests during the last two decades (Figure 3a). As a result of increasing NPP and on average nearly constant disturbances, the resulting biomass of Russian forests is estimated to have increased during the 1980s and 1990s. This increase is explained by both changes in atmospheric CO2 content and temperature (Figure 3b). The simulated increase in soil and litter carbon densities is due to increased litter fall (Figure 3c). Simulations with constant meteorological input under rising CO2 show the same behavior while trends in temperature lead to decreasing model estimates of soil and litter carbon densities, due to the temperature dependence of Rh [Lloyd and Taylor, 1994]. This effect acts to counter-balance the temperature-dependent carbon gain in vegetation. The resulting net carbon balance of Russian forests in vegetation, litter and soil is seen to have been mostly a result of increasing carbon dioxide concentration in the atmosphere (Figure 3d).

Figure 3.

(a) Anomalies of simulated NPP and carbon density of different pools: (b) biomass, (c) litter and soil (d) and vegetation, litter and soil in Russian forests. Model results are shown for full inputs (LPJ) and runs with constant climatic inputs at 1901–2003 averages but variable temperature (LPJtmp), precipitation (LPJpre), cloudiness (LPJcld) or CO2 content (LPJco2). LPJall represents model results with only constant climatic inputs.

[14] The carbon sink in the Russian forests between 1981 and 1999 is quantified from these computations to have been 131 TgC/a. The largest uncertainty in this figure is caused by the fire model used, which is based on climatically dependent ignition probabilities rather than actual ignitions, producing fire emissions with reduced interannual variability. If the actual occurrence of fires should have been skewed toward later years in the period (for example, a doubling of burned area between 1990–1994 and 1995–1999 has been reported for Siberia [Conard et al., 2002]), carbon emissions from fires may have increased by 110 TgC/a rather than 40 TgC/a (mean modeled emission of 220 TgC/a, and assuming a factor of two increase for the whole of Russia from 1981–1999). The residual terrestrial sink would then have been only 61 TgC/a. This figure does not take into account, however, that more fire would have simultaneously reduced Rh, and it would be reconcilable with the observed [Shvidenko and Nilsson, 2003] biomass increase (see above) if the model underestimated NPP increase or overestimated mortality. In addition, forest expansion and management could compensate higher disturbance-induced losses but are not represented by this model analysis.

[15] Considering the area of these forests (21% of the northern-hemisphere extra-tropical land), this carbon sink of between 7 and 13% of the total terrestrial carbon sink of 1–2 PgC/a in the Northern Hemisphere in the 1990s [e.g., Prentice et al., 2000] signals that other ecosystems or processes in the Northern Hemisphere have been intensively contributing to the global carbon balance, and that the Russian forests have a limited ability (∼7%) to offset CO2 emissions from fossil fuel use in Eurasia (European and FSU states).


[16] We are indebted to H. Balzter and C. George from the Centre for Ecology and Hydrology Monks Wood, UK for providing satellite-derived burned area over the SIBERIA-II region from 1992 until 2003. Furthermore, we thank S. Sitch, B. Smith, D. Gerten, S. Schaphoff, K. Thonicke, I. McCallum and S. Quegan for valuable comments, data and contributions to the LPJ code. Financial support came from the European Commission for the SIBERIA-II project (EVG2-2001-00008) and the German Ministry for Education and Researcher DEKLIM/CVECA project (W.L., partly).