Ecosystems

, Volume 10, Issue 3, pp 380–401

Projected Changes in Terrestrial Carbon Storage in Europe under Climate and Land-use Change, 1990–2100

Authors

    • Potsdam Institute for Climate Impact Research (PIK)
    • Laboratoire des Sciences du Climat et de l’Environnement
  • Alberte Bondeau
    • Potsdam Institute for Climate Impact Research (PIK)
  • Timothy R. Carter
    • Finnish Environment Institute
  • Wolfgang Cramer
    • Potsdam Institute for Climate Impact Research (PIK)
  • Markus Erhard
    • Potsdam Institute for Climate Impact Research (PIK)
    • European Environment Agency
  • I. Colin Prentice
    • Max-Planck-Institute for Biogeochemistry
    • Department of Earth SciencesUniversity of Bristol
  • I. Reginster
    • Department of GeographyUniversite Catholique de Louvain
  • Mark D. A. Rounsevell
    • Department of GeographyUniversite Catholique de Louvain
  • Stephen Sitch
    • Potsdam Institute for Climate Impact Research (PIK)
    • Met Office (JCHMR), Crowmarsh-Gifford
  • Benjamin Smith
    • Geobiosphere Science Centre, Physical Geography and Ecosystems AnalysisLund University
  • Pascalle C. Smith
    • Potsdam Institute for Climate Impact Research (PIK)
    • Laboratoire des Sciences du Climat et de l’Environnement
  • Martin Sykes
    • Geobiosphere Science Centre, Physical Geography and Ecosystems AnalysisLund University
Article

DOI: 10.1007/s10021-007-9028-9

Cite this article as:
Zaehle, S., Bondeau, A., Carter, T.R. et al. Ecosystems (2007) 10: 380. doi:10.1007/s10021-007-9028-9

Abstract

Changes in climate and land use, caused by socio-economic changes, greenhouse gas emissions, agricultural policies and other factors, are known to affect both natural and managed ecosystems, and will likely impact on the European terrestrial carbon balance during the coming decades. This study presents a comprehensive European Union wide (EU15 plus Norway and Switzerland, EU*) assessment of potential future changes in terrestrial carbon storage considering these effects based on four illustrative IPCC-SRES storylines (A1FI, A2, B1, B2). A process-based land vegetation model (LPJ-DGVM), adapted to include a generic representation of managed ecosystems, is forced with changing fields of land-use patterns from 1901 to 2100 to assess the effect of land-use and cover changes on the terrestrial carbon balance of Europe. The uncertainty in the future carbon balance associated with the choice of a climate change scenario is assessed by forcing LPJ-DGVM with output from four different climate models (GCMs: CGCM2, CSIRO2, HadCM3, PCM2) for the same SRES storyline. Decrease in agricultural areas and afforestation leads to simulated carbon sequestration for all land-use change scenarios with an average net uptake of 17–38 Tg C/year between 1990 and 2100, corresponding to 1.9–2.9% of the EU*s CO2 emissions over the same period. Soil carbon losses resulting from climate warming reduce or even offset carbon sequestration resulting from growth enhancement induced by climate change and increasing atmospheric CO2 concentrations in the second half of the twenty-first century. Differences in future climate change projections among GCMs are the main cause for uncertainty in the cumulative European terrestrial carbon uptake of 4.4–10.1 Pg C between 1990 and 2100.

Keywords

terrestrial carbon balanceclimate changeland-use changeSRES-scenariosLPJ-DGVM

Introduction

The terrestrial biosphere in Europe currently acts as a small carbon (C) sink, sequestering annually up to 12 ± 19% of the European 1995 fossil fuel emissions, and thereby reduces the impact of these emissions on the global climate system (Janssens and others 2003). This ‘climate protection’ service is of societal, but specifically of political interest, because it may help the European countries to meet their emission reduction targets under the Kyoto protocol, and climate mitigation policies beyond Kyoto. The future of the European terrestrial carbon balance, however, is largely uncertain. Past land-use and land-management changes, such as increasing forest area and decreasing use of forests for wood production, are believed to contribute substantially to the present-day net C uptake in European terrestrial ecosystems, but it is unclear, whether, and to what extent this uptake can continue in the future (Nabuurs and others 2003). Environmental changes may have profound impacts on the amount of C stored in the terrestrial biosphere (Prentice and others 2001). Deposition of reactive nitrogen and fertilization from increasing atmospheric carbon dioxide concentrations [CO2], may stimulate plant production (Amthor 1995; Vitousek and others 1997), leading to enhanced net C uptake in terrestrial ecosystems (Karjalainen and others 1999). On the other hand, climate change resulting from fossil fuel emissions may significantly alter the capacity of terrestrial ecosystems, in particular soils, to sequester C (Cramer and others 2001; Fang and others 2005). There is a risk of net carbon losses from the terrestrial biosphere, which in turn act to amplify rather than dampen the climatic change (Cox and others 2000; Friedlingstein and others 2003).

Only few studies have so far addressed the interactions between future human transformations of the landscape, and changes in climate and atmospheric [CO2] on the global terrestrial C cycle (Leemans and others 2002; Levy and others 2004; Sitch and others 2005). These studies have shown that important feedbacks may be expected from both climate and land-use change in regional carbon budgets. Substantial uncertainty in future global land–atmosphere fluxes arises from uncertainty in climate change projections (Cramer and others 2004; Schaphoff and others 2006). However, no study has so far systematically evaluated the combined impacts of land-use change and climate change projections, and the associated uncertainties, on regional-scale land–atmosphere fluxes using a consistent set of regionally-specific future land-use and climate change scenarios.

The present study aims to assess the magnitude of the terrestrial carbon fluxes that can be expected from these changes in the twenty-first century for a domain including the member states of EU15, as well as Norway and Switzerland (hereafter EU*). Trajectories of future land–atmosphere fluxes are analyzed using a consistent set of highly resolved regional specific land-use and climate change scenarios. The study also aims to provide a comprehensive analysis of the uncertainty that arises in such projections from uncertainty about climatic changes under a particular emission scenario. The assessment is based on an advanced version of the widely used Lund-Potsdam-Jena dynamic global vegetation model (LPJ-DGVM; Smith and others 2001; Sitch and others 2003) that has been adapted to represent actual land-cover as well as land-use changes. Plausible future development of greenhouse gas emissions and regional land-use changes are based on four of the six illustrative scenarios from the IPCC Special Report on Emission Scenarios (SRES-storylines: A1FI, A2, B1, B2; Nakicenovic and others 2000; see Box 1). The effect of these scenarios on the terrestrial carbon balance is evaluated under a range of regional climate change projections derived from downscaling four different coupled ocean–atmosphere general circulation models (GCMs: CGCM2, CSIRO2, HadCM3, PCM2; Mitchell and others 2004, see Table 1) to gain understanding of the individual and combined effects of climate, atmospheric [CO2] and land-use changes, as well as the uncertainties associated with climate change projections.
Box 1.

The Four Storylines of the Special Report on Emission Scenarios (SRES) Used in This Study

The SRES-scenario narratives are differentiated along two axes: one describing ‘material consumption’ (denoted with A) and ‘sustainability and equity’ (B), and the second differentiating either ‘globalization’ (1) and ‘regionalization’ (2). Along these axes, the four of the six illustrative storylines used in this study are A1FI (world markets: fossil fuel intensive), A2 (provincial enterprise), B1 (global sustainability) and B2 (local stewardship). A1FI envisages rapid economic growth, and high technological advance, with strong reliance on fossil fuel based energy. Materialist-consumerist values are predominant, with low population growth. This narrative describes a strong globalization, and reduced differences in affluence between different regions. A2 describes a heterogeneous, market-led world, with emphasis of regional social and economic development, and a continuous increase in population. The B1 narrative has a similar low population growth as A1, but represents a more convergent world emphasizing global solutions towards a more environmentally sustainable pathway, including the introduction of clean technologies. B2 represents a world with local emphasis, intermediate economic development, and a population that stabilizes at the end of the twenty-first century. As such, no probabilities can be attributed to scenarios.

 

See Nakicenovic and others (2000) for a detailed description.

Materials and Methods

Overview

SRES-storylines give plausible paths of future development for socio-economic driving forces, fossil-fuel emissions, atmospheric greenhouse gas concentrations, as well as estimates of global land-use changes (Nakicenovic and others 2000, see Box 1). The global driving forces for land-use changes, as described in the SRES-storylines, were interpreted qualitatively in the European context to derive European level driving forces for the sectors ‘urban area’, ‘arable land’, and ‘forests’ (Kankaapää and Carter 2004; Ewert and others 2005; Rounsevell and others 2005; Reginster and Rounsevell 2006). The qualitative, European specific interpretations of SRES-storylines were then translated into input parameters for sector specific land-use modelling tools. The land-use models were run with the same climate change scenarios as used in this study, to derive future extent and geographical distribution of the land-use sectors for 2020, 2050 and 2080 (Rounsevell and others 2006). A detailed description of the method to derive the land-use change scenarios is given in the Data section.

The emission scenarios based on the four selected storylines cover 68% of the range in projected CO2 emissions (1990–2100) from all 40 SRES scenarios (Nakicenovic and others 2000). Climate models differ in their climate sensitivity, that is, their change in equilibrium surface air temperature following a unit change in radiative forcing, and in particular with respect to the spatial pattern of changes in temperature and precipitation. Output from four different climate models, for the same SRES-storylines, form the basis of climate change scenarios to represent this uncertainty (see Mitchell and others 2004 and Data section). The set of 16 future climatologies (four GCMs times four emission scenarios) covers 93% of the range in global warming in 2100 reported in Houghton and others (2001). For this study, a subset of seven scenarios was selected to illustrate the range of land–atmosphere flux resulting from: (1) uncertainties among the four storylines of future socio-economic development as they affect land-use changes, [CO2] trajectories, and differing magnitudes (though similar patterns) of climate changes based on one GCM (HadCM3), (2) uncertainties in the rate and spatial pattern of projected climate change, using all four GCMs assuming a single storyline (A2) and fixed [CO2] trajectory. The aim in selecting this subset was to capture a representative proportion of the two sources of uncertainty in projections without having to run all scenarios. See Mitchell and others (2004) for a detailed discussion.

Figure 1 gives an overview of the experimental setup of this study. For each scenario, three sets of simulations were performed (see details in Data section):
  • a control simulation accounting only for detrended interannual climate variability (S1),

  • a simulation accounting for historic and predicted changes in climate and atmospheric [CO2] assuming fixed (historical) interannual climate variability (S2),

  • a simulation accounting for all of the above, but also land-use changes (S3).

By doing so, the marginal effect of climate and atmospheric [CO2] change can be inferred by subtracting the effect of interannual climate variability (S1) from S2, and the marginal land–atmosphere flux from land-use change by subtracting the fluxes resulting from interannual climate variability, climate and atmospheric [CO2] change from the simulations considering all forcing (S3 minus S2).
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-007-9028-9/MediaObjects/10021_2007_9028_Fig1_HTML.gif
Figure 1.

Overview of the experimental setup to simulate the combined effects of climate variability, climate change, increasing atmospheric [CO2], and land-use changes on the European terrestrial carbon balance between 1901 and 2100. Data-sets of CO2 concentration, climate and land-use for the historic period (1901 and 2000), and future scenarios (2001–2100) are used to force LPJ on a 10′ × 10′ grid. Scenarios are based on the SRES-storylines A1FI, A2, B1, B2, and the climate models HadCM3, CSIRO2, CGCM2, PCM2. Simulated land–atmosphere C flux (NBE) results from the difference between terrestrial net primary production (NPP), and C releases from heterotrophic respiration (Rh), biomass burning, losses from land-use conversions, and human appropriation of biomass. The latter flux is separated into two product pools with different residence times: a 1-year decay pool (agricultural products and short-lived forest products), and a 25-year decay pool (paper, pulp, wood and other long-lasting products).

Data

Soil and Land-use Data

For each grid cell, the proportions of land occupied by cropland, grassland, forests, urban area and ‘other land uses’ were derived from the PELCOM database (Mücher and others 2000) to construct the baseline land-use data set, assumed to correspond to 2000. Fractions of individual crop functional types (CFTs) within the agricultural area were prescribed from the IMAGE2.2 data-set (RIVM 2001). Partitioning of the forest area into different plant functional types (PFTs) was based on a combination of the European tree species map (Köble and Seufert 2001), an area corrected estimate of broadleaved and coniferous forest cover (Schuck and others 2002), as well as bioclimatic limits (Sitch and others 2003), as in Zaehle and others (2006). Soil properties were derived from soil texture (IGPB-DIS:2000) using transfer functions for hydraulic (Saxton and others 1986) and thermal properties (Melillo and others 1995). Maps of historic land uses (1901–2000) were a spatial refinement of historic land-use data sets from Ramankutty and Foley (1999) and Klein-Goldewijk (2001) mainly in terms of more spatial detail in land-use change trends after 1960 (Zaehle and others 2006 Appendix D).

A three step approach was used to construct the scenarios (10′ × 10′ grid) of future land-use patterns in the EU* domain for the years 2020, 2050, and 2080 (see Kankaapää and Carter 2004; Ewert and others 2005; Rounsevell and others 2005; Reginster and Rounsevell 2006, for more information). First, global constraints to the future development of European land-use (in terms of the demand for agricultural goods and timber) were set using results of the integrated assessment model IMAGE 2.2 (Alcamo and others 1998; RIVM 2001) driven with the SRES-storylines. Secondly, the global driving forces were translated into European-level driving forces for each land-use sector (urban, arable, forest) in a qualitative way based on a knowledge of land-use change processes interpreted from past and present national trends and land-use policies. The subsequent step differs between the different land-use sectors: for arable land and grasslands, sector specific, spatially explicit modelling tools were applied to estimate future supply for the land-use classes agriculture, bio-fuels, and grasslands, taking account of technological advances and regional environmental changes under the same climate change scenarios as applied in this study (Ewert and others 2005; Rounsevell and others 2005). Changes in urban land-use were based on the application of a simple statistical model of urban area development, as well as land-use planning policies (Reginster and Rounsevell 2006). Demand for forest area, either for timber production, conservation or recreation, was estimated based on country-level forest planning information, taking account of demographic, economic, institutional, social and environmental factors (Kankaapää and Carter 2004). The resulting aggregated level land-use changes were then allocated geographically using a land-use allocation model, in which competition between the different land-use sectors was considered assuming near-optimality in the choice of land-use for any one location, given the general land-use trends and policies, and following the hierarchy (Rounsevell and others 2006):

Protected area > urban > agriculture > grassland > bioenergy crops > forests > not actively managed.

The two land-use change data-sets were linked by the baseline land-use data to ensure consistency in the trajectories of land-use change between 1901 and 2100. Gridded time-series for each land-use change scenario, suitable for the use in terrestrial biosphere models, were constructed by linear interpolation between the respective time slices (2000–2020, 2020–2050 and 2050–2080), and extrapolation to 2100 using the trend between 2050 and 2080. The fractions of individual CFTs within ‘croplands’ and PFTs in ‘managed forests’ were maintained constant throughout the simulation. Irrigation was assumed to have increased linearly between 1901 and 2000, and to remain constant thereafter, owing to the lack of a reliable trend estimate for the different scenarios. In other words, the scenarios do not account for changes in land management, for example, changes in forest species selection, or crop types.

Land-use Change Scenarios

The globally-oriented scenarios A1FI and B1 show heteorogeneous patterns of land use, based on the assumption that land use is optimized for production across the entire domain. On the contrary, in the A2 and B2 world, trends are more homogeneous because of the tendency for regional subsistence. Gross changes between different land-use types differ strongly amongst the scenarios because of the heterogeneous spatial patterns of land-use change. Key features of the land-use change scenarios are described below and summarized by Figure 2.
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-007-9028-9/MediaObjects/10021_2007_9028_Fig2_HTML.gif
Figure 2.

Changing land cover fractions for arable land, grassland, and managed forests, as well as the extent of abandoned land (‘regrowth’) under the four land-use change scenarios (A1FI, A2, B1, B2, for the HadCM3-GCM) for 2000–2020 (2020), 2020–2050 (2050) and 2050–2080 (2080) (104 km2). Positive values imply a gain in land cover of the given land-use type at the expense of either ‘arable’, ‘grass’, managed ‘forest’, or abandoned area (forest ‘regrowth’), and vice versa. ‘Arable’ includes cropland area used for food-production and non-woody bioenergy crops, ‘Grass’ comprises pastures, and grasslands on agricultural ‘surplus’ land, ‘Forest’ include managed forests, and those used for woody-bioenergy production, whereas ‘regrowth’ denotes the areas in natural succession following abandonment. Gray bars denote gross conversions, whereas colored bars denote the resulting net conversions.

A reduction in the area used for traditional agriculture, that is, mainly for food production, is common to all four land-use scenarios. The decrease results mainly from technological advances and beneficial climate change and CO2 fertilization effects on agricultural productivity. It is thus more pronounced under the technologically oriented A scenarios. Bioenergy production increases in importance under all scenarios, but particularly so in the environmentally friendly B scenarios. Some of the arable surplus land is used for bioenergy production, and remains under agricultural management. Increases in forest area are more widespread under the environmentally oriented B scenarios. In all but the B2 scenario, increases in forest and bioenergy production area do not completely occupy the surplus area from agricultural intensification. The scenario storylines do not give any suggestion as to the fate of these surplus areas. In this study, it is assumed that woody encroachment would occur on these areas, simulated by the natural vegetation dynamics of LPJ following disturbance. Urban areas are projected to expand under all scenarios, mostly on former agricultural land. This trend is most pronounced in A2 and least in B2, but the changes are small compared to the changes in other land uses, as depicted in Figure 2.

Climate and Atmospheric [CO2] Data

Atmospheric [CO2] was based on Keeling and Whorf (2003) for the observed and on IMAGE 2.2 (RIVM 2001) for the scenario period. The scenario [CO2] data include an estimate of the net effect of global land-use change. Monthly fields of temperature, diurnal temperature range, precipitation, number of rain-days and cloudiness for each 10′ × 10′ grid cell were provided by the Climatic Research Unit (CRU), University of East Anglia (Mitchell and others 2004). Data for the period 1901–2000 were based on spatially interpolated meteorological station observations. Climate change scenarios for the period 2001–2100 were those constructed by Mitchell and others (2004). Scenario data for number of rain days per month are not available from this dataset, and were held constant at average 1971–2000 levels during the scenario period.

The four GCMs used in this study differ in their spatial resolution, and in the degree with which present-day climate variability is reproduced. Interannual climate variability is an important driver of variability in the terrestrial carbon cycle (Kindermann and others 1996; Prentice and others 2000). For a consistent analysis of present and future carbon fluxes based on realistic climate variability, only the monthly anomaly fields for each climate variable were taken from the GCMs. These monthly anomalies of each climate variable and GCM for the period 2070–2099 were superimposed on the 10′ × 10′ CRU climatology for 1961–1990 to arrive at a high-resolution scenario data-set suitable for regional scale analyses. A smooth trend for each monthly climate variable, GCM and scenario in the twenty-first century was derived using MAGICC (Hulme and others 2000). The observed, linearly detrended interannual variability of each monthly climate variable for 1951–2000 was superimposed onto this smooth trend. In other words, the interannual and decadal climate variability of the 1951–2000 period was repeated twice in the twenty-first century, however, with a climate model derived long-term trend of climate change for each monthly climate variable. Thereby, all scenarios share the same, realistic climate variability, but account for changes in the seasonality as predicted by the individual GCMs. To control for the effect of choosing this particular realization of interannual and decadal variability, and to factor out the influence from short-term variability and long-term trends, a run only forced with the detrended climate variability was performed (S1). The term ‘interannual climate variability’ in this paper refers to the preserved historic interannual and decadal climate variability with the longer time-scale trends removed.

Climate Change Scenarios

Key characteristics of the climate change scenarios used are summarized in Table 1 and an example of the spatial patterns of climate change obtained from the different climate models for the A2 scenario is given in Figure 3. Spatial patterns of climate change are relatively similar between different storylines for each GCM. Generally, the HadCM3 climate model shows the most pronounced rates of warming amongst these GCMs, whereas PCM2 depicts the most modest warming rate. All four climate models project their strongest warming over the high-latitudes of Europe, with a larger rate of change in the winter months. Different emission scenarios for one model (HadCM3) lead to an almost twofold difference in the change in annual temperature, however, the spatial pattern of change is fairly similar. Overall annual precipitation over Europe changes only slightly under all four climate models, however, notable regional trends can be seen in Figure 3. All four models show a decline in precipitation over the Mediterranean, which is most pronounced in summer. Amongst the suite of climate models, HadCM3 shows the strongest drying trend both with respect to the seasonal decrease as well as the increase in the length of the dry period, whereas changes with the other three models are more moderate.
Table 1.

Key Characteristics of the Climate Change Scenarios used in This Study

Scenario

Europe

Finland temperature

Iberian peninsula precipitation

SRES

GCM

[CO2]

ppmv

Temp.

°C

Precip.

% (mm)

DJF

(°C)

JJA

(°C)

DJF

% (mm)

MAM

% (mm)

JJA

% (mm)

SON

% (mm)

A2

HadCM3a

870

4.7

−1 (−4)

8.0

4.7

−5 (−12)

−31 (−57)

−40 (−36)

−19 (−37)

A2

CSIRO2b

870

4.2

6 (45)

5.8

3.7

3 (7)

−11 (−19)

−27 (−25)

1 (2)

A2

CGCM2c

870

3.7

3 (19)

6.1

3.0

−6 (−14)

−16 (−30)

−29 (−21)

−11 (−22)

A2

PCM2d

870

2.8

3 (23)

7.7

2.4

3 (6)

−10 (−18)

−23 (−20)

−13 (−25)

A1FI

HadCM3a

958

5.8

−1 (−7)

9.5

5.6

−3 (−7)

−35 (−63)

−41 (−37)

−25 (−48)

A2

HadCM3a

870

4.7

−1 (−4)

8.0

4.7

−5 (−12)

−31 (−57)

−40 (−36)

−19 (−37)

B1

HadCM3a

607

3.3

−0 (−1)

6.4

3.3

5 (10)

−19 (−34)

−32 (−29)

−9 (−18)

B2

HadCM3a

516

3.0

−1 (10)

6.2

3.1

−9 (−20)

−24 (−43)

−28 (−25)

−20 (−40)

Carbon dioxide concentrations are for the year 2100, annual mean temperature and annual precipitation values are anomalies referring to the difference between 1971–2000 and 2071–2100. Seasonal anomalies are given for monthly temperature in Finland (broadly defined as inside EU* north of 60°N and east of 20°E), and seasonal precipitation sum for the Iberian Peninsula. The A2 HadCM3 scenario is repeated for ease of comparison.

aMitchell and others (1998).

bGordon and O’Farrell (1997).

cFlato and Boer (2001).

dWashington and others (2000).

DJF = December, January, February; MAM = March, April, May; JJA = June, July, August; SON = September, October, November.

https://static-content.springer.com/image/art%3A10.1007%2Fs10021-007-9028-9/MediaObjects/10021_2007_9028_Fig3_HTML.gif
Figure 3.

Anomaly (2071–2100 vs. 1971–2000) in A annual mean temperature (°C) and B annual precipitation (% change) for the HadCM3, CSRIO2, CGCM2 and PCM2 GCMs, and the A2-SRES emission scenario.

Model

LPJ-DGVM

The Lund-Potsdam-Jena dynamic global vegetation model (LPJ, Smith and others 2001; Sitch and others 2003) is a model derived from the BIOME family (Prentice and others 1992; Haxeltine and Prentice 1996b). The version used in this study has been modified in terms of a representation of human induced fires (Venevsky and others 2002), a more detailed treatment of hydrological processes (Gerten and others 2004), a representation of croplands (Bondeau and others 2007, see below), and a module to account for land cover changes (this study). A generic representation of forest management influences the average age and size of the forest population, without explicitly modelling forest age structure and age-dependent effects on tree growth (this study).

Gross primary production (GPP) is calculated using a modified form of the Farquhar scheme (Farquhar and others 1980; Collatz and others 1992) with canopy-level optimized nitrogen allocation (Haxeltine and Prentice 1996a; modified by F.-W. Badeck, unpublished data) and an empirical convective boundary layer (Monteith 1995) to couple the C and H2O cycles. Soil hydrology is simulated using two soil layers (Haxeltine and Prentice 1996b). Net primary production (NPP), that is, GPP reduced by C losses to autotrophic respiration, is allocated to plant tissues daily for crop functional types (CFTs, Bondeau and others 2007) and annually for woody plant functional types (PFTs, Sitch and others 2003) satisfying a set of allometric and functional relationships. LPJ distinguishes ten PFTs with different photosynthetic (C3, C4), phenological (deciduous, evergreen), and physiognomic (tree, grass) attributes. Of these, the eight PFTs that exist in Europe are the temperate and boreal needleleaved evergreen, temperate and boreal broadleaved summergreen, boreal needleleaved summergreen and temperate broadleaved evergreen, as well as C3 and C4 herbaceous PFTs. Turnover of plant tissues, plant mortality and/or management redistribute C from living biomass to above- and belowground litter pools, which in turn provide input to a fast and a slow decomposing soil C pool. Decomposition rates depend on a modified Arrhenius formulation (Lloyd and Taylor 1994) that implies a decline in apparent Q10 values with temperature and an empirical soil moisture relationship (Foley 1995). Forest fires are simulated using the regional fire module of LPJ, taking account of aboveground litter stocks and moisture and PFT-specific fire resistances (Venevsky and others 2002).

LPJ has been shown to simulate C and H2O fluxes in good agreement with observations of land–atmosphere fluxes at different scales, including field-scale eddy covariance measurements (Sitch and others 2003; Zaehle and others 2005), the seasonal cycle of atmospheric [CO2] at different latitudes (Sitch and others 2003; Zaehle and others 2005), the observed trend in the seasonal amplitude of global atmospheric [CO2] since the 1960s (McGuire and others 2001), and the interannual variability in its growth rate (Prentice and others 2000; Peylin and others 2005).

Changes to the Original Model Formulation

Each 10′ × 10′ grid cell is subdivided into fractions of different land uses, ‘croplands’, ‘grasslands/pastures’, ‘managed forest’, ‘other land uses’ and ‘urban/barren’ based on the land-use data described in the Soil and Land-use Data section. ‘Croplands’ and ‘managed forests’ are further subdivided into homogeneous patches for different CFTs or PFTs, respectively. Urban area is treated as barren ground, for which only soil processes are calculated. For each of these subdivisions, fluxes and balances of water and carbon are calculated independently on the basis of the common LPJ physiology. Natural vegetation dynamics are calculated for the ‘other land use’ grid cell fraction, as in LPJ-DGVM (Sitch and others 2003).

Cropland and Grassland Management

The cropland/grassland module, described in more detail by Bondeau and others (2007), considers 11 crop and 2 grass functional types with different photosynthetic (C3, C4), phenological and morphological characteristics, representing temperate or tropical cereals, maize, rice, pulses, temperate or tropical roots and oil crops, as well as C3 and C4 pastures and meadows. Sowing dates are determined by temperature and/or water availability constraints for climate-sensitive CFTs, or prescribed otherwise. Phenological development and carbon allocation to leaves, storage organs, roots, stems and mobile reserves follows a heat unit approach, with distinct parameterizations applying for the juvenile and senescent phase. The parameterization of the phenological development depends on the local climate, which implicitly features the choice of different cultivars under different climatic conditions. Harvesting occurs as soon as maturity is reached. Storage organs, and 90% of the aboveground biomass are removed, and assumed to respire within the same year. The remaining biomass is added to the surface or belowground litter pool, respectively. Cropping systems are assumed to rotate within any one grid cell, leading to average soil C pool stocks in the agricultural land-cover corresponding to the difference of average crop plant production and removal by harvest. In this study, irrigation, where possible, and intercropping, that is, a grass CFT covering the bare ground between two cycles of a crop rotation for cereals, are enabled, whereas changes in management over time that could affect productivity have not been considered, owing to the lack of consistent data to parameterize such a change. Comparisons with the FAO statistics for the European dominant crop types show that the simulated yields agree reasonably with observations (Bondeau and others 2007).

Grazing is modelled as consumption of 50% of aboveground biomass once a certain LAI threshold is exceeded (Bondeau and others 2007). This fraction, contributing to animal metabolism, is modelled as a carbon flux to the atmosphere.

Forest Management

The effect of forest management on carbon stocks and fluxes is calculated for stands with a prescribed PFT composition. Vegetation dynamics in ‘managed forests’ allow for structural changes in response to changing environmental conditions or management, resulting from changes in growth efficiency and self-thinning related mortality and success of natural reestablishment, however, not in terms of changed PFT composition. The total annual fellings from forest management are estimated from an approximation of the average felling in a evenly structured forest landscape as a function of maximal stand biomass (Dewar 1991), as well as characteristic (logistic) growth curves (described by the parameters b and r) and rotation-times Trot based on (Nabuurs and Mohren 1995). Biomass at the time of felling is estimated as:
$$ B(T_{{\text{rot} }} ) = \frac{{B_{{\max }} }} {{1 + b \times \exp ( - r \times T_{{\text{rot} }} )}} $$
(1)
where Bmax is calculated from LPJs biomass estimates, and fixed to the level of 1900, when the model is in equilibrium. The felling rate, P, is then determined by
$$ P = B(T_{{\text{rot} }} )/T_{{\text{rot}}}$$
(2)
The biomass per square meter felled per year is the product of this felling rate, and the actual biomass density of that PFT. The fraction of the biomass felled that is actually removed from the site is calculated taking account of the typical ratio of whole-tree biomass to roundwood volume and the typical losses of timber in the forest during harvesting activity (together 70% of the felled wood mass, UN-ECE/FAO 2000). Biomass removed is partitioned into a 1- and a 25-year wood-products pool as in Nabuurs and others (2003) (Table 2), the remainder is added to the above- and belowground litter pools, respectively. Although such an approach accounts for the timber removed from forests (and its storage in wood-products), changes in forest age-structure due to changes in harvest regimes or forest area are not modeled, nor are their effects on average forest growth rate. Wood demand changes have not been considered in this study.
Table 2.

The Fate of Carbon upon Conversion, and Ecosystem Management for Different Terrestrial Ecosystems

Ecosystem

Fate upon conversion

Fate as product (as percentage of aboveground biomass)

Belowground biomass (left dead in soils) (%)

Aboveground biomass (slash left on site) (%)

1-year pool (crops, fuel wood) (%)

25-year pool (paper, pulp, wood) (%)

Croplands

100

10 (if any)

90

NA

Grasslands

100

10

90

NA

Temperate and Boreal forest

100

30

67

33

Woodlands (other land uses)

100

30

67

33

Conversion and partitioning coefficients for forest and woodlands are based on UN-ECE/FAO (2000); McGuire and others (2001); Nabuurs and others (2003). Removal of C in croplands and grasslands is based on the agricultural module as described in Bondeau and others (2007) and the Model section of this study.

Land-use Change

For each grid cell, fractions for each land-use type are updated annually. The fate of the living C following land conversion and management is summarized in Table 2. Aboveground C of trees on converted land is treated as in McGuire and others (2001), however, partitioning coefficients for the wood products pool are taken from Nabuurs and others (2003), which are more appropriate for a European-scale simulation. Slash and belowground litter are added to the respective litter pools of the converted land.

Modelling Protocol

LPJ was spun up to equilibrium in terms of pre-industrially stable C pools and vegetation dynamics using the reconstructed land-use patterns of 1901. Thirty years of climate (1901–1930) and atmospheric [CO2] (1901) seed this spin-up. In S1, LPJ was forced only with the detrended interannual climate variability, and constant atmospheric [CO2] and land-use patterns at 1901 levels. In S2, historic (1901–2000) and projected (2001–2100) climatic changes were superimposed on the interannual variability of S1, and [CO2] levels increase as observed or projected in the scenarios, respectively. In S3, in addition to the above, land-use patterns are varied based on the reconstruction for the period 1901–2000, and scenario projections for 2001–2100.

The land–atmosphere carbon flux, the so called net biome exchange (NBE), is calculated separately for each fraction of a grid cell as the difference between net primary production (NPP), and carbon losses to the atmosphere from heterotrophic respiration (Rh), biomass burning, or by human appropriation. The latter term is the sum of decaying wood products (as described above), and the harvesting flux from croplands and pastures, added to the annual fluxes of the grid cell fraction of production.
$$ \text{NBE} = R_{\text{h} } + \text{BiomassBurning} + \text{HumanAppropriation} - NPP $$
(3)
where a negative sign denotes a C flux into the terrestrial biosphere, seen as depletion of [CO2] in the atmosphere, and vice versa. The land–atmosphere flux of the entire grid cell is then calculated as the area-weighed mean of the fractional land–atmosphere exchanges.

Results

Present-day European Carbon Balance

Europe’s terrestrial NPP averages 1,670 Tg C/year in the 1990s (Figure 4). About a third of the NPP is lost to the atmosphere by human appropriation, resulting either from food consumption, or decaying wood products. The remainder of this uptake is, to a large extent, offset by carbon losses through Rh. Biomass burning, although locally very important, is estimated to play only a minor role in the European scale carbon budget. The modelled land–atmosphere flux of the terrestrial biosphere of EU* averages at −29 Tg C/year for the 1990s (Figure 5A, E), with a large interannual variability between −106 Tg C/year (net uptake) and +45 Tg C/year (net loss). This net flux is the sum of carbon losses from urbanization (3.3 ± 0.8 Tg C/year), agriculture (19.3 ± 9.1 Tg C/year) and grasslands (14.5 ± 13.2 Tg C/year) and carbon uptake in forests and wood products of 59.1 ± 31.4 Tg C/year.
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-007-9028-9/MediaObjects/10021_2007_9028_Fig4_HTML.gif
Figure 4.

Constituent fluxes of the net land–atmosphere flux (Tg C/year) considering climate, atmospheric [CO2] and land-use change for 1901–2100. Increase in NPP is denoted by a more negative contribution to the land–atmosphere flux. Left Results for the four different SRES-storylines (A1FI, A2, B1, B2) with climate change projections derived from the Hadley climate model. Right Results for the A2-SRES storyline, with climate change projections based on four different GCMs (HadCM3, CSIRO2, CGCM2, PCM2).

https://static-content.springer.com/image/art%3A10.1007%2Fs10021-007-9028-9/MediaObjects/10021_2007_9028_Fig5_HTML.gif
Figure 5.

Land–atmosphere flux (Tg C/year) for EU* in 1960–2100. Left (AD) Results for the four different SRES-storylines (A1FI, A2, B1, B2) with climate change projections derived from the Hadley climate model. Right (EH) Results for the A2-SRES storyline, with climate change projections based on four different GCMs (HadCM3, CSIRO2, CGCM2, PCM2). A, E Land–atmosphere flux resulting from climate, atmospheric [CO2] and land-use change (S3); B, F land–atmosphere flux attributable to land-use change (S3-S2); C, G land–atmosphere flux attributable to climate and atmospheric [CO2] change (S2-S1); D, H land–atmosphere flux attributable to detrended interannual climate variability (S1). Natural variability denotes here the 10-year running average of the land–atmosphere flux resulting from detrended interannual climate variability. Negative sign denotes a carbon flux towards the terrestrial biosphere.

Reconstructed historical land-use data suggest that agricultural area increased between 1900 and the 1950s, replacing ‘other land-uses’, which are assumed to be occupied by natural vegetation. Stagnation and subsequent reversal of this trend around the 1950s, as well as concomitant increases in forest area are the main causes for the decreasing net loss of C from land-use change. In the 1980s, the net loss of C associated with land-use change averages at +9 Tg C/year (Figure 5B, F), reaching a zero balance in about 1990.

Effects of Future Land-use Change

Land-use changes in the scenario period are projected to result in net C uptake in all four storylines when averaged over the entire domain (Figure 5B), as an effect of agricultural land abandonment (see Data section). Even though trajectories of land–atmosphere flux from land-use change for the scenarios are similar for certain periods, there is considerable spatial heterogeneity in the land–atmosphere flux between these scenarios (Figure 6).
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-007-9028-9/MediaObjects/10021_2007_9028_Fig6_HTML.gif
Figure 6.

Land–atmosphere fluxes associated with the effects of land-use change, averaged over 2021–2050 for the four SRES storylines (A1FI, A2, B1, B2) and implemented with HadCM3-GCM.

In the A1FI world, agricultural activity is centered in regions of prime productivity, leading to surplus land from agricultural land abandonment in less favored areas. These areas are abandoned and allowed to return to the naturally existing land-cover type. This results in increasing vegetation carbon storage, as in most regions of Europe the potential natural vegetation is a forest ecosystem and a C increase in soils, as C from NPP is no longer removed from the ecosystem. Planned afforestation and forest regrowth on previously agriculturally used soils are the main cause for the land–atmosphere flux of −44 (2020) and −53 (2080) Tg C/year under the A1FI scenario (Figure 5B). Between 2020 and 2050, C uptake is substantially reduced, as re-expansion of agriculture in intensively used agricultural areas replaces some grasslands and previously abandoned areas (Figure 2). This is seen in Figure 6 as areas of net C loss, for example, in Eastern England and Southern Finland.

Homogeneous decreases in agricultural area across Europe in the A2 scenario make land available either for surplus or afforestation, both of which contribute to the fairly evenly distributed land–atmosphere flux of about −40 Tg C/year between 2000 and 2050 (Figure 6). Net forest expansion stops in this scenario at around 2050. However, carbon sequestration due to land-use change continues at a similar rate in the second half of the twenty-first century (Figure 5B) resulting from the previous increases in forest area and decreasing use of forest for wood production. Urban expansion, strongest in the A2 scenario, results in strong net C losses locally, however, the magnitude is too small to affect continental scale land–atmosphere fluxes.

In B1, agricultural productivity is centered in prime locations as in A1FI, however, the decline in arable area is less, thus less surplus area becomes available. Afforestation is assumed to be more widespread across Europe and most pronounced in Central and Southern Europe, however, before 2020 at half the rate as in A1FI. These changes result in a land–atmosphere flux of −22 Tg C/year before 2020 (Figure 5B). The effect of the strong afforestation on the total net land–atmosphere flux after 2020 is reduced to a net −10 Tg C/year until 2050 because the area used for cropping, mostly for bioenergy production, increases again at the expense of former grasslands and previously abandoned area. After 2050, even larger increases in forest and grassland area, lead to a land–atmosphere flux of about −35 T gC/year in 2080 (Figure 5B).

The B2 world is considerably different from the other three scenarios, as no surplus land becomes available from declines in agricultural area. In fact, the decline in area for food production is more than compensated for by increasing bioenergy production. A net conversion of grasslands into arable land is required to allocate the land area for bioenergy production between 2000 and 2020 (Figure 2). These changes and rapid afforestation on grassland soils with large stocks of soil organic matter, for example, in Ireland (Figure 6), reduces the net effect of the increasing carbon storage in vegetation until 2050 with land–atmosphere fluxes of around −7 Tg C/year. Although the increase in forest area in the B2 scenario is the largest among the four scenarios, more wide-spread afforestation begins to increase biospheric uptake (−35 Tg C/year in 2080, Figure 5B) only after 2050, when afforestation occurs mainly on previously agriculturally used area.

Effects of Future Climate and Atmospheric [CO2] Change

Climate change and increasing atmospheric [CO2] lead to a rise in NPP under all scenarios (Figure 4). However, in most scenarios this increase levels off around the year 2070. Initially, increases in NPP are faster than the rate of increase C release from Rh, thereby sustaining a net C uptake of the terrestrial biosphere between the 1950s and at least 2040 (Figure 5C, G). All scenarios show a decline in the C flux towards the biosphere attributed to climate and [CO2] change in the second half of the twenty-first century. The terrestrial biosphere becomes a C source to the atmosphere for the final few years of the scenario period under all HadCM3 scenarios. With moderate predicted climate warming, as for instance under the A2 PCM2 scenario, increasing respiration losses only balance increasing NPP, halving the net land–atmosphere carbon flux due to climate change at around 2100. The simulated trends in land–atmosphere flux resulting from climate change and increasing atmospheric [CO2] are of a similar magnitude as the variability in land–atmosphere fluxes from ‘natural’ climate variability alone (Figure 5D, H). There is considerable spread in the magnitude of the land–atmosphere C flux, as well as the timing of the reduction in net terrestrial C uptake, between the different scenarios towards the end of the scenario period (Figure 5C, G).

These differences are less pronounced between the runs using four different SRES storylines and the HadCM3 climate model (HadCM3–4 SRES storylines). All four ‘HadCM3’ scenarios show a stabilization of NPP in the last 20 years of the scenario (Figure 4), despite substantial differences between the magnitude of NPP between the high (A1FI and A2) and low (B1 and B2) atmospheric [CO2] scenarios. Trends in land–atmosphere flux are fairly similar for all four scenarios. Biospheric C uptake until 2040 turns into a net C release towards the end of the scenario period (Figure 5G). The A1FI scenario is noteworthy because it exhibits the most rapid warming trend, leading already in the 2040s to a net C loss to the atmosphere from climate change and rising atmospheric [CO2] alone. In general, initially larger gains from increased NPP in high atmospheric [CO2] scenarios (A) are lost due to the larger rates of climate warming in these two scenarios.

Figure 7 illustrates the spatial coherence in land–atmosphere flux anomalies in 2021–2050 and 2071–2100 resulting from differential climate change projections based on the four GCMs, forced with the same emission scenario (A2), relative to 1971–2000. In 2021–2050, most of Europe’s terrestrial biosphere sequesters C at a similar rate to the present-day, with only slight differences among the projections based on the four GCMs (Figure 7B). Between 2071 and 2100, the ensemble average land–atmosphere flux for large parts of Central Europe is close to zero, resulting in a small positive anomaly because the biosphere acts as a small net sink in 1971–2000. Scenarios agree relatively well over large parts of Central Europe, however, disagree in Eastern Fennoscandia and the Mediterranean region; most notably for the mountain ranges (Figure 7D). These are the areas with the most prominent differences in climate change projections between the four GCMs (see Table 1).
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-007-9028-9/MediaObjects/10021_2007_9028_Fig7_HTML.gif
Figure 7.

Land–atmosphere fluxes associated with climate change and CO2-fertilization under the A2-SRES scenario, as anomalies for 2021–2050 and 2071–2100, relative to the period 1971–2000. The left-hand maps (MEAN) show the average over all four S2 simulations (A2-storyline, HadCM3, CSIRO2, CGCM2, PCM2), the right-hand maps (SD) the standard deviation between these.

In Fennoscandia, all four climate models show their strongest projected warming in Europe, with anomalies in annual mean temperature between 4.4°C (A2-CGCM2) and 5.9°C (A2-HadCM3; Figure 3A). Three out of the four models show the most pronounced warming in winter (Table 1), affecting respiratory processes more than photosynthesis. Only subtle differential changes in rates of ecosystem C uptake and release can have a substantial effect on the net land–atmosphere flux, as boreal forest soil carbon densities are amongst the largest soil C stock densities in Europe. Projected growth enhancement and increasing storage of C in vegetation are offset by increasing losses from soil respiration in all four A2 scenarios, however, this effect is more pronounced in those scenarios with greater warming (that is, A2-HadCM3 and A2-CSIRO).

All four climate models generally predict a decline in summer precipitation over large parts of the Mediterranean (Figure 3B, Table 1). A concomitant increase in monthly temperatures is more pronounced in summer in HadCM3 and CGCM2, but spread evenly over the year in two models (PCM2 and CSIRO2). HadCM3 is the most extreme climate model in terms of the increase in seasonal temperatures as well as the decline in annual precipitation and the prolongation of the summer dry-season (see Table 1). The differences in land–atmosphere flux shown in Figure 7 in the Mediterranean result mainly from differences between the A2-HadCM3 run and the other three scenarios. On a regional scale, CO2 induced increases in water-use efficiency more than compensate for the effect of water limitation on photosynthesis in three out of four scenarios (A2-PCM2, A2-CSIRO2, and A2-CGCM2), but not in A2-HadCM3. In A2-HadCM3, drought stress leads to a decline in NPP in the last 25 scenario years, partly masked by substantial interannual variability. Stabilization of NPP and increasing Rh as a response to increasing temperature reduce the biospheric C uptake most strongly under HadCM3. In addition, the pronounced drying trend projected with the Hadley model leads to a nearly twofold increase in fire danger (increase by 83% between the 30-year averages in 1971–2000 and 2071–2100) relative to the 34–53% increase in fire risk in the other three scenarios. As a result, C releases from biomass burning increase by 87% (29–59%) towards the end of the scenario period.

Cumulative Land–Atmosphere Flux Between 1990 and 2100 Under Climate, Atmospheric [CO2] and Land-use Change

The differences in magnitude and spatial pattern of land-use change related fluxes across the four land-use change scenarios between 1990 and 2100, result in a 2.2 Pg C difference in cumulative land–atmosphere fluxes from land-use change (Figure 8C, Table 3). The average yearly net uptake from land-use change (1990–2100) corresponds to 1.9–2.9% of the EU*'s CO2 emissions over the same period. Interactions of the land-use change flux with climate and [CO2] change lead to only subtle differences in the projected uptake attributed to land-use change (Figure 5F).
https://static-content.springer.com/image/art%3A10.1007%2Fs10021-007-9028-9/MediaObjects/10021_2007_9028_Fig8_HTML.gif
Figure 8.

A Cumulative land–atmosphere fluxes (Tg C/year) between 1990 and 2100; split into the components, B attributable to climate and [CO2] change, and C to land-use change.

Table 3.

Cumulative Changes in Terrestrial Vegetation and Soil C Pools (1990–2100) in Pg C under the Different Scenarios Analyzed, Attributed to Climate and Atmospheric [CO2] or Land-use Change

 

A1FI

A2

B1

B2

HadCM3

HadCM3

CSIRO2

CGCM2

PCM2

HadCM3

HadCM3

Climate and [CO2] change (S2)

 Vegetation (Pg C)

2.87

3.06

3.83

3.52

3.72

2.30

2.74

 Soil (Pg C)

−1.61

−0.33

0.21

1.05

2.21

−0.26

0.12

 Total (Pg C)

1.26

2.73

4.04

4.57

5.93

2.03

2.85

Land-use change (S3-S2)

 Vegetation (Pg C)

3.39

4.07

4.36

4.32

4.42

1.93

2.51

 Soil (Pg C)

0.41

0.01

−0.19

−0.09

−0.24

0.39

−0.68

 Total (Pg C)

3.81

4.08

4.17

4.23

4.18

2.33

1.83

All forcing (S3)

 Vegetation (Pg C)

6.26

7.13

8.18

7.84

8.14

4.23

5.52

 Soil (Pg C)

−1.20

−0.32

0.03

0.96

1.97

0.13

−0.57

 Total (Pg C)

5.06

6.81

8.21

8.89

10.11

4.36

4.68

Average land–atmosphere flux (1990–2100) as percentage of the EU* CO2 emissions (1990–2100)a

 Land-use change (S3-S2)

2.31

2.81

2.87

2.92

2.88

2.40

1.87

 All forcing

3.07

4.69

5.66

6.06

6.97

4.50

4.80

aBased on projections of IMAGE 2.2, as in RIVM (2001).

A1FI = 1,493 Tg C/year, A2 = 1,389 Tg C/year, B1 = 881 Tg C/year, B2 = 887 Tg C/year.

Uncertainties in cumulative land–atmosphere flux attributable to climate change and future atmospheric CO2 concentrations are more pronounced than those in the land-use change related fluxes in this particular set of scenarios. Differences in land–atmosphere flux resulting from the choice of a particular storyline (that is, the four different storylines and one climate model) are smaller than those resulting from uncertainty in climate change under the same emission scenario (that is, four different climate models and the A2-storyline) (Figure 8B).

Uncertainty in climate change under the same storyline (land-use change and CO2 scenario) propagates to a 4.8 Pg C difference in cumulative land–atmosphere fluxes by the end of the scenario period (Table 3). Changes in soil C stock differ widely between these four scenarios (a loss of approximately 0.3 Pg C in A2-HadCM3 versus an uptake of approximately 2.2 Pg C in A2-PCM2), whereas vegetation stock increases are more similar (see Table 3). All four HadCM3 scenarios predict the strongest decline in soil C stocks, and the smallest increase in vegetation C, related to the pronounced drought in the Mediterranean. These scenarios also show the smallest land C uptake over the scenario period. The most extreme warming scenario, A1FI-HadCM3, shows the smallest cumulative net carbon uptake, resulting from substantial net losses of soil C by 2100.

The combined effect of land-use, climate change and rising atmospheric CO2 content on the terrestrial biosphere results in a cumulative net uptake of C between 1990 and 2100 for all scenarios considered in this study (Figure 8A). Climate change, however, weakens the marginal effects of land-use change on terrestrial C sequestration in all but the PCM2 scenarios because of increasing soil C losses in the second half of the twenty-first century. Uncertainty in the spatial pattern of climate change is the largest contributor to the range of projected land–atmosphere fluxes of 4.3–10.1 Pg C.

Discussion

Simulated present-day carbon stocks in soil and vegetation, as well as their temporal changes are in reasonable agreement with independent estimates based on the extrapolation of soil surveys, forest inventories and modelling efforts. Average modelled living biomass in forests is only slightly larger than reported from forest inventories (6.7 vs. 6.2 kg C/m², Goodale and others 2002). Modelled forest soil carbon density is larger than estimated from forest inventories and soil surveys (13.8 vs. 9.6 kg C/m², Goodale and others 2002; Smith and others 2006), probably because past forest management practices like litter raking have not been taken into account (Mather 1990; Zaehle and others 2006). Average soil carbon densities for crop- and grasslands of 10.0 kg C/m² compare to averages of 9.1 kg C/m² based on soil carbon modelling with RothC (Smith and others 2005), however, modelled equilibrium soil carbon stocks differ substantially between different CFTs. For the two European dominant crop types (temperate cereals and maize), comparisons of the present day crop yields simulated by the model for global runs agree well with the FAO statistics (Bondeau and others 2007). However, no specific tests have been conducted to evaluate the simulated yields obtained from the European-wide simulations that use specific input data of this study.

The estimated sectorial carbon fluxes for the 1990s, that is, carbon losses from agriculture (19 ± 9 Tg C/year) and grasslands (14 ± 13 Tg C/year) and a net carbon uptake of forests and wood products of 59 ± 31 Tg C/year compare to independent data and model based estimates of carbon losses from agriculture (46 ± 30 Tg C/year) and carbon uptake in grasslands (40 ± 26 Tg C/year) and forests plus wood products (84 ± 34 Tg C/year) (Janssens and others 2005). The notable discrepancy in sectorial land–atmosphere fluxes for grasslands probably reflects changes in pasture management practices that lead to soil carbon sequestration, whereas LPJ, in the absence of such effects, predicts a small, mostly temperature related decline in soil carbon stocks. The net loss of C from land-use change in the 1980s averages at 9 Tg C/year, which is within the range of uncertainty of the estimated uptake of about 20 ± 200 Tg C/year based on book-keeping of land-use changes for geographical Europe, excluding the former Soviet Union (Houghton 1999).

Effects of Land-use Changes

The results presented here are strongly influenced by the fate of the abandoned agricultural area and the representation of forest regrowth. A supplemental simulation not reported here, in which surplus land is assumed to be converted into managed grassland, shows substantially lower carbon uptake in soil and, even more pronounced, in vegetation. This is particularly true for the A scenarios with a large surplus area. In these alternative simulations, cumulative carbon uptake amounts to approximately 1.7 Pg C for each of the four land-use change scenarios. This indicates that there is a potential for land-use policy, affecting the fate of this surplus area, to exert notable effects on the terrestrial carbon balance. The surplus area in this study results from increasing cropland productivity (due to the net effect of climate and [CO2] changes and technological advances) that is larger than the concurrent growing demand for agricultural goods in any of the scenarios applied (Ewert and others 2005). Thus leakage, that is, land-use related carbon losses in other parts of the world resulting from production shifts to countries outside Europe, which would reduce the magnitude of the carbon sequestration realised though set-aside land, is not relevant in the context of the scenarios applied in our study.

Accompanying studies on the change in forest vegetation C stocks using different modelling approaches but the same forcing show qualitatively similar changes in forest growth, however, the estimates of cumulative carbon uptake differ by up to approximately 35% (Schröter and others 2005; J. Meyer, unpublished results; Zaehle 2005). The magnitude of the estimated carbon uptake rates, is sensitive to the way in which the vegetation dynamics are modelled, in particular the response of self-thinning to changes in forest productivity and the age-related changes in forest growth rates (Zaehle 2005; Zaehle and others 2006), and how inventory data are scaled to total vegetation biomass. It should be noted that the estimates provided here assume no change in management, which could alter trends in forest vegetation carbon build-up additionally.

Estimates of land-use change related fluxes depend on past land-use changes because of the long response-time of both vegetation and soil C. Consistent data on a continental scale on historical land uses are very sparse, so that substantial uncertainty is inherent in any backward projection of land-use patterns (House and others 2003). In addition, it is not net change in land use sectors, as documented for instance in FAO statistics (FAO 2004), but the gross conversions between land-use types that determine the land-use change carbon flux. Past land-management changes that effect soil C inputs or turnover times are poorly quantified for larger regions, however, may have an important role in estimating correctly the carbon stocks and their change (Glatzel 1999). This lack of a reliable land-use and land-management history contributes to the difference between the modelled present-day carbon fluxes and independent estimates. Changes in present-day soil carbon stocks in grasslands and forests are probably to a certain degree underestimated because the soils were assumed to be in equilibrium in 1900, whereas in reality these soils are recovering from past degradation or experiencing enhanced present-day C inputs (Janssens and others 2005).

The change in soil C stock as a consequence of land-use conversion depends on the differential sizes of the soil C pool between the land-use types at equilibrium. Generally, cropland soils have lower C stocks than grasslands, whereas forest soils have similar pool sizes to grasslands (Guo and Gifford 2002). Such differences are modelled by LPJ, and also conversions between these three land-use types show similar trends to observed studies (see Figure 1 of Guo and Gifford 2002, and references therein). The accuracy with which the magnitude of the continental scale soil C flux following conversion can be estimated depends on how well the differences between soil stocks of different land-use types are represented in the model. In particular, reliable estimates of forest soil C stocks are required, which are currently poorly quantified at a larger scale. This study focuses on changes of the land–atmosphere flux after 1990. The bias resulting from the initialization of the soil carbon stocks is therefore not strongly relevant for the general conclusions of this study. Nevertheless, reliable modelling of present-day and future carbon soil C remains a challenge to reduce the uncertainty in the simulations.

It should be noted that uncertainty in the projections of future soil C arises from potential changes in land management, which could alter the C returns to the soil, or C turnover in the soil, such as tillage. Such management effects have not been considered in the present study. A result of a previous simulation without intercropping and alternative management options for harvest residues (see Methods; Bondeau and others 2007) showed that alternative management regimes alter the soil C stock size and also the loss of soil C from soil warming, thereby affecting the modelled change in net C flux. Smith and others (2005, 2006) have demonstrated that alternative management regimes can be expected to significantly affect soil C stock changes, and potentially more than offset the temperature related soil C losses simulated in this study.

Effects of Climate Change and Rising Atmospheric CO2

CO2-fertilization is the main cause for the increase in terrestrial carbon sequestration before 2050, affecting carbon storage in plants by increasing net primary production as well as water efficiency of plants (Amthor 1995). This result is typical for the response of terrestrial biosphere models (Cramer and others 2001; McGuire and others 2001; Levy and others 2004; Schaphoff and others 2006), and consistent with measurements of ecosystem responses to CO2-elevation (Norby and others 2005). Long-term effects of increased atmospheric [CO2] on terrestrial carbon storage are still a subject of scientific debate (Prentice and others 2001), mainly because nutrient availability may in the long-term limit the effect of enhanced [CO2] on ecosystem carbon uptake. An analysis of growth trends across several European forest plots—including process-based models that link C to N cycles—suggests that N-deposition has been a major contributor to increased forest growth in the past, whereas in the future elevated atmospheric [CO2] and climate change will likely dominate the environmental effects of forest growth (Spieker and others 1996; Karjalainen and others 1999). Nevertheless, because of the complex interactions between C and N cycles (for example, Lloyd and others 1999), the interaction of N-availability with elevated [CO2] remains one of the key uncertainties in this study.

Global warming will likely offset some of the sequestration from land-use and increased productivity, as increasing heterotrophic respiration is not counterbalanced by increasing litter fall. This is a robust finding in this study under all scenarios for larger parts of boreal Europe, however, the magnitude of this effect is subject to considerable uncertainty deriving from uncertainty in boreal warming and implies a positive feedback of the terrestrial biosphere to the climate system, as observed in a range of other modelling studies (Cox and others 2000; Cramer and others 2001; Dufresne and others 2002; Friedlingstein and others 2003; Schaphoff and others 2006). Some studies have suggested that the resistant pools of soil C might not respond to increasing temperatures (Liski and others 1999; Giardina and Ryan 2000; Thornley and Cannell 2001), contrary to the assumptions made in LPJ. However, recent analyses of experimental evidence (Fang and others 2005; Knorr and others 2005; Davidson and Janssens 2006) show that soil respiration responses to temperature as modelled by LPJ are compatible with the data. Similar to Jones and others (2005), we find sensitivities of soil C stocks to increasing temperatures that are comparable to results of a more sophisticated model of soil C turnover (RothC) forced with the same set of climate change scenarios (Smith and others 2005, 2006). Fang and others (2005) showed that the effect of uncertainty in the temperature sensitivity of soil C led to some uncertainty in temperate regions, whereas the response in boreal regions, in which we simulate the strongest signal was less affected.

Pronounced droughts or extreme temperatures, when occurring at a large scale, can lead to substantial shifts in the annual terrestrial carbon balance of Europe (Ciais and others 2005), potentially offsetting the carbon sequestration of several years in one single year. The results presented here show that year-to-year changes in the net carbon balance are substantial and can lead to an annual land–atmosphere flux opposite to the longer term trend (see Figure 5D), resulting from the differential response of NPP, soil respiration and the magnitude of wild fire carbon losses to seasonal climate. The climate scenarios used here describe changes in seasonal climate, for instance, the length of dry periods (see Table 1), and are subject to the same interannual variability as observed in 1951–2000. These two factors together can, for certain regions such as the Mediterranean region, result in a higher frequency of periods with temperatures above critical thresholds, or low water-availability relative to the water demand. As such, the simulations account for the possibility that seasonal phenomena alter the annual carbon balance as described by Ciais and others (2005) with the limitation that potential changes in the interannual variability are not accounted for, and that monthly climates might not adequately reflect the effects of extreme, but short-term meteorological events, such as spells of extremely high temperatures.

The climate model related uncertainty is twice as large as the difference in land–atmosphere flux projections under alternative scenario storylines derived from a particular climate model. Additional simulations for different GCM and SRES storyline combinations not presented in this study suggest that the range in cumulative land–atmosphere flux depicted in Figure 8 encompasses the response of the terrestrial biosphere under all 16 scenarios of the scenario set. Pronounced differences resulting from uncertainty in climate change projections based on the same emission scenario have been also obtained in a global study of five different GCMs forcing LPJ-DGVM (Schaphoff and others 2006), and in a study using two GCMs to analyze the carbon storage of tropical rainforests under climate and land-use changes (Cramer and others 2004). These results demonstrate that regional detail of climate projections are an important determinant of terrestrial biosphere responses to climate change.

Conclusions

Abandonment of agricultural area and subsequent increases in forest (managed or unmanaged) lead to a net carbon uptake in Europe’s terrestrial ecosystems under all four plausible future path-ways of land-use change between 1990 and 2100. The differences in the magnitude of the uptake between these scenarios is primarily influenced by the extent of agricultural abandonment, and thus more pronounced under the technologically oriented A scenarios. The cumulative carbon sequestration resulting from land-use change is equivalent to about 1.9–2.9% of the EU* fossil fuel related CO2 emissions between 1990 and 2100, even when taking account of the considerable forest regrowth in the A scenarios.

The C fluxes resulting from climate and [CO2] change are—on average—of a similar magnitude to the land-use change related fluxes. On the European scale, the cumulative carbon sequestration between 1990 and 2100 associated with climate and atmospheric [CO2] changes is equivalent to about 0.7–3.8% of the EU* fossil fuel related CO2 emissions over the same time period. Climate and atmospheric [CO2] changes lead to enhanced biospheric uptake rates before 2040, and a weakening of the terrestrial uptake rate thereafter in all scenarios, turning the European terrestrial biosphere into a net C source for the climate change scenarios that exhibit the strongest warming.

Uncertainty in the future European terrestrial carbon balance associated with uncertainty in the rate and spatial pattern of climate change among GCMs for a particular emission scenario is larger than the differences between alternative scenarios of consistent land-use and climate changes interpreted by one particular climate model. This implies that for a sound assessment of climate change impacts, not only different SRES-scenarios, but also a suite of climate models have to be considered.

The 1990–2100 average EU* land–atmosphere flux under the combined land-use, climate and [CO2] change projections corresponds to 3.1–6.9% of the total EU* fossil–fuel emissions during that period. In agreement with earlier global studies (House and others 2002; Sitch and others 2005), the terrestrial C uptake—and thus its impact on mitigating climate change—is likely to be small.

ACKNOWLEDGEMENTS

We are grateful to Tim Mitchell (CRU) for providing the climate data and thank all participants of the ATEAM project for three-and-a-half years of constructive discussion. This work contributes to the EU-funded project ATEAM (Advanced Terrestrial Ecosystem Assessment and Modelling; http://www.pik-potsdam.de/ateam; EVK2-CT−2000–00075). SZ was supported by the HSB-programme of the Federal State of Brandenburg, Germany (AZ: 24–04/323;200) and the EU-funded CarboEurope-IP (GOCE-CT-2003–505572).

Copyright information

© Springer Science+Business Media, LLC 2007