Predicting pan-tropical climate change induced forest stock gains and losses—implications for REDD

Marlies Gumpenberger¹, Katrin Vohland^1,3, Ursula Heyder¹, Benjamin Poulter^1,4, Kirsten Macey², Anja Rammig¹, Alexander Popp¹ and Wolfgang Cramer¹

¹ Potsdam Institute for Climate Impact Research (PIK), Telegraphenberg A 62, D-14473 Potsdam, Germany
² Climate Analytics, Telegraphenberg, 14412 Potsdam, Germany

³ Present address: Museum für Naturkunde, Leibniz Institute for Research on Evolution and Biodiversity at the Humboldt University Berlin, Invalidenstraße 43, 10115 Berlin, Germany

⁴ Present address: Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), Zürcherstrasse 111, CH-8903 Birmensdorf, Switzerland

E-mail: Marlies.Gumpenberger@pik-potsdam.de

Received 14 November 2009
Accepted 3 February 2010
Published 16 February 2010

Contents

• 1. Introduction
  • 1.1. Deforestation and climate change
  • 1.2. Policy incentives to reduce deforestation
  • 1.3. Predicting future forest carbon stocks
• 2. Data and methods
  • 2.1. LPJmL model
  • 2.2. Land-use change
  • 2.3. Climate change and CO₂ projections
  • 2.4. Simulation protocol
  • 2.5. Analysis of model output
• 3. Results
  • 3.1. Impact of climate and land-use change on pan-tropical carbon balances
  • 3.2. Regional differentiation of carbon stocks projections
  • 3.3. Comparison with other estimates of carbon stocks and emissions
• 4. Discussion
  • 4.1. Carbon winners and losers under contrasting climate and land-use change scenarios
  • 4.2. Uncertainties in the estimation of future tropical carbon stocks
  • 4.3. Implications for REDD
• 5. Conclusions
• Acknowledgments
• Appendix
• References

1. Introduction

1.1. Deforestation and climate change

Deforestation is the largest source of emissions from the LULUCF (land use, land-use change and forestry) greenhouse gas inventory sector within the UNFCCC (United Nations Framework Convention on Climate Change) and accounts for 12–20% of global anthropogenic greenhouse gas emissions [1–4]. Land-use change related fluxes to the atmosphere from the tropics have been estimated to be as high as 2.2 ± 0.6 Pg C yr ^{– 1} for the 1990s [5]. Recent estimates for carbon emissions from deforestation and forest degradation show lower rates of 1.2 Pg C yr ^{– 1} over the period 1997–2006, with additional 0.3 Pg C yr ^{– 1} from tropical peatland oxidation [3]. Forest loss in Latin America accounts for 60% of total tropical biome clearing (Brazil 48%). Over one-third of clearing occurs in Asia (Indonesia 13%) and Africa contributes 5% to the estimated loss of humid tropical forest cover [6]. Agriculture, logging and mining are the direct drivers of tropical deforestation and result from or are amplified by population growth, agricultural subsidies and infrastructure investment [7, 8].

1.2. Policy incentives to reduce deforestation

Proposals to finance deforestation reduction have been debated for some years [9]. More recently, opportunities have arisen to provide incentives for developing countries to reduce emissions from deforestation and forest degradation. While the severity of the expected impacts of climate change has increased as described by the IPCC Fourth Assessment Report [10, 11], reducing emissions from deforestation is a cost-effective option for mitigating climate change (although over time marginal costs would rise) [12–14]. The Bali Action Plan provided a mandate to consider the policy incentives to reduce emissions from deforestation and forest degradation (REDD) as part of the post-2012 climate regime.

Full success of REDD would mean halting deforestation immediately. However, even a reduction in deforestation rates is considered as progress [15]. Without successful implementation of forest protection, tropical deforestation is likely to continue throughout this century. According to a study by Kindermann et al [16] today's forest cover would shrink by around 500 million hectares until 2100 without carbon price incentive schemes on deforestation. However, there are various methodological challenges in the implementation of an effective regime on REDD. This includes establishing reference emission levels which do not reward business-as-usual, address leakage or emissions displacement, ensuring policies resulting in permanent emission reductions and developing an effective measuring, reporting and verification system (MRV) [17–22].

1.3. Predicting future forest carbon stocks

While losses due to ongoing deforestation prevail in the international discussion on policy schemes, climate change increasingly is acknowledged as a possible risk for forest carbon stocks [23]. The aim of this study is to give a first assessment of risks arising from climate change in combination with a successful REDD scheme. Since future changes in forest integrity and carbon storage cannot be extrapolated linearly from current observations, we use the advanced dynamic global vegetation model LPJmL [24–26] to disentangle the success of REDD in terms of reduced deforestation against the background of different climate change scenarios on a country scale. The different projections of reducing deforestation success are assessed by applying two extreme land-use change scenarios. In the first scenario, forests are completely protected in every country from 2012 onwards. In the second scenario half of the forest area existing in 2012 is deforested by the end of the twenty-first century, with a constant area deforested every year. We set the year 2012 as earliest possible start point to stop deforestation, because REDD mechanisms will not be implemented beyond pilot studies before the expiration of the Kyoto Protocol. We run the LPJmL model with IPCC AR4 climate change projections of five different general circulation models (GCMs) under forcing from SRES A2 emissions. The results from this study could be of use for policy makers who need to evaluate climate change induced risks for REDD schemes.

2. Data and methods

In this study we investigate the role of climate change and deforestation on the development of future tropical forest carbon stocks. We applied the dynamic global vegetation model LPJmL (described in section 2.1) with two contrasting land-use change scenarios (section 2.2) and five climate change scenarios under SRES A2 emission trajectories (section 2.3). Simulations were conducted for the historic period and the 21st century (section 2.4). The analysis was performed with a focus on tropical countries (more details on selected countries in section 2.5).

2.1. LPJmL model

Process-based dynamic global vegetation models provide an important perspective for understanding the combined effects of increasing levels of atmospheric CO₂, water cycling, and global warming on plant productivity and their component fluxes of water and carbon at spatially differentiated scale. The process-based LPJmL DGVM is a global, grid-based biogeography–biogeochemistry model, which has been comprehensively validated for a broad range of conditions and quantities [24–30]. LPJmL realistically reproduces terrestrial carbon pool sizes and fluxes and the biogeographical distribution of vegetation [26]. The water balance computed by the model performs on the level of state-of-the-art global hydrological models [25]. The representation of agricultural land allows for the quantification of the impacts of land use on water and carbon cycles [24].

The simulation in any grid cell is driven by input of monthly climatology, soil type, atmospheric CO₂ concentration, and agricultural land use. No ecosystem features are prescribed: plant type presence and the associated carbon stocks arise as a function of the environment. In our calculations, LPJmL is run off-line, therefore no feedback mechanisms from vegetation to the atmosphere are considered. Natural vegetation is represented by nine different plant functional types (PFTs), of which two are herbaceous and seven woody. Different PFTs coexist within each grid cell, but their abundance is constrained by climatic conditions and competition. Vegetation structure responds dynamically to changes in climate, including invasion of new habitats and dieback. For the tropics the prevailing PFTs are `tropical broad-leaved evergreen' trees, `tropical broad-leaved raingreen' trees, and the C4 photosynthetic grasses. LPJmL simulates processes as photosynthesis and transpiration, maintenance and growth respiration and reproduction cost. Net primary production (NPP) is allocated to the different plant compartments (vegetation carbon pool) and enters the litter and soil carbon pools due to litter-fall and mortality. Fire disturbance is driven by a threshold litter load and a soil moisture function [31].

As this study focuses on forests carbon stocks we do not simulate the 11 different crop functional types (CFTs) contained in LPJmL, instead we use only one type of agricultural land, which is rain-fed managed grassland. Natural vegetation and managed grasslands are simulated as separate stands in each grid cell, each having its own soil carbon and water pools. The annual fractional coverage of agricultural land in each grid cell is provided by the land-use input to LPJmL. If deforestation occurs, natural vegetation is reduced and the deforested carbon is allocated to the litter pool, eventually entering the soil carbon pool from where it is respired back to the atmosphere. The occurrence of fire leads to an alternative pathway, allowing carbon to return to the atmosphere directly from standing biomass or litter. If agricultural land is abandoned, forest regrowth occurs.

2.2. Land-use change

Several global gridded datasets for historic land use have been developed in recent years [32–35]. The HYDEv3.0 historic land-use dataset [33, 36] comprises cropland and pasture areas from the years 1700 to 2000 with decadal time-steps and was used in this study to determine the fractions of natural vegetation and agricultural land in each grid cell of LPJmL for the historic period. The land-use dataset is based on satellite data and agricultural statistics from the United Nations Food and Agriculture Organization (FAO) and other sub-national land-use data. Distribution of population density, land suitability, distance to major rivers and natural land cover are used as weighting maps to allocate historical cropland. (The HYDE dataset is available at ftp://ftp.mnp.nl/hyde/.) We aggregated the 5 ' × 5 ' (longitude/latitude) resolution data to 30 ' (0.5°), which is the spatial resolution of the LPJmL input drivers. Between the time-slices of each decade, land-use change was linearly interpolated for each grid cell to provide a quasi-continuous yearly historical dataset. We retained deforestation rates from 1990 to 2000 for the period from 2001 to 2012, as for example Hansen et al [6] showed, that rates of clearing from 2000 to 2005 in the humid tropical biome remained comparable with those observed in the 1990s. Post-2012 we applied two extreme land-use scenarios, a forest protection and a deforestation scenario. In the protection scenario we assume full forest protection, where the share of natural vegetation in each grid cell is kept constant from 2012 onwards. In the deforestation scenario every year an equal fraction of natural vegetated land is converted to managed grassland until only 50% of the natural coverage in 2012 is left at the end of the 21st century, which corresponds to a pan-tropical forest loss of 555 million hectares by 2100 (defining forest with a minimum tree canopy cover of 30%) [37]. The deforestation scenario after 2012 does not include regionally differentiated deforestation rates and land abandonment was not taken into account.

2.3. Climate change and CO₂ projections

Climate projections from five general circulation models (GCMs), ECHAM5/MPI-OM, CONS/ECHO-G, UKMO-HadCM3, GFDL-CM2.1 and NCAR/CCSM3.0 under forcing from the SRES A2 emission scenario were used [38]. The models have been used in the World Climate Research Programme's (WCRP's) Coupled Model Intercomparison Project phase 3 (CMIP3) (available from https://esg.llnl.gov:8443/) carried out for the IPCC Fourth Assessment Report [39]. A documentation of all GCMs can be found at www-pcmdi.llnl.gov/ipcc/model_documentation/ipcc_model_documentation.php. Predicted climate anomalies of monthly fields of precipitation and surface air temperature for the years 1860–2100 are calculated for each of the five climate models with respect to the reference period (1960–1990). Those anomalies are interpolated to 0.5° resolution and are combined with the mean climatology for the reference period of an extended CRU TS2.1 climate dataset [40, 41]. Table 1 gives an overview of the GCMs used in this study including bias-corrected projections for temperature and precipitation in the tropical zone. For the SRES A2 scenario all models simulate a temperature increase over land surfaces and broad spatial patterns of increase are similar between GCMs. In contrast, there are major differences between GCMs in projected changes in precipitation, in which the regional patterns vary greatly (figure 1).

We ran the LPJmL model with CO₂ concentrations increasing as they did for the IPCC SRES A2 emission scenario, which is 395 ppm in 2012 rising to 532 ppm in 2050 and reaching 847 ppm in 2099. The SRES A2 scenario includes anthropogenic CO₂ emissions from fossil-fuel consumption and land-use change projections for the 21st century, with a relative contribution from each source of about 95% and 5%, respectively [38]. The SRES A2 is one of the highest emission scenarios of the IPCC range of projections, with increasing growth rates of greenhouse gas emissions during the course of the 21st century. However, recent observations show that growth rates of greenhouse gas emissions are extending beyond the upper boundary of the envelope of IPCC emissions scenarios [42].

2.4. Simulation protocol

In most ecosystems, carbon pools in soil and vegetation reach equilibrium only after a long time. Therefore, a 1000 year spin-up simulation with natural vegetation was carried out. The first spin-up was followed by a second spin-up for 398 years with natural vegetation and managed grassland using land-use patterns from 1860. In the spin-ups LPJmL was driven with climate data from the University of East Anglia's Climatic Research Unit (CRU) [40] with repeating cycles from 1901 to 1930 and with pre-industrial CO₂ concentrations. After the spin-ups the simulations from 1871 to 2099 were conducted with five IPCC AR4 climate change projections, SRES A2 CO₂ concentrations and the two land-use scenarios described above.

2.5. Analysis of model output

The countries selected for this study are the same as listed in the study by Gibbs et al [43] (see table A.1). We added Argentina, Pakistan and Sudan, because these countries had requested participation in the Forest Carbon Partnership Facility (FCPF, www.carbonfinance.org/fcpf; whereas only Argentina has been selected as a REDD country). Except for Bhutan, Nepal and Pakistan all countries are at least partially located within the tropics of Cancer and Capricorn. All countries except French Guiana are listed as non-Annex I parties to the UNFCCC convention. The countries Brunei and Gambia contained less than eight grid cells and were excluded from the analysis (grid cell at 0.5° × 0.5° resolution corresponding  ~  50 km × 50 km) because of inaccuracies in area calculation.

We evaluated LPJmL outputs for vegetation carbon of natural vegetation by comparing with forest carbon estimations given in [43]. They synthesized, mapped and updated prominent forest biomass carbon databases to create a set of national-level forest carbon stock estimates for the year 2000. In addition we compared the coverage of tree PFTs simulated by LPJmL with country-based forest area referenced in the Forest Resources Assessment (FRA) of the FAO [44]. A validation of soil pools simulated by LPJmL is more difficult. Literature data on tropical soil depths and carbon contents are limited and differ strongly. Some datasets include carbon contents for a soil depth of one metre, e.g. the Soil Organic Carbon Map of NRCS (http://soils.usda.gov/use/worldsoils/). The LPJmL version we used has a uniform soil depth of 2 m. However, tropical soils can be much deeper, even if it is difficult to estimate the real extent. Nevertheless, soil carbon is an important component in the ecological system and for the Brazilian Amazon estimates are as high as 27–32 Pg C [45]. Milne et al [45] used detailed geo-referenced datasets of soils, climate, land use and management information and a modelling system to produce soil organic carbon stocks. We compare LPJmL output for the Brazilian Amazon region and for Kenya with these estimates.

We analysed future changes in carbon stocks by summing up simulated carbon pools for each country and comparing the output of the LPJmL model for the mid (2041–2050) and the end of the 21st century (2090–2099) with a reference period (1991–2000). We also looked at trends over the simulated period and for different carbon pools spanning the tropical countries we selected. We include all carbon pools simulated by LPJmL, i.e. vegetation, litter and soil pools of natural vegetation and managed land, if not specified otherwise. Given the uncertainty of tropical soil carbon pools, and in order to allow comparison with other data, we present results of this study in part for above-ground carbon stocks only.

3. Results

3.1. Impact of climate and land-use change on pan-tropical carbon balances

In total, vegetation carbon stocks in the pan-tropics are ranging between 154 and 291 Pg C during the historical period from 1901 to early 21st century (figure 2). Under the GFDL-CM2.1 climate scenario, the lowest carbon pools are projected, while the other four models are in the same range. Overall, tropical carbon stocks decreased during the 20th century, reaching a minimum around 1990, increasing thereafter until 2012. From 2012 on, the effects of the two contrasting land-use change scenarios become evident. Generally, under the forest protection scenario, carbon stocks in the tropics are increasing in our simulations due to the effects of CO₂ fertilization. Simulations with CONS/ECHO-G, GFDL-CM2.1 and NCAR/CCSM3.0 climate projection showed higher gains in carbon stocks with forest protection in comparison to simulations with ECHAM5/MPI-OM or UKMO-HadCM3 climate change projections. Under the deforestation scenario, carbon stocks generally decrease. Stronger decreases in carbon stocks can be observed for the ECHAM5/MPI-OM, UKMO-HadCM3, CONS/ECHO-G and NCAR/CCSM3.0 climate, the scenarios for which LPJmL projects higher carbon stocks under current conditions. For the low carbon stock GFDL-CM2.1 scenario, pan-tropical vegetation carbon stocks show almost no decrease (–24 Pg C).

The simulated tropical vegetation carbon pool (as shown in figure 2) was higher than the soil carbon pool, which held between 204 and 236 Pg C during the historical period from 1901 to early 21st century. Soil and litter pool combined contained about one half of all carbon stocks simulated by LPJmL. The high variability in changes of carbon stocks between different climate projection and land-use scenarios was mainly due to the high variability in the simulated vegetation carbon pool, soil and litter carbon pools were much less affected. When simulated vegetation, soil and litter carbon pools are accounted for, deforestation was reflected by diminishing carbon pools in tropical countries between – 35 Pg C (GFDL-CM2.1) to – 134 Pg C (UKMO-HadCM3) until the end of the 21st century. Without deforestation, tropical carbon pools stabilized to even higher levels than today with an increase ranging from  +  7 Pg C (UKMO-HadCM3) to  +  121 Pg C (NCAR/CCSM3.0).

The sensibility of LPJmL for CO₂ fertilization was tested in order to estimate its effect on simulated carbon stocks. We found that, without an increase in CO₂ concentration during the course of the 21st century, rising temperatures under the SRES A2 climate projection trigger high tree mortality rates from heat stress in LPJmL, causing drastic break downs of pan-tropical carbon stocks (–54 Pg C GFDL-CM2.1 to – 172 Pg C UKMO-HadCM3) without deforestation (see section 4.2. for discussion on the CO₂ fertilization effect).

3.2. Regional differentiation of carbon stocks projections

The changes in carbon stocks were regionally differentiated (figures 3 and 4, table A.1). In Africa and in Asia, and when the forest protection scenario was applied, carbon stocks mainly increased, whereas in Latin America carbon stocks decreased or increased according to the different climate projection. Under the UKMO-HadCM3 climate projection, the LPJmL model simulated a strong reduction of carbon stocks in the Amazon region.

The Asian countries Bangladesh, Cambodia, Sri Lanka and Thailand showed the largest relative increase of their carbon stocks under the forest protection scenario with a high agreement between the different climate scenarios. In Bangladesh carbon stocks increased even under the deforestation scenario (up to  +  10.3%). Malaysia was one of the countries with the highest relative loss under the deforestation scenario (up to – 32.6%). For Indonesia, the country with the highest carbon stock resources in this region, the model simulated carbon uptakes with forest protection (up to  +  24.8%) and carbon stock decreases under the deforestation scenario (up to – 28.0%) under all climate projections.

On the African continent Cameroon, Central African Republic, D.R. Congo, Ethiopia, Gabon and Kenya showed the largest relative increase of carbon stocks under the forest protection scenario. On the other hand, Madagascar and Sudan lost carbon stocks even under the protection scenario (up to – 13.0%, – 15.6% respectively). Burundi showed a carbon loss under the forest protection scenario in simulations with four out of five climate scenarios (–8.8% to  +  10.8%). In contrast, Ethiopian carbon stocks increased even under the deforestation scenario ( + 14.2% to  +  40.5%), likewise in Kenya carbon stocks increased in simulations with four climate change scenarios (–4.2% to  +  30.7%). In D.R. Congo, the country with the largest carbon stocks in Africa, carbon stocks increased ranging from  +  21.9% to  +  58.6% under the forest protection scenario and decreased under the deforestation scenario with four climate scenarios (–22.9% to  +  6.3%). In Senegal and with forest protection the highest variability between the different climate change scenarios was found (–33.7% to  +  37.1%).

In Latin America, the variability in carbon stocks changes resulting from different climate scenarios was higher, especially in Costa Rica, El Salvador, French Guiana, Guyana, Honduras, Nicaragua, Suriname and Venezuela. Despite forest protection and under the UKMO-HadCM3 climate projection the LPJmL simulated a vegetation dieback (more than – 45% carbon loss) in Costa Rica, El Salvador, Guyana, Nicaragua and Suriname. However, in the same countries and under different climate scenarios carbon uptakes were possible, for example in Suriname and Guyana, with more than  +  50% under the GFDL-CM2.1 climate projection. In Brazil and with forest protection, simulated gains in carbon stocks increased under the CONS/ECHO-G, NCAR/CCSM3.0 and GFDL-CM2.1 climate projections (up to  +  38.1%) and decreased under UKMO-HadCM3 and ECHAM5/MPI-OM (up to – 24.8%). Under the deforestation scenario and the UKMO-HadCM3 climate projection there was a simulated loss of – 45.1% in carbon stocks.

3.3. Comparison with other estimates of carbon stocks and emissions

To evaluate how well simulated carbon stocks compare with literature values, we used the country-based estimates for forest biomass carbon stocks for the year 2000 given by Gibbs et al [43]. Simulated vegetation carbon stocks were well within the ranges for most of the tropical countries (figure 5, table A.1). For soil carbon stocks we compared LPJmL output with values given in [45] for the Brazilian Amazon and for Kenya for the year 2000. LPJmL simulated soil carbon stocks were underestimated for the Brazilian Amazon and overestimated for Kenya but within the same order of magnitude. For the Brazilian Amazon the simulated soil carbon stocks without coarse roots were 17 Pg C (21 Pg C including litter) compared to 27–32 Pg C given in [45]. For Kenya simulated carbon stocks were 2.4 Pg C (2.7 Pg C including litter) compared to 1.4–2.0 Pg C. In addition we analysed how well the LPJmL simulated coverage of tree PFTs, constrained by land use, compares with country-based forest inventory data for 2005 by the FAO [44] and found a positive correlation (R² = 0.52, p  <  0.0001).

We show a range of deforestation losses for the tropics from – 35 to – 134 Pg C and gains from forest protection from 7 to 121 Pg C by the end of the 21st century for all carbon pools simulated by LPJmL (forested and not forested land, above and belowground carbon stocks). In a study by Gullison et al [46] estimated losses from tropical deforestation ranged from – 87 to – 130 Pg C by 2100. Estimates by Cramer et al [47] using an earlier version of the LPJ model ranged from – 101 to – 367 Pg C for the tropics by 2100. For the SRES A2 scenarios the cumulative emissions from land-use from 1990 to 2100 range from 49 to 181 Pg C. For comparison, the emissions from fossil fuels range from 1303 to 1860 Pg C [38].

4. Discussion

Generally, we found a high interregional variability between carbon losses and gains for the different scenarios. In consequence, countries may benefit differentially from forest protection, which can be attributed to changing of regional climate regimes. In our simulations forest protection strongly increased carbon stocks in many regions which is mainly due to growth enhancing effects of CO₂. Deforestation, on the other hand, leads to strong carbon stock reduction in most regions. Below, we discuss (1) the potential future impacts on tropical carbon stocks under contrasting climate and land-use change scenarios, (2) the uncertainties in the estimation of future tropical carbon stocks, and (3) the implications for a successful REDD mechanism.

4.1. Carbon winners and losers under contrasting climate and land-use change scenarios

During recent decades, old-growth and intact forests in the tropics were carbon sinks, accumulating approximately 0.8–1.6 Pg C yr ^{– 1} [48]. In Africa, the increasing carbon storage of intact tropical forests has been attributed to an increase in resource availability, including fertilization by atmospheric CO₂, changes in solar radiation at the Earth's surface, increases in nutrient deposition and changes in rainfall [48]. How the carbon storage potential of tropical forests will change under future climate conditions is nevertheless highly uncertain. Changes in precipitation patterns and temperature increase among other factors could strongly alter vegetation dynamics. Over the past two decades air temperatures in the tropical forest biome have increased on average by 0.26 °C/decade [49]. There has been a strong and significant decline in rainfall in the northern African tropics, but no significant trend in other tropical regions. Similarly, strength and intensity of the dry season have significantly increased in Africa but not in Latin America or Asia [49]. Despite some recent progress in global climate model development [50], climate scenarios continue to contain substantial uncertainties. In terms of their ability to forecast long-term trends there are important differences between climate models, especially on a regional scale [51, 52]. Most climate models project increasing temperatures with similar spatial patterns. More pronounced differences exist for projected changes in precipitation (table 1, figure 1).

For tropical Asia most GCMs simulate a general increase in precipitation until the end of the century, although the seasonal distribution remains uncertain. In Africa, the prediction for changes in precipitation patterns is not uniform. For central Africa four out of five climate models predict an increase in precipitation (figure 1). In Asia and Africa climate change in combination with increasing CO₂ concentrations had an overall positive effect on carbon storage potentials in simulations with LPJmL. For some regions, e.g. parts of the African highlands (Ethiopia, Kenya), gains in carbon stock were simulated despite a reduction of 50% of the countries naturally vegetated area under the deforestation scenario. Carbon losses from deforestation were overcompensated by the combined effects of CO₂ fertilization and climate change. However, simulated carbon stocks in the reference period are overestimated for these countries, which might be due to missing disturbance processes in the LPJmL model. Nevertheless, the simulated abundance of tree PFTs was still very low in this region. Climatic change increased tree cover (replacing C4 grasses) and there was vegetation growth in previously non-vegetated areas. In addition, the CO₂ fertilization effect increased NPP and both effects were leading to the relatively strong carbon sink.

In Latin America GCMs vary greatly in their projections of future climate change [53–55], accordingly, the congruence in simulated changes of carbon stocks between different climate scenarios was particularly low for this region (figure 3). A high inter-annual variability in precipitation in the GFDL-CM2.1 climate projection caused an underestimated net primary production (NPP) in tropical Latin America, consequently reducing pan-tropical vegetation carbon stocks, with relatively little changes in the 21st century under the deforestation scenario (figure 2). This demonstrates the relative importance of tropical rainforests in Latin America for pan-tropical carbon stocks. In simulations with UKMO-HadCM3 climate projection, where a strong decrease in precipitation is projected for the Amazon region, the LPJmL model simulated a vegetation dieback, even without the additional pressure of increasing land use (figure 3). This result is in accordance with findings of other studies, in which for parts of the Amazon basin a tipping for the rain forest into savannah is shown [56–58]. Other recent studies on the Amazonian rainforest emphasize the high vulnerability of this region due to climate change in combination with land-use change [54, 59, 60]. Land-use change including large-scale deforestation and fragmentation might trigger or strongly enhance climatic change effects. For carbon stocks and the net carbon exchange land-use change may well be more important than climatic change [30, 47]. Tropical Latin America has a higher risk to lose large amounts of its carbon stocks during the course of this century.

4.2. Uncertainties in the estimation of future tropical carbon stocks

Generally, our simulated carbon stocks are in the range of other studies (figure 5, table A.1). In the model, land use constrains the area of natural vegetation, which is forested if climate conditions allow it. Thus, the size of the forested area determines the natural vegetation carbon balances. We used the HYDE3.0 gridded dataset to constrain historic and current land use in LPJmL. However, different land-use datasets are not consistent and can differ especially regionally, because of the differences in the methods applied, the use of different input data, and definitions (e.g. for pasture land) [61]. One of the most important reference dataset for forests and deforestation trends is the Forest Resources Assessment (FRA) of the FAO [44]. But changing classification schemes over time, adjustments in the presentation of trends, as well as in aggregating algorithms, make the data an inconsistent source of global deforestation rates and trends [62]. The inconsistencies in different datasets may explain that the correlation we found between simulated forest areas and country-based forest areas given by the FAO was not high (R² = 0.52). As it is difficult to determine current land use and land-use change rates, large uncertainties exist over the changing rate of deforestation in the future. The IMAGE model has been used to project future land-use changes under different SRES scenarios [63]. IMAGE land-use projections have been applied to study the effects of climate and land-use change on the global terrestrial carbon cycle for the 21st century using the LPJmL model [64]. The current study mainly focuses on changes in tropical forest carbon stocks by comparing hypothetical land-use scenarios with climate scenarios, temporal and regional differentiated land-use scenarios were not used or developed.

Our study shows that under the protection scenario, in some countries the carbon gain is large (figures 3 and 4, table A.1). This is due to the model's assumption of enhanced water use efficiency by CO₂ fertilization. There is no consensus in the scientific community about the magnitude of the CO₂ fertilization effect with rising CO₂ concentrations under climate change. The sensibility towards CO₂ in LPJ might be rather over-than underestimated [47]. Hickler et al [28] showed that the LPJ-GUESS dynamic vegetation model reproduces the magnitude of the NPP enhancement at temperate forest FACE experiments, but in tropical forests predicted NPP enhancement was more than twice as high as in boreal forests, suggesting that currently available FACE results are not applicable to tropical ecosystems. It has been argued that the availability of nutrients will constrain NPP responses to CO₂ enhancement [28]. However, in LPJmL CO₂ fertilization is limited only by the availability of water, and processes for nitrogen and phosphorus limitation, which are especially important in the tropics [65, 66] are not represented.

Other factors influencing the estimation of changes in future carbon stocks are selective logging, fire, forest grazing and edge effects in fragmented landscapes [54]. Forest degradation is difficult to detect at large scale and is not necessarily stopped with deforestation [62, 67]. Fire in the tropics is primarily associated with human activity and influence on land cover; lightning strikes rarely lead to forest fires, as these events are usually associated with heavy rainfall [68]. Fire as a disturbance factor is causing biomass loss and modified site conditions might delay or prevent regeneration of the vegetation. In the LPJmL model, fire disturbance is included by a process-based fire-module, which allows for fires in natural vegetation ignited only by lightning [31]. Deforestation and forest degradation frequently lead to nutrient depletion, soil degradation or erosion—processes that reduce a region's growth potential irreversibly on a timescale of centuries. Most processes of forest or soil degradation are not represented in LPJmL, so that future carbon gains might be overestimated.

4.3. Implications for REDD

Our results show that tropical forests have the potential to increase their carbon stocks substantially, if they are protected. In contrast, climate change possesses risks for forest carbon stocks to decrease without any direct human influence. The challenge in a policy context lies in determining how incentives will be given to countries for reducing emissions and protecting forests. In providing incentives to countries for increases in carbon stocks, natural and indirect human induced effects such as CO₂ fertilization as well as the risks of climate change impacts must also be taken into account. Thus it will be important to understand the processes that govern current greenhouse gas emissions and future projections [69]. As with developed countries in the Kyoto Protocol, it will be necessary to improve how to factor out the impacts of CO₂ fertilization effects and the impacts of climate change [69, 70]. Incentives should be restricted to direct human induced increases in carbon stocks and reductions in deforestation emissions below business-as-usual. Therefore, it must be considered to include not only carbon stocks alone but also other criteria that refer to policy implementation combating the drivers of deforestation as a calculation basis to pay for successful forest protection [71].

5. Conclusions

Climate change will have regionally differentiated impacts on tropical carbon stocks. Countries in tropical South East Asia and Africa could profit from higher carbon densities mainly due to changes in precipitation patterns, increase in temperature, and CO₂ fertilization effects. Also positive effects due to CO₂ fertilization might prevail in the coming decades, latest at the end of the century severe losses due to climate change induced forest degradation could be expected at least for some parts of the tropics, e.g. for Latin America. There is a higher risk that large parts of the tropical Amazonian rainforest could degrade due to a strong reduction in rainfall. Limiting deforestation and the spread of fires may be successful tools to maintain Amazonian forest resilience under the risk of future climate change [54, 72].

Based on the findings of this study, we suggest that factors such as future changes of climate, water availability, as well as CO₂ fertilization effects must be taken into account in order to achieve an effective and fair REDD mechanism. Continuing to gain an understanding of the different interactions affecting carbon stocks and related emissions from the land-use sector will become increasingly important in identifying the direct human induced reductions from deforestation.

Acknowledgments

This study was financially supported by the EU Marie Curie Research Training Network GREENCYCLES (MRTN-CT-2004-512464) and by the German BMBF (Bundesministerium für Bildung und Forschung). Results benefitted from discussions within the context of the Klima-und-Gerechtigkeit Project (www.klima-und-gerechtigkeit.de). We thank two anonymous referees for valuable comments on the manuscript. We acknowledge the modelling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP's Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multi-model dataset. Support of this dataset is provided by the Office of Science, US Department of Energy.

Appendix

References

[1]

Achard F, Eva H D, Mayaux P, Stibig H-J and Belward A 2004 Improved estimates of net carbon emissions from land cover change in the tropics for the 1990s Glob. Biogeochem. Cycles 18 GB2008 
CrossRef

[2]

Schimel D S et al 2001 Recent patterns and mechanisms of carbon exchange by terrestrial ecosystems Nature 414 169–72 
CrossRefPubMed

[3]

van der Werf G R, Morton D C, DeFries R S, Olivier J G J, Kasibhatla P S, Jackson R B, Collatz G J and Randerson J T 2009  CO₂ emissions from forest loss Nat. Geosci. 2 737–8 
CrossRef

[4]

IPCC 2000 Special Report on Land Use, Land-Use Change and Forestry ed R T Watson, I R Noble, B Bolin, N H Ravindranath, D J Verardo and D J Dokken (Cambridge: Cambridge University Press) p 377 available at www.ipcc.ch/ipccreports/sres/land_use/index.php?idp = 0 

[5]

Houghton R A 2003 Revised estimates of the annual net flux of carbon to the atmosphere from changes in land use and land management 1850–2000 Tellus B 55 378–90 
CrossRef

[6]

Hansen M C et al 2008 Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data Proc. Natl Acad. Sci. USA 105 9439–44 
CrossRefPubMed

[7]

Lambin E F, Geist H J and Lepers E 2003 Dynamics of land-use and land-cover change in tropical regions Ann. Rev. Environ. Resour. 28 205–41 
CrossRef

[8]

Santilli M, Moutinho P, Schwartzman S, Nepstad D, Curran L and Nobre C 2005 Tropical deforestation and the Kyoto Protocol Clim. Change 71 267–76 
CrossRef

[9]

Fearnside P M 2001 Saving tropical forests as a global warming countermeasure: an issue that divides the environmental movement Ecol. Econ. 39 167–84 
CrossRef

[10]

IPCC 2007 Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change ed M L Parry, O F Canziani, J P Palutikof, P J van der Linden and C E Hanson (Cambridge: Cambridge University Press) p 976 available at www.ipcc.ch/publications_and_data/ar4/wg2/en/contents.html 

[11]

Smith J B et al 2009 Assessing dangerous climate change through an update of the Intergovernmental Panel on Climate Change (IPCC) `reasons for concern' Proc. Natl Acad. Sci. USA 106 4133–7 
CrossRefPubMed

[12]

Kindermann G, Obersteiner M, Sohngen B, Sathaye J, Andrasko K, Rametsteiner E, Schlamadinger B, Wunder S and Beach R 2008 Global cost estimates of reducing carbon emissions through avoided deforestation Proc. Natl Acad. Sci. USA 105 10302–7 
CrossRefPubMed

[13]

H M Treasury 2006 Stern Review on the Economics of Climate Change (London: H M Treasury) available at www.hm-treasury.gov.uk/stern_review_report.htm 

[14]

Strassburg B, Turner R K, Fisher B, Schaeffer R and Lovett A 2009 Reducing emissions from deforestation—the `combined incentives' mechanism and empirical simulations Glob. Environ. Change 19 265–78 
CrossRef

[15]

Gurney K R and Raymond L 2008 Targeting deforestation rates in climate change policy: a `Preservation Pathway' approach Carbon Balance Manag. 3 doi:10.1186/750-0680-3-2 

[16]

Kindermann G, Obersteiner M, Rametsteiner E and McCallum I 2006 Predicting the deforestation-trend under different carbon-prices Carbon Balance Manag. 1 doi:10.1186/750-0680-1-15 

[17]

Fry I 2008 Reducing emissions from deforestation and forest degradation: opportunities and pitfalls in developing a new legal regime Rev. European Community Int. Environ. Law 17 166–82 
CrossRef

[18]

Karsenty A 2008 The architecture of proposed REDD schemes after Bali: facing critical choices Int. Forest. Rev. 10 443–57 
CrossRef

[19]

Angelsen A 2008 REDD models and baselines Int. Forest. Rev. 10 465–75 
CrossRef

[20]

Dutschke M and Wolf R 2007 Reducing emissions from deforestation in developing countries: the way forward Deutsche Gesellschaft für Technische Zusammenarbeit (GTZ) available at www.gtz.de/de/dokumente/en-climate-reducing-emissions.pdf 

[21]

Murray B C 2008 Leakage from an avoided deforestation compensation policy: concepts, empirical evidence, and corrective policy options Working Paper available at http://nicholas.duke.edu/institute/wp-leakage.pdf 

[22]

UNFCCC 2008 Results of the work on scientific and methodological aspects of the proposal by Brazil The Twenty-Eighth Session of the Subsidiary Body for Scientific and Technological Advice, United Nations Framework Convention on Climate Change Bonn, Germany available at http://unfccc.int/resource/docs/2008/sbsta/eng/misc01.pdf 

[23]

Ebeling J and Yasue M 2008 Generating carbon finance through avoided deforestation and its potential to create climatic, conservation and human development benefits Phil. Trans. R. Soc. B 363 1917–24 
CrossRef

[24]

Bondeau A et al 2007 Modelling the role of agriculture for the 20th century global terrestrial carbon balance Glob. Change Biol. 13 679–706 
CrossRef

[25]

Gerten D, Schaphoff S, Haberlandt U, Lucht W and Sitch S 2004 Terrestrial vegetation and water balance—hydrological evaluation of a dynamic global vegetation model J. Hydrol. 286 249–70 
CrossRef

[26]

Sitch S, Smith B and Prentice I C 2003 Evaluation of ecosystem dynamics, plant geography and terrestrial carbon cycling in the LPJ dynamic global vegetation model Glob. Change Biol. 9 161–85 
CrossRef

[27]

Cowling S A and Shin Y 2006 Simulated ecosystem threshold responses to co-varying temperature, precipitation and atmospheric CO₂ within a region of Amazonia Glob. Ecol. Biogeogr. 15 553–66 
CrossRef

[28]

Hickler T, Smith B, Prentice I C, Mjofors K, Miller P, Arneth A and Sykes M T 2008  CO₂ fertilization in temperate FACE experiments not representative of boreal and tropical forests Glob. Change Biol. 14 1531–42 
CrossRef

[29]

Lucht W, Prentice I C, Myneni R B, Sitch S, Friedlingstein P, Cramer W, Bousquet P, Buermann W and Smith B 2002 Climatic control of the high-latitude vegetation greening trend and Pinatubo effect Science 296 1687–9 
CrossRefPubMed

[30]

Poulter B, Aragao L, Heyder U, Gumpenberger M, Heinke J, Langerwisch F, Rammig A, Thonicke K and Cramer W 2009 Net biome production of the Amazon Basin in the 21st century Glob. Change Biol. doi:10.1111/j.365-2486.009.02064.x 

[31]

Thonicke K, Venevsky S, Sitch S and Cramer W 2001 The role of fire disturbance for global vegetation dynamics: coupling fire into a dynamic global vegetation model Glob. Ecol. Biogeogr. 10 661–77 
CrossRef

[32]

Erb K-H, Gaube V, Krausmann F, Plutzar C, Bondeau A and Haberl H 2007 A comprehensive global 5 min resolution land-use data set for the year 2000 consistent with national census data J. Land Use Sci. 2 191–224 
CrossRef

[33]

Goldewijk K K, van Drecht G and Bouwman A F 2007 Mapping contemporary global cropland and grassland distributions on a 5 × 5 minute resolution J. Land Use Sci. 2 167–90 
CrossRef

[34]

Ramankutty N, Evan A T, Monfreda C and Foley J A 2008 Farming the planet: 1. Geographic distribution of global agricultural lands in the year 2000 Glob. Biogeochem. Cycles 22 GB1003 
CrossRef

[35]

Ramankutty N and Foley J A 1999 Estimating historical changes in global land cover: croplands from 1700 to 1992 Glob. Biogeochem. Cycles 13 997–1027 
CrossRef

[36]

Goldewijk K K and van Drecht G 2006 HYDE 3: current and historical population and land cover Integrated Modelling of Global Environmental Change. An Overview of IMAGE 2.4 ed A F Bouwman, T Kram and K K Goldewijk (Bilthoven: Netherlands Environmental Assessment Agency) available at www.rivm.nl/bibliotheek/rapporten/500110002.pdf 

[37]

FAO 2006 Choosing a forest definition for the Clean Development Mechanism Forests and Climate Change Working Paper 4 ed T Neeff, H von Luepke and D Schoene (Rome: Food and Agriculture Organization of the United Nations) available at www.fao.org/forestry/11280-1-0.pdf 

[38]

IPCC 2000 Special Report on Emissions Scenarios: A Special Report of Working Group III of the Intergovernmental Panel on Climate Change ed N Nakicenovic and R Swart (Cambridge: Cambridge University Press) p 599 available at www.ipcc.ch/ipccreports/sres/emission/index.php?idp = 0 

[39]

IPCC 2007 Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change ed S Solomon, D Qin, M Manning, Z Chen, M Marquis, K B Averyt, M Tignor and H L Miller (Cambridge: Cambridge University Press) p 996 available at www.ipcc.ch/publications_and_data/ar4/wg1/en/contents.html 

[40]

Mitchell T D and Jones P D 2005 An improved method of constructing a database of monthly climate observations and associated high-resolution grids Int. J. Climatol. 25 693–712 
CrossRef

[41]

Österle H, Gerstengarbe F W and Werner P C 2003 Homogenisierung und Aktualisierung des Klimadatensatzes der Climate Research Unit der University of East Anglia, Norwich Terra Nostra 6 326–9 

[42]

Richardson K et al 2009 Climate Change: Global Risks, Challenges and Decisions http://climatecongress.ku.dk 

[43]

Gibbs H K, Brown S, Niles J O and Foley J A 2007 Monitoring and estimating tropical forest carbon stocks: making REDD a reality Environ. Res. Lett. 2 045023 
IOPscience

[44]

FAO 2006 Global Forest Resources Assessment 2005: Progress Towards Sustainable Forest Management (Rome: Food and Agriculture Organization of the United Nations) available at www.fao.org/DOCREP/008/a0400e/a0400e00.htm 

[45]

Milne E et al 2007 An increased understanding of soil organic carbon stocks and changes in non-temperate areas: national and global implications Agric. Ecosyst. Environ. 122 125–36 
CrossRef

[46]

Gullison R E et al 2007 Tropical forests and climate policies Science 316 985–6 
CrossRefPubMed

[47]

Cramer W, Bondeau A, Schaphoff S, Lucht W, Smith B and Sitch S 2004 Tropical forests and the global carbon cycle: impacts of atmospheric carbon dioxide, climate change and rate of deforestation Phil. Trans. R. Soc. B 359 331–43 
CrossRef

[48]

Lewis S L et al 2009 Increasing carbon storage in intact African tropical forests Nature 457 1003–6 
CrossRefPubMed

[49]

Lewis S L, Malhi Y and Phillips O L 2004 Fingerprinting the impacts of global change on tropical forests Phil. Trans. R. Soc. B 359 437–62 
CrossRef

[50]

Reichler T and Kim J 2008 How well do coupled models simulate today's climate? Bull. Am. Meteorol. Soc. 89 303–11 
CrossRef

[51]

Giorgi F 2006 Climate change hot-spots Geophys. Res. Lett. 33 L08707 
CrossRef

[52]

Gleckler P J, Taylor K E and Doutriaux C 2008 Performance metrics for climate models J. Geophys. Res. 113 D06104 
CrossRef

[53]

Cook K H and Vizy E K 2008 Effects of twenty-first-century climate change on the Amazon rain forest J. Clim. 21 542–60 
CrossRef

[54]

Malhi Y, Aragão L E O C, Galbraith D, Huntingford C, Fisher R, Zelazowski P, Sitch S, McSweeney C and Meir P 2009 Exploring the likelihood and mechanism of a climate-change-induces dieback of the Amazon rainforest Proc. Natl Acad. Sci. USA 106 20610–5 
CrossRefPubMed

[55]

Vera C and Silvestri G 2009 Precipitation interannual variability in South America from the WCRP-CMIP3 multi-model dataset Clim. Dyn. 32 1003–14 
CrossRef

[56]

Cowling S A, Betts R A, Cox P M, Ettwein V J, Jones C D, Maslin M A and Spall S A 2004 Contrasting simulated past and future responses of the Amazon forest to atmospheric change Phil. Trans. R. Soc. B 359 539–47 
CrossRef

[57]

Cox P M, Betts R A, Collins M, Harris P P, Huntingford C and Jones C D 2004 Amazonian forest dieback under climate-carbon cycle projections for the 21st century Theor. Appl. Climatol. 78 137–56 
CrossRef

[58]

Phillips O L et al 2009 Drought sensitivity of the Amazon rainforest Science 323 1344–7 
CrossRefPubMed

[59]

Senna M C A, Costa M H and Pires G F 2009 Vegetation-atmosphere-soil nutrient feedbacks in the Amazon for different deforestation scenarios J. Geophys. Res. 114 D04104 
CrossRef

[60]

Nepstad D C, Stickler C M, Soares-Filho B and Merry F 2008 Interactions among Amazon land use, forests and climate: prospects for a near-term forest tipping point Phil. Trans. R. Soc. B 363 1737–46 
CrossRef

[61]

Goldewijk K K and Ramankutty N 2004 Land cover change over the last three centuries due to human activities: the availability of new global data sets GeoJournal 61 335–44 
CrossRef

[62]

Grainger A 2008 Difficulties in tracking the long-term global trend in tropical forest areas Proc. Natl Acad. Sci. USA 105 818–23 
CrossRefPubMed

[63]

Strengers B, Leemans R, Eickhout B, de Vries B and Bouwman L 2004 The land-use projections and resulting emissions in the IPCC SRES scenarios as simulated by the IMAGE 2.2 model GeoJournal 61 381–93 
CrossRef

[64]

Müller C, Eickhout B, Zaehle S, Bondeau A, Cramer W and Lucht W 2007 Effects of changes in CO₂, climate, and land use on the carbon balance of the land biosphere during the 21st century J. Geophys. Res. 112 G02032 
CrossRef

[65]

Sanchez P 2002 Soil fertility and hunger in Africa Science 295 2019–20 
CrossRefPubMed

[66]

Zougmoré R, Zida Z and Kamboua N F 2003 Role of nutrient amendments in the success of half-moon soil and water conservation practice in semiarid Burkina Faso Soil Tillage Res. 71 143–9 
CrossRef

[67]

Foley J et al 2007 Amazonia revealed: forest degradation and loss of ecosystem goods and services in the Amazon Basin Front. Ecol. Environ. 5 25–32 
CrossRef

[68]

Cochrane M A 2003 Fire science for rainforests Nature 421 913–9 
CrossRefPubMed

[69]

Canadell J G, Kirschbaum M, Kurz W, Sanz M-J, Schlamadinger B and Yamagata Y 2007 Factoring out natural and indirect human effects on terrestrial carbon sources and sinks Environ. Sci. Policy 10 370–84 
CrossRef

[70]

IPCC IPCC meeting on current scientific understanding of the processes affecting terrestrial carbon stocks and human influences upon them, 2003 Expert Meeting Report (Geneva, July 2003) available at www.ipcc.ch/pdf/supporting-material/ipcc-meeting-2003-07.pdf 

[71]

Motel P C, Pirard R and Combes J-L 2009 A methodology to estimate impacts of domestic policies on deforestation: Compensated Successful Efforts for `avoided deforestation' (REDD) Ecol. Econ. 68 680–91 
CrossRef

[72]

Cochrane M A and Laurance W F 2008 Synergisms among fire, land use, and climate change in the Amazon Ambio 37 522–7 
CrossRefPubMed