Global Biogeochemical Cycles

Exploring the range of climate biome projections for tropical South America: The role of CO2 fertilization and seasonality

Authors

  • David M. Lapola,

    1. Centro de Previsão de Tempo e Estudos Climáticos, Instituto Nacional de Pesquisas Espaciais, Cachoeira Paulista, Brazil
    2. Now at Center for Environmental Systems Research, University of Kassel, Kassel, Germany.
    3. Also at International Max Planck Research School on Earth System Modeling, Hamburg, Germany.
    Search for more papers by this author
  • Marcos D. Oyama,

    1. Instituto de Aeronáutica e Espaço, Divisão de Ciências Atmosféricas, São José dos Campos, Brazil
    Search for more papers by this author
  • Carlos A. Nobre

    1. Centro de Previsão de Tempo e Estudos Climáticos, Instituto Nacional de Pesquisas Espaciais, Cachoeira Paulista, Brazil
    Search for more papers by this author

Abstract

[1] Tropical South America vegetation cover projections for the end of the century differ considerably depending on climate scenario and also on how physiological processes are considered in vegetation models. In this paper we use a potential vegetation model (CPTEC-PVM2) to analyze biome distribution in tropical South America under a range of climate projections and a range of estimates about the effects of increased atmospheric CO2. We show that if the CO2 “fertilization effect” indeed takes place and is maintained in the long term in tropical forests, then it will avoid biome shifts in Amazonia in most of the climate scenarios, even if the effect of CO2 fertilization is halved. However, if CO2 fertilization does not play any important role on tropical forests in the future or if dry season is longer than 4 months (projected by 2/14 GCMs), then there is replacement of large portions of Amazonia by tropical savanna.

1. Introduction

[2] Amazonia, along with the circumpolar boreal and Artic regions, has been identified as an especially vulnerable region for biome shifts caused by future climate change [Lenton et al., 2008; Scholze et al., 2006]. The stability and resilience of the Amazonian climate-vegetation dynamics have long been investigated through modeling studies [e.g., Nobre et al., 1991; Zeng and Neelin, 1999; Cox et al., 2000, 2004] and also by experimental studies [e.g., Brando et al., 2008]. Early studies concentrated on the climatic impacts of Amazonian-wide deforestation, which could lead to a drier and warmer regional climate [e.g., Nobre et al., 1991; Zeng et al., 1996]. In the last years focus has changed to the impacts of climate change on forest dynamics, and its potential long-term replacement by drier biomes such as tropical savanna, C4 grasslands or even desert [Cox et al., 2000, 2004; Huntingford et al., 2004; Scholze et al., 2006; Schaphoff et al., 2006; Salazar et al., 2007; Cook and Vizy, 2008; Huntingford et al., 2008]. However, these latter assessments differ considerably in their outcomes, depending mostly on the prescribed climate scenarios [Huntingford et al., 2004; Scholze et al., 2006; Schaphoff et al., 2006; Salazar et al., 2007], but also, though to a smaller extent, on the dynamic global vegetation model (DGVM) used [Cramer et al., 2001; Sitch et al., 2008; Huntingford et al., 2008]. That is also valid in coupled climate–carbon cycle simulations, even though there are indications toward amplification of global warming due to the additional increase in atmospheric [CO2] by soil respiration and reduced carbon uptake by land and ocean [Cox et al., 2000, 2004; Friedlingstein et al., 2006]. While increased temperatures and decreased precipitation would cause loss of vegetation biomass [Scholze et al., 2006; Schaphoff et al., 2006; Brando et al., 2008], higher atmospheric [CO2] could increase carbon uptake through the so-called CO2 fertilization effect [Curtis and Wang, 1998; Prentice et al., 2001; Norby et al., 2005]. Furthermore, modeling studies point that net primary productivity (NPP) enhancement due to increased [CO2] in tropical forests might be 60% higher than in temperate forests [Hickler et al., 2008]. In fact, little is known about the effects of CO2 increase on tropical forest ecosystems since no open air experiment of CO2 enrichment has been conducted so far in tropical forests, but only in extratropical ecosystems [Ainsworth and Long, 2005, Norby et al., 2005]. However, there is some evidence suggesting that it might have a strong effect on primary productivity: rising [CO2] has been suggested as one of the main causes for the observed increases in recruitment and growth rates over the Amazon and African forests over the last 3 decades [Lewis et al., 2004, 2009]. Nevertheless, whether such an increase is likely to continue or not in the long term is still subject of debate.

[3] In theory (i.e., it has been shown to be valid for temperate ecosystems but not for tropical forests yet), higher [CO2] optimizes water use by plants, reducing canopy transpiration, but also increases NPP, which increases canopy transpiration [Field et al., 1995]. This water released through plant transpiration is particularly relevant for the Amazonian forest since it accounts for 25 to 50% of local rainfall [Nobre et al., 1991; Eltahir and Bras, 1994; Betts et al., 2004]. Therefore, quantifying the causes and thresholds responsible for future biome shifts in tropical South America involves multiple driving factors and as yet not fully understood mechanisms. Large-scale vegetation models are useful to address these questions since the effect of different drivers of changes can be assessed separately, even though some mechanisms that might be important for the Amazonian forest, like phosphorous cycling [Chambers and Silver, 2004], are not currently considered in these models.

[4] In this study we analyze the critical climatic thresholds and plant physiological mechanisms acting in a range of future climate and CO2 fertilization scenarios for South America using a potential vegetation model driven by different climate scenarios. Along with a consensual biome projection derived from 14 different climate scenarios, we comprehensively analyze the outputs of four GCMs which have better performance over tropical South America to simulate current climate and project different extremes in future climate change (low- to high-temperature increase; increase, decrease, or minor changes in precipitation). Differently from previous studies [Cramer et al., 2001; Sitch et al., 2008; Schaphoff et al., 2006; Scholze et al., 2006] the separate effects of climate and increased [CO2] on biome distribution are assessed for climate scenarios other than the Hadley Center family of global climate models, considering that NPP enhancement by CO2 fertilization depends on temperature increase (Figure 1) [Lloyd and Farquhar, 2008; Hickler et al., 2008]. Also differently from the studies by Salazar et al. [2007] and Cook and Vizy [2008], CO2-plant interactions through plant physiological processes and their interactions with the water cycle are now considered. Although not studied as deeply as for the Amazonian forest biome, biome shifts in the semiarid northeast Brazil (NEB) are also projected to happen [Oyama and Nobre, 2003; Salazar et al., 2007; Cook and Vizy, 2008] and therefore are analyzed here as well.

Figure 1.

Enhancement of NPP under three atmospheric [CO2] according to (a) different temperatures and (b) soil water levels in comparison to a baseline NPP (25°C and nonwater-stressed conditions) in the CPTEC-PVM2 with full CO2 fertilization effect.

2. Material and Methods

2.1. CPTEC-PVM2

[5] We develop the CPTEC-PVM2, a new version of the CPTEC (global) Potential Vegetation Model [Oyama and Nobre, 2004], which is fully described and evaluated in the auxiliary material. As its predecessor, CPTEC-PVM2 shows a particularly good performance over South America due to the consideration of seasonality as a determinant for the delimitation of forests and savannas. However, the use of the original CPTEC-PVM for future climate-vegetation simulations [as Salazar et al., 2007; Cook and Vizy, 2008] is limited, since the quantification of seasonality is done solely through climate-hydrological variables and, unlike CPTEC-PVM2, does not take into account plant physiological responses to this seasonality (such as primary productivity) under varying atmospheric [CO2]. The CPTEC-PVM2 biome allocation rules are almost completely different from the original version and rely mainly on the optimum net primary productivity (NPP) values for a given grid cell, which demands calculation of gross primary productivity (GPP) and plant respiration (Ra), based (in some aspects) on those used by TRIFFID-DVM [Cox, 2001] and Simple TRIFFID-DVM [Huntingford et al., 2000]. The determination of biome distribution through NPP is done here (tentative) on the basis of numerous studies showing that different biomes have different average NPP [e.g., Sahagian and Hibbard, 1998; Turner et al., 2006]. However, in some cases NPP can be quite similar among biomes (such as for boreal forest and grassland) and, in these cases, other variables than NPP (e.g., coldest month temperature) are used for biome allocation. As a nondynamic model, it calculates only equilibrium solutions based on long-term mean monthly climate variables. This is done concomitantly with a water balance submodel using climatologies of surface temperature and precipitation (1961–1990 (C. J. Willmott and K. Matsuura, University of Delaware, Newark, Terrestrial air temperature and precipitation: Monthly and annual climatologies, available at http://climate.geog.udel.edu/∼climate/html_pages/archive.html, hereinafter referred to as Willmott and Matsuura, data set, 1998)), incident photosynthetically active radiation (IPAR) (1986–1995 [Raschke et al., 2006]) and atmospheric [CO2] (1961–1990, 350 ppmv) as inputs.

[6] Water balance routine is nearly the same as in CPTEC-PVM [Oyama and Nobre, 2004] based on the work by Willmott et al. [1985], though canopy resistance rc (1/canopy conductance gc) is now calculated in terms of NPP and atmospheric [CO2], based on the formulation by Collatz et al. [1991], which is used by several DGVMs [Sitch et al., 2008] and GCM surface schemes [e.g., Sellers et al., 1996]. The canopy resistance is used to calculate evapotranspiration (hereafter E) according to Penman-Monteith equation. This formulation enables a two-way interaction of water cycle and plant physiology. CPTEC-PVM2 also considers a simple parameterization of lightning-induced fires in savannas based on the study by Cardoso et al. [2008]. Fire occurrence is regarded as dependent on the availability of natural ignition source (using 850 hPa zonal wind as a proxy to lightning) and fuel moisture (through soil water level).

[7] Global and especially South America NPP simulated by CPTEC-PVM2 are quite comparable to that from observations and also from other NPP models. Biome distribution is evaluated against a new geographical analysis of natural vegetation [Lapola et al., 2008] and results in a global kappa statistics [Monserud and Leemans, 1992] of 0.53, and agreement fraction of 57% (see the auxiliary materials).

2.2. Climate Scenarios

[8] CPTEC-PVM2 is forced by precipitation and surface temperature inputs derived from 14 coupled ocean-atmosphere global climate models of the IPCC AR4. They are BCCR-BCM2.0, CCCMA-CGCM3.1, CNRM-CM3, CSIRO-Mk3.0, ECHAM5/MPI-OM, GFDL-CM2.0, GFDL-CM2.1, GISS-ER, INM-CM3.0, IPSL-CM4.0, MIROC3.2(MedRes), MRI-CGCM2.3.2, NCAR-CCSM3 and UKMO-HadCM3 (for model details, see http://www.ipcc-data.org). These models have distinct horizontal resolution and project the climate for the 21st century according to changes in several climate forcings, the most important being the increase of atmospheric GHG concentrations. We consider two GHG emissions scenarios: SRES-A2 and B1. As input to CPTEC-PVM2, we consider mean values of atmospheric [CO2] of 730 ppmv for A2 scenario and 535 ppmv for B2 in the 2070–2099 time slice [Intergovernmental Panel on Climate Change, 2000]. The 1961–1990 model climatology for each GCM is used to derive the model's anomalies. To filter out the effect of GCM's systematic errors in calculating the input fields of precipitation, surface temperature and zonal winds, these anomalies are added to the 1961–1990 observed climatology and this sum is used as input to CPTEC-PVM2, following an anomaly coupling procedure [Kutzbach et al., 1998].

[9] In a first analysis, we look for the grid points where more than 75% of the GCMs (>10/14 GCMs) coincide in projecting the same biome with CPTEC-PVM2 for the 2070–2099 period [see Salazar et al., 2007]. In addition to that, another analysis is carried out taking into consideration that GCMs differ considerably in their ability to represent global or regional current climate accurately. Therefore, the results of using four GCMs that better simulate current climate globally and/or over tropical South America [Covey et al., 2003; Li et al., 2006; Vera et al., 2006] to drive CPTEC-PVM2 are analyzed more thoroughly: ECHAM5/MPI-OM, GISS-ER, NCAR-CCSM3 and UKMO-HadCM3. Mean seasonal climatologies of temperature and precipitation for these four GCMs under SRES-A2 in 2070–2099, as well as under the current climate data set used by CPTEC-PVM2, are shown in Figure 2 for selected areas of Amazonia (70°W–50°W, 15°S–0°N; the same area analyzed by Cox et al. [2004]) and NEB (45°W–38°W, 15°S–6°S, indicated as the area most susceptible to biome changes in NEB [Oyama and Nobre, 2003; Salazar et al., 2007; Cook and Vizy, 2008]). Projected temperature increases in both areas in UKMO-HadCM3 and ECHAM5/MPI-OM are higher in comparison to the other two GCMs (annual means (°C): Amazonia–current (1961–1990) = 25.2; future (2070–2099): GISS-ER = 28.5, NCAR-CCSM3 = 29.1, ECHAM5/MPI-OM = 30.1, UKMO-HadCM3 = 31.7; NEB – current = 24.5; future: GISS-ER = 27.5, NCAR-CCSM3 = 27.9, ECHAM5/MPI-OM = 28.5, UKMO-HadCM3 = 28.6). Precipitation varies more between the GCMs. In Amazonia, HadCM3 projects strong precipitation reduction for the whole year (4.66 mm d−1) in comparison to the 1961–1990 observed values (5.96 mm d−1). Values slightly increase during the rainy season but decrease during the dry season in ECHAM5/MPI-OM (annual mean 5.94 mm d−1). NCAR-CCSM3 and GISS-ER both project increase in precipitation, with small differences in the onset of rainy season (annual means of 6.41 and 6.59 mm d−1, respectively). Precipitation is drastically reduced in NEB's rainy season in UKMO-HadCM3, with no changes in the dry season (annual mean of 1.25 mm d−1 compared to current observed 2.16 mm d−1). NCAR-CCSM3 and GISS-ER both project small reductions for every month, which in the end result in annual means of 1.99 and 1.89 mm d−1, respectively. ECHAM5/MPI-OM presents an increase in mean annual value (2.25 mm d−1), though dry season precipitation is decreased. For further description of these climatologies, see Li et al. [2006] and Vera et al. [2006].

Figure 2.

Seasonal cycle in (a, c) temperature and (b, d) precipitation over Amazonia (70°W–50°W, 15°S–0°N) and northeast Brazil (45°W–38°W, 15°S–6°S) for observational 1961–1990 climatology (Willmott and Matsuura, data set, 1998) (black line) and for four GCMs in 2070–2099 under SRES-A2: GISS-ER (green line), NCAR-CCSM3 (yellow line), ECHAM5/MPI-OM (blue line), and UKMO-HadCM3 (red line).

[10] In order to comprehensively assess the impact of the increased atmospheric [CO2] in our simulations, we carry out six simulations with different assumptions for the CO2 fertilization effect:

[11] 1. In the “CO2-only” simulation, only the effect of increased [CO2] is considered, while climate is kept unchanged and taken as 1961–1990 average.

[12] 2. In the “climate-only” simulation, only the effect of 2070–2099 projected climate (temperature and precipitation) is considered, while [CO2] is kept at 350 ppmv. This simulation represents the low edge of the probable CO2 effect (no effect).

[13] 3. In the “climate and CO2” simulation, 2070–2099 projected climate and increased [CO2] are both considered. This simulation represents the upper range of the probable CO2 effect.

[14] 4. The “climate and CO2 on gc-only” simulation is that same as simulation 3 but CO2 fertilization effect affects only canopy conductance (gc).

[15] 5. The “climate and CO2 on GPP-only” simulation, is same as simulation 3 but CO2 fertilization effect affects only GPP (gross primary productivity).

[16] 6. The “climate and 1/2 CO2 fertilization” simulation is the same as simulation 3 but CO2 fertilization effect is halved (delta between future and current [CO2] is halved).

[17] Therefore, the results of our study are based on a range of climate scenarios and also on a range of estimates about the possible effect of CO2 on biome distribution.

3. Results

3.1. Consensus

[18] Figure 3 shows that there are no substantial changes in biome distribution in Amazonia for the end of the century if the CO2 fertilization is fully considered. Shrubland in the north part of NEB is consensually substituted by savanna, due to the dominant effect of CO2 fertilization on NPP. A small (big) area of uncertainty occurs in southeast Amazonia (central NEB) associated with uncertainties in future precipitation anomalies. Therefore, CO2 fertilization effect plays the key role on NPP in areas where consensus occurs, such as in Amazonia. In contrast, larger uncertainties on the tendency of future precipitation anomalies in NEB lead to uncertainties in the sign (increase or decrease) of NPP change in this region for the future (see Figure S10 of auxiliary materials).

Figure 3.

(a) Natural vegetation reference map [Lapola et al., 2008]; (b) potential vegetation simulated by CPTEC-PVM2 under 1961–1990 climate; (c) grid points where more than 10/14 of the GCMs coincide in projecting the same biome with CPTEC-PVM2 in 2070–2099 under SRES-A2 climate and CO2 scenario; (d) the same as Figure 3c but for climate-only (white grid points denote nonconsensus); and (e) 1961–1990 climate and atmospheric [CO2] of 730 ppmv.

[19] However, if the CO2 fertilization is inexistent and climate change takes place, there is a pronounced shift to drier biomes in Amazonia (Figure 3d). Although nonconsensus white areas cover large portions of Amazonia, it is clear that the future biome would not be tropical evergreen forest: the uncertainty there involves drier biomes like tropical seasonal forest, savanna, shrubland and even semidesert. Western Amazonia consensually turns to tropical seasonal forest. The increase in temperature alone (projected by all the 14 GCMs) is sufficient to drive the biome shift through reduced NPP. However, the identity of the replacing biome would depend then on the degree of NPP reduction caused by future precipitation changes. In NEB biome shifts are more dependent on precipitation: if it increases then shrubland remains, otherwise it is replaced by semidesert in this climate-only simulation.

[20] The CO2-only simulation shows that at 730 ppmv and current climate, the CO2 fertilization effect strongly favors the occurrence of more productive biomes like tropical forests (Figure 3e and Tables 1 and 2). Tropical evergreen forests expand on to southeastern Brazil and most of NEB's shrubland is converted to savanna. Stimulation of the Rubisco-limited photosynthesis causes a GPP increase of more than 5 times higher than the corresponding Ra increase, causing an overall increase of over 70% in NPP. In Amazonia, this year-round increased NPP results in a decrease of NPP seasonality (SNPP) by 27% while E and gc are decreased by ∼5%. In NEB E is reduced by 1.6%, though gc increased by 40%. Increasing [CO2] to 535 ppmv enhances Amazonia NPP by approximately 40%, (Figure S8 in the auxiliary materials), a result close to the 35% increase for tropical areas simulated by Hickler et al. [2008].

Table 1. CPTEC-PVM2 Outputs for a Selected Area in Amazoniaa
GCMNPPGPPRaSNPPEgcw
  • a

    NPP, net primary productivity; GPP, gross primary productivity; Ra, plant respiration (kg C m−2 a−1); SNPP, NPP seasonality (dimensionless); E, evapotranspiration (mm d−1); gc, canopy conductance (cm s−1); and w, soil water (dimensionless). Area covered is 70°W–50°W, 15°S–0°N for current and future (four GCMs) climate conditions and different assumptions regarding CO2 fertilization effect (see text). Enhancement in comparison to current climate simulation (%) is shown in parentheses.

Current Climate and CO2Simulation
 1.072.971.900.063.160.430.71
CO2-Only Simulation
 1.88 (+75.8)3.87 (+30.3)1.99 (+4.7)0.04 (−27.2)3.02 (−4.6)0.41 (−5.0)0.72 (+0.9)
Climate-Only Simulation
GISS-ER0.77 (−27.4)3.03 (+1.9)2.25 (+18.3)0.07 (+12.4)3.21 (+1.7)0.35 (−19.1)0.73 (+3.0)
NCAR-CCSM30.70 (−34.2)2.97 (+0.1)2.27 (+19.4)0.08 (+40.4)3.13 (−0.9)0.32 (−26.7)0.73 (+2.5)
ECHAM5/MPI-OM0.58 (−45.8)2.82 (−5.2)2.24 (+17.7)0.11 (+82.9)2.89 (−8.5)0.26 (−39.4)0.72 (+0.5)
UKMO-HadCM30.38 (−64.4)2.51 (−15.5)2.13 (+12.0)0.15 (+148.6)2.41 (−23.9)0.17 (−59.9)0.68 (−4.2)
Climate and CO2Simulation
GISS-ER1.7 (+58.9)3.98 (+34.0)2.28 (+20.0)0.06 (+0.3)3.32 (+5.0)0.33 (−22.9)0.72 (+1.0)
NCAR-CCSM31.61 (+50.5)3.9 (+31.3)2.29 (+20.5)0.07 (+16.7)3.29 (+4.2)0.30 (−30.2)0.71 (−0.5)
ECHAM5/MPI-OM1.45 (+35.5)3.66 (+23.2)2.21 (+16.3)0.09 (+50.0)3.22 (+1.8)0.26 (−40.5)0.67 (−5.4)
UKMO-HadCM31.19 (+11.2)3.25 (+9.4)2.06 (+8.4)0.11 (+83.3)3.05 (−3.4)0.21 (−51.9)0.60 (−15.8)
Climate and 1/2 CO2Fertilization Simulation
GISS-ER1.27 (+18.3)3.55 (+19.5)2.28 (+20.2)0.06 (−1.9)3.31 (+4.8)0.37 (−14.2)0.72 (+1.4)
NCAR-CCSM31.19 (+10.9)3.48 (+17.1)2.29 (+20.6)0.07 (+22.5)3.27 (+3.5)0.34 (−19.8)0.71 (+0.3)
ECHAM5/MPI-OM1.04 (−2.4)3.28 (+10.5)2.24 (+17.8)0.09 (+54.6)3.16 (−0.1)0.30 (−29.5)0.68 (−3.8)
UKMO-HadCM30.81 (−24.1)2.94 (−1.1)2.12 (+11.8)0.11 (+91.2)2.94 (−6.9)0.24 (−45.3)0.62 (−13.0)
Climate and CO2on GPP-Only Simulation
GISS-ER1.57 (+46.5)3.71 (+25.0)2.15 (+13.0)0.08 (+33.3)3.91 (+23.6)0.70 (+63.0)0.68 (−5.0)
NCAR-CCSM31.49 (+39.4)3.64 (+22.4)2.14 (+12.9)0.09 (+47.1)3.87 (+22.6)0.67 (+54.9)0.67 (−6.4)
ECHAM5/MPI-OM1.32 (+23.3)3.36 (+13.2)2.04 (+7.5)0.11 (+80.5)3.74 (+18.3)0.59 (+36.7)0.63 (−10.9)
UKMO-HadCM31.03 (−3.9)2.88 (−3.1)1.85 (−2.6)0.14 (+130.0)3.50 (+10.6)0.46 (+6.0)0.55 (−22.3)
Climate and CO2on gc-Only Simulation
GISS-ER0.83 (−22.0)3.19 (+7.4)2.35 (+23.9)0.04 (−25.6)2.41 (−23.7)0.18 (−58.0)0.78 (+9.7)
NCAR-CCSM30.76 (−29.0)3.14 (+5.6)2.38 (+25.1)0.06 (−1.1)2.33 (−26.2)0.16 (−61.8)0.78 (+9.0)
ECHAM5/MPI-OM0.63 (−41.2)2.99 (+0.7)2.36 (+24.3)0.09 (+48.5)2.12 (−33.0)0.14 (−68.4)0.76 (+7.3)
UKMO-HadCM30.43 (−59.5)2.72 (−8.5)2.28 (+20.2)0.13 (+111.8)1.71 (−45.9)0.09 (−78.2)0.73 (+3.0)
Table 2. CPTEC-PVM2 Outputs for a Selected Area in Northeast Brazila
GCMNPPGPPRaEgcw
  • a

    See Table 1 footnote. Area covered is 45°W–38°W, 15°S–6°S) for current and future (four GCMs) climate conditions and different assumptions regarding CO2 fertilization effect (see text). Enhancement in comparison to current climate simulation (%) is shown in parentheses.

Current Climate and CO2Simulation
 0.591.691.101.950.160.44
CO2-Only Simulation
 1.08 (+85.0)2.25 (+33.2)1.17 (+6.4)1.92 (−1.6)0.22 (+40.0)0.44 (−0.5)
Climate-Only Simulation
GISS-ER0.35 (−40.5)1.37 (−19.2)1.02 (−7.3)1.70 (−13.1)0.16 (−2.9)0.40 (−7.4)
NCAR-CCSM30.36 (−37.7)1.48 (−12.5)1.11 (+0.9)1.79 (−8.2)0.16 (+0.6)0.42 (−4.6)
ECHAM5/MPI-OM0.42 (−28.8)1.69 (−0.1)1.27 (+15.5)1.98 (+1.2)0.18 (+13.9)0.44 (+1.0)
UKMO-HadCM30.17 (−71.5)0.94 (−44.6)0.77 (−30.0)1.19 (−39.1)0.08 (−50.1)0.36 (−18.6)
Climate and CO2Simulation
GISS-ER0.75 (+28.0)1.80 (+6.5)1.05 (−4.05)1.71 (−12.5)0.07 (−57.3)0.39 (−11.0)
NCAR-CCSM30.78 (+33.1)1.89 (+11.8)1.11 (+0.9)1.81 (−7.4)0.08 (−48.9)0.40 (−9.5)
ECHAM5/MPI-OM0.91 (+55.2)2.19 (29.6)1.28 (+16.4)2.01 (+3.0)0.13 (−18.9)0.42 (−4.0)
UKMO-HadCM30.39 (−33.5)1.16 (−31.4)0.77 (−30.0)1.20 (−38.7)0.03 (−79.1)0.35 (−20.8)
Climate and 1/2 CO2Fertilization Simulation
GISS-ER0.56 (−5.0)1.59 (−6.1)1.03 (−6.4)1.71 (−12.6)0.16 (−1.1)0.40 (−9.6)
NCAR-CCSM30.58 (−0.6)1.69 (+0.2)1.11 (+0.9)1.81 (−7.6)0.16 (+1.9)0.40 (−7.6)
ECHAM5/MPI-OM0.67 (+14.8)1.96 (+15.7)1.28 (+16.4)2.00 (+2.6)0.19 (+17.3)0.43 (−2.0)
UKMO-HadCM30.28 (−51.9)1.06 (−37.3)0.78 (−29.1)1.20 (−38.8)0.08 (−49.3)0.36 (−18.3)
Climate and CO2on GPP-Only Simulation
GISS-ER0.48 (−18.4)1.20 (−29.1)0.72 (−34.5)1.78 (−8.8)0.21 (+29.8)0.34 (−21.9)
NCAR-CCSM30.50 (−14.3)1.28 (−24.3)0.78 (−29.1)1.88 (−3.7)0.22 (+35.3)0.35 (−19.5)
ECHAM5/MPI-OM0.63 (+7.0)1.55 (−8.3)0.92 (−16.4)2.12 (+8.6)0.27 (+66.8)0.37 (−15.7)
UKMO-HadCM30.19 (−66.9)0.66 (−61.1)0.46 (−58.2)1.21 (−37.9)0.09 (−43.6)0.30 (−31.7)
Climate and CO2on gc-Only Simulation
GISS-ER0.53 (−9.8)1.95 (+15.2)1.42 (+29.1)1.53 (−21.9)0.11 (−30.6)0.47 (+6.9)
NCAR-CCSM30.56 (−5.0)2.09 (+23.8)1.53 (+39.1)1.62 (−17.1)0.12 (−27.2)0.48 (+9.1)
ECHAM5/MPI-OM0.58 (−0.7)2.27 (+34.1)1.68 (+52.7)1.72 (−12.2)0.12 (−24.1)0.51 (+16.7)
UKMO-HadCM30.33 (−44.0)1.55 (−8.3)1.22 (+10.9)1.14 (−41.8)0.07 (−56.7)0.42 (−4.5)

3.2. Selected GCMs

[21] In our simulations, a dry season (months with less than 100 mm rainfall) in Amazonia with a mean length longer than 4 months (considering that the mean contemporary extension of dry season in the region is 3.45 months) would trigger the shift from tropical (evergreen or seasonal) forest to a drier biome. The identity of this drier biome (savanna or shrubland or semidesert) would then depend on the assumptions about CO2 fertilization and on the given climate scenario. In NEB a shift from shrubland to semidesert or to savanna depends even more on the magnitude of the CO2 effect and namely on the precipitation anomaly. Following is a detailed account for each of the simulations.

3.2.1. Climate and CO2

[22] All the four GCMs show an increase of NPP in Amazonia, ranging from 11% enhanced NPP with UKMO-HadCM3 to 58% with GISS-ER in the climate and CO2 simulation. Tropical evergreen forest cover in Amazonia remains nearly the same for GISS-ER and NCAR-CCSM3 and most of NEB shrubland is replaced by savanna (Figure 6b). Northeast Amazonia and most of NEB are replaced by savanna with ECHAM5/MPI-OM, after increase of SNPP (NPP) in the former (latter) region. For UKMO-HadCM3 most of Amazonia shifts to savanna, while NEB's shrubland turns to semidesert. GPP experiences a higher increase in comparison to Ra, resulting in higher NPP values in comparison to current climate simulation. However, for ECHAM5/MPI-OM and UKMO-HadCM3 in Amazonia SNPP increases by 50% and 83%, respectively, due to the increased length of the dry season projected in these GCM scenarios (Table 1 and Figure 4). For this experiment, CO2 has a dominant effect on gc in Amazonia, which is drastically reduced throughout the year for all the four GCMs (Figure 4d), ranging from 22 to 51% of reduction (yearly average). Soil water shows little change for GISS-ER and NCAR-CCSM3, even though precipitation increases. However, higher temperatures increase E throughout the year, reducing soil water availability, which in the end is not significantly different from the 1961–1990 values. For UKMO-HadCM3, soil water availability is smaller throughout the year in comparison to that modeled under current climate (Figure 4e), because of the strong decrease in precipitation for all months. NPP, GPP, and Ra are increased in NEB resulting in the replacement of shrubland by savanna (ECHAM5/MPI-OM), or in the permanence of shrubland (GISS-ER), except for UKMO-HadCM3 (Table 2 and Figure 5). Reductions in E and gc are closely linked to the magnitude of changes in precipitation.

Figure 4.

Seasonal cycle in (a) net primary productivity, (b) gross primary productivity, (c) plant respiration, (d) evapotranspiration, (e) canopy conductance, and (f) soil moisture (dimensionless) in Amazonia (see map in Figure 2) modeled by CPTEC-PVM2 (climate and CO2 simulation) for observational 1961–1990 climatology (Willmott and Matsuura, data set, 1998) (black line) and for four GCMs in 2070–2099 under SRES-A2: GISS-ER (green line), NCAR-CCSM3 (yellow line), ECHAM5/MPI-OM (blue line), and UKMO-HadCM3 (red line).

Figure 5.

Seasonal cycle in (a) net primary productivity, (b) gross primary productivity, (c) plant respiration, (d) evapotranspiration, (e) canopy conductance, and (f) soil moisture (dimensionless) in northeast Brazil (see map in Figure 2) modeled by CPTEC-PVM2 (climate and CO2 simulation) for observational 1961–1990 climatology (Willmott and Matsuura, data set, 1998) (black line) and for four GCMs in 2070–2099 under SRES-A2: GISS-ER (green line), NCAR-CCSM3 (yellow line), ECHAM5/MPI-OM (blue line), and UKMO-HadCM3 (red line).

3.2.2. Climate Only

[23] Climate-only simulations project a pronounced shift to less productive biomes (Figure 6a). In this simulation, changes in GPP are always negative and/or lower than the changes in Ra, resulting in decreased NPP for all the four GCMs scenarios for both areas. For GISS-ER and NCAR-CCSM3 scenarios, the Amazonian tropical evergreen forests give place to tropical seasonal forests (NPP is lower but SNPP increase does not surpass a 40% increase), while in NEB there is replacement of shrubland by semidesert in some grid cells. For ECHAM5/MPI-OM scenario, most of the Amazonia shifts to savanna and shrubland (lower NPP and SNPP), and NEB shrubland coverage expands. For UKMO-HadCM3 scenario, most of the Amazonia turns to shrubland and semidesert (lower NPP and SNPP), the latter also substituting NEB shrubland. Changes in E are tied to changes in precipitation projected by each of the four GCMs, though the magnitude of changes is higher in NEB. Canopy conductance in Amazonia is reduced in all four GCMs simulations. On the other hand, gc showed little variation in NEB for GISS-ER and NCAR-CCSM3 and an increase (decrease) of 13% (50%) for ECHAM5/MPI-OM (UKMO-HadCM3). Soil water shows little changes in comparison to the current climate simulation in Amazonia, with a maximum increase (decrease) of 3.0% (−4.2%) for GISS-ER (UKMO-HadCM3). On the other hand, it is remarkable the reduction in soil water in NEB for GISS-ER (−7.4%) and UKMO-HadCM3 (−18.6).

3.2.3. Climate and CO2 on gc Only

[24] Figure 6c shows that less productive biomes are favored in Amazonia in the climate and CO2 on gc-only simulations, though changes are slightly less dramatic than in climate only. In NEB, there is an expansion or permanence of shrubland. Therefore, despite the decrease in NPP, the reduction or small increase of SNPP causes the substitution of tropical evergreen forest by tropical seasonal forest in large parts of Amazonia. Because of the water savings caused by decreased gc, GPP reduction is not as severe as in the climate-only simulations. However, this water not evapotranspired stays in the soil, and increases Ra rates, which are slightly (GISS-ER) to strongly (UKMO-HadCM3) higher in comparison to climate-only simulation. This simulation yields the strongest reductions in E and gc due to the combination of strong reduction in NPP and increased atmospheric CO2.

Figure 6.

Potential vegetation simulated by CPTEC-PVM2 driven by four GCM's climatologies: (a) only 2070–2099 climate; (b) 2070–2099 climate plus 730 ppmv [CO2]; (c) 2070–2099 climate and CO2 fertilization affecting only canopy conductance; (d) 2070–2099 climate and CO2 fertilization affecting only gross primary productivity; and (e) 2070–2099 climate and CO2 fertilization is halved.

3.2.4. Climate and CO2 on GPP Only

[25] If the effects of increased CO2 are limited only to GPP, then impacts are somewhat stronger than in climate and CO2 though much weaker than in the climate-only simulations. Larger extensions of Amazonian tropical forests are substituted by savanna, starting in southeast Amazonia when compared to the climate and CO2 simulations. In NEB, there is more replacement of shrubland by semidesert when compared to the climate and CO2 case, denoting the importance of stomata-related water savings for the climate-vegetation dynamics of this region (Figure 6d). On average, NPP is 12% (43%) lower than in the climate and CO2 simulation in Amazonia (NEB), but SNPP is increased from 33% (GISS-ER) to 103% (UKMO-HadCM3) over Amazonia. These simulations show the strongest increases in E and gc over Amazonia and NEB, while soil water has the strongest decrease among the scenarios. That is due to the combination of substantial increase in NPP affecting gc, which is not affected by the direct effect of increased [CO2] in this simulation.

3.2.5. Climate and 1/2 CO2 Fertilization

[26] Halving the CO2 fertilization does not yield considerably different results for biome distribution over Amazonia in comparison to the climate and CO2 simulation (Figure 6e). In NEB there are more cells with permanence of shrubland in comparison to climate and CO2 (for UKMO-HadCM3 results are nearly the same as in climate and CO2). In this simulation NPP enhancement (in comparison to current climate simulation) does not surpass 20% in Amazonia. In NEB NPP is reduced for all the four GCMs, since now the CO2 fertilization effect is balanced with the effect of increased temperature. Although NPP is considerably lower than in the climate and CO2 simulation in both regions, SNPP is only 2% higher over Amazonia, which constrain more substantial biome shifts in this region. Changes in E, gc and soil water are also small in comparison to climate and CO2 in Amazonia. On the other hand, the low values of NPP found in NEB leads to more pronounced changes in gc with a maximum increase of 132% for GISS-ER. Effects on gc are different for both regions because of limitations of soil water (w) (i.e., w is lower than 0.5) found in NEB and not in Amazonia (see Figure 1b).

4. Discussion

[27] Main findings of our simulations are that in tropical South America:

[28] 1. If the effect of CO2 fertilization does not happen or is not sustained in tropical forests in the long term and climate change takes place there is, in all the cases, substantial shifts to drier and less productive biomes.

[29] 2. Projections are considerably ameliorated when CO2 fertilization is fully or half considered, i.e., biomes are kept the same or are substituted by wetter and more productive biomes due to the sustained long-term CO2 fertilization effect.

[30] 3. Independently of the magnitude of CO2 fertilization effect, if dry season is longer than 4 months, then Amazonian forests are replaced by drier and less productive biomes like savanna, as is the case for ECHAM5/MPI-OM and UKMO-HadCM3.

[31] Our results for the climate and CO2 simulation indicate less biome shifts than simulated by Salazar et al. [2007] and Cook and Vizy [2008]. Both studies used CPTEC-PVM1, which lacked the consideration of plant-CO2 interactions and had different biome allocation rules. In the current simulations, CO2 fertilization effect, when fully or half considered, overwhelms the impacts arising from temperature (in agreement with Lloyd and Farquhar [2008]) and even some of the precipitation changes projected by most of the GCMs, resulting in higher NPP by the end of the century. There are no considerable differences in the calculated biome distribution over Amazonia between simulations where the CO2 fertilization effect is halved and those where the effect is fully considered. NPP is somewhat lower, but seasonality (SNPP) shows almost no change. On the other hand, in NEB, halving the effect of CO2 fertilization yields considerable changes in comparison to the case that takes full account of CO2 fertilization. In the other extreme of our projections, climate-only simulations yield quite drastic projections for both Amazonia and NEB. In this simulation almost all tropical evergreen forest in Amazonia disappears, giving place to less productive biomes, like tropical seasonal forest or even semidesert, depending on the GCM. The disentanglement of the effects of increased CO2 between its influences on GPP and gc reveals that CO2 fertilization influences differently the vegetation of Amazonia and NEB. In Amazonia the stimulation of Rubisco carboxylation and consequently of GPP by increased CO2 accounts for most of the NPP increase modeled when CO2 fertilization is fully considered. Contrastingly, the effect of increased CO2 on reducing gc and increasing water use efficiency plays a key role for the increase in NPP modeled under climate and CO2 projections in NEB, where vegetation is more water limited.

[32] Therefore, the biome projections for the end of the century over tropical South America simulated here are closely tied to the effect of CO2 fertilization and also on future climate change. And, unfortunately the current knowledge on both processes for tropical South America is an issue for which there is no consensus. Following we discuss the plausibility of our results in view of the uncertainties involving the effect of CO2 fertilization, the impact of CO2 on stomatal conductance, climate scenarios and parameterization of vegetation.

4.1. Plausibility of Results: CO2 Fertilization

[33] It is long known, after a number of small-scale experiments in laboratory and field conditions (reviewed by Curtis and Wang [1998] and Norby et al. [1999]), that increased atmospheric [CO2] stimulates carbon assimilation by plants. The general mechanism behind this assimilation enhancement is summarized by Long et al. [2004, p. 596]: “Increased [CO2] increases the rate of carboxylation at Rubisco while inhibiting the oxygenation reaction and thus decreasing photorespiratory loss of carbon. Increased production allows increased leaf area development, providing positive feedback on the plant photosynthetic rate. This is further reinforced by decreased transpiration and improved leaf water status, which also favor increased leaf area growth.” Nevertheless, the magnitude of this enhancement is quite variable, depending on plant identity and age, and the experimental design, with plants grown in growing chambers presenting lower response than those grown in greenhouses or open-top chambers for example [Curtis and Wang, 1998; Long et al., 2004]. These uncertainties are also reflected in model simulations, mainly due to different parameterizations of the effect of increasing [CO2] on plant water-use efficiency and, therefore, on NPP [Friedlingstein et al., 2006; see Denman et al., 2007, Figure 7.14; Sitch et al., 2008]. Free-Air CO2 Enrichment (FACE) experiments, conducted under fully open-air field conditions, generally confirm the results of the previous small-scale experiments, showing an average 23% increase in NPP at 550 ppmv [Norby et al., 2005; Nowak et al., 2004]. However, there are no FACE experiments being carried out in tropical ecosystems. By using a global dynamic vegetation model, Hickler et al. [2008] reproduced FACE results for temperate forests and extrapolated it to the tropics, finding that the NPP enhancement (at 550 ppmv) in tropical forests (35–40%) would be ∼60% higher than the observed for temperate forests (25%). CPTEC-PVM2 results in Amazonia are in agreement with this enhancement (see Figure S8 in the auxiliary material). Therefore, while climate change would likely increase NPP more in higher latitudes due to the increase of growing season, CO2 effects on NPP would be stronger in tropical regions because of the dependence of the CO2 fertilization effect on temperature (Figures 1 and S2) [see also Hickler et al., 2008].

[34] Rising [CO2], along with increased radiation incidence, provides the most common explanation to the observed increases in recruitment and growth rates of tropical forests over the last 3 decades [Lewis et al., 2004; Lewis et al., 2009] (see also Nemani et al. [2003] for satellite estimates) and Tian et al. [2000] for modeling) (the study by Rice et al. [2004] on the Tapajós National Forest, even though for a 2-year time period and somewhat influenced by the 1998 El Niño episode, suggests that the hypothesized sequestration flux from CO2 fertilization is comparatively small in comparison to carbon losses due to heterotrophic respiration). Notwithstanding, there is no observational evidence suggesting that CO2 fertilization effect is likely to play an important role on tropical forests on longer time scales (decades to centuries). Obviously, there is a limit to the maximum size a tropical forest can attain, but little is known on how long the current observed carbon sequestration is likely to last in practice. Some evidences suggest that photosynthesis enhancement by way of Rubisco activity is constrained by the rate of utilization and export of carbohydrates produced in the leaves, the so-called limitation by carbon sink capacity [Körner, 2003; Long et al., 2004], which applicability for tropical trees is still under debate (see discussion by Lloyd and Farquhar [2008]). Others argue that nutrient availability, especially phosphorous in the Amazonian forests, would become more limiting for tropical forest growth [Hungate et al., 2003; Chambers and Silver, 2004; Powers et al., 2005]. That seems to be more valid for seedling stands, which, differently from mature forests, have limited ability to increase root-to-shoot ratio in response to higher [CO2] in order to absorb more nutrients [Curtis and Wang, 1998; Powers et al., 2005]. On the other hand, it is argued that numerous mechanisms exist that allow extra phosphorous to be taken up from the soil solution to support increased NPP in response to higher [CO2] [Lloyd et al., 2001], especially if [CO2] increase is gradual [Chambers and Silver, 2004], which is not the case for any of the FACE experiments. Additionally, there are strong variations in the response of Ra to increasing [CO2] between species: while some do not respond at all (as is the case in CPTEC-PVM2), others show a decrease of up to 60% in respiration [Norby et al., 1999; Amthor, 2000; see also Huntingford et al., 2004].

4.2. Plausibility of Results: CO2 and Stomatal Conductance

[35] Increased CO2 reduces the width of stomatal pores, which is explained as an optimization strategy of plants to maximize carbon gain per unit of water [Collatz et al., 1991]. However, this behavior is not uniform between plant species: while some (grassland species and many, but no all broadleaved deciduous species) show this response, others (especially conifers) do not [Körner, 2004]. And the magnitude of the response is also variable between those species that do respond to increased [CO2] [Ainsworth and Long, 2005]. Although there is a lack of data on this issue for tropical ecosystems, it is thought as almost certain that such interespecific differences also occur in tropical forests [Körner, 2004]. Moreover, complex biophysical feedbacks at the canopy level (difficult to capture in open-top chamber experiments) may overcome the water savings due to decreased stomatal opening. For example, higher soil moisture and lower vapor pressure deficit could increase leaf-to-atmosphere moisture gradient enhancing transpiration and reducing the consequences of the leaf level response [Körner, 2004, Körner et al., 2007].

[36] As indicated by our results, this mechanism is of potential importance for both the Amazonia and NEB regions. In NEB, most of the increased NPP comes from this relation between increased [CO2] and water savings. For example, a decrease in soil water from 0.45 to 0.4 (as is projected with GISS-ER and NCAR-CCSM3 for climate and CO2 simulation), would reduce NPP by ∼25%. However, this decrease is almost halved if [CO2] is increased from 350 to 535 ppmv (Figure 1), not considering temperature changes. On the other hand, this mechanism plays a secondary role in the NPP increase over Amazonia in our simulations. This is probably not true in reality, in view of the importance of water recycling over the Amazonian forests [Nobre et al., 1991; Eltahir and Bras, 1994; Betts et al., 2004], and could be verified by a coupled simulation between CPTEC-PVM2 and a GCM. Furthermore, changes in the export of moisture from Amazonia to other regions would certainly cause changes in plant productivity and biome distribution in other regions, including NEB [Nobre et al., 1991; Sampaio, 2008]. However, our parameterization of gc is in agreement with the results obtained in the FACE experiments. On average NPP is stimulated by 25.7% and gc is reduced by 21% with a 550 ppmv [CO2] in CPTEC-PVM2 (CO2 fertilization fully considered) in the grid cells corresponding to FACE sites ORNL, FACTS-I, FACTS-II and PopFACE (the same analyzed by Hickler et al. [2008]). FACE results show that NPP is stimulated on average by 23% and gc is reduced by 20% [Norby et al., 2005; Ainsworth and Long, 2005]. Once again, considering that FACE experiments are on average 13-year old, longer-term effects of increased CO2 on ecosystems are yet to be assessed. Therefore, in view of the current still large uncertainties associated with the long-term (>3 decades to centuries) response of tropical ecosystems to increased [CO2], one should consider the biome projections originated from different assumptions on the CO2 fertilization effect shown here as equally probable.

4.3. Plausibility of Results: Climate Scenarios and Vegetation Parameterization

[37] When analyzing extreme climate-biome scenarios (ECHAM5/MPI-OM and UKMO-HadCM3) within CPTEC-PVM2, we see that enhanced NPP alone is not sufficient to maintain Amazonian forests by the end of the century. If climate seasonality is increased, that means NPP seasonality (SNPP) is also increased and the occurrence of savanna in detriment of tropical forest is favored. This is more evident for the results using the El Niño-like future climate projected by UKMO-HadCM3, in general agreement with other climate-vegetation simulations which used the climate projections from this GCM [Scholze et al., 2006; Schaphoff et al., 2006; Sitch et al., 2008]. Studies on the sensitivity of the Amazon dieback projected by the TRIFFID DGVM showed that different parameterizations of respiration, canopy light interception, or replacement of TRIFFID by a more sophisticated vegetation model, as well as the replacement of HadCM3 control climate by observed climatology may delay, but not avoid the forest dieback under the HadCM3 projected climate [Huntingford et al., 2004, 2008]. However, the responses of other DGVMs (e.g., LPJ and ORCHIDEE) to the future climate projected by UKMO-HadCM3 are different from that of TRIFFID in tropical South America, depending mainly on the way that plant responses to water stress are formulated in the DGVMs [Sitch et al., 2008]. Specifically, the weak response (i.e., strong resilience) shown by the LPJ-DGVM in Amazonia in the study by Sitch et al. [2008] is still a subject for further discussion. That is because the parameterization of tropical plant functional types in LPJ may need refinement/readjustments as suggested by Cowling and Shin [2006], who found that only an 80% reduction in precipitation causes a considerable reduction of tropical trees in a region of Amazonia with LPJ. It is argued that the increased growth and recruitment rates observed in Amazonia [e.g., Lewis et al., 2004] might result into a long-term loss of biomass by favoring of shorter-lived climbing plants (lianas) with low woody biomass [Körner, 2003, 2004]. In fact, an increase in the density, basal area and mean size of lianas, which can enhance tree mortality and suppress tree growth, has been observed during the 1980s and 1990s in western Amazonia [Phillips et al., 2002]. Such hypothesis could be tested when “biodiversity” and/or more flexible parameterization of plants are considered in vegetation models. Furthermore, the photosynthesis model by Farquhar et al. [1980] (used by CPTEC-PVM2 as well as in all the other DGVMs [Sitch et al., 2008]) could be viewed as representing the short-term effects of climate and CO2 on photosynthesis. A long-term adaptation, if indeed possible in this time scale of only some decades, would affect for example the parameterization of photosynthetic capacity on leaf level (e.g., k21 or k24 could have higher values in equation (S12) in Text S1 in the auxiliary materials).

[38] Obviously our results for UKMO-HadCM3 are more similar to those of the TRIFFID since CPTEC-PVM2 formulation is based on that DGVM. However, consideration of NPP seasonality as the determining factor for tropical forests and savanna boundaries is the most important difference between the CPTEC-PVM2 and TRIFFID formulation (besides the first being an equilibrium model while the latter is a dynamical model). One should notice that if the allocation of tropical forest and savanna in CPTEC-PVM2 was based solely on NPP there would be no changes in the Amazonian forests even with the UKMO-HadCM3 extreme climate change scenario. But some studies suggest that the dry season length is indeed an important factor in shaping the forest-savanna boundary in tropical regions [Sternberg, 2001; Maslin, 2004; Hutyra et al., 2005], and the consideration of such feature improved the representation of tropical forest and savanna distribution in both CPTEC-PVM versions, though in CPTEC-PVM1 this is a function of soil moisture [Oyama and Nobre, 2004] while in CPTEC-PVM2 it is derived from NPP.

[39] Our result, showing that the maximum length of the dry season to sustain tropical forests is 4 months, is in very good agreement with other studies on climate-ecological thresholds. These studies show that rain forests are unable to survive a dry season longer than 4 months, whatever the rainfall is during the rest of the year [Sternberg, 2001; Maslin, 2004]. This supports CPTEC-PVM2 results with UKMO-HadCM3, which shows that irrespective of what the NPP is throughout the year, or irrespective if its annual value is higher than the current standard values for evergreen forests, if NPP is low for more than 4 months, then Amazonian forests are replaced by savanna (Figures 4 and 6). Moreover, even in UKMO-HadCM3 scenario, which projects widespread replacement of forest by savanna, mean annual precipitation of 4.66 mm d−1 was quite above the critical threshold value pointed by Lenton et al. [2008] of 3 mm d−1 for the Amazonian region. While recent short-term experiments or observations reveal a more resilient Amazonian forest than previously thought [Saleska et al., 2007; Brando et al., 2008], the recurrence of such dry spells for longer time (>30 years) could result in rather different responses by the vegetation. More frequent drought events (such as the 2005 drought in southwestern Amazonia [Marengo et al., 2008]) have the potential to further induce this savannization of Amazonia [Hutyra et al., 2005] Moreover, higher frequency and/or amplitude of El Niño episodes, which bring prolonged dry seasons to parts of Amazonia and NEB (e.g., the 1997–1998 mega-El Niño episode), must also be a matter of concern for vegetation shifts in both regions [Williamson et al., 2000]. It is remarkable that most of the differences in NPP and biome distribution between GCMs are due to different precipitation anomalies for the end of the century [see Schaphoff et al., 2006; Scholze et al., 2006]. This lack of consensus in precipitation is particularly relevant in the tropics where ecosystems are more limited by water availability and/or radiation (the latter controlled by cloudiness) [Nemani et al., 2003; Betts et al., 2004; Brando et al., 2008]. Soil water availability (w) has important and interrelated influences on GPP and Ra, as well as on the magnitude of the CO2 fertilization effect, E and gc, especially where w is lower than 0.5, as is the case for NEB in our simulations (Figures 1 and 5). Therefore, it would be more valuable if climate change risk assessments for the tropical ecosystems could be classified according to different levels of precipitation changes (as partially done here) rather than temperature classes like done by Scholze et al. [2006].

4.4. Plausibility of Results: Longer-Term Projections and Direct Human Perturbations

[40] It is also noteworthy to think on biome redistribution on longer time scales, responding to climate change beyond this century. It is likely that with continued climate change the effect of increased temperature (higher than that projected by the IPCC simulations for 21st Century) might have a stronger influence on vegetation than the CO2 fertilization effect, which is thought to stabilize after 800–900 ppmv [Lloyd and Farquhar, 2008]. One should also consider that the present study does not take into account direct human perturbations, like changes in land use and anthropogenic fire frequency [Soares-Filho et al., 2006; Sampaio et al., 2007; Malhi et al., 2008; Barlow and Peres, 2008]. Large-scale deforestation, for example, can reduce the amount of rainfall recycled within the Amazonian basin and increase dryness [Baidya Roy and Avissar, 2002]. However, it is likely that the critical deforestation threshold of 50% pointed by Sampaio [2008] as a tipping point for regional climate change would be higher if CPTEC-PVM2 was used, given the potential effect of CO2 fertilization over vegetation productivity.

5. Conclusions

[41] Biome projections for the end of the century in tropical South America are quite variable, depending not only on the climate scenario, but also on the effect of CO2 fertilization on photosynthesis and gc. Our simulations show that if, in the future, CO2 fertilization effect does not play any role in tropical ecosystems then there must be substantial biome shifts in the region, including substitution of the Amazonian forest by savanna. Otherwise, if CO2 fertilization indeed enhances NPP in the future, then impacts could be less catastrophic, while most of Amazonia would remain the same. Notwithstanding, no matter whether the CO2 fertilization effect is considered or not, if future climate has a dry season longer than 4 months (as is the case for the GCMs ECHAM5/MPI-OM and UKMO-HadCM3), then Amazonian tropical evergreen forests are substituted by drier and less productive biomes like savanna, shrubland or even semidesert. Biome changes in NEB are more dependent on the signal of future precipitation anomalies and on the magnitude of the CO2 fertilization effect on gc. In view of the uncertainties associated with long-term response of tropical ecosystems to increased [CO2], and the different biome distribution projections arising from different assumptions on the magnitude of the CO2 fertilization effect, this study points out the urgent need of long-term experimental studies on the effects of rising [CO2] on the productivity and canopy conductance of these ecosystems.

Acknowledgments

[42] This paper is part of first author's M.Sc. thesis, under the guidance of the second and third authors and under a scholarship by the São Paulo State Research Funding Agency (FAPESP) (Process 04/12235-3). We are grateful to M. H. Costa, H. R. Rocha, J. Marengo, A. Arneth, and two anonymous reviewers for useful comments and suggestions on the manuscript. We also acknowledge the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and the WCRP's Working Group on Coupled Modeling (WGCM) for their roles in making available the WCRP CMIP3 multimodel data set.