Geophysical Research Letters

Predicting land cover changes in the Amazon rainforest: An ocean-atmosphere-biosphere problem

Authors

  • Marcos Paulo Santos Pereira,

    Corresponding author
    1. National Institute for Amazonian Research, Manaus, Brazil
    2. Escritório Central do Programa LBA, Amazon State University, Manaus, Amazonas, Brazil
      Corresponding author: M. P. S. Pereira, National Institute for Amazonian Research, Av. André Araújo, 2936, Bairro Aleixo, 69060-000, Manaus, Amazonas, Brazil. (marcos@inpa.gov.br)
    Search for more papers by this author
  • Ana Cláudia Mendes Malhado,

    1. Institute of Biological Sciences and Health, Federal University of Alagoas, Prado, Brazil
    Search for more papers by this author
  • Marcos Heil Costa

    1. Department of Agricultural and Environmental Engineering, Federal University of Viçosa, Viçosa, Brazil
    Search for more papers by this author

Corresponding author: M. P. S. Pereira, National Institute for Amazonian Research, Av. André Araújo, 2936, Bairro Aleixo, 69060-000, Manaus, Amazonas, Brazil. (marcos@inpa.gov.br)

Abstract

[1] Accurate studies of the impacts of climate change on the distribution of major vegetation types are essential for developing effective conservation and land use policy. Such studies require the development of models that accurately represent the complex and interacting biophysical factors that influence regional patterns of vegetation. Here we investigate the impacts of Sea Surface Temperature (SST) on the vegetation of the Amazon, testing the hypothesis that changes in Amazonian vegetation structure are a consequence of an ocean-atmosphere-biosphere interaction. We design a numerical experiment in which we force a coupled climate-biosphere model by 10 SST patterns produced by different IPCC AR4 models, for the A2 scenario for the period 2000–2050. Simulations for 2011–2050 show that certain patterns of SST are likely to decrease the ensemble for tropical evergreen rainforest and savanna, and that these areas will be occupied mainly by tropical deciduous rainforest, emitting an average of 0.53 Pg-C.yr−1 during the transition.

1. Introduction

[2] One of the most important challenges in environmental science is to understand and predict the influence of climate change on the major ecosystems of the world. The focus of much of this research (reviewed by Malhi et al. [2008]) has been the relatively pristine tropical forests of the Amazon basin in South America - the largest remaining area of continuous rainforest in the world and a vital component in maintaining global ecosystem services [Costanza et al., 1997] such as hydrological cycles that have potential impacts on the regional and global climate system [Chahine, 1992; Cox et al., 2002; Werth and Avissar, 2002].

[3] However, despite decades of empirical research and considerable advances in modeling there is still uncertainty regarding the fate of the Amazon under global warming. For example, it has been proposed that the Amazon rainforest may experience a dieback during the 21st century as a consequence of anthropogenically induced global warming [Cox et al., 2000]. Specifically, Cox and his colleagues predicted that forests of Amazonia would begin to be lost due to a drying and warming of the atmosphere based on the results of a fully coupled, three-dimensional carbon-climate model. However, the authors accept that the forest loss scenario rests upon “uncertain aspects of regional climate change, and may be ‘short-circuited’ by direct human deforestation” [Cox et al., 2000, p. 186]. Huntingford et al. [2008] also predict die back based on the results of perturbed physics simulations covering a wide range of climate sensitivity. The effects of forest die back could, in turn, release more CO2 into the atmosphere leading to further warming [Betts et al., 2004].

[4] By contrast, some studies [e.g., Levis et al., 2000] have found that rising CO2 would not significantly modify land cover in Amazonia. Alo and Wang [2008] examined the geographical responses of natural vegetation to future changes in atmospheric CO2 concentration and climate at a global level. To achieve this they used an offline version of the National Center for Atmospheric Research Community Land Model's (NCAR CLM) dynamic global vegetation model, forced by the climate simulated by eight general circulation models. The eight simulations showed considerable uncertainty, especially with respect to the coverage of evergreen trees in South America with some models predicting dieback while others predicted an increase in forest cover in some regions. One of the key variables generating much of the uncertainty is precipitation, the prediction of which has remained problematic. As Malhi et al. [2009] point out, much of the discussion about the possibility of Amazon dieback has been based on either a single model, perturbed physics ensembles of a single model, or the uncritical examination of the results of a number of models. To remedy this situation Malhi et al. [2009]assessed the output of 19 IPCC AR4 Global Climate Models under the medium-high range A2 Emissions Scenarios. The results of the models varied considerably (a few studies suggested an increase in precipitation in Amazonia during 21st century, others suggested a decrease, and the model mean was close to zero). However, in this case there was reasonable evidence to suggest that changes in the rainfall regime in East Amazonia may cause a transition to seasonal forest (deciduous rainforest) in this area [Malhi et al., 2009].

[5] The huge variation in the forecasts for precipitation in the Amazon region under global warming strongly suggests that the models are failing to capture an important component of the hydrological cycle. A strong candidate for this missing factor is sea surface temperature (SST), patterns of which in the Pacific and Atlantic oceans are known to strongly influence the South American climate [Nobre and Shukla, 1996; Diaz et al., 1998; Grimm et al., 2000; Haylock et al., 2006]. Although SST has been shown to strongly influence the predictions of coupled climate-vegetation models [Jiang et al., 2011], at a global scale, its potential influence on the future climate-vegetation coupled system in Amazonia has not been addressed.

[6] Here we hypothesize that the vegetation in Amazonia in the 21st century will be a consequence of an ocean-atmosphere-biosphere interaction with the objective of producing more accurate regional forecasts of precipitation under global warming. To test this hypothesis, ideally, we should use several different coupled dynamic atmosphere-ocean-vegetation models, a task beyond the capabilities of most world laboratories today. Instead, we design a much simpler numerical experiment in which we force a coupled climate-biosphere model by 10 SST patterns produced by different IPCC AR4 models, for the A2 scenario for the period 2000–2050. While this design does not capture the full ocean-atmosphere-biosphere interactions, it captures all the bi-directional interactions between atmosphere and biosphere, and the one-way interaction between ocean and atmosphere, expressed by the several SST scenarios. Finally, we analyze the results for the period 2011–2050 in order to demonstrate the relationship between land cover (tropical evergreen rainforest, tropical deciduous rainforest and savanna) and specific patterns of SST.

2. Model Description and Experiment Design

[7] In this work, we performed simulations using the CCM3 (Community Climate Model) general circulation model coupled to the surface model IBIS (Integrated Biosphere Simulator); a combination known as CCM3-IBIS and commonly used in simulation studies [e.g.,Senna et al., 2009]. Detailed information on the models is documented in Kiehl et al. [1998] and Kucharik et al. [2000], and will not be repeated here. The simulations were performed at T42 resolution (2.81° × 2.81°) and 18 levels in the vertical in the atmosphere and used a hybrid sigma-pressure coordinate system with a 15 minute time step interval. CCM3-IBIS incorporates vegetation dynamics scheme and parameterized physics for the Amazon rainforest region in order to reproduce bi-directional interactions between vegetation and climate [Foley et al., 2000; Senna et al., 2009] making it a useful tool for studying biome distribution, ecosystem functioning, physiological response to rising CO2 and climate feedbacks in face of global climate change and land use change.

[8] The performance of the CCM3-IBIS model in simulating the regional climate for the Amazon rainforest is described in detail bySenna et al. [2009]. They reported that annual mean precipitation is within 10% of five datasets precipitation mean, and precipitation seasonality is also robustly simulated.

[9] CCM3-IBIS was run for the period 1951–2050, to investigate the influence of variability in SST on vegetation cover in the Amazon rainforest through ten experimental simulations. SST data was obtained from PCMDI (Program for Climate Model Diagnosis and Inter-comparison). The 20th century period (1951–2000) was used to spin up the model. Ten representative outputs of the coupled climate models were chosen out of those used the 21st century SST under the A2 emission scenario by the IPCC in AR4 on the basis of spatial resolution and including a representative sample of meteorological research centers: CCCMA CGM3.1 T47, CNRM CM3, CSIRO MK3.0, GFDL CM2.1, GISS MODEL E_R, IPSL CM4, MIROC3.2 MEDRES, MPI ECHAM5, NCAR CCSM3.0 and UKMO HadCM3. More information about the models can be found at www-pcmdi.llnl.gov/ipcc. In addition to SST, all simulations were forced by atmospheric concentrations of CO2 and CH4 in accordance with the IPCC A2 scenario. Each simulation was spun up with dynamic vegetation and three repetitions (ensembles) per treatment, with runs starting on 17–19 January 2001, to obtain a more realistic representation of vegetation and of climate situations for the same time period. The total experiment included 1650 years of simulation.

3. Results and Discussion

[10] There is a clear pattern of change and replacement of major ecosystems in Amazonia in the first half of the 21st century in response to increased temperature and CO2 (Figure 1). Specifically, the annual net ecosystem carbon exchange (NEE) of the Amazon rainforest shows positive values in terms of carbon released to the atmosphere (Figure 1a), with an annual mean of about 79 g-C.m−2.year−1in the period 2011–2050, or about 0.53 Pg-C.year−1for the study area. The probability of decadal vegetation occurrence generated by CCM3-IBIS simulation ensembles reveals a clear temporal trend of decreasing tropical evergreen rainforest and savanna and increasing tropical deciduous rainforest (Figure 1). This result suggests that increases of tropical deciduous rainforest are the main feature of the changes in vegetation composition under future global warming in the Amazon rainforest – this is in concordance with other regional modeling studies in the region [Malhi et al., 2009; Jiang et al., 2011]. Such increases in NEE levels (Figure 1) are a result of foliage loss to conserve water and prevent senescence during prolonged dry periods [Bullock and Solis-Magallanes, 1990]. Thus, one important consequence of global warming in Amazonia could be to turn the “sinks” into sources.

Figure 1.

(a) Decadal mean of NEE (Pg-C.year−1) in Amazonia (11°N to 12°S by 49°W to 80°W). (b) Probability of decadal vegetation occurrence in Amazonia as predicted by the CCM3-IBIS simulation ensembles (3 ensembles × 10 SSTs patterns × 10 years). Vegetation types = tropical evergreen rainforest (green), tropical deciduous rainforest (blue) and savanna (red). Shaded area indicates the uncertainty range (see text for explanation).

[11] The rate of transition decreases in the second half of the study period and the uncertainties increase (represented by the internal variability amplitudes). The increase in uncertainty is associated with the natural variability of internal processes within the climate system in the model and variability in the ten future patterns of SST (Figure 1). The range of vegetation modeling uncertainty is expected to be smaller than the range of uncertainty found in climate predictions from different GCMs, as vegetation response depends mainly on the combination of temperature and precipitation changes and such combination tends to be the same for the same GCM. However, this study underestimates the uncertainty in vegetation response due to the use of a single model.

[12] Present-day climate conditions in Amazonia favor the occupation of the dominant tropical evergreen rainforest in almost 60% of the study area (11°N to 12°S by 49°W to 80°W), while savanna takes up approximately 35% and tropical deciduous rainforest occupies less than 5%. Global warming is predicted to affect vegetation in the next 40 years by reducing the cover of tropical evergreen rainforest by around 8% (approximately 50% of the total land cover) and savanna by around 7% (approximately 30% of the total land cover).Salazar et al. [2007] estimated a 9% reduction in Amazon rainforest cover during the period 2050–2059 – a value similar to that reported here. However, Salazar's study predicted replacements of rainforest by savanna vegetation while the present study supports the hypothesis that increasing CO2 and changes in variability of rainfall patterns that will most likely expose region to changes in frequencies, magnitude and durations of droughts, will stimulate the growth of tropical deciduous rainforest which is predicted to occupy the space lost by the other two vegetation types: Deciduous rainforest is predicted to occupy approximately 18% of the total Amazonian land cover by 2050 (Figure 1). Regions of transitional vegetation between savanna and tropical evergreen rainforest are predicted to become tropical deciduous rainforest.

[13] We then select the ensembles with the highest (top 10%) levels of tropical evergreen rainforest, tropical deciduous rainforest and savanna and calculate the mean SST pattern associated with them (Figures 2a–2c) and the difference among them (Figures 2d–2f). The results consist of an analysis of 1200 years (3 ensembles × 10 SST × 40 years). This analysis suggests that specific patterns of SST can be identified that favor (through their influence on Amazonian precipitation) each of the three ecosystems. Thus, a pattern of SST of below 29°C reaching temperatures of 23°C in the Equatorial Pacific produces precipitation patterns that promote tropical evergreen rainforest coverage (Figure 2a).

Figure 2.

SST characteristics that produce climate conditions that favor certain vegetation types: (a) tropical evergreen rainforest, (b) tropical deciduous rainforest and (c) savanna in Amazon rainforest and (d–f) the differences between them as labeled. Only areas significant at the 95% level of a Student's t-test are illustrated.

[14] Tropical deciduous rainforest is favored when there are SST conditions of around 29°C in the Equatorial Pacific and SST above 28°C across the tropical North Atlantic region (i.e. characteristic of the Atlantic Multidecadal Oscillation, AMO positive phase) (Figure 2b). Savanna is associated with SST patterns in Equatorial Pacific with regions of around 29°C (i.e. characteristic of the El Niño event), and across Tropical Atlantic regions with SST above 27°C (Figure 2c). The difference between SST patterns for the tropical deciduous rainforest and tropical evergreen rainforest shows positive anomalies in the North Atlantic and North Pacific Equatorial Oceans, and negative anomalies in the South Pacific (Figure 2d). The difference between SST patterns associated with savanna and tropical evergreen rainforest ensembles shows positive anomalies in the Equatorial Pacific region and negative anomalies in the South Atlantic and South Pacific equatorial Oceans (Figure 2e). Furthermore, SST difference patterns between savanna and tropical deciduous rainforest produce positive anomalies in the Equatorial Pacific, and negative anomalies in the North Atlantic and North Pacific Equatorial Oceans and the South Atlantic and South Pacific Equatorial Oceans (Figure 2f).

[15] The relationship between SST and transitional vegetation in the Amazon rainforest under global warming scenarios can be best addressed through a statistical analysis of the patterns of variability. This spatial variation in SST is based on a series of bi-dimensional correlations between 10 future patterns of SST and the probability of occurrence of Amazonian vegetation types (tropical evergreen rainforest, tropical deciduous rainforest and savanna) estimated from different SST scenarios for 2011–2050. These data are analyzed for 1200 years (3 ensembles × 10 SST patterns × 40 years) (Figure 3). Previous evaluations of the modeled interannual variability (not shown here), indicate that this climate model simulates well the Amazonia rainfall patterns associated with warm/cold SST anomalies in central/eastern Pacific and northern/southern tropical Atlantic, so we are confident that observed tropical ocean-atmosphere interaction are well reproduced in our simulated results.

Figure 3.

Spatial variation in SST based on a series of bi-dimensional correlations between 10 future patterns of SST and probability of occurrence of Amazonian vegetation types: (a) tropical evergreen rainforest, (b) tropical deciduous rainforest and (c) savanna estimated from different SST scenarios for 2011–2050 (n = 1200: 3 ensembles × 10 SSTs patterns × 40 years). It should be noted that these figures only represent significant values at the 0.01 level that show a positive or negative correlation of greater than 0.08.

[16] It has been well documented that anomalous warming or cooling of SST over the surrounding Pacific and Atlantic basins plays an important role in changes in precipitation in the Amazon region [e.g., Ronchail et al., 2002; Marengo, 1992; Yoon and Zeng, 2010]. For example, an anomalously southward migration of the Intertropical Convergence Zone (ITCZ) during May–June 2009, due to the warmer than normal surface waters in the tropical South Atlantic, was responsible for abundant rainfall in large regions of eastern Amazonia from May to July 2009 [Marengo et al., 2011a]. The model demonstrates that tropical South Atlantic SSTs are positively correlated with rainfall in tropical evergreen rainforest and tropical deciduous (Figures 3a and 3b), and are negatively correlated with precipitation in the savanna region (Figure 3c). This means that the development of tropical evergreen rainforest and tropical deciduous rainforest over the Amazon region is associated with anomalously warm SSTs in the tropical South Atlantic, and that the development of the savanna is associated with anomalously cold SSTs.

[17] A warming of the tropical Atlantic in the north relative to the south leads to a northwestwards shift in the ITCZ and compensating atmospheric descent over Amazonia [Fu et al., 2001]. Tropical North Atlantic SSTs exert a large influence on dry season rainfall in western Amazonia. For example, the Amazonian drought in 2005 was associated with the persistent warm anomaly in the tropical North Atlantic that caused a delayed onset of the South American monsoon [Cox et al., 2008]. Our analysis of the probabilities of annual occurrence of Amazonian vegetation types shows that North Atlantic SSTs are negatively correlated with tropical evergreen rainforest and savanna (Figures 3a and 3c), and positively correlated with tropical deciduous rainforest (Figure 3b).

[18] Simulations for the 21st century show a strong tendency for the SST conditions associated with the 2005 drought to become much more common. In other words, the North Atlantic region will warm more rapidly than the South Atlantic leading to a northwards movement of the ITCZ and suppression of July–October rainfall in western Amazonia [Cox et al., 2008]. If this happens, climatic conditions will produce long dry seasons of the characteristics associated with the tropical deciduous rainforest environment. In the CCM3-IBIS model, drought resistant deciduous plants (including tropical deciduous trees, grasses) are assumed to respond to changes in the net canopy carbon budget. The mean (using a 10-day running average) net photosynthesis rate calculated by IBIS becomes negative (indicating that leaf respiration now exceeds gross photosynthesis), which causes the model to drop canopy leaves [Kucharik et al., 2000].

[19] In 2010, the drought over the Amazon rainforest started in early austral summer during El Niño and subsequently intensified as a consequence of the warming of the tropical North Atlantic [Marengo et al., 2011b]. Xu et al. [2011]showed the warming of SST along the tropical North Atlantic and equatorial of the Pacific during the 2010 drought, as measured by rainfall deficit, affected an area 1.65 times larger than the 2005 drought. Moreover, the decline in greenness during the 2010 drought spanned an area that was four times greater and more severe than in 2005. Notably, 51% of all drought-stricken forests showed greenness declines in 2010 compared to only 14% in 2005. Overall, the widespread loss of photosynthetic capacity of Amazonian vegetation due to the 2010 drought may represent a significant perturbation to the global carbon cycle [Xu et al., 2011].

[20] Previous studies have shown that rainfall patterns over the northern part of the Amazon change significantly depending on the phase of El Niño-Southern Oscillation (ENSO) [Kousky et al., 1984; Ropelewski and Halpert, 1987]. By extension, ENSO also plays an important role in the vegetation types in the Amazon rainforest. The probabilities of annual occurrence of tropical evergreen rainforest over the Amazon rainforest region are positively correlated with SST in the North Subtropical and Equatorial Pacific regions (proportional to the PDO and ENSO face) (Figure 3a). However, the probability of annual occurrences of the savanna and tropical deciduous rainforest are negatively correlated with SST anomalies in the Equatorial Pacific (Figures 3b and 3c).

4. Conclusions

[21] Forecasting of vegetation cover under climate change scenarios at a regional scale is important for effective land use planning, adaptation and mitigation measures. The results of this study using the CCM3-IBIS coupled model clearly demonstrate the significant influence of SST on precipitation patterns and, therefore, the dynamics of Amazonian vegetation over time. Thus, the incorporation of the SST patterns in the coupled climate-vegetation models is clearly important for improving projections of future vegetation cover in this region. Moreover, further improvements can be achieved by the use of an ensemble of fully coupled ocean-atmosphere-biosphere models.

[22] The probabilities of occurrence of the vegetation types in the Amazon rainforest is linked to anomalous warming or cooling of SST over the surrounding ocean basins of the Pacific and Atlantic. SSTs are especially important in the tropics because the atmosphere is sensitive to the oceanic and continental surface conditions, which greatly influence the variability of the climate. Our quantification of the potential changes in vegetation under global warming in first half of 21st century predicts a suite of changes that favor a decrease in both tropical evergreen rainforest and savanna vegetation and an increase in tropical deciduous rainforest that leads to increasing NEE levels, with an average loss of carbon on the order of 0.53 Pg-C.year−1. Under this scenario atmospheric CO2 increases thereby contributing to additional global warming.

Acknowledgments

[23] We thank Amazonas and Minas Gerais State Research Funding Agencies (FAPEAM, FAPEMIG) for financial support. We also thank Richard J. Ladle for reviewing the English and commenting on the manuscript.

[24] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.