Previous Article |
Table of Contents
| Next Article 
Ecology
Carbon emissions from tropical deforestation and regrowth based on satellite observations for the 1980s and 1990s
*Department of Geography and Earth System Science
Interdisciplinary Center, University of Maryland, College Park, MD
20742; Woods Hole Research Center, P.O. Box 296, Woods
Hole, MA 02543; ¶Department of Global Ecology, Carnegie
Institution of Washington, Stanford, CA 94305-1297; and||
Department of Geography, Michigan State University, East
Lansing, MI 48824
Contributed by Christopher B. Field and approved September 13, 2002
 |
Abstract |
|---|
 Top Abstract Methods and ResultsDiscussion and ConclusionsReferences |
|---|
Carbon fluxes from tropical deforestation and regrowth are
highly uncertain components of the contemporary carbon budget,
due in part to the lack of spatially explicit and consistent
information on changes in forest area. We estimate fluxes for the 1980s
and 1990s using subpixel estimates of percent tree cover derived from
coarse (National Oceanic and Atmospheric Administration's Advanced
Very High Resolution Radiometer) satellite data in combination with a
terrestrial carbon model. The satellite-derived estimates of change in
forest area are lower than national reports and remote-sensing surveys
from the United Nations Food and Agriculture Organization Forest
Resource Assessment (FRA) in all tropical regions, especially for the
1980s. However, our results indicate that the net rate of tropical
forest clearing increased 
10% from the 1980s to 1990s, most notably
in southeast Asia, in contrast to an 11% reduction reported by the
FRA. We estimate net mean annual carbon fluxes from tropical
deforestation and regrowth to average 0.6 (0.3–0.8) and 0.9 (0.5–1.4)
petagrams (Pg)•yr–1 for the 1980s and 1990s,
respectively. Compared with previous estimates of 1.9 (0.6–2.5)
Pg•yr–1 based on FRA national statistics of changes
in forest area, this alternative estimate suggests less "missing"
carbon from the global carbon budget but increasing emissions from
tropical land-use change.
Abbreviations: Pg, petagram(s); FAO, United Nations Food and Agriculture Organization; FRA, Forest Resource Assessment; IPCC, Intergovernmental Panel on Climate Change; AVHRR, National Oceanic and Atmospheric Administration's Advanced Very High Resolution Radiometer; PTC, percent tree cover; PTCA, PTC-corrected area
Future atmospheric carbon-dioxide concentrations and consequent climate change depend to a large extent on the future course of the terrestrial uptake (9). If the underlying mechanisms are no longer able to sequester carbon at some point in the future, as for example would be the case once regrowing forests mature, a larger proportion of emitted carbon dioxide would remain in the atmosphere, and carbon-dioxide concentrations would increase at a greater rate for the same level of emissions.
Atmospheric inversion studies, which calculate net sources and sinks of carbon dioxide from the spatial distribution of atmospheric concentrations, indicate a net land sink of 0.6–2.3 petagrams (Pg)•yr–1 in the extra tropics (6). In the tropics, inverse models are poorly constrained but indicate that the region, overall, is neutral or a small source of carbon to the atmosphere (10). Although inversion studies locate and quantify the net terrestrial sources or sinks, the attribution to mechanisms and their possible future trajectories depend on quantifying the gross sources and sinks. For a net sink, the mechanisms responsible for uptake of carbon dioxide must be powerful enough to offset the sources from fossil fuel and deforestation. The carbon dioxide emitted from fossil-fuel combustion is well quantified (11), but the emission from tropical land-use change is highly uncertain. Without more precise estimates of this source term, deciphering possible mechanisms sequestering the missing carbon remains problematic.
The flux of carbon to the atmosphere from tropical land-use change is one of the largest uncertainties in the contemporary carbon budget (6, 12) because of the difficulties in quantifying deforestation and regrowth rates, initial biomass, and fate of carbon in areas where vegetation has been cleared. Estimates of carbon fluxes from tropical deforestation as reported by the Intergovernmental Panel on Climate Change (IPCC; ref. 12) from refs. 5 and 13 range from 0.6 to 2.5 Pg•yr–1 for the 1980s, based primarily on calculations using cropland statistics from the United Nations Food and Agriculture Organization (FAO) and deforestation rates from the FAO Forest Resource Assessment (FRA).
The FRA information is obtained through national reporting supplemented by limited satellite analysis in the assessment for the 1990s (14–16). Participation of individual countries through national reporting is a strength from some perspectives, but it generates problems from varying definitions of forest cover among countries and time intervals (17). These problems are particularly acute in developing countries, where most tropical deforestation occurs.
Comparisons of national statistics from the FRA with other country-level analyses suggest that the FRA overestimated changes in forest cover in some African countries (18), Bolivia (19), and other developing countries (20, 21). For the 1990–2000 interval, the FRA also conducted a remote-sensing survey, analyzing 10% of all tropical land area (15, 21). Forest area and deforestation rates from the FRA remote-sensing survey are generally lower than the FRA (15, 22) country reports for the 1990–2000 interval for Latin America and tropical Asia, although the differences are not statistically significant. For tropical Africa, the difference is very large (3 million ha/yr), suggesting exaggerated deforestation rates in the country data (15). For the 1980–1990 interval, on which the IPCC estimates of carbon fluxes from tropical deforestation are based, the country reports are the sole source of information for the FRA analysis.
Satellite data offer the possibility of spatially and temporally
consistent estimates of forest cover to complement national reports.
Data acquired by the Landsat platform, with a pixel resolution of 30
m for the thematic mapper sensor and 60 m for the multispectral
scanner sensor before the early 1980s, have provided estimates of
deforestation rates for individual regions such as the Amazon basin
(23). However, because of cloud coverage and limited acquisitions over
the past several decades, it has not been possible to obtain
comprehensive coverage for the entire tropics. Global data from the
early 1980s to present acquired by the National Oceanic and Atmospheric
Administration's Advanced Very High Resolution Radiometer (AVHRR)
provide daily coverage but at a coarse spatial resolution of 8 km (24).
AVHRR data at the sensor resolution of 1 km are not available for
the full time series with adequate spatial coverage. In this study we
estimate changes in forest area by using an approach to estimate
subpixel changes in tree cover within the coarse spatial resolution of
the AVHRR data. This analysis thus provides a spatially explicit
alternative to the FAO's nationally reported changes in forest area
and an alternative estimate for carbon fluxes over the past two
decades.
|
Methods and Results |
|---|
|
TopAbstractMethods and ResultsDiscussion and ConclusionsReferences |
|---|
PTC Estimates. Based on the Pathfinder AVHRR Land data set (24, 33), we estimate PTC within each pixel (27, 29) for each year in the 1982–2000 record. The method uses regression tree analysis to estimate subpixel PTC individually for each year in the time series. Inputs to the regression tree are multitemporal metrics derived from monthly values of the five AVHRR spectral bands ranging from visible to thermal wavelengths (0.58–12.5 µm) and the normalized difference vegetation index (NDVI) calculated from the red and near-infrared bands. Monthly composites are generated from the date with maximum NDVI to reduce cloud contamination. Metrics characterize the vegetation's spectral reflectance and phenology and include annual mean, maximum, minimum, and amplitude for each of the five AVHRR bands. We grow the regression tree using a global network of training sites derived from over 200 Landsat scenes and aggregated to the 8-km resolution of the AVHRR data (27, 29). For each year in the AVHRR time series, we grow and apply a regression tree using the same training data (61,222 8-km pixels).
Because the algorithm is applied independently to each year based on training data derived from over 200 Landsat scenes, the estimate is relatively insensitive to sensor degradation and other calibration problems in the AVHRR record (34). However, misregistration between years and spurious data in any single year generates noise that complicates the interpretation of year-to-year differences in PTC. To minimize the noise, we derived estimates of PTC for three 5-yr intervals (1982–87, 1988–92, and 1992–99) by using the median value for the interval to represent PTC for the time interval. The standard error (standard deviation of residuals) for the three 5-yr intervals as compared with the training data is 11.03%.
We label a grid cell as "change" if the difference in PTC between
time periods exceeds a threshold value. The threshold value, 14%,
corresponds to 2 standard deviations from the mean difference in PTC
between time periods for the training-site locations. The training
sites were selected in locations with no change based on expert
knowledge of the locations. It is possible, of course, that a small
percentage of the training sites have experienced change. Differences
in PTC for training pixels are assumed to represent noise if they are
less than the 2 standard-deviation threshold. Because the changes in
PTC are converted to changes in forest area (PTCA) and carbon (PTC
carbon) by using corrections based on comparison with high-resolution
data (see below), the estimates of deforestation rates do not depend on
the precise selection of the threshold value.
Changes in PTC between intervals indicate extensive forest loss in the well known arc of deforestation in the Amazon basin and in southeast Asia (Fig. 1). Both decreases and increases in tree cover were observed in Africa in a patchy distribution (data not shown).
|
|
1.5 (Table 1). In
Africa, where clearing typically occurs in a more patchy distribution
for small-scale agriculture, the only available correction factor is
much larger: 3.7 (Table 1). In the absence of wall-to-wall Landsat analysis with which to develop correction factors for each country, we consider a range for each continent. For Latin America, with relatively extensive Landsat-based analyses, the central estimate is the mean of the three sources listed in Table 1, with the low and high estimates from Bolivia and Instituto Nacional de Pesquisas Espacias (INPE), respectively. For southeast Asia, where Landsat-based analyses are much more limited, we use a central estimate that is the mean for all Latin American and Asian data (Table 1). The low estimate is the sum for the four Indonesian islands, and the high estimate is from Sumatra, where the most extensive deforestation has occurred. For Africa, Landsat-based analyses cover only a small fraction of the forest area. In the absence of better constraints, our central estimate is the mean based on all Landsat-based studies in Table 1. The high end of the range is estimated by comparison with the sparse Landsat analyses carried out over the Democratic Republic of Congo (http://glcf.umiacs.umd.edu), and the low end of the range is based on the correction factor for Bolivia. The FAO FRA remote-sensing survey, a 10% sample taken over each region, was not used to adjust the change in PTC estimates. Rather, it is used to assess the results of estimated deforestation and regrowth rates derived by correcting the PTC estimates with the other higher resolution analyses.
For the 1990s, the net change in forest area derived from AVHRR-based analysis corrected with high-resolution analyses (PCTA) is modestly lower than the FRA remote-sensing survey for tropical Latin America (24.3%) and tropical Asia (12.7%) (Table 2). For tropical Africa, however, the estimates are more than 80% lower. Some of this difference may be due to the limited availability of Landsat-based studies for developing appropriate correction factors, especially for sites where change occurs in very small patches across the landscape. The estimates for the 1990s from PTCA are also substantially lower than the FRA country reports for the 1990s for tropical Latin America (27.8%) and tropical Asia (16.3%), and they are dramatically lower for tropical Africa (93%). For the 1990s, total tropical net change in forest area from the central estimate of the PTCA analysis is 5.563 x 106 ha•yr–1, 35.3% less than the 8.600 x 106 ha•yr–1 from the FRA remote-sensing survey, and 53.6% less than the 12.000 x 106 ha•yr–1 from the FRA country data.
|
Because the PTCA analysis applies the same methodology over the length of the AVHRR record, it provides an alternative to the FRA data for assessing trends in deforestation and regrowth rates from the 1980s to 1990s. The PTCA and FRA country data indicate contrasting trends. PTCA indicates a 10.3% increase in the rate of net forest loss (accelerating forest loss) from the 1980s to the 1990s, whereas the FRA country data indicate a 10.9% decrease in the rate (decelerating loss).
Carbon Fluxes from Tropical Deforestation and Regrowth. The PTCA approach provides a means to estimate carbon fluxes in the 1980s and 1990s independent of the FRA country statistics. We calculate the carbon fluxes from deforestation and regrowth using a bookkeeping model (13, 31, 32) applied to each 8-km grid cell. The bookkeeping model accounts for forest clearing and regrowth by tracking (i) the immediate release of carbon to the atmosphere from plant material burned at the time of clearing, (ii) slower release of carbon from decay of slash, (iii) accumulation of carbon during regrowth, and (iv) changes in soil carbon. The fluxes are calculated on the basis of areas of clearing and regrowth, initial biomass values specified in the model, decay rates of dead plant material, and carbon uptake rates by regrowing vegetation. The rates of clearing and regrowth were based on the estimates of changes in subpixel tree cover.
Initial biomass values were set by using the values of Houghton and Hackler (30), with the forest type from a 1-km global land-cover classification (37) aggregated to 8 km according to the dominant vegetation type. For undisturbed forest, initial biomass values specified in the model were reduced linearly in proportion to the PTC in the grid cell. To test the sensitivity to assumptions about initial biomass, we ran the model with the initial biomass ±25%. Error bars represent the extreme values using the range of correction factors and the sensitivities to initial biomass assumptions. The high extreme values represent the largest correction factor applied to the high biomass assumption and the low extreme values represent the lowest correction factor applied to the low biomass assumption.
The total carbon-flux estimate includes, in addition to the fluxes from our estimated changes in PTC, fluxes from land-use change that are not likely observable by even high-resolution satellite data. Nepstad et al. (38) estimate that carbon fluxes from "cryptic" logging activities not detectable with Landsat data account for 4–7% of carbon fluxes from deforestation in Amazonia. Houghton (39) estimates that these cryptic fluxes from logging and shifting cultivation total 0.041 Pg•yr–1 for the 1990s (mostly occurring in Asia). Although the total fluxes from these processes remain uncertain, we add 7% to our estimates to crudely account for these sources.
To test whether the implementation of our bookkeeping model provides carbon-flux estimates in line with other published estimates, we first consider the best studied tropical country, Brazil. We estimate the net mean annual carbon flux for Brazil to be 0.15 (0.085–0.29) Pg•yr–1 in the 1980s and 0.28 (0.17–0.49) Pg•yr–1 in the 1990s. Most of the carbon flux is attributable to burning and decay of vegetation and slash, with only a small uptake from regrowth, which is in agreement with Houghton et al. (32). These estimates from PTC carbon analysis are generally within the range of previous estimates: 0.18 Pg•yr–1 (32) mean net flux for 1989–99 and 0.26 Pg•yr–1 (40), in the Brazilian Amazon only, and 0.174–0.336 Pg•yr–1 for the entire land area of Brazil (13, 41, 42).
We apply the model to estimate carbon fluxes from deforestation and regrowth throughout the tropics. Initial biomass values and decay and uptake rates are identical to other estimates using the same model (13, 39) such that differences in carbon fluxes are attributable to our alternate estimates of areas undergoing deforestation and regrowth. We estimate that net carbon fluxes from tropical deforestation and regrowth in the 1980s and 1990s are 0.6 (0.3–0.8) and 0.9 (0.5–1.4) Pg•yr–1, respectively (Fig. 2 and Table 3). The largest flux occurs in Latin America, although emissions increased most rapidly between the 1980s and 1990s in tropical Asia. Relative to the 1980s, PTCA estimates for the 1990s indicate lower rates of forest loss in Latin America and Africa and higher rates in Asia. However, the carbon-flux estimates suggest increasing emissions to the atmosphere in all continents. The increased fluxes can be attributed to (i) increased clearing in higher biomass forests in the 1990s relative to the 1980s (Fig. 1) and (ii) decreased areas reforested in the 1990s relative to the 1980s (Tables 2 and 3). These factors illustrate the importance of spatial information about the location of clearing and regrowth, not available with national level statistics, to reduce uncertainties about carbon fluxes from land-use change.
|
|
|
Discussion and Conclusions |
|---|
|
TopAbstractMethods and ResultsDiscussion and ConclusionsReferences |
|---|
The result is consistent with other studies suggesting that the
national statistics may be overestimating changes in forest cover in
the few countries for which data are available. Overestimated
deforestation rates in the 1980s mask increased deforestation rates
from the 1980s to the 1990s. The FRA statistics indicate an 11%
decrease in the annual rate of change in forest area between the two
decades for the tropical countries considered in this analysis, whereas
the satellite-derived PCTAs suggest an 10% overall increase (Table
2). In Latin America, the two estimates both indicate declines in
deforestation rates, but the FRA indicates a larger decline (38%) than
the PTCA estimates (11%). The large increase in tropical Asia (68%
from PTCA in contrast to 2% decline from the FRA reports) occurs
mostly in Indonesia due at least partially to the drought-related fires
of 1997–98 (43) and reduced areas of regrowth relative to the 1980s.
The estimate of carbon fluxes from tropical land-use change is on the low end of the range reported in the IPCC, indicating that the missing carbon sink may be smaller than estimated previously. Although our results indicate that tropical forest clearing and carbon emissions accelerated from the 1980s to 1990s, the magnitude of the flux is lower than previous estimates based on changes in forest-area rates reported by the FRA. Of the three continental regions, we regard our carbon-flux estimates for tropical Africa to be the most uncertain because of difficulties in detecting patchy clearings and sparse data sources. It is unlikely, however, that this uncertainty accounts for the overall difference between our pantropics and the IPCC estimates.
The bookkeeping model to estimate carbon fluxes from changes in forest cover assumes an initial biomass for above-ground vegetation and soils and a known fate for carbon that is burned, deposited as slash, or stored in products. Although satellite data can improve the estimates of areas undergoing clearing and regrowth, the lack of spatially explicit information on the other factors contributes to the large uncertainties in carbon fluxes from tropical land-use change. Linkage of process studies with carbon models in a spatially explicit framework is needed to reduce these uncertainties.
In summary, this study suggests:
10% from the 1980s to 1990s in contrast to FRA
statistics that report declining rates. The increase is largely in
southeast Asia, with only slight decreases in clearing rates in Latin
America and Africa. 0.6
and 0.9 Pg•yr–1 for the 1980s and 1990s,
respectively, meaning that the missing terrestrial carbon is less than
half of the current IPCC estimate.
|
Acknowledgements |
|---|
|
Footnotes |
|---|
To whom correspondence should be addressed at:
University of Maryland, 2181 Lefrak Hall, College Park, MD 20742.
E-mail: rd63{at}umail.umd.edu. 
|
References |
|---|
|
TopAbstractMethods and ResultsDiscussion and ConclusionsReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg What's this?
This article has been cited by other articles in HighWire Press-hosted journals:
|
|
 |
|
  H. Nagendra Going Beyond Panaceas Special Feature: Drivers of reforestation in human-dominated forests PNAS, September 25, 2007; 104(39): 15218 - 15223. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. B. Stephens, K. R. Gurney, P. P. Tans, C. Sweeney, W. Peters, L. Bruhwiler, P. Ciais, M. Ramonet, P. Bousquet, T. Nakazawa, et al. Weak Northern and Strong Tropical Land Carbon Uptake from Vertical Profiles of Atmospheric CO2 Science, June 22, 2007; 316(5832): 1732 - 1735. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. C. Morton, R. S. DeFries, Y. E. Shimabukuro, L. O. Anderson, E. Arai, F. del Bon Espirito-Santo, R. Freitas, and J. Morisette Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon PNAS, September 26, 2006; 103(39): 14637 - 14641. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. M. Souza Jr. Mapping land use of tropical regions from space PNAS, September 26, 2006; 103(39): 14261 - 14262. [Full Text] [PDF]
|
|
|
|
|
|
G. P. Asner, D. E. Knapp, E. N. Broadbent, P. J. C. Oliveira, M. Keller, and J. N. Silva Selective Logging in the Brazilian Amazon Science, October 21, 2005; 310(5747): 480 - 482. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. Friedlingstein and S. Solomon Contributions of past and present human generations to committed warming caused by carbon dioxide PNAS, August 2, 2005; 102(31): 10832 - 10836. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
