© 2000 Macmillan Magazines Ltd letters to nature NATURE | VOL 403 | 20 JANUARY 2000 | www.nature.com 301 3. Anderson, J. G., Toohey, D. W. & Brune, W. H. Free radicals within the Antarctic vortex: The role of CFCs in the Antarctic ozone loss. Science 251, 39±46 (1991). 4. Wever, R. Ozone destruction by algae in the Arctic atmosphere. Science 355, 501 (1988). 5. Butler, J. H. et al. A record of atmospheric halocarbons during the twentieth century from polar ®rn air. Nature 399, 749±755 (1999). 6. Gribble, G. W. Naturally occurring organohalogen compoundsÐa survey. J. Natural Prod. 55, 1353± 1395 (1992). 7. Lovelock, J. E. Natural halocarbons in the air and in the sea. Nature 256, 193±194 (1975). 8. Singh, H. B., Salas, L. J. & Stiles, R. E. Methylhalides in and over the Eastern Paci®c (408 N-328 S). J. Geophys. Res. 88, 3684±3690 (1983). 9. Class, T. & Ballschmiter, K. Sources and distribution of bromo- and bromochloromethanes in marine air and surface water of the Atlantic Ocean. J. Atmos. Chem. 6, 35±46 (1988). 10. Moore, R. M., Groszko, W. & Niven, S. J. Ocean-atmosphere exchange of methyl chloride: Results from NW Atlantic and Paci®c Ocean studies. J. Geophys. Res. 101, 28529±28538 (1988). 11. Sturges, W. T., Sullivan, C. W., Schnell, R. C., Heidt, L. E. & Pollack, W. H. Bromoalkane production by Antarctic ice algae. Tellus B 45, 120±126 (1993). 12. Laturnus, F. & Adams, F. C. Methylhalides from Antarctic macroalgae. Geophys. Res. Lett. 25, 773±776 (1998). 13. Harper, D. B. Halomethane from halide ionÐa highly ef®cient fungal conversion of environmental signi®cance. Nature 315, 55±57 (1985). 14. Andreae, M. O. et al. Methylhalide emissions from savanna ®res in southern Africa. J. Geophys. Res. 101, 23603±23613 (1996). 15. Rasmussen, R. A., Rasmussen, L. E., Khalil, M. A. K. & Dalluge, R. W. Concentration of methyl chloride in the atmosphere. J. Geophys. Res. 85, 7350±7356 (1980). 16. Hoekstra, E. J., Lassen, P., van Leeuwen, J. G. E., De Leer, E. W. B. & Carlsen, L. in Naturally Produced Organohalogens (eds Grimvall, A. & De Leer, E. W. B.) 149±158 (Kluwer Academic, Dordrecht, 1995). 17. Desjardins, S., Landry, J. A. & Farant, J. P. Effects of water and pH on the oxidative oligomerization of chloro and methoxyphenol by a montmorillonite clay. J. Soil Contamin. 8, 175±195 (1999). 18. Lovley, D. R. Dissimilatory Fe(III) and Mn(IV) reduction. Microbiol. Rev. 55, 259±287 (1991). 19. Post, W. M., Emanuel, W. R., Zinke, P. J. & Stangenberger, A. L. Soil carbon pools and world life zones. Nature 298, 156±159 (1982). Acknowledgements We thank D. Schlo Èsser and K. Kratz for the instrumental neutron activation analysis (INAA) measurements and I. Fahimi and L. Warr for reviewing the manuscript. Correspondence and requests for materials should be addressed to F.K. (e-mail: fkeppler@ix.urz.uni-heidelberg.de). ................................................................. Annual ¯uxes of carbon from deforestation and regrowth in the Brazilian Amazon R. A. Houghton*, D. L. Skole², Carlos A. Nobre³, J. L. Hackler*, K. T. Lawrence* & W H. Chomentowski² * Woods Hole Research Center, PO Box 296, Woods Hole, Massachusetts 02543, USA ² Department of Geography, Michigan State University, East Lansing, Michigan 48824, USA ³ Instituto Nacional de Pesquisas Espaciais, Caixa Postal 515, Sa Äo Jose  dos Campos, SP, CEP 12201-970, Brazil .............................................................................................................................................. The distribution of sources and sinks of carbon among the world's ecosystems is uncertain. Some analyses show northern mid- latitude lands to be a large sink, whereas the tropics are a net source 1 ; other analyses show the tropics to be nearly neutral, whereas northern mid-latitudes are a small sink 2,3 . Here we show that the annual ¯ux of carbon from deforestation and abandon- ment of agricultural lands in the Brazilian Amazon was a source of about 0.2 Pg C yr -1 over the period 1989±1998 (1 Pg is 10 15 g). This estimate is based on annual rates of deforestation and spatially detailed estimates of deforestation, regrowing forests and bio- mass. Logging may add another 5±10% to this estimate 4 , and ®res may double the magnitude of the source in years following a drought 4 . The annual source of carbon from land-use change and ®re approximately offsets the sink calculated for natural eco- systems in the region 5,6 . Thus this large area of tropical forest is nearly balanced with respect to carbon, but has an interannual variability of 6 0.2 PgC yr -1 . We determined the annual ¯ux of carbon with a `bookkeeping' model 7,8 that tracks the annual emission and uptake of carbon that follow the clearing of forest for agriculture and the regrowth of secondary forests on abandoned agricultural land. Changes in carbon include (1) the immediate loss of carbon to the atmosphere from plant material burned at the time of clearing, (2) the slower release of carbon from decay of dead plant material left on site (slash) and removed for wood products, and (3) the accumulation of carbon during forest growth. Changes in soil carbon were not included in this analysis, as they are small relative to the changes in biomass and are inconsistent in direction 9±12 . We used two estimates of deforestation, three estimates of biomass and two estimates of the rate of decay of organic matter to calculate a range of net carbon emissions attributable to land-use change. The ®rst estimate of deforestation was obtained from the Brazilian Space Agency (INPE), where data from the Landsat satellite are delineated manually for each state to determine both annual rates of deforestation and cumulative areas deforested for each year between 1988 and 1998 (except 1993). The annual and cumulative data are not entirely consistent, and we used the cumulative areas deforested to calculate annual rates of change (Table 1). INPE also determined the area deforested in 1978; before 1960 rates of deforestation were negligible 13 . The second estimate of deforestation was based on a map of land cover derived from classi®cation of 1986 Landsat multi-spectral scanner data (Fig. 1). Areas classi®ed as deforested in 1986 were consistently lower than INPE's 1988 estimate of deforested area. Because the dates were different, we interpolated a rate for 1988 based on maps of land cover derived from 1986 and 1992 Landsat data. The interpolated area deforested in 1988 was still about 25% lower than INPE's estimate, although the actual percentage varied among states (Fig. 2). We used this lower estimate for a second estimate of deforestation, varying it annually in proportion to the rates from INPE. According to the Landsat-derived land-cover classi®cation, about 30% of the deforested area was in secondary forest in 1986Ð presumably as a result of the abandonment of agricultural land 14±17 . The percentage varied from 5% in Gois to 65% in Maranhao. As we lacked data to suggest temporal trends in abandonment, we assumed that cleared lands were abandoned annually at the rate de®ned by the ratio of secondary forest to deforested area in 1986. Forest Deforested Regrowing Cerrado Water Cloud & shadow 500 km Figure 1 Land cover in Brazilian Amazonia as of 1986, based on a classi®cation of Landsat MSS data. The classi®cation identi®es seven classes of land cover: forest, deforested land, regrowing forest, water, clouds, cloud shadow and cerrado (savanna). Here data for cloud and cloud shadow are grouped together.