`
`http://science.sciencemag.org/
`
`
`
`on July 5, 2017
`
`R E P O R T S
`
`38. C. Elschenbroich, A. Salzer, Organometallics (Wiley-
`VCH, Weinheim, Germany, ed. 1, 1989).
`39. A. Haaland, Acc. Chem. Res. 12, 415 (1979).
`40. M. Bochmann, Organometallics (Oxford Chemistry
`Primers, Oxford Univ. Press, Oxford, New York,
`1994), vol. 2.
`41. K. L. T. Wong, H. H. Brintzinger, J. Am. Chem. Soc. 97,
`5143 (1975).
`42. F. W. S. Ben(cid:222)eld, M. L. H. Green, J. S. Ogden, D. Young,
`J. Chem. Soc. Chem. Commun. 1973, 866 (1973).
`43. D. W. Blackburn, D. Britton, J. E. Ellis, Angew. Chem.
`Int. Ed. 31, 1495 (1992).
`44. M. Baudler, D. Du‹ster, D. Ouzounis, Z. Anorg. Allg.
`Chem. 544, 87 (1987).
`
`45. M. Lein, J. Frunzke, A. Timoshkin, G. Frenking, Chem.
`Eur. J. 7, 4155 (2001).
`46. Supplementary data are available on Science Online
`at www.sciencemag.org/cgi/content/full/295/5556/
`832/DC1.
`47. Dedicated to Professor Marianne Baudler on the occa-
`sion of her 80th birthday. We are grateful to the Na-
`tional Science Foundation (J.E.E.) and the donors of the
`Petroleum Research Fund administrated by the Ameri-
`can Chemical Society (J.E.E.) for (cid:222)nancial support of this
`work. We acknowledge B. Chen for the solid-state 31P-
`NMR spectral measurements.
`
`22 October 2001; accepted 27 December 2001
`
`Aerosol Effect on Cloud Droplet
`Size Monitored from Satellite
`Francois-Marie Bre«on,1* Didier Tanre«,2 Sylvia Generoso1
`
`Aerosol concentration and cloud droplet radii derived from space-borne measure-
`ments are used to explore the effect of aerosols on cloud microphysics. Cloud
`droplet size is found to be largest (14 micrometers) over remote tropical oceans
`and smallest (6 micrometers) over highly polluted continental areas. Small droplets
`are also present in clouds downwind of continents. By using estimates of droplet
`radii coupled with aerosol load, a statistical mean relationship is derived. The
`cloud droplet size appears to be better correlated with an aerosol index that
`is representative of the aerosol column number under some assumptions than
`with the aerosol optical thickness. This study reveals that the effect of aerosols
`on cloud microphysics is signi(cid:222)cant and occurs on a global scale.
`
`Aerosols may reduce the degree of Earth
`global warming resulting from the increase of
`greenhouse gases in the atmosphere (1, 2).
`They directly impact the radiative balance of
`Earth through a net increase of its albedo,
`particularly over the oceans (3, 4). Aerosols
`can also act as cloud condensation nuclei
`(CCN), increasing the number of droplets in
`clouds, which tends to decrease the mean
`droplet size and may increase the cloud albe-
`do (5), depending on the aerosol absorption
`and cloud optical thickness (6). This process,
`referred to as the “Twomey effect” or the
`“first indirect” aerosol radiative forcing, has a
`net cooling effect on climate. A direct dem-
`onstration of the aerosol effect on cloud al-
`bedo was provided by the observation of lines
`of larger reflectance in cloud fields identified
`as tracks of ship exhaust (7). Indirect obser-
`vations of this effect can also be made by
`comparing cloud droplet size and aerosol
`concentration. Cloud droplet effective radii
`were derived by using global scale AVHRR
`(advanced very high resolution radiometer)
`measurements (8). The results of a global
`
`1Laboratoire
`de
`et
`du Climat
`Sciences
`des
`l(cid:213)Environnement, Commissariat a‘ l(cid:213)Energie Atomique,
`Gif sur Yvette, France. 2Laboratoire d(cid:213)Optique Atmo-
`sphe«rique, CNRS, Universite« des Sciences et Technol-
`ogies de Lille, Villeneuve d(cid:213)Ascq, France.
`
`*To whom correspondence should be addressed. E-
`mail: fmbreon@cea.fr
`
`application (9) indicate a contrast in cloud
`droplet size of about 2 mm over land and
`ocean surfaces, as well as a hemispheric con-
`trast of 1 mm, both of which support the
`Twomey hypothesis. Similar patterns of the
`aerosol optical thickness and the cloud drop-
`let effective radius, derived from AVHRR
`measurements, have been observed over the
`oceans (10). Cases of reduced droplet radii
`and suppression of rain—the second indirect
`aerosol effect—in areas of high aerosol load
`were identified on satellite imagery (11, 12).
`Furthermore, several in situ measurements
`have shown a relationship between the aero-
`sol concentration and the cloud droplet size
`distribution (13–15).
`The polarization and directionality of the
`earth reflectances
`(POLDER)
`instrument
`(16) is well suited for assessing the Twomey
`hypothesis globally, because its measure-
`ments provide a unique opportunity to mea-
`sure cloud droplet effective radius (hereafter
`referred to as CDR) (17), as well as aerosol
`loading (18), over both land and ocean sur-
`faces. The
`POLDER radiometer was
`launched aboard the Advanced Earth-Observ-
`ing Satellite (ADEOS) in August 1996. Con-
`tinuous monitoring of the solar radiation re-
`flected by the earth, including its polarization
`and directional signatures, started on 30 Oc-
`tober 1996, and ended on 30 June 1997, with
`the unexpected failure of the satellite solar
`panel. Monthly maps were generated of an
`
`may also play an important role in the stabi-
`lization of monomers containing [(P5)2M]z
`units. On this basis, it seems possible that a
`wide variety of other charged decaphospha-
`metallocenes and related carbon-free sand-
`wich complexes (45) may be accessible
`species.
`
`References and Notes
`1. T. J. Kealy, P. L. Pauson, Nature 168, 1039 (1951).
`2. S. A. Miller, J. A. Tebboth, J. F. Tremaine, J. Chem. Soc.,
`632 (1952).
`3. R. B. Woodward, M. Rosenblum, M. C. Whiting, J. Am.
`Chem. Soc. 74, 3458 (1952).
`4. A. Togni, R. L. Halterman, Eds., Metallocenes: Synthe-
`sis, Reactivity, Applications (Wiley-VCH, Weinheim,
`Germany, 1998).
`5. N. J. Long, Metallocenes (Blackwell Science, Oxford,
`UK, 1998).
`6. A. D. Garnovskii, A. P. Sadimenko, M. I. Sadimenko,
`D. A. Garnovskii, Coord. Chem. Rev. 173, 31 (1998).
`7. K. B. Dillon, F. Mathey, J. F. Nixon, Phosphorus: The
`Carbon Copy (Wiley, Chichester, UK, 1998).
`8. R. Bartsch, P. B. Hitchcock, J. F. Nixon, J. Chem. Soc.
`Chem. Commun. 1987, 1146 (1987).
`
`9. iiii , J. Organomet. Chem. 356, C1 (1988).
`
`10. P. B. Hitchcock, J. F. Nixon, R. M. Matos, J. Orga-
`nomet. Chem. 490, 155 (1995).
`11. T. Clark et al., Angew. Chem. Int. Ed. 39, 2087 (2000).
`12. R. Bartsch et al., J. Organomet. Chem. 529, 375
`(1997).
`13. F. G. N. Cloke, J. R. Hanks, P. B. Hitchcock, J. F. Nixon,
`Chem. Commun. 1999, 1731 (1999).
`14. F. G. N. Cloke, J. C. Green, J. R. Hanks, J. F. Nixon, J. L.
`Suter, J. Chem. Soc. Dalton Trans. 2000, 3534 (2000).
`15. R. Bartsch et al., J. Chem. Soc. Dalton Trans. 2001,
`1013 (2001).
`16. M. Al-Ktaifani, J. C. Green, P. B. Hitchcock, J. F. Nixon,
`J. Chem. Soc. Dalton Trans. 2001, 1726 (2001).
`17. M. Baudler et al., Angew. Chem. Int. Ed. Engl. 27, 280
`(1988).
`18. O. J. Scherer, Angew. Chem. Int. Ed. Engl. 29, 1104
`(1990).
`19. E. J. P. Malar, J. Org. Chem. 57, 3694 (1992).
`20. A. Dransfeld, L. Nyulaszi, P. v. R. Schleyer,
`Chem. 37, 4413 (1998).
`21. M. Baudler, T. Etzbach, Angew. Chem. Int. Ed. Engl.
`30, 580 (1991).
`22. O. J. Scherer, Acc. Chem. Res. 32, 751 (1999).
`
`23. iiii , H. Swarowsky, G. Wolmersha‹user, W. Kaim,
`
`Inorg.
`
`S. Kohlmann, Angew. Chem. Int. Ed. Engl. 26, 1153
`(1987).
`24. J. E. Ellis, D. W. Blackburn, P. Yuen, M. Jang, J. Am.
`Chem. Soc. 115, 11616 (1993).
`25. Details of the syntheses, isolation, and characteriza-
`tion of the [K(18-Crown-6)]1, (Ph3P)2N1, and Ph4P1
`salts of 1 are available on Science Online (46).
`26. Electrochemical characterization of [CpxFeP5]2 has
`been reported by R. F. Winter and W. E. Geiger
`[Organometallics 18, 1827 (1999)].
`27. D. F. Evans, J. Chem. Soc., 2003 (1959).
`28. Crystallographic details are available on Science On-
`line (46).
`29. L. Weber, Chem. Rev. 92, 1839 (1992).
`30. O. J. Scherer, T. Hilt, G. Wolmersha‹user, Organome-
`tallics 17, 4110 (1998).
`31. T. P. Hamilton, Henry F. Schaefer III, Angew. Chem.
`Int. Ed. Engl. 28, 485 (1989).
`32. M. D. Fryzuk, T. S. Haddad, D. J. Berg, Coord. Chem.
`Rev. 99, 137 (1990).
`33. J. Perdew, Y. Wang, Phys. Rev. B 45, 13244 (1992).
`34. K. Burke, J. P. Perdew, Y. Wang, Electronic Density
`Functional Theory. Recent Progress and New Direc-
`tions, J. F. Dobson, G. Vignale, M. P. Das, Eds. (Plenum,
`New York, 1998).
`35. C. Adamo, V. Barone, J. Chem. Phys. 108, 664 (1998).
`36. W. J. Hehre, L. Radom, P. v. R. Schleyer, J. A. Pople, Ab
`Initio Molecular Orbital Theory (Wiley, New York,
`1986).
`37. R. Stowasser, R. Hoffmann, J. Am. Chem. Soc. 121,
`3414 (1999).
`
`834
`
`1 FEBRUARY 2002 VOL 295 SCIENCE www.sciencemag.org
`
`R.J. Reynolds Vapor
`IPR2016-01268
`R.J. Reynolds Vapor v. Fontem
`Exhibit 1029-00001
`
`
`
`Downloaded from
`
`http://science.sciencemag.org/
`
`
`
`on July 5, 2017
`
`R E P O R T S
`
`index” that quantifies the atmo-
`“aerosol
`spheric load by small particles (19). Under
`some assumptions, the aerosol index is ex-
`pected to be proportional to the aerosol col-
`umn number when the widely used optical
`thickness is proportionally more sensitive to
`the large particle fraction (20). Spatial and
`temporal distribution of the index indicate
`that it is mostly sensitive to aerosols gener-
`ated by biomass burning and human-generat-
`ed pollution (21). Similarly, the polarization
`signature of liquid water clouds was used to
`derive monthly mean estimates of CDR (22).
`
`A seasonal (March-May) average of these
`two parameters, aerosol index and CDR, is
`illustrated in Figs. 1 and 2. The highest values
`of the aerosol index (Fig. 1) are observed
`over southeast Asia and India, as well as
`Central Africa. These presumably reflect the
`presence of anthropogenic aerosols from in-
`dustrial activity and biomass burning. Rela-
`tively large values are also observed over the
`sub-Sahelian region, Central America, and
`Eastern Europe. Very low values are apparent
`over the open oceans. There is a significant
`land-ocean contrast. The influence of conti-
`
`nental aerosols over the ocean is notable,
`particularly downwind of regions with high
`aerosol loading (winds are mostly westward
`in the tropics, and eastward in mid-latitude
`regions). Over the oceans, the highest values
`are observed east of China, over the northern
`part of the Indian Ocean, downwind of the
`Sahel, and surrounding Central America.
`Note also that the atmosphere over southern
`oceans appears cleaner than over the basins
`of the Northern Hemisphere.
`Cloud droplet radius estimates (Fig. 2) are
`based on the angular signature of the polar-
`
`index
`Fig. 1. Aerosol
`derived from POLDER
`measurements during
`the spring of 1997. The
`aerosol index quanti(cid:222)es
`the atmospheric load-
`ing by small particles.
`
`Fig. 2. Cloud droplet
`radius
`derived
`from
`POLDER measurements
`during the spring of
`1997. The units are mi-
`crons. White areas cor-
`respond
`to
`regions
`where no successful es-
`timate was possible.
`This image is based
`upon a compilation of
`19,500 estimates.
`
`www.sciencemag.org SCIENCE VOL 295 1 FEBRUARY 2002
`
`835
`
`R.J. Reynolds Vapor Exhibit 1029-00002
`
`
`
`Downloaded from
`
`http://science.sciencemag.org/
`
`
`
`on July 5, 2017
`
`R E P O R T S
`
`the sets were obtained from measurements clos-
`er than 100 km in space. The study area was
`restricted to 60°N to 45°S to limit the impact of
`the latitudinal gradient observed in the Southern
`Hemisphere that clearly results from a different
`cloud process than the one analyzed here. Be-
`cause most significant sources of aerosols are
`within these boundaries as confirmed by previ-
`ous studies (25), this region is well suited to
`explorations of the effect of aerosols on cloud
`microphysics.
`As a result of this procedure a total of
`44,066 valid (CDR, aerosol index) sets were
`available for
`further analysis, of which
`28,686 aerosol estimates were derived from
`oceanic measurements, and 15,380 from land
`observations. Recall that the respective algo-
`rithms make use of different techniques, and
`that the results derived over the ocean are
`more accurate than the ones obtained over
`land.
`The mean and standard deviation of CDRs
`were computed for sets with a given aerosol
`index (bins of 0.025; 0.0125 for the first two
`bins) and while distinguishing between aero-
`sol estimates derived over land and over the
`ocean (Fig. 3). The curves confirm the rela-
`tionship apparent on the seasonal maps. On
`average, CDR is smaller when the aerosol
`load is larger. Because there is a clear phys-
`ical process that predicts such a decrease, this
`observed correlation can be interpreted as a
`causal connection. The aerosols clearly im-
`pact the cloud microphysics, not only for
`specific cases, as in situ and airborne mea-
`surements have already shown, but also on a
`global scale. The aerosol effect on cloud mi-
`crophysics appears nonlinear: the change in
`CDR with aerosol index (the derivative) is
`larger for clean atmospheres than for those
`with a significant load as already noted in
`specific cases (11).
`The statistical significance of the CDR
`mean is illustrated by the error bars on the y
`axis. They represent the confidence level of
`the mean value, i.e., s/=n – 2, where n and
`s are the number of CDR measurements
`
`within the bin and their standard deviation, if
`one assumes independent data. Although the
`observations show a wide range of CDR, the
`large number of measurements for relatively
`clean atmospheric areas make the corre-
`sponding mean values highly significant.
`In the interpretation of the results, one
`must emphasize that the uncertainty on indi-
`vidual measurements of aerosol index is sig-
`nificant in regard to the x-axis scale in Fig. 3.
`This uncertainty is larger over land than over
`the oceans because of the large and variable
`land surface contribution to the measured
`radiance. The noise in the aerosol load esti-
`mates tends to smooth the curves in Fig. 3
`and thus to decrease the apparent impact of
`the aerosol to the cloud microphysics (26).
`This may explain the apparent lower impact
`than over oceanic surfaces. It may also ex-
`plain the divergence in CDR between ocean
`and land curves for the smallest values of
`aerosol index. On the other hand, it is possi-
`ble that land surfaces naturally generate a low
`level of aerosol that is not within the capa-
`bilities of space-borne passive remote sens-
`ing. Thus a “clean” continental atmosphere
`would not be as pristine as an oceanic one,
`resulting in a smaller CDR.
`index–
`The mean slope of the aerosol
`CDR relationship is highly significant for the
`clean cases, up to an aerosol index of 0.15 to
`0.20, as shown by the small error bars. The
`constant mean CDR observed for aerosol in-
`dices greater than 0.15 indicates a saturation
`effect, although an extension of Twomey’s
`hypothesis as suggested by (27) cannot be
`rejected. Fluctuations observed above the
`threshold of 0.15 are not statistically signifi-
`cant because of a limited number of samples.
`Because the optical thickness is widely
`used to characterize the particle concentra-
`tion, we show the result of a similar process-
`ing when the optical thickness rather than the
`aerosol index is used (Fig. 3). This is limited
`to ocean observations, because an accurate
`estimate from POLDER measurements is not
`possible over land (19). Recall that the aero-
`
`Fig. 3. Mean CDR as a function
`of aerosol load. The two lower
`curves show the mean CDR as a
`function of the aerosol
`index
`(lower scale) for land (red) and
`ocean (blue) retrievals. The up-
`per curve is the same as the lat-
`ter but as a function of optical
`thickness (upper scale). The error
`bars represent the con(cid:222)dence
`level of the mean value,
`i.e.,
`s/= n — 2, where n and s are the
`number of CDR measurements
`within the bin and their standard
`deviation.
`
`ized radiance reflected by a cloud field (17).
`This estimate requires a specific viewing ge-
`ometry and is only possible when the cloud
`field is relatively uniform over an area of
`approximately 150 3 150 km2 and when the
`cloud droplet size distribution is narrow.
`When these conditions are met, the retrieval
`is highly reliable with no identified cause for
`bias (23). The seasonal composite of CDR for
`the spring of 1997 is shown in Fig. 2. Despite
`the natural variability, coherent patterns are
`clearly depicted. Many spatial features are
`similar to those found in the aerosol index;
`namely, the smallest CDR values are found
`over regions of high aerosol index, and the
`largest values are over the open oceans where
`the atmosphere is very clean: CDR is be-
`tween 6 and 10 mm over land surfaces, values
`over the oceans vary between 12 and 14 mm
`in remote areas down to the same low values
`of 6 mm in regions with the highest aerosol
`index. CDR over areas downwind of land
`surfaces is clearly affected by continental
`influences.
`A latitudinal gradient is apparent in the
`southern ocean with smaller droplets occur-
`ring toward the pole. This result agrees with
`previous in situ measurements that showed
`CDR much smaller in polar stratocumulus
`clouds than in mid-latitude clouds (24). This
`effect, however, cannot be presently attribut-
`ed to the presence of aerosols, because this
`area is known to be very clean, as confirmed
`by the low aerosol index values. The gradient
`is apparent during other months, in particular
`during the December-to-February period
`when the solar elevation allows observations
`in the southernmost latitudes (22). The cor-
`responding latitudinal gradient is not appar-
`ent in the Northern Hemisphere, presumably
`because the aerosol impact is then dominant.
`The similarities between the spatial distribu-
`tions of aerosol index and CDR suggest that a
`more quantitative analysis can be performed.
`The two estimates are not fully coincident, be-
`cause the presence of clouds makes the quanti-
`tative monitoring of aerosol impossible with a
`passive technique. Thus, the individual mea-
`surements that are used to derive the two sea-
`sonal fields may correspond to different mete-
`orological situations. A quantitative analysis re-
`quires the use of individual estimates rather
`than temporal means. In this study, for each
`individual estimate of CDR, a back-trajectory
`was computed up to a location where an aerosol
`index value was available. This procedure
`yields sets (CDR, aerosol load) that are based
`on individual and quasi coincident observa-
`tions. The back-trajectory makes use of the six
`hourly wind fields from the European Center
`for Medium-Range Weather
`Forecast
`(ECMWF) at 850 hPa, and was limited in time
`(24 hours) and space (1000 km). Here, we
`assume that the aerosol estimate remains valid
`over this time span. Note that more than half of
`
`836
`
`1 FEBRUARY 2002 VOL 295 SCIENCE www.sciencemag.org
`
`R.J. Reynolds Vapor Exhibit 1029-00003
`
`
`
`Downloaded from
`
`http://science.sciencemag.org/
`
`
`
`on July 5, 2017
`
`the effect’s
`allow the quantification of
`magnitude over land and ocean. The ob-
`served mean relationship can be used to
`validate the aerosol-cloud physical mecha-
`nisms implemented in models.
`Whether the observed impact on cloud
`microphysics is of anthropogenic origin is a
`question of importance. The satellite mea-
`surement cannot unambiguously distinguish
`natural and human-generated aerosols. How-
`ever, the analysis of the spatial and temporal
`patterns in the aerosol index monthly maps
`strongly suggests that the bulk of the aerosol
`load originates from slash-and-burn agricul-
`ture practices and from highly polluted areas
`(25). A large fraction of the observed aerosol
`effect on clouds is probably an anthropogenic
`impact.
`
`References and Notes
`1. J. E. Penner et al., Bull. Am. Meteorol. Soc. 75, 375
`(1994).
`2. R. J. Charlson et al., Science 255, 423 (1992).
`3. J. Haywood, V. Ramaswamy, B. Soden, Science 283,
`1299 (1999).
`4. O. Boucher, D. Tanre«, Geophys Res. Lett. 27, 1103
`(2000).
`5. S. Twomey, J. Atmos. Sci. 34, 1149 (1977).
`6. Y. J. Kaufman, R. S. Fraser, Science 277, 1636 (1997).
`7. J. A. Coakley, R. L. Bernstein, P. A. Durkee, Science
`237, 1020 (1987).
`8. T. Nakajima, M. D. King, J. Atmos. Sci. 47, 1878
`(1990).
`9. Q. Han, W. B. Rossow, A. A. Lacis, J. Clim. 7, 465
`(1994).
`10. M. A. Wetzel, L. L. Stowe, J. Geophys. Res. 104, 31287
`(1999).
`11. D. Rosenfeld, Science 287, 1793 (2000).
`
`12. iiii , Geophys. Res. Lett. 26, 3105 (1999).
`
`13. R. J. Vong, D. S. Covert, J. Atmos. Sci. 55, 2180 (1998).
`14. G. M. Frick, W. A. Hoppel, Bull. Am. Meteorol. Soc. 74,
`2195 (1993).
`15. J. R. Snider, J.-L. Brenguier, Tellus 52, 828 (2000).
`16. P.-Y. Deschamps et al., IEEE Trans. Geosci. Remote
`Sens. 32, 598 (1994).
`17. F.-M. Bre«on, Ph. Goloub, Geophys. Res. Lett. 25, 1879
`(1998).
`18. M. Herman et al., J. Geophys. Res. 102, 17039 (1997).
`19. Over land surfaces, the aerosol
`index is inferred
`from the polarized radiance that is not fully related
`to the aerosol optical thickness. Because large
`particles yield negligible polarization, the index
`quanti(cid:222)es the load of the accumulation mode, with
`an effective radius of 0.18 mm. Over the oceans,
`the aerosol index is de(cid:222)ned as the product of the
`aerosol optical thickness and the (cid:129)ngstro‹m coef(cid:222)-
`cient a. a is close to zero for large particles such as
`dust, and on the order of 1 to 1.5 for smaller
`particles such as those generated by biomass burn-
`ing or industrial activity. The consistency of the
`index over land and oceans is demonstrated by the
`observed continuity at the coastal boundaries.
`20. T. Nakajima et al., Geophys Res. Lett. 28, 1171
`(2001). For a given optical thickness, the number of
`particles is much larger for a sub-micronic aerosol
`(typical for biomass burning smoke and pollution)
`than for larger sized particles such as dust. On the
`other hand, the (cid:129)ngstro‹m coef(cid:222)cient decreases as
`the particle size increases. Thus, the aerosol index
`(optical thickness 3 angstro‹m coef(cid:222)cient) partly
`compensates the size effects and is better related
`than the optical thickness to the number of particles.
`21. J.-L. Deuze« et al., J. Geophys. Res. 106, 4913 (2001).
`22. F.-M. Bre«on, S. Colzy, Geophys. Res. Lett. 27, 4065
`(2000).
`23. No ground truth validation has been performed, but
`the agreement between the measured and modeled
`polarized radiances suggests an RMS accuracy better
`than 0.5 mm on individual estimates. The noise in the
`
`R E P O R T S
`
`impact of aerosols on
`strate a significant
`cloud microphysics at the global scale; how-
`ever, several limitations discussed below pre-
`vent a definite quantitative interpretation of
`their indirect radiative forcing.
`1) The constraints on the CDR retrieval,
`in particular the spatial homogeneity and the
`narrow size-distribution criteria, may select
`particular cloud types (stratiform rather than
`convective) and/or a specific time in the
`cloud life cycle. Therefore, the relationship
`may not be valid for all cloud types and may
`be biased toward sensitive or
`insensitive
`clouds.
`2) The parameter of importance for the
`Twomey effect is the concentration of CCN
`in the atmospheric layer where cloud droplets
`are formed. The POLDER retrievals measure
`an aerosol
`load index integrated over the
`vertical, not the CCN concentration. This is a
`source of uncertainty in the aerosol index–
`CDR relationship, although it is reasonable to
`assume that the CCN concentration and the
`aerosol index are well correlated, because the
`latter is proportional to the column aerosol
`number.
`The observed relationship of Figs. 3 and 4
`is clearly not a spurious result of the inver-
`sion algorithms. Although over land surfaces
`both the aerosol and cloud inversions make
`use of the polarized radiance measurements,
`the former is based on the magnitude of the
`radiance; the latter uses its directional signa-
`ture. We cannot identify any mechanism in
`the satellite estimates that would generate a
`misleading relationship between CDR and
`the aerosol index. Moreover, the aerosol in-
`version technique over the oceans is based on
`the radiance measurement, with limited con-
`tribution from polarization.
`Because of the various limitations of the
`remote sensing techniques and the inherent
`statistical aspect of our results, it is not
`possible to reliably quantify the aerosol
`first
`indirect radiative effect on climate.
`Nevertheless,
`the present results provide
`evidence of a strong effect of aerosols on
`cloud microphysics at the global scale and
`
`sol index is roughly proportional to the aero-
`sol column number when the optical thick-
`ness is affected by the presence of large
`particles even in limited number, which puts
`a large weight to the large particle fraction.
`As shown by the comparison of the blue and
`green curves (28) in Fig. 3, CDR is more
`sensitive to the aerosol index than to the
`optical thickness, which is to be expected,
`because the aerosol index is a function of the
`CCN concentration. Aerosol characteristics
`other than typical size (for instance, their
`hygroscopicity) may affect the statistical re-
`sults of Fig. 3.
`We now compare the observed statistical
`relationship to the simple theory originally
`proposed by Twomey. The number of aerosol
`particles that may act as CCN, Na, and the
`number of cloud droplets, Nd, are approxi-
`mately related through (29):
`Nd ’ (Na)a
`
`(1)
`
`Cloud process models and measurements in-
`dicate that a is on the order of 0.7. If one
`assumes a constant
`liquid water content,
`these numbers are related to the cloud droplet
`effective radius through (5, 6, 27):
`]log(Nd)
`3
`
`]log(r) 5
`
`]log(AI )
`
`(2)
`
`a 3
`
`]log(Na) 5
`
`a 3
`
`5
`
`Thus, a slope of ’0.23 is expected between
`the mean radius and the aerosol index on a
`log-log scale. Figure 4 shows the correspond-
`ing plot. Our analysis can be seen as support-
`ing Twomey’s hypothesis: There is a linear
`relationship (in log scale) between a change
`in aerosol concentration and a change in
`cloud droplet radius. On the other hand, the
`value of the slope is much smaller than the
`0.23 expected: 0.085 over the oceans and
`0.04 over land. The value found here for the
`oceanic cases is very similar to the one that
`can be derived from AVHRR data (20) and in
`the range of recent modeling results (27).
`The results of this study clearly demon-
`
`Fig. 4. Same as Fig. 3 on a log-
`log scale. Only the curves as a
`function of aerosol
`index are
`shown.
`
`www.sciencemag.org SCIENCE VOL 295 1 FEBRUARY 2002
`
`837
`
`R.J. Reynolds Vapor Exhibit 1029-00004
`
`
`
`R E P O R T S
`
`monthly averages also results from the natural vari-
`ability when few measurements are available.
`24. G. F. Herman, J. Curry, J. Clim. Appl. Meteorol. 23, 5
`(1984).
`25. D. Tanre« et al., Geophys. Res. Lett., 28, 4555 (2001).
`26. Each point in Fig. 3 shows the average of individual
`CDR measurements that correspond to a given bin of
`aerosol load estimates. Because there is a signi(cid:222)cant
`uncertainty on the aerosol load, this average is de-
`rived from cases with actual loads that may be higher
`or lower than the bin value. For the lowest load bin,
`
`only higher loads (smaller CDR) contaminate the
`estimate. This tends to decrease the average CDR all
`the more that the uncertainty on the aerosol load is
`larger.
`27. G. Feingold et al., J. Geophys. Res. 106, 22907
`(2001).
`28. Because the curve as a function of optical thickness
`(green) is signi(cid:222)cantly above that as a function of
`aerosol index (blue), one may get the false impres-
`sion that they have been obtained from different
`sets of CDR and aerosol retrievals. In fact, more
`
`than half of aerosol index estimates fall into the
`lowest load bin when the corresponding optical
`thicknesses are distributed over a wider range of
`values. The blue and green curves have been de-
`rived from the same sample and have the same
`overall mean CDR.
`29. S. Twomey, in Atmospheric Aerosols (Elsevier Sci-
`ence, New York, 1977); see in particular equations
`4 to 15.
`
`20 September 2001; accepted 3 January 2002
`
`Evidence for Strengthening of
`the Tropical General Circulation
`in the 1990s
`Junye Chen,1,2* Barbara E. Carlson,2 Anthony D. Del Genio2
`
`Satellite observations suggest that the thermal radiation emitted by Earth to
`space increased by more than 5 watts per square meter, while re(cid:223)ected sunlight
`decreased by less than 2 watts per square meter, in the tropics over the period
`1985—2000, with most of the increase occurring after 1990. By analyzing
`temporal changes in the frequency of occurrence of emitted thermal and
`re(cid:223)ected solar (cid:223)uxes, the effects of El Nin(cid:247)o—Southern Oscillation are minimized,
`and an independent longer-time-scale variation of the radiation budget is
`identi(cid:222)ed. Similar analyses of upper tropospheric humidity, cloud amount,
`surface air temperature, and vertical velocity con(cid:222)rm that these (cid:223)ux changes
`are associated with a decadal-time-scale strengthening of the tropical Hadley
`and Walker circulations. Equatorial convective regions have intensi(cid:222)ed in up-
`ward motion and moistened, while both the equatorial and subtropical sub-
`sidence regions have become drier and less cloudy.
`
`Downloaded from
`
`http://science.sciencemag.org/
`
`
`
`on July 5, 2017
`
`of the well-known shift of the convection
`center between the two regions in different
`ENSO phases (10). But the decadal LW flux
`increase (Fig. 1) cannot be identified by the
`spatial-temporal EOF decomposition. Al-
`though in 1998 both the ENSO index and the
`tropical mean LW flux anomalies reach their
`maximum, the two time series are poorly
`correlated with each other (Fig. 2A). This
`implies that the mechanism behind the long-
`term average LW flux increase is distinct
`from the ENSO phenomenon. The elevated
`CERES Terra LW flux in 2000, a non-ENSO
`year, relative to the ERBE period is further
`evidence for a longer term flux variation.
`Because ENSO primarily involves a spa-
`tial redistribution, frequency histograms of
`SW and LW fluxes covering the entire tropics
`exhibit much less interannual variability than
`do the geographical distributions of the fluxes
`within the tropics. We therefore minimize the
`ENSO signal by constructing a SW-LW joint
`frequency distribution ( JFD) matrix (11) for
`each month. We then compute anomalies of
`the JFD with respect to the climatology and
`decompose these with EOF analysis (12).
`The high correlation between the first PC and
`the tropical mean LW flux anomaly time
`series means that the first mode of the SW-
`LW JFD EOF represents the long-term in-
`crease in LW flux (Fig. 2A). The first PC
`shows that the increase begins at the start of
`the 1990s, and elevated flux levels are still
`present in 2000. The first JFD EOF pattern
`(Fig. 2B) shows that occurrences of LW flux
`values of ;255 W m22 and SW flux values
`of ;75 W m22 decrease over the 1985–2000
`time period and are replaced by higher LW
`fluxes of ;290 W m22 and lower SW fluxes
`of ;60 W m22. To mimic the effect of a
`worst-case hypothetical calibration shift in
`
`Local LW changes are weakly positive, about 4
`W m22 over 10 years, in most of the tropics
`(6). Much larger changes occur adjacently with
`opposite sign (for example, the West Pacific
`warm pool and Central Pacific region).
`Empirical orthogonal
`function (EOF)
`analysis decomposes temporal variations into
`orthogonal spatial patterns that sometimes
`reveal independent physical mechanisms (7).
`When EOF analysis is applied to the radiation
`flux anomalies (8), the first principal compo-
`nent (PC) describing the time evolution of the
`first spatial pattern is strongly correlated with
`the NINO3 El Nin˜o–Southern Oscillation
`(ENSO) index (9), and the first spatial pattern
`explaining the largest fraction of the temporal
`variance resembles the West Pacific–Central
`Pacific dipole. This mode is an embodiment
`
`The energy exchange between Earth and its
`environment is determined by the emitted ther-
`mal [longwave (LW)] flux and the reflected
`part of the solar irradiance [shortwave (SW)]
`flux at the top of the atmosphere (TOA). Equi-
`librium of Earth’s climate requires that the
`global annual mean net radiation flux at the
`TOA be approximately zero.
`Clouds and the Earth’s Radiant Energy Sys-
`tem (CERES) (1) instruments on the Tropical
`Rainfall Measuring Mission (TRMM, begun in
`1998) satellite and the Earth Observing System
`Terra (begun in 2000) satellite have observed
`LW fluxes 5 to 10 W m22 (;2 to 4%) higher
`than those from the Earth Radiation Budget
`Experiment (ERBE) (2) scanner data (1985–
`1989). These differences cannot be fully ex-
`plained by known changes of the satellite ob-
`servation systems (3–5). The ERBE wide field
`of view (WFOV) data (1985–1995) span a
`longer time period though at lower spatial res-
`olution than the ERBE scanner data and clearly
`show a decadal increase of LW flux, primarily
`during the first half of the 1990s (3) (Fig. 1).
`
`1Department of Earth and Environmental Sciences,
`Columbia University, Palisades, NY 10964, USA.
`2NASA/Goddard Institute for Space Studies, 2880
`Broadway, New York, NY 10025, USA.
`
`*To whom correspondence should be addressed. E-
`mail: jchen@giss.nasa.gov
`
`838
`
`1 FEBRUARY 2002 VOL 295 SCIEN