Aerosols from biomass burning over the tropical South Atlantic region: Distributions and impacts

. The NASA Global Tropospheric Experiment (GTE) Transport and Atmospheric Chemistry Near the Equator--Atlantic (TRACE A) expedition was conducted September 21 through October 26, 1992, to investigate factors responsible for creating the seasonal South Atlantic tropospheric ozone maximum. During these flights, fine aerosol (0.1-3.0 gm) number densities were observed to be enhanced roughly tenfold over remote regions of the tropical South Atlantic and greater over adjacent continental areas, relative to northern hemisphere observations and to measurements recorded in the same area during the wet season. Chemical and meteorological analyses as well as visual observations indicate that the primary source of these enhancements was biomass burning occurring within grassland regions of north central Brazil and southeastern Africa. These fires exhibited fine aerosol (N) emission ratios relative to CO (dN/dCO) of 22.5 + 9.7 and 23.6 + 15.1 cm -3 parts per billion by volume (ppbv) -1 over Brazil and Africa, respectively. Convection coupled with counterclockwise flow around the South Atlantic subtropical anticyclone subsequently distributed these aerosols throughout the remote South Atlantic troposphere. We calculate that dilute smoke from biomass burning produced an average tenfold enhancement in optical depth over the continental regions as well as a 50% increase in this parameter over the middle South Atlantic Ocean; these changes correspond to an estimated net cooling of up to 25 W m -2 and 2.4 W m -2 during clear-sky conditions over savannas and ocean respectively. Over the ocean our analyses suggest that modification of CCN concentrations within the persistent eastern Atlantic marine stratocumulus clouds by entrainment of subsiding haze layers could significantly increase cloud albedo resulting


Introduction
Recently, the effect of atmospheric aerosols on the Earth's radiative budget has received attention as several studies suggest particulate matter contributes a radiative forcing approximately equal but opposite in sign to that produced by the greenhouse gases accumulating within the Earth's atmosphere [Charlson et al., 1992a, b;Kiehl and Briegleb, 1993;Penner et al., 1994]. The aerosol forcing is produced by a number of mechanisms. In a direct sense, aerosols reflect solar radiation back into space which increases planetary albedo, hence lowering surface temperatures [Coakley et al., 1983]. Aerosols, particularly the carbonaceous type, also absorb solar radiation [Ackerman and Air mass trajectories were used as a qualitative aid to identify the processes influencing sampled air masses. These were obtained from 5-day backward looking isentropic air mass trajectory calculations based on National Meteorological Center (NMC) wind field grids as described by Bachmeier and Fuelberg [this issue]. The general procedure was to calculate the trajectories for a cluster of locations surrounding the point of interest and to only accept those whose groupings that did not diverge considerably during the previous 5 days.
Aerosol number densities and size diameters over the range from 0.1 to 3.0 gm were determined as a function of time using a passive cavity aerosol scattering probe (PCASP) (Particle Measuring Systems, Inc., Boulder, Colorado) which was mounted on a pilon extending 0.5 m below the aircraft's left wingtip, a locatior•. calculated to be minimally effected by aircraft-induced flow distortion. The PCASP provides 15 bins of size information, with bins of progressively increasing width (e.g., bin 1 is 0.02 mm wide, whereas bin 15 is 0.5 txm wide). This probe has a resistive heater on the inlet which prevents ice formation during penetration of clouds and acts to dehydrate aerosol samples before measurement [Strapp et al., 1992]. This heater was operated at all th]aes during flight, so that measured number densities and size distributions were unaffected by changes in ambient relative humidity.
The PCASP was calibrated by the manufacturer just prior to the experiment using nonabsorbing, spherical latex particles with real and imaginary refractive indices, n r and n/, of 1.59 and 0.0, respectively. This calibration is highly sensitive to deviations of the sampled aerosol optical properties from that of the calibration aerosol [Kim and Boatman, 1990]. For example, Pueschel et al. [1990] found that sulfuric acid particles with n r = 1.44 were undersized by up to 33% in diameter and an average of 71% in volume by a similarly calibrated optical scattering probe. By inspecting the appropriate Mie scattering curves, we estimate that particles <1 grn in diameter with n r = 1.55 and ni = 0.03 (the values adopted below for biomass buming aerosols) would be undersized by 2 to 3% in diameter resulting in 6-10% and 10-16% underestimates of mass and extinction. However, because the exact composition and hence refractive index of particles sampled during TRACE A is uncertain, we have chosen not to apply corrections to PC ASP factory calibration. Data from the PCASP were recorded at 2 s intervals but have been averaged over varying periods of time for purposes of the following presentation. Total aerosol number densities, N, were obtained by surm]aing the counts from all size bins of the PCASP and are presented, except where specified, in the units of number per cubic centimeter at standard temperature and pressure (20øC and 760 torr). Particle volumes were calculated by multiplying the number of counts in each size bin of the PCASP by the corresponding bin volume, then sununing over all pertinent bins. Particle volume measurements were converted to mass assuming a nominal aerosol mass density of 1.0 gcm -3 [Radke et al., 1988].
Regression statistics for the relationship between CO 2 and aerosols and CO in haze layers and plumes are presented in the following tables and discussions. Because of the large number of and 19 were tramsits to and from the experiment area and included encounters with both clean North Atlantic and smoke-tainted South Atlantic air masses. Flights 5 and 8 and 13, 14, and 15 investigated continental outflow over the Atlantic from Brazil and Africa, respectively. Missions 6 and 7 over Brazil and 10 and 12 over south central Africa focused upon obtaining source emission signatures for the two source regions. Flight 17 and portions of 16 and 18 were conducted near Ascension Island to examine the impact of biomass burning upon the composition of rmnote South cases, rather than editing and extracting the individual data seg-Atlantic air masses. ments for analysis, we calculated "running", 30-point (5 min) linear regressions with 25-point overlap on the entire data. Time series exhibiting CO standard deviations > 5 ppbv were judged to contain plumes, and statistics for only such series with r 2 values ß between the variables of interest exceeding 0.45 (p > 0.99) were used in preparing the tables and figures.
Optical properties of aerosols were calculated using a standard Mie scattering algorithm for spherical particles as described by Bohren and Huffman [1983]. This calculation is highly sensitive to the value selected for the index of refraction, and a wide range of values have been reported for the 500-to 550-nm wavelength (e.g., solar maximum) region for biomass burning aerosols. This

Large-Scale Observations
The TRACE A field deployment took place between September 21 and October 26, 1992, near the end of the burning season over both continents. At that time, fire count statistics indicate burning activity was greatly reduced in Brazil because of rainy weather over the interior agricultural regions [Fishman et al., this issue]. Also, a severe draught over southern Africa had limited the growth of vegetation, so that only a fraction of normal parameter is comprised of two terms. The real part, nr, which fuel was available for combustion [Bachmeier and Fuelberg, this controls loss due to scattering, has been estimated to lie between issue]. However, despite this reduced frequency of fires, fine 1.38 and 1.55, with the majority of values grouped between 1.52 aerosol number densities were greatly enhanced, particularly and 1.55 [Li and Mao, 1990; Lenoble, 1991; Westphal and Toon, within the middle to upper troposphere, over the entire tropi-1991]. We have chosen to make our calculations using the value cal/subtropical South Atlantic basin relative to northern of 1.55. A sensitivity study indicates that if 1.52 were correct hemisphere observations (Figures 2 and 3). Indeed, a rough instead of 1.55, the scattering would decrease by 6%. The imagi-comparison of Figure 2 data suggests that values at altitudes nary term, ni, reportedly varies from 0.01 to 0.04 [Li and Mao, above the marine boundary layer were at least an order of 1990; Lenoble, 1991; Westphal andToon, 1991]. We have, based magnitude greater over the southern regions. These loadings, on the reported fraction of elemental carbon in tropical biomass which are also approximately a factor of 10 higher than seen burning aerosols [Andreae et al., 1988], chosen to use 0.03 in our during the wet season over coastal and interior Brazil [Gregory calculation. This gives a single-scatter albedo, OJ(•'afire, where •:a et al., 1990], are consistent with previous South Atlantic dry is the aerosol absorption and Xe is the aerosol extinction), of about 0.83. In addition, we note that a variation of n i from 0.03 to 0.015 increases the scattering by 5%, since the absorption decreases.
The high positive correlation observed between aerosol number and mass concentrations and CO allows us to calculate source emission ratios for these parameters. Table 2   boundary layer (PBL) (the lowest 1-2 km above ground level) by -20% if 1-to 3-grn particles were included in the mass measurements over the Brazilian and African source regions. The calculations (see Table 1). peak aerosol number density and mass mixing ratios reported in the  Table 2, we calculate that the combustion efficiencies of emissions smnpled over both locations ranged from about 91 to 99%, which suggests the respective source fires were generally in the flaming stage of combustion.  (Figure 6a) provide further evidence that vertical transport was more efficient within the Brazilian regime as values above 6 km were, in general, a factor of 2 greater there than over Africa. However, note that even over Africa, the cleanest layers generally contained -10 times more aerosols than recorded within North Atlantic and U.S. continental air masses observed during TRACE A ferry flights (Figures 2 and 3).
Moreover, a comparison between the N and CO profiles suggests that a considerable portion of the smoke particles were lost during the vertical transport process (Figure 6a and 6b). This was not due to humidity effects on the measurement, e.g., a reduction in aerosol diameters below the instrument's size range sensitivity, because, as noted above, the PCASP probe has deicing heaters which dry the particles prior to measurement. More likely, the aerosols were scavenged by precipitation during passage tlxrough convective clouds [Chuang et al., 1992] ~1000 cm -3 anddN/dCO ratios were ~25 cm -3 ppbv -1, which is about the same as seen over continental source regions (Table 2), suggesting very little aerosol is lost during horizontal transport within ttfis preferred outflow band.
Aerosol size distributions within continental outflow ( Figure 8) were very similar to those recorded over source regions (Figure 4). In fact, close inspection reveals that the African volume spectra are almost identical despite the large difference in time since emission (several days). As to the Brazilian data, the accumulation mode peak remains at the same diameter but is somewhat broadened. Also, a slightly greater volume of aerosols are found at diameters >0.7 gm which is likely related to the sampling of African emissions along the northern portion of the flight tracks on flights 4 and 5. Long range effects. Streamline and trajectory analyses indicate air sampled in the region around Ascension Island arrived from diverse locations, depending on altitude and position relative to the island. At high elevations, air along most flight paths origi-nated over South America, whereas near the surface, clean mari-time air from the subtropical South Atlantic was prevalent. Air at midlevels (-2 to 6 km) appeared to originate over southern Africa although most 5-day backward trajectories (the limit of our cal-culations) fell short of reaching the continent.
Because of these convoluted flow patterns, trace species concentrations exhibited wide ranges of values over the region. For example, CO varied from ~70 to 140 ppbv and average 90 ppbv at 2.5 km altitude (Figure 10a) To make the problem more tractable, we assumed that aerosol loadings at all locations were dominated by biomass burning particulates mad that particles of all sizes were of uniform composition and spherical shape (see experimental section above for further details). Although comparisons between calculated and lidar-observed aerosol extinction profiles were generally quite good, we estimate that these assumptions coupled with uncertainties in refractive index and probe calibration produce a maximum uncertainty of +_50% in the calculated optical properties.
Sample results are seen in Figure 11 for a profile recorded through a regional smoke pall over African savanna on flight 10. Simultaneous meteorological soundings (not shown) indicated that inversions at 2.5 and 4 km, a wet layer between 6.5 and 7.5 km altitude, and that, within 50-100 m below the higher inversion, RH exceeded 90%. Aerosol

for 'lJext, Tab s, and Xbs, respectively, and a single-scattering albedo, (•½xt-'rabs)/%xt, of 0.88 which compares reasonably well with that reported for biomass burning aerosols in previous studies
(0.83 +0.11) [Radke et al., 1991]. Recalling that I = I o e -v, where I o and I are the radiation intensities entering and exiting tl•e optical medium, we calculate that for this case, of all 500-nm light incident at 10 l•n, 45% was scattered and 9% was absorbed before reaching 1.5-km altitude; of the amount scattered over this height interval, about 2% was returned directly to space.
The above calculations were repeated upon all TRACE A profiles which covered a significant altitude range and for which RH and aerosol data were available. Results are shown in Table 3 from flight 17 show total mass and optical depths 2 to 3 times greater than the background case. The maximum optical extinction and absorption depths of Table 3, 1.45 and 0.147, respectively, are for data acquired during flight 10, profile 8 through a smoke pall over an African savanna fire region. These values suggest a total loss of 77% of 500-nm radiation within the column, about 10% of which is absorbed and converted to heat energy. Although quite high, these values were, for several reasons, likely much lower than might be measured at a ground site with an upward looking radiometer. In addition to the fact that only aerosols in the 0.1-to 3.0-grn size range were measured and included in the calculation, the DC-8 was generally diverted around large, intense regions of smoke to prevent contaminating sample inlets and chemical instruments which were, for the most part, optimized for measuring trace species ordinarily present at substantially lower concemrations. The aircraft was also restricted to a minimum flight level of 300 m above the surface and higher in cases of reduced visibility. At these altitudes, the low-height vegetation fire plumes were substantially diluted with background air. Indeed, the average DC-8-measured CO mixing ratio within the mixed layer corresponding to flight 10, profile 8 was about 500 ppbv; values much greater than this are  The data of Table 3 can be used to calculate mass-dependent total optical cross sections for biomass burning aerosols. Figure 12 shows a plot of optical extinction depth calculated from both dry and humidity-corrected size distributions versus total dry aerosol loading. The slopes of these plots, 7.0 m 2 g-1 and 9.6 m 2 g-l, represent the mass extinction cross sections for the dry and humidity-corrected cases, respectively. For comparison, Ferrare et at [i990] calculated a value for dry backscatter cross sections for wavelengths of 350, 500, 600, and 1064 nm for dry and humidity-corrected biomass burning aerosol profiles are given in Table 4; these can be used in conjunction with Table 3 (and Table 5) integrated aerosol mass loadings to calculate optical depths at wave, lengths other than the 500-ran shown. The cross section values increase with decreasing wavelengths due to the dominance of submicron particles in total aerosol mass loadings. The lfigh extinction values at the shorter wavelengths suggest that •moke aerosols may greatly reduce the downward flux of photosynthetically active radiation and thus negatively impact the productivity of ecosystems within effected regions: The average 500-nm optical characteristics of South Atlantic basin air masses are summarized in Table 5. These values were computed from averaged vertical profiles of aerosol number/size distributions and integrated from the surface to 12 kin. The height-dependent aerosol size/number distributions were derived from all data recorded over the given region, including that from horizontal flight legs and vertical as, cents/descents and regardless of whether sampling took place within clean background air or pollution-tainted continental plumes. The table is broken into categories of "fine" and "total" and "dry" and "wet" to illustrate the relative contribution of fine (0.i-0.9 grn) aerosols and the effect of humidity on calculated P•ameters. For these cases the fine aerosol comprised a mass fraction ranging from 47% (African outflow region) to 71% (Brazilian source region) of the common a.t surface sites within active burning regions [Kirchhoff, total. This is less than the 80% seen in relatively fresh smoke 1991]. Thus• we suggest that the Table 3 source region values plumes and may be caused by several factors, including a greater may significantly underestimate tile local radiative imp0ct of axe solubility (and hence wet removal rate) for the fine aerosols, a fires.
higher loading of coarse aerosols in the background air due to the dry conditions over the continents, and, particularly over the oceanic regions, input of sea-salt pmticles. Yet despite its reduced relative presence, the fine aerosols accounted for a large majority of the 500-nm optical extinction depth; their fractional contribution to fixis parameter ranged from about 77% to 90%. As noted above, the optical parameters are highly wavelength dependent and the coarse particles are expect. ed to become of more relative importance to extinction of longer-wavelength radiation. For smoke of about 4.5 m 2 g-1 and Radke et al.
[1988] measured a 500-nm light, the coarse mode particles do make a higher relative value of 5 m 2 g-1 in northem midlatitude forest fire plumes; This contribution to the optical absorption depth as the fine aerosol parameter is., of course, inversely dependent upon the aerosol only account for about 60 to 80% of the total. We note, however, mass density. If we had used a dry aerosol density of 1.5 gcm -3, that the fractional absorption of the coarse aerosols may be overwhich is typical for sulfate-type aerosols, the above mass estimated in Table 5 due to our computationally simplifying extinction cross sections would have been 4.7 and 6.4 m 2 g-l, assumption that the index of refraction remained constant across respectively. Values of total mass extinction, absorption, and the entire aerosol size distribution. Although the coarse (or fine) The influence of ambient water vapor on aerosol optical extinction and backscatter depths was significant even though RH in the sampling regions seldom exceeded 60% and all data from within and near clouds were deleted from t_he data set under consideration. These results suggest that the direct radiative impact of tt•e biomass burning aerosols could be greatly amplified in regions of high relative humidity. Indeed, using the RH corrections adopted in this paper and the mass cross-section information of Table 4, we calculate that the optical extinction depth of a parcel containing 0.1 g dry aerosol would increase from about 0.3 to 1.7 in going from 40 to 95% relative humidity. Such mixtures of aerosol loadings and humidities may be common over and within the persistent stratocumulus cloud decks of the eastern South Atlantic and could explain the large optical depths observed by remote sensors over that region during the tropical dry season [Cahoon et al, 1995]. Table $ results, when compared to background values (i.e., Table 3, flight 8, profile 2) suggest incident radiation was influenced by biomass burning aerosols at all locations within the South Atlantic basin during the TRACE A experhnent. The average effect upon total extinction was relatively small in the clearsky region of the middle South Atlantic, where optical depths were perhaps only 50% greater than background levels but significant over continental regions where this parameter was enhanced about an order of magnitude. The increased aerosol absorption depths perhaps had a greater impact upon regional radiation budgets because this parameter is, due to the low value of n i for noncarbonaceous aerosols, ~0 in background air.

Direct Impact on Regional Radiation Budgets
The status of understanding the impact of biomass burning aerosols on the Earth's radiation budget as of 1994 is sununarized by Penner et al. [1994]. Table 3 in that paper has values for a number of the key parameters required for calculating the radiative effects of biomass burning aerosols. While the values were thought to be representative of both temperate and tropical biomass burning, the numbers seem now to be more representative of temperate biomass burning than tropical mixed savanna/cerrado/forest burning. On the basis of recent results in the literature and the measurements and calculations reported here, we would like to propose a new set of values appropriate for Africa and Brazil. The parameters, Penner et al. [1994] values, and our values are presented in Table 6. The aerosol emission factor for the tropics is much lower in our estimate since the savannas and cerrados are characterized more by flaming fires than smoldering fires. The aerosol mass scattering and absorption efficiencies are based on the recent results from biomass burning in southern Africa [Le Canut et al., 1996;Scholes et al., 1996]. They are more representative of burning of savannas than are the results from temperate forests, although emissions from various ecoregions were included in making the determinations.
Using Tables 5 and 6   where Ras is the reflectivity of the combined aerosol surface system; Ra(s ) is the reflectivity of the aerosol layer (surface); and T a is the transmission of the aerosol layer as given by Ta = 1 -R a -Aa, where Aa is absorption by the aerosol layer. We include aerosol layer absorption in the expression since biomass burning aerosols absorb, as opposed to sulfate aerosols, which do not [Charlson et al., 1991]. Sample evaluations of (1) show that for low surface reflectivity compared with the aerosol reflectivity, there is cooling, as would be expected, but that for high surface reflectivity, there can be warming, since the aerosol layer absorbs some of the incident radiation. We can solve (1) using the above data to esthnate the change in radiation budgets for the TRACE A study area, an aerosol layer in central Africa. We first multiply the coefficients of Table 5 by a factor of 0.73 which is an estimate of the ratio of integrated average extinction over the entire solar spectrum to that calculated for 500 nm; this correction is necessary because, as pointed out by Kiehl and Briegleb [1993], use of the values appropriate for the solar maximran tends to overestimate the impact of biomass burning aerosols on the Earth's radiation budget. For surface albedos we use values for black earth or sea (0.06), green savanna (0.15), brown savanna (0.2), and stratus clouds (0.6) from Budyko [1956] and Peixoto and Oort [1992]. For the aerosol layer we assume that 30% of the scattered radiation is back into the hemisphere from which the incoming radiation came [Penner et al., 1994] Table 7, where negative values mean more radiation is reflected by the combined system. In Table 8 the changes in the atmospheric radiation budget are given, where absorption by the biomass burning aerosols as well as absorption by the atmosphere for the extra reflected radiation are combined. For the changes in atmospheric radiation balance we assumed that the atmospheric absorption was 10% of the change in Earth-surface insolation: for dry atmospheres, such as over Africa, water vapor absorption would be low; for moist atmospheres, such as over Brazil, absorption by water vapor for incoming solar radiation would reduce the mnount of radiation that could be absorbed on the way back to space.
Thus we see that tlm average biomass burning plume reduces surface radiation by 10-25 W m -2 over land and sea but increases' surface insolation over moderately thick clouds. The reduction over land and sea surfaces varies to 2.4 W m -2 for light aerosol loading. In addition, the aerosol layer absorbs some of the radiation that would have been absorbed and/or reflected by the Earth's surface, as does the entire atmosphere. The sum of the atmospheric and aerosol layer absorption always more than compensates for reduced surface absorption, except over very low albedo surfaces, due to the absorption by the aerosols.

Indirect Impact on Regional Radiation Budgets
In addition to the direct effects discussed above, we have also noted that changes in aerosol (CCN) abundance can indirectly where x is cloud optical depth, N is the CCN number density, and R s is the cloud reflectance as described above. Pollution-related albedo changes may, in particular, have serious radiative consequences for the central/eastern South Atlantic region where large spatial regions (>10,000 kan 2) are frequently covered by a low-level stratocumulus cloud deck. As shown in Plate 1, aerosol-enhanced, biomass burning plumes from Africa were often observed to overlie these clouds and the relatively rapid subsidence prevalent along the eastern and northern periphe W of the subtropical oceanic high coupled with radiative cooling

Summary and Conclusion
The above presentation indicates that aerosol loadings over all regions of the tropical South Atlantic basin visited during TRACE A were significantly enhanced as a result of biomass burning occurring both in South America and in Africa. Results of simple calculations strongly suggest that these enhancements had a significant direct and indirect impact on regional radiation budgets. We note, however, that TRACE A was conducted at a time when burning activity was greatly reduced in Brazil because of unusually wet conditions and Africa because of draught. Thus it is likely that the observed fine aerosol loadings may be significantly lower than seen during the peak of typical dry seasons over the South Atlantic basin. Thus it seems reasonable to anticipate that our results could significantly underestimate aerosol spatial distributions and impacts at the peak of fire activity in a typical burning season.
Clearly, additional, longer-term research is warranted to investigate the physical/chemical characteristics of these aerosols and their influence on incident radiation fields. Compositional, morphological, optical, and more extensive size distribution measurements of these aerosols as a function of fuel-type and fire conditions are particularly needed in order to reduce uncertainties in the mass and direct radiation budget calculations. A comprehensive investigation of biomass burning emission influence on cloud microphysical properties should also receive high priority in order to place constraints on the indkect impact of t_he aerosol on hydrologic cycles and surface albedo.