Influence of plumes from biomass burning on atmospheric chemistry over the equatorial and tropical South Atlantic during CITE 3

. During all eight flights conducted over the equatorial and tropical South Atlantic (27ø-35øW, 2øN-11øS; September 9-22, 1989) in the course of the Chemical Instrumentation Test and Evaluation (CITE 3) experiment, we observed haze layers with elevated concentrations of aerosols, 03, CO, and other trace gases related to biomass burning emissions. They occurred at altitudes between 1000 and 5200 m and were usually only some 100-300 m thick. These layers extended horizontally over several 100 km and were marked by the presence of visible brownish haze. These layers strongly influenced the chemical characteristics of the atmosphere over this remote oceanic region. Air mass trajectories indicate that these layers originate in the biomass burning regions of Africa and South America and typically have aged at least 10 days since the time of emission. In the haze layers, 03 and CO concentrations up to 90 and 210 ppb were observed, respectively. The two species were highly correlated. The ratio AO3/ACO (A, concentrations in plume minus background concentrations) is typically in the range 0.2-0.7, much higher than the ratios in the less aged plumes investigated previously in Amazonia. In most cases, aerosol (0.12-3/zm diameter) were also strongly correlated with elevated CO levels. Clear correlations between of the in which most would have already reacted away within of to variable sinks for NOy. The average enrichment of ANOy/ACO was quite high, consistent with the efficient production of ozone observed in the plumes. The chemical characteristics of the haze layers, together with remote sensing information and trajectory calculations, suggest that fire emissions (in Africa and/or South America) are the primary source of the haze layer components.


Introduction
Satellite studies have shown high levels of ozone in the troposphere over the tropical and equatorial South Atlantic [Fishman and Larsen, 1987;Fishman et al., 1990Fishman et al., , 1991. Ozone soundings made at Natal on the east coast of Brazil also show the presence of layers with elevated ozone concentrations in the middle and upper troposphere [Kirchhoff and Nobre, 1986;Kirchhoffet al., 1990Kirchhoffet al., , 1991. At Ascension Island, in the center of the tropical South Atlantic, the same enhancement of ozone in the middle and upper troposphere has been observed . This phenomenon is most pronounced during the southern hemisphere dry sea-  [Hoell et al., 1993]. In this paper we present the results of our airborne measurements of trace gases and aerosols in tropospheric haze layers over the equatorial and tropical South Atlantic. We use air mass trajectories and other meteorological data to determine the origin of these haze layers.

Methods
In situ measurements (ozone, carbon monoxide, nitrogen oxides, and aerosol particles) and grab sampling (nonmethane hydrocarbons (NMHC)) were conducted on the NASA Electra research aircraft, which was based at Natal, Brazil, for the second half of the CITE 3 experiment (September 9-22, 1989). The research flights took place in the region between longitude 27 ø and 35øW and latitude 2øN and 11øS. Data from the transit flight to Natal (flight 12) also showed evidence of plumes from biomass burning in the region from 50 ø to 35øW and will therefore be included in this paper. The tracks of flights 12-19 are shown in Figure 1; the shaded areas along the flight tracks indicate the regions where our measurements indicated the presence of elevated amounts of 0 3 , CO, and aerosol from biomass burning.
The sampling and analytical methods employed for the collection of data on the haze layers are described in detail in other publications by the members of the CITE 3 science team; therefore they will be outlined only briefly in the following paragraphs.

Ozone
Ozone concentrations were determined using a C2H 4 chemiluminescence instrument which had been modified to compensate for altitude variations [Gregory et al., 1983]. The instrument was calibrated using the NO gas phase titration technique with National Institute for Standards and Technology (formerly National Bureau of Standards) traceable calibration gases. It has a response time of about 2 s to 90% of reading, a precision of 2 ppb or 2% (whichever is largest) for a 10-s average, and an absolute accuracy of 5 ppb or 5%. Sample air was delivered to the instrument via a Teflon-lined inlet which extended about 30 cm out from the aircraft skin into relatively undisturbed airflow. Data were recorded at 1-s intervals; 10-s averages were used for the profiles and analyses presented in this paper.
Carbon Monoxide CO measurements were provided by the differential absorption CO measurement (DACOM), a diode-laser-based instrument designed to measure CO aboard an aircraft platform. The instrument determines CO by modulating the laser wavelength across an isolated absorption line in the 4.7-ttm band and then detecting the periodic attenuation experienced by the laser beam due to the absorption by CO molecules in the White cell. The pressure in the White cell is maintained at 50 torr to narrow the CO absorption line and other potentially interfering absorption lines. An inlet probe is extended past the slipstream of the aircraft to capture the sample air, which is then passed through a chemical dryer to remove H20 vapor by reaction with Mg(C104). The sample air then enters the pressure-controlled sample cell which contains a 12.5-m folded optical path and then is continuously pumped overboard. Approximately 20% of the diode laser power is directed to a wavelength stabilization leg which provides an error signal when the laser wavelength sweep drifts from the CO absorption line center. The instrument is calibrated approximately every 15 min by passing a calibration gas of known CO concentration through the sample system. Precision and accuracy are improved by interpolating between these frequent calibrations. Power fluctuations are removed by dividing the differential absorption signal by the transmitted power from the laser, which can be measured using an optical chopper and a lock-in amplifier.

Nitrogen Oxides
Nitric oxide, NO2, and NOy were measured simultaneously with the Georgia Tech two-photon/laser-induced fluorescence (TP/LIF) instrument [Bradshaw et al., 1985;Sandholm et al., 1990]. This spectroscopically selective NO technique was used to simultaneously determine NO, NO produced from the photolysis of NO2, and NO produced from the reduction of NOy compounds using a 300øC gold catalytic surface with 0.3% CO as a reducing agent. A 1-kW photolytic converter was operated with a photolysis passband of 350 nm < A < 410 nm, a photolytic yield ranging from 30 to 60%, and sample residence times ranging from 2 to 4.5 s. A porcelain-glass-coated inlet was used to sample ambient air in an orientation perpendicular to the airstream. These NxOy measurements were reported using an integration time of 180 s. Accuracy of the instrument calibration is estimated to be -+ 16% for NO and _+ 18% for NO 2 and NOy at the 95% confidence limit. Limits of detection for a 180-s signal integration were ---3 ppt for NO and ---10 ppt for NO 2 with a signal-to-noise ratio of 2:1. The typical measurement precision for NOy (at 95% confidence limit) was approximately _+ 10% at 500 ppt increasing to approximately _+20% at 200 ppt [Sandholm et al., 1990[Sandholm et al., , 1992. The NOy measurements included some fractional component of fine particulate nitrate-containing aerosols. The plume layers generated a memory effect in the unheated portion of the NOy sampling line. Measurements adversely affected by this phenomenon were not included in the analysis of the data.

Aerosol densities as a function of time and size diameter
were measured with two optical scattering probes mounted externally to the aircraft. Both probes were manufactured and calibrated by particle measuring systems (Boulder, Colorado). An active scattering aerosol spectrometer probe (ASASP) was used to monitor 0.12-to 3.1-txm-diameter particles, while a forward scattering spectrometer probe (FSSP) was used for the 0.5-to 8.0-txm range. According to the manufacturer's specifications, the ASASP, or "small" aerosol probe, classifies particles into 15 size bins of progressively increasing width: bin 1 is 0.025 txm wide, bin 2 is 0.050 txm wide, and bin 3 is 0.075 txm wide. The FSSP, or "large" aerosol probe, also provides 15 channels of information, but it sizes particles into 0.5-txm wide bins. In both probes the relative humidity inside the measurement cavity is close to the ambient value. Profile and size distribution plots were prepared from 10-and 60-s averaged data, respectively. In the case of the size distribution plots, ASASP data were used for the 0.12 to 0.5 txm range while FSSP data filled the 0.5-to 8.0-txm range.

Nonmethane Hydrocarbons
The analytical details for the determination of nonmethane hydrocarbons wer•e described in detail by Blake et al. [ 1992]. Samples for the determination of nonmethane hydrocarbons were collected in specially manufactured and cleaned stainless steel canisters. The air sample in the canister was Compressed to a pressure of 2.7 bar using a metal-bellows pump. Samples were analyzed by gas chromatography within 4-7 days after collection. A dried air sample from Niwot ridge was used as a secondary standard for daily calibrations. This sample had been calibrated against artificially prepared reference standard mixtures. The accuracy of the determinations was about 5%.

Meteorological Products
A variety of meteorological products, including GOES and METEOSAT satellite imagery, isentropic back trajectory analysis, streamline and wind vector fields, and soundings, were used to describe salient meteorological features observed during CITE 3. The trajectory analysis procedure was developed by Danielsen [1961] and further refined by Haagensen and Shapiro [1979]. The package inputs both NMC (National Meteorological Center) derived fields and soundings and was used to generate 7-to 10-day back trajectories for this and other papers.

Meteorological Overview
The meteorological environment during the CITE 3 experiment is discussed in detail by Shipham et al. [1993]; only a brief overview will be presented here. The dominant synoptic feature over the South Atlantic Ocean during the study period (September 9-22, 1989) was the South Atlantic subtropical anticyclone, which generally transported air across the tropical Atlantic toward eastern Brazil. In the lower troposphere (850 hPa, --• 1.5 km), the mean axis of the South Atlantic subtropical anticyclone was located between 18 ø and 26øS latitude, while in the middle troposphere (700-500 hPa, --•3-6 km) the mean ridge axis was situated farther north, between 12 ø and 20øS latitude. Pronounced subsidence along the northern periphery of the South Atlantic subtropical anticyclone contributed to the well-defined trade wind inversion and the highly stratified nature of the atmosphere which was frequently observed over the study area.
With this circulation regime a prevailing deep easterly to southeasterly trade wind flow was present in the vicinity of Natal, with a long low-level fetch over the tropical South Atlantic. A westerly flow off the South American continent dominated south of the ridge axis. Some high-altitude recirculation of air from South America was observed, as was cross-equatorial transport which had come from northern Africa.
Variations in the relative positions and intensities of circulation features over South America, Africa, and the South Atlantic basin led to pronounced differences between the transport pathways of air sampled during the seven Natal flights. Intense midlatitude cyclones propagating eastward across the South Atlantic perturbed the location and strength of individual subtropical anticyclone centers. Disturbances also moved westward across the tropical Atlantic, initiating episodes of cross-equatorial flow from Africa.

Air Mass Origin and Transport
As discussed in detail below, haze layers originating from biomass burning plumes were identified during flights 12-19 based on elevated levels of aerosol particles, CO, 03, and other trace gases. In the following paragraphs we will describe the origin and transport paths of these smoke-laden air masses using some typical examples. The first such layer was encoun-  I  I  I  I  I  I  I  I  I  I  I  I   20  40  60  80  100  120 03, co, ppb Streamline analysis [Shipham et al., 1993] suggests that these air masses may have originated in southern Africa.

Visual Observations
Visible haze layers were present on all flights in the study region, with the obvious exception of flights 15 and 17, which were conducted at night. The color of these layers ranged from almost pure white to brown (Figure 9). They could be seen throughout a wide range of altitudes, from about 1 km to above 5.2 km, the highest level reached by the aircraft. They were most frequent just above the tops of the trade wind cumulus, in the altitude range from 1.5 to 3 km. Seen from the side, the layers appeared quite thin, frequently with sharp edges above and below, and seemed to extend to the horizon. Individual layers could often be traced over some tens of kilometers during horizontal flight legs. The chemical fine structure of the layers, which is evident in the CO, 03, and aerosol soundings discussed below as a series o7 multiple peaks, is seen as multiple layers when viewed from the side.

Ozone and Carbon Monoxide
From the chemical measurements the presence of the biomass-burning-derived plumes was most clearly evident in the closely correlated peaks of 03 and CO concentr'ations observed during soundings obtained during climbs or descents of the aircraft. In the course of the study period we identified 34 cases where haze layers were penetrated during soundings, sometimes with several sublayers present within one major layer. Figure 10  suggesting that convective activity has distributed combustion-derived material throughout the planetary boundary layer.
In addition to the vertical profiles discussed above, measurements on the plumes were also made during horizontal flight legs. A total of 15 horizontal flight segments were conducted within the haze layers. Figure 11 shows some examples of the type of data obtained. Again, we find CO and 0 3 to be highly correlated. To quantify the observed correlation between 0 3 and CO, we have performed regression analyses on our data. Here, we have used the same procedure as we have applied previously in our study on haze layers over the Amazon basin [Andreae et al., 1988]. In this analysis, it is assumed that the air in the plume mixes with the air above and below. The chemical composition of the plume can then be characterized by the regression slope of 03 versus CO, the "ozone production ratio" (AO3/ACO), which represents the number of 0 3 molecules formed in the plume per molecule of CO emitted in the combustion process (assuming that only a negligible amount of CO has been photooxidized during transport). To correct for the gradual increase of 03 concentration with height in the background atmosphere, a line is fitted to the background, and the background values are subtracted from the ambient concentrations. Where appropriate, an analogous correction is performed on the CO profile. No such correction is necessary for the data from level flight legs.
The results of this analysis are summarized in Table 1 Delany [ 1990] shows that 03 concentrations in diluted smoke from biomass burning in the midtroposphere keeps increasing over 14 days, while the CO concentration remains almost constant. The plumes investigated here have been in the atmosphere for at least 7 days in the case of the plumes recycled from South America, but in most cases they have been exposed to photochemical processing for over 10 days. In contrast, the transport times of the plumes we sampled in Amazonia was of the order of 1 day, and those sampled in the northern Congo were about 4 days old. An equally important factor, however, may be the NOx/CO ratio in the emissions. In fact, Jacob et al. [1992] have suggested that the NOx/ NMHC ratio in the emissions is likely to be the controlling factor in the photochemical production of 03 in the plumes and that for typical NMHC/CO ratios,/XNOx//XCO emission ratios of the order of 0.1 or greater are required to lead to the levels of 03 production observed here. This issue will be discussed in more detail in the section on nitrogen oxides.

Aerosols
Aerosol particles were also found to be enriched in the plumes, consistent with the presence of visible haze. The enrichment is most clearly seen in the data from the ASASP probe, which determines particles in the 0.12-to 3.1-/am diameter range. The sum of the particles present in this size range, Np, is plotted together with the 03 and CO data in several micrometers. The presence of a mode in a similar size range is also evident in the measurements over the fresh savanna fires and may be related to entrainment of soil and ash particles due to the turbulence created by the flames. Entrainment of sea-salt aerosols during convective events over the ocean may also have contributed to this mode. Since particles smaller than 0.12 tam, which are not detected by the ASASP, are also expected to be present, we compared the results from the ASASP probe with the available data on total particle (condensation nucleus (CN)) concentrations as determined with a TSI 3020 Aitken nuclei counter (R. J. Ferek, personal communication, 1989). The ratio between ASASP counts and CN counts, Np/NcN, ranges from 0.024 to 1.23 and depends strongly on relative humidity (Figure 13). At the lowest humidities observed in the haze layers (below ---10%), the ASASP probe sees only a small fraction of the particles present (---2-20%), and consequently there is no significant correlation between ASASP count and CO concentration (Table 1)    From the particle number versus size distributions provided by the ASASP instrument, we have calculated the aerosol mass concentration. We assume a density of 1 g cm -3 for the aerosol particles, which is a reasonable approximation for particles consisting to a large degree of organic material. Densities near 1 g cm -3 were observed for smoke aerosol particles by Stith et al. [1981] and Radke et al. [1991]. Statistical analysis of the data obtained in the haze layers shows that the aerosol mass in the size fraction below 1 /xm diameter is highly correlated with CO concentration ( Table 1)  is absent in the haze layers which are well separated from the boundary layer, most of the aerosol mass is in the submicron fraction, and the regression results become almost identical for total or submicron aerosol against CO. This is consistent with the aerosol size spectra observed in the plumes from CITE 3 and with literature data on the size distribution of aerosols from biomass fires [Radke et al., 1978[Radke et al., , 1988[Radke et al., , 1991Stith et al., 1981;Holben et al., 1991].

Nonmethane Hydrocarbons
Six of the hydrocarbon canister samples collected during CITE 3 were taken within plume layers. Unfortunately, they all missed the more concentrated parts of the plumes, and they contain relatively low concentrations of hydrocarbons. CO concentrations during the NMHC sampling times are only 10-40 ppb over the background values. Nevertheless, slight enrichment of a series of hydrocarbons relative to the unpolluted boundary layer and to the free troposphere above 3000-m altitude are evident ( Table 2). The hydrocarbon distribution shows the acetylene, ethene, and ethane enrichment typical of pyrogenic emissions [Greenberg et al., 1984;Rudolph et al., 1992]. The alkenes are less enriched relative to the alkanes than in fresh smoke, consistent with an advanced plume age.

Nitrogen Oxides and Ozone Formation
The concentrations of NO, NO2, and NO r in the plumes are given in Table 3. Given the elevated concentrations of other pyrogenic pollutants in these plume layers, the concentrations of NO and NO2 are quite low, 8 -6 and 37 -17 ppt, respectively. Clear correlations between pyrogenic tracer species and NOx enrichments were not apparent. The low NOx levels and the absence of correlations between NOx and CO is probably due to the age of the plumes, in which most NO x would have already reacted away within 1-2 days. The low ratio of NO to 03 measured in most of the plumes suggests that ozone production has ceased, or that ozone may even be consumed in some cases.
In contrast, NO r concentrations were high in most of the plume layers, and in spite of the limitations imposed by the time resolution of the NO r data, correlations with other pyrogenic species could in several cases be obtained. Table   3 contains ANOr/ACO enhancement factors for these cases (average 0.053 -0.027). Since most of the NO r in the plumes presumably originates from the conversion of NOx emitted by the fires, this ratio provides an indication for the original emission ratio ANOx/ACO. However, given the age of the CITE 3 plumes, the original ratio is difficult to deduce with certainty from the observed ANOr/ACO ratios. Assuming that some loss of NO r has occurred due to aerosol formation and precipitation removal, these ANOr/ACO ratios would be consistent with initial ANOx/ACO of up to 0.1. Such high NOx enrichments would explain the efficient ozone formation observed [Jacob et al., 1992].
In general, our ANOr/ACO values are substantially larger than most of the values previously reported from forest burning [Andreae, 1993a, and references therein], e.g., those found for smoldering tundra fires in the subarctic (0.006 -0.006 [Wofsy et al., 1992]) and deforestation fires in Amazonia (0.016 [Andreae et al., 1988] [Wofsy et al., 1992]. This would be consistent with the high ozone production efficiency observed in the CITE 3 plumes. Indeed, the ozone production efficiency, expressed by the ratio AO3/ACO, increases with increasing ANOy/ACO in the plumes (Figure 14). Around September, savanna burning in southern Africa, cerrado (savanna) fires in Brazil, and grassland fires in the pampas of South America reach their seasonal maxima [Kirchhoff and Rasmussen, 1990;Andreae, 1993b] ment), consisting of the two components TRACE A (Transport and Atmospheric Chemistry near the Equator, Atlantic) and SAFARI (Southern Africa Fire/Atmosphere Research Initiative), showed that smoke plumes containing high amounts of ozone, aerosols, and carbon monoxide were indeed present throughout the troposphere in the South Atlantic region (J. Fishman et al., personal communication, 1993). These results suggest that photochemical ozone production in the smoke plumes emitted from the vast areas of savannas and tropical forest, which are subjected to burning every year, represents a major source of ozone to the tropical troposphere. Furthermore, the smoke aerosol particles contained in these plumes may make a significant contribution to the radiative properties of the atmosphere in the tropics and to the concentrations of cloud condensation nuclei (CCN). The perturbation of the radiation budget due to aerosol backscatter and due to the modification of cloud albedo by increased CCN concentrations may have a considerable effect on regional and global climate [Penner et al., 1991[Penner et al., , 1992Andreae, 1994].

Overview and Conclusion
The preceding discussion has shown that a number of atmospheric species which originate in biomass burning (carbon monoxide, nonmethane hydrocarbons, oxides of nitrogen, and smoke aerosol particles) were present over a large fraction of the area investigated during CITE 3. The analysis of air mass trajectories and streamline charts points to savanna and grassland fires, predominantly in Africa but also in South America, as the source of these pyrogenic species. As a result of photochemical reactions during the prolonged transport of these smoke-laden air masses, substantial amounts of ozone are formed, resulting in a considerable enrichment of ozone in the troposphere.
In the same region, satellite analyses have shown an enhancement of tropospheric ozone by about 15-20 Dobson units (DU) over the tropical background value during the September-October period [Fishman et al., 1991]. From a statistical analysis of the O• data collected during CITE 3, Anderson et al. [1993b] derive an average ozone column concentration of 13.$ DU in the lower troposphere (up to 3.3 kin, the maximum altitude for most flights). Based on a comparison with wet season data, they deduce that during CITE 3, ozone in this layer was enhanced by about 8-9 DU over the average wet season value and that about $0% of this enhancement can be attributed to ozone formation from pyrogenic precursors. Clearly, this photochemical ozone source and the resulting plumes in the lower troposphere alone cannot explain the l$-20-DU ozone enrichment observed by remote sensing. Ozone sondes launched during CITE 3 showed that elevated O• levels were present throughout the troposphere, resulting in a tropospheric ozone column of 39-$1 DU, in good agreement with the satellite results of Fishman et al. [1991]. Since no pyrogenic tracers are available to establish the source of these ozone enriched layers (photochemical production or downward mixing from the stratosphere), the role of biomass burning in the formation of the ozone maximum over the South Atlantic cannot be determined on the basis of the results from CITE 3. Recent work during the 1992 STARE (Southern Tropical Atlantic Regional Experi-