Origin of ozone and NO x in the tropical troposphere: A photochemical analysis of aircraft observations over the South Atlantic basin

,


Thompson et al. [this issue], the tropospheric 03 column over the South Atlantic during the dry season maximum is only 50% higher than the wet season background when biomass burning
influence is minimal. This 03 background, over the South Atlantic and elsewhere in the tropical troposphere, plays a critical role in determining the oxidizing power of the atmosphere.
Our analysis is based on a combination of photochemical modeling and statistical interpretation of the TRACE A data. The modeling approach is described in section 2 and is evaluated in sections 3 and 4 by simulation of H202, CH3OOH, and CH20 concentrations measured aboard the aircraft. Regional budgets for 03 and NOn over the South Atlantic basin are derived in sections 5 and 6, respectively. A general representation of the 03 budget in the tropical troposphere is proposed in section 7. Conclusions are in section 8.

Modeling Approach
The TRACE A expedition used a DC size distributions were also measured. Table 1 gives median values for the various measurements at different altitudes.
We base our analyses on merged time series of the aircraft data where all measurements have been averaged over a common time interval. The choice of averaging interval must balance the need for high temporal resolution with the danger of misinterpret- Statistics are for the ensemble of southern hemisphere data collected in TRACE A. The median concentrations of >C5 alkanes, >C3 alkenes, xylenes, isoprene, and terpenes were below the detection limit of 2 pptv. Concentrations of NO2 were measured in TRACE A, but the measurement is considered unreliable [Crawfin'd et al., 1996] and is not used in the paper.
a Measured with an Eppley radiometer (Eppley, Inc., Newport, Rhode Island). Twenty-four hour average model values computed as described in section 2. The concentration of NOx is calculated in the model using as constraint the NO measurement for the time of day of observation.
ing coarse-resolution data. We choose to average the data over the sampling time of the critical variable with the coarsest time resolution. Because this critical variable depends on the problem at hand, we will use over the course of this paper four different merged time series matched to the sampling times of peroxides, CH20, NO, and HNO3 (Table 2). Ozone columns are included in the merged time series using daily observations with 1øxl.25 ø resolution from the total ozone mapping spectrometer (TOMS). Calculation of Ox production and loss rates requires knowledge of the concentrations of HOx radicals (OH, HO2, RO?) along the flight tracks. These species were not measured from the aircraft. We estimate their concentrations for each point in the merged time series by using a zero-dimensional photochemical model constrained with the ensemble of measurements for that point (Table 3). The model is described in Appendix A.

The median 03 column over the south tropical Atlantic during TRACE A was 281 Dobson units (DU) (1 DU = 2.687x10 •6 molecules cm -2). A major objective of this paper is to derive a chemical budget for 03 in the tropospheric column over the South Atlantic basin (section 5). Our approach is to calculate 24-hour average 03 production and loss rates for individual points in the merged
It calculates the diel steady state concentrations of HOx radicals, short-lived NOy species (NO, NO2, NO3, N205, HNO2, HNO4, organic nitrates), and other photochemical intermediates (peroxides, carbonyls). Production and loss of species are by chemical reactions only (loss by deposition is riot considered). "Diel steady state" is defined by reproducibility of concentrations over a 24-hour solar cycle, as obtained in a time-dependent calculation with periodic boundary conditions Ci(t) = Ci(t + T) where C• is the concentration of species •, t •s time, and T = 24 hours. In this calculation, all input variables in Table 3 Table 3 be available for at least a fraction of the time interval. We also give in parentheses the number of model points obtained in the fixed-hydrocarbon approximation, which relaxes the constraint on availability of data for hydrocarbons and acetone (see section 2). a Comparison of simulated and observed peroxide concentrations (section 3). b Comparison of simulated and observed CH20 concentrations (section 4). c 03 production and loss rates and NO2 concentrations (sections 5 and 6). d Chemical budget of NOx (section 6). varies quadratically with the concentration of HO2, the principal peroxy radical responsible for 03 production. At low altitudes where H202 is the principal sink of HOx, the concentration of H202 reflects the rate of the O(ID)+H20 reaction which is the principal source of HOx and the principal sink of Ox. We will also apply the model to study the chemical cycling of NOx over the South Atlantic basin in light of the concurrent measurements of NO, HNO3, and PAN made aboard the aircraft (section 6). Oxidation of NOx to HNO3 and PAN takes place on a timescale of the order of 1 day: We compute the 24~hour average rates of reactions (8)-(14) in the model at individual points using the observed concentrations of NO, HNO3 and PAN as constraints, as discussed above. The resulting mass balance between sources and sinks of NOx allows us to estimate the degree to which NOx concentrations are maintained by chemical recycling from HNO3 and PAN through the above set of reactions. In such a NOx budget calculation it is necessary to correct the rates of conversion between NOx and PAN to account for cycling within the (CH3CO3+PAN) family; the correction is described in Appendix B. Table lb gives the median concentrations of OH, HO2, CH•O 2, NOx, and HNO4 computed in the model for the southern hemisphere TRACE A data set. Diel steady state for these species and for CH20 is typically approached to within 10% after 2 days of integration in the model. The peroxides, which have longer lifetimes, require typically 2~3 days at 2 km and 1 week at 8-12 km for approach to steady state. The steady state assumption is prone to large errors in pollution plumes and in the continental boundary layer, where concentrations of the input variables in Table 3 may change greatly over timescales of less than 1 day. It is more adequate in the free troposphere and in the marine boundary layer. To conduct the model calculation for a given point in the merged time series we require that data be available for all variables in Table 3. Because of lack of temporal overlap in some of the measurements (notably NO, hydrocarbons, acetone), model calculations can be done for only a small fraction of the data points (Table 2). To increase the number of points, we use in certain applications a "fixed-hydrocarbon" version of the model in which we assume 630 parts per trillion by volume (pptv) C2H 6 and 650 pptv acetone (mean values in TRACE A) and neglect the other nonmethane hydrocarbons. By relaxing the constraint on the availability of hydrocarbon and acetone data, the number of model points is greatly increased ( Table 2). The errors incurred by the fixed hydrocarbon approximation will be discussed in the context of each application. Data for HNO3 and PAN are critical only for the NOx mass balance analysis presented in section 6. For all other applications, when data for HNO3 are missing a low value is assumed; and when data for PAN are missing the PAN concentration is assumed from chemical steady state. We verified that these assumptions have a negligible effect on HOx levels and other relevant model results. The successful simulation of the peroxides below 8-km altitude lends confidence in the 03 production and loss rates computed by the model. At 8 to 12-km altitude the-40% underestimate of H202 could imply an -20% underestimate of Considering that HO 2 accounts for-90% of total peroxy radicals in that altitude range (Table lb), we may expect the O3 production and loss rates above 8 km to be underestimated by about 20%.

Peroxides
Singh et al. [ 1995] pointed out the importance of acetone photolysis (included in our model) as a primary source of HOx in the upper troposphere. We conducted a sensitivity calculation without acetone for the point representative of median conditions at 8-12 km (Table la). The H202, CH3OOH, HO 2, and OH concentrations computed for that point decreased by 27% , 36%, 14%, and 16%, respectively, relative to the simulation including acetone. Neglecting acetone as a source of HOx makes the model underestimate of peroxide concentrations in the upper troposphere significantly worse. The model captures 62% of the observed variance. Model values tend to be too high in the lower troposphere (0-4 km) and too low in the middle troposphere (4-8 km). Most observations above 8 km are below the detection limit of 40 pptv. Our ability to interpret model results for CH20 is evidently hampered by the small number of model points. To address this problem, we turned to the fixed-hydrocarbon version of the model which provides 926 model points for comparison with observations. The fixed-hydrocarbon approximation gives only a lower limit for CH20, because it does not allow for the possibility of high hydrocarbon concentrations, but it still allows investigation of model overestimates. We find that the worst overestimate is in the marine boundary layer over the tropical South Atlantic (mean observed conditions 26 parts per billion by volume (ppbv) 03, 66 ppbv CO, 3 pptv NO, 0.015 vol/vol H20 ), where the mean observed concentration of CH20 is 0.11 ppbv (range <0.04-0.25 ppbv) while the corresponding model mean is 0.41 ppbv (range 0.28-0.55 ppbv). The mean observed H202 and CH3OOH concentrations in that region are 2.3 ppbv and 1.1 ppbv, respectively, while the corresponding model means are 3.1 ppbv and 1.7 ppbv. The model shows some overestimate for the peroxides, which in the case of H202 could be due to deposition [Heikes et al., this issue], but these discrepancies are small compared to CH_•O. We repeated the model calculation for the marine boundary layer points using observed concentrations of H20: and CH3OOH as constraints, and assuming a deposition rate constant of 4x10 -6 s -I for CH:O as representative of a deposition velocity of 0.4 cm s -• over the ocean [Thompson and Zafiriou, 1983]. The mean CH:O concentration in this calculation was 0.29 ppbv (range 0.21-0.48 ppbv), still 3 times higher than observed. The root of the discrepancy must therefore lie in the simulation of CH20 chemistry.
Our model results for CH20 can be viewed as typical of the current generation of photochemical models. We participated recently in an intercomparison of 20 models organized by the Intergovernmental Panel on Climate Change [Prather et al., 1995]. In a test simulation representative of the remote marine boundary layer we obtained a CH20 concentration of 0.21 ppbv, while other models obtained values ranging from 0.13 to 0.33 ppbv. The models at the low range had unusually high CH20 photolysis frequencies.
We compared the marine boundary layer measurements of CH•_O in TRACE A to prior shipboard measurements made over the tropical Atlantic in the same season. Lowe and Schmidt  [ 1980]. These authors found evidence that CH20 in the aerosol is predominantly present as CH20 complexes, but they also found that the aerosol contributes only a small fraction of total atmospheric CH20.
In summary, our results point to a major gap in current understanding of CH20 concentrations in the marine boundary layer. Either some of the CH20 measurements are flawed or a large CH•_O sink is missing from the models under some conditions. Liu et al. [1992] previously noted that CH20 concentrations measured in free troposphere air at Mauna Loa, Hawaii (3.4-km altitude) were a factor of 3 lower than calculated from a standard photochemical model and attributed the discrepancy to a model overestimate of OH. However, later measurements by Zhou et al. [1996] at Mauna Loa indicated that the discrepancy is driven by a model overestimate of CH3OOH and that CH20 concentrations calculated in a model constrained with observed CH3OOH are in good agreement with observations. A global three-dimensional photochemical model study by Brasseur et al. [1996] shows, in fact, good simulation of both CH3OOH and CH20 at Mauna Loa. We cannot draw a parallel between Mauna Loa and the marine boundary layer in TRACE A, because the discrepancy between model and observations for CH20 in TRACE A is far greater than for CH.•OOH; also, as discussed above, there is no evidence in TRACE A for a model overestimate of CH20 in the tree troposphere. We searched the TRACE A data set for conditions that would approximate the We see from Plate 1 that the gross production of 03 decreases on average with altitude while the net production increases with altitude. Median vertical profiles of Po• and Lo• for the southern hemisphere TRACE A data are plotted in Figure 4, and a summary 03 budget for the 0 to 12-km column is given in Table 4. Both production and loss decrease with altitude, due to decreasing humidity, but production decreases slower than loss above 6 km because The above analysis implies a strong sensitivity of tropospheric 03 over the South Atlantic basin to the supply of NOx. We show in Figure 6 the dependence of the net 03 production rate on NOx concentrations in the model for points representing the median conditions observed in TRACE A at 0 to 4 km, 4 to 8 km, and 8 to 12-km altitude (Table l a). Even at the relatively high NOx concentrations observed in TRACE A, 03 production is still NOx limited throughout the 0 to 12 km column. A number of authors have noted that NOx levels in the remote troposphere are near the turnover point for net 03 production versus net loss [Chameides et al., 1987[Chameides et al., , 1990; Jacob et al., 1992; Liu eta!., 1992; Mauzerall et al., 1996;Carroll and Thompson, 1995], and this is also found in the TRACE A data below 8 km (Figure 6). Based on the discussion above, we explain this general result as reflecting the tendency of 03 to approach a chemical steady state determined by the NOx concentration.

Several authors have examined the primary sources of NOx over the South Atlantic basin in TRACE A by correlating the observed NO concentrations with chemical variables and with back-trajectories [Smyth et al., this issue; Singh et al., this issue; Talbot et al., this issue; Pickering et al., this issue
]. They conclude that biomass burning was a major source of NOx throughout the tropospheric column, and that lightning was also a major and perhaps dominant source in the upper troposphere. Soil emissions would provide an additional source of NOx in the continental boundary layer [Jacob and Wofsy, 1988Wofsy, , 1990] but the TRACE A data are not well suited for assessing the importance of this source. Transport from the stratosphere was a negligible contributor to the NOx budget even in the upper troposphere [Smyth eta!., this issue]. from combustion, soils, and lightning; during deep convective events over the continents, NO,, from combustion and soils is injected to the upper troposphere, and additional NO,, is produced by lightning. However, above 4 km we find no significant decrease of NOx from continents to oceans, even though the primary sources of NOx are continental (there is little lightning over the oceans) and the lifetime of NOx is short. It appears that the high concentrations over the oceans must be maintained by chemical recycling. We used our photochemical model to diagnose the degree to which NOx levels in TRACE A could be maintained by chemical recycling based on current understanding of tropospheric chemistry. For this purpose we computed the chemical production and loss rates of NOx at individual points in the merged time series matched to the HNO3 sampling times, using the model con-strained with measurements of NO, HNO3, and PAN (details of the approach are given in section 2). Figure 7 shows the ratio of chemical loss to chemical production of NO,, for individual points as a function of altitude, and Table 5  budget for NOx (Figure 7), with production of NOx balancing on average only 16% of the loss. This deficit is due to insufficient recycling of HNO3 to NOx; recycling of PAN is not an issue because rates for NOx-PAN cycling are slow and PAN is near chemical steady state (Table 5). The median rate of conversion of NO, to HNO• at 8-12 km in the model is 57 pptv day -• , while the median observed HNO• concentration at that altitude is 59 pptv. A HNO• lifetime of 1 day would be necessary to balance the HNO• budget, but the median lifetime of HNO• against pho-tolysis and reaction with OH is 12 days. It thus appears that a major mechanism for high-altitude conversion of HNO3 to NO,, may be missing from the model. The problem would remain even if N 20.s conversion to HNO.• in aerosols did not proceed, as this reaction accounts, on average, for only 30% of the conversion of NOx to HNO3 (Table 5). The above analysis suggests that HNO3 in the upper troposphere must be consumed on a timescale of 1 day to eventually regenerate NOx. Chatfield (1994) proposed that rapid reaction of HNO3 with CH20 in acid aerosols might take place and yield NO, and formic acid as immediate products. We find, however, no correlation between HNO3 and NO concentrations in the TRACE A data above 8 km (Figure 8), which argues against direct conversion of HNO3 to NOx on a short time scale. Fan et al. [ 1994] proposed that the reaction of HNO3 with CH20 in acid aerosols may not produce NOx but hydroxymethylnitrate H2C (   a Formation of all organic nitrates other than PAN. These organic nitrates are assumed inert in the model.

ooo
Our representation of the budget of O3 in the tropical troposphere goes back to the earliest papers arguing that the tropospheric O3 budget is dominated by in situ production and that transport from the stratosphere is small in comparison [Chameides and Walker, 1973;Crutzen, 1973  would result in acetic acid providing a major source of CH3OOH and CH20 in the upper troposphere; at 230 K, 3 ppbv of acetic acid would produce as much CH20 as 1700 ppbv of CH4. However, the TRACE A data show no correlation between measured acetic acid and CH20 in the upper troposphere, even when measured acetic acid is as high as 8 ppbv ( Figure A1) When NO concentrations are low, formation of the organic peroxide CH2(OOH)COOH provides a temporary reservoir but does not change the ultimate fate of the carbon. With this mechanism, acetic acid has little effect on the model photochemistry even at the high concentrations measured in TRACE A.
The origin of the high acetic acid concentrations observed in TRACE A remains a puzzle. There are to our knowledge no previous observations in the upper troposphere that can be used for comparison. Figure A2 shows the relationship between acetic acid and CO in TRACE A, and compares to the relationship found in 10 dry season measurements taken at 0-4 km over the

Appendix B: Treatment of NOx-PAN Interconversion in NOx Budget Calculations
When assessing the importance of NOx-PAN interconversion in a regional NOx budget calculation constrained with observations for NOx (or NO) and PAN, and with CH3CO3 taken to be at steady state, one must correct the rates of reactions (10), (13), and (14) to account for cycling within the chemical family PANx = CH3CO3 + PAN. To illustrate the need for this correction, consider a situation where reactions (10), (13), and (14) dominate the production and loss of CH3C'O3. Under this situation, given any specified input pair of NOx and PAN concentrations, the steady state CH3CO3 concentration computed in the model yields an apparent steady state between NOx and PAN; this steady state plays, however, no role in regulating NOx since loss of NOx and CH3CO3 by reaction (10)