A Photostationary State Analysis of the NO2-NO System Based on Airborne Observations From the Subtropical/Tropical North and South Atlantic

The Chemical Instrumentation Test and Evaluation 3 (CITE 3) NO-NO 2 database has provided a unique opportunity to examine important aspects of tropospheric photochemistry as related to the rapid cycling between NO and NO 2. Our results suggest that when quantitative testing of this photochemical system is based on airborne field data, extra precautions may need to be taken in the analysis. This was particularly true in the CITE 3 data analysis where different regional environments produced quite different results when evaluating the phototchemical test ratio (NO0(cid:127)t/(NO0c (cid:127), designated here as RE/R e. The quantity (NOz)c(cid:127) was evaluated using the following photostationary state expression: [NO2]c(cid:127) = (k(cid:127)[O3] + Iq[HO2] + ks[CH302] + 1%[RO2])[NO](cid:127)la/J 2. The four most prominent regional environmental data sets identified in this analysis were those labeled here as free-tropospheric northern hemisphere (FTNH), free-tropospheric tropical northern hemisphere (FIWNH), free-tropospheric southern hemisphere (FTSH), and tropical-marine boundary layer (plume) (TMBL(P)). The respective I(cid:127)/R c mean and median values for these four data subsets were 1.74, 1.69; 3.00, 2.79; 1.01, 0.97; and 0.99, 0.94. Of the four data subsets listed, the two that were statistically the most robust were FTNH and FTSH; for these the respective 1(cid:127)/1(cid:127) mean and standard deviation of the mean values were 1.74 _+ 0.07 and 1.01, +_ 0.04. The FTSH observations were in good agreement with theory, whereas those from the FTNH data set were in significant disagreement. An examination of the critical photochemical parameters 03, UV(zenith), NO, NO2, and non-methane hydrocarbons (NMHCs) for these two databases indicated that the most likely source of the RE/Rc bias in the FTNH results was the presence of a systematic error in the observational data rather than a shortcoming in our understanding of fundamental photochemical processes. Although neither a chemical nor meteorological analyses of these data identified this error with complete certainty, they did point to the three most likely possibilities' (1) an NO 2 interference from a yet unidentified NOy species; (2) the presence of unmeasured hydrocarbons, the integrated reactivity of which would be equivalent to - 2.7 parts per billion by volume (ppbv) of toluene; or (3) some combination of points (1) and (2). Details concerning hypotheses (1) and (2) as well as possible ways to minimize these problems in future airborne missions are discussed.


INTRODUCTION
Since the early 1970s when photochemically generated free radicals became recognized as major atmospheric chemical drivers, there has been an increasing effort to quantify various aspects of what is now referred to as "fast photochemical theory." One of the more experimentally amenable components of this theory that lends itself to testing is the three-reaction sequence that couples NO to NO•., i.e., Although the addition of reactions (R4) -, (R6) leads to an improvement in the overall chemical description of the NO-NO 2 system, an obvious shortcoming is the fact that any analysis using this equation requires reliable independent measurements of the transients HO 2, CH302, and RO 2. At this time, however, the instrumentation required for making reliable measurements of these species is still under development. Because of this experimental shortfall, most field sampling programs have focused on recording simultaneous measurements of the variables NO, NO2, 03 and the UV flux and/or J(NO2) via chemical actinometry; the free radical species HO2, CH302, and RO 2 have been evaluated using photochemical models. The early urban studies of this type [$tedtnan and Jackson, 1975;Calvert, 1976;$hetter et al., 1983] tended to show that the ratio of NO 2 to NO could be reasonably well represented by equation (1). For example, the dominant reaction involved in the conversion of NO to NO 2 was in these eases reaction (R1).
For rural or remote locations, where much lower NO x levels lead to much higher total peroxy radical to 03 ratios, experimental measurements of the NO2/NO ratio were found to be in better agreement with equation ( Trainer et al., 1987]. However, in nearly all of these cases, the major exception being the work reported by Trainer et al. [1987], to explain the observed NO2/NO ratios required pcroxy levels well in excess of model predictions. Following several years of further development and testing of NO, NO x and NO 2 measurement techniques [Hoell et al., 1987;Gregory et al., 1990a, b], more recent NO-NO2-O 3 photostationary state studies appear to be giving more encouraging results [Chameides et al., 1990;. In National Aeronautics and Space Administration's (NASA's) Chemical Instrumentation Test Evaluation 2 (CITE 2) airborne field program, two independent NO-NO 2 data sets were generated using instrumentation which involved two different measurement techniques [Charneides et al., 1990]. The focus of the latter study was on relatively clean tropospheric air. The analysis, as presented by Charneides et a1. [1990], indicated that "simple photostationary theory" predicted NO2/NO ratios that were a factor of 1.6 to 2.0 lower than the observations. By contrast, equation (2) resulted in an average calculated NO2/NO ratio that was only a factor of 1.25 lower than the observed ratio. The (Mauna Los Observatory Photochemical Experiment) (MLOPEX) on the island of Hawaii represents perhaps one of the most comprehensive photochemical studies carried out in recent years [Ridley and Robinson, 1992 Although collectively the above results can be viewed as quite encouraging, both from an experimental and from a theoretical perspective, we believe potentially important unresolved issues may still remain. This is particularly true in the ease of the aforementioned CITE 2 airborne data sets where the reported average value for the observed NO2/NO ratio was given as approximately 1.25 times larger than the value estimated from equation (2). This difference, although modest in magnitude, is nevertheless well outside of the quoted standard deviation of the mean, e.g., + 59[. Furthermore, because Chameides et al. [1990] based their analysis on what was defined as a "calculated ratio difference" approach, the analysis tended to underweight high values of the ratio (NO2/NO)mpt. Following an analysis approach similar to that taken in this paper, these authors have found that the level of disagreement between (NO2/NO)oe•t and (NO2/NO)cal c was approximately a factor of 1.5, the experimentally measured ratio being higher. Possible reasons for this level of disagreement include that (1) theory, as defined by equation (2), is chemically incomplete; (2) the peroxy radical concentrations calculated from the model are in error due to systematic errors in the J and k values employed, or (3) that there are undefined sources of systematic error in the input observational data.
Concerning the latter possibility, it is important to recognize that because of the limited number of individual sample runs recorded during an airborne field experiment, the typical data analysis approach has involved an evaluation of an "ensemble" mean value for the NO2/NO ratio. The potential difficulty with this approach is that in airborne field sampling it is quite common to encounter a wide range of air mass types over a relatively short time period. Thus a systematic error in any one of the critical photochemical parameters in any one of these different environment regimes can potentially lead to a systematic shit• in either the experimental or the calculated ratio of NO2/NO. In principle, if all chemical species required for interpreting the NO2-NO chemistry were measured and measured correctly and all processes influencing the NO2/NO ratio were accounted for in the form of some version of equation (2), the need for segregating data according to the type of environment sampled would be small. In actual fact, the first condition is almost never strictly satisfied either for airborne or for ground-based studies. But in the case of airborne studies, this problem has the potential for becoming acute.
The purpose of this work will be to further explore how well we understand the NO-NO2-O 3 photostationary state system in the context of some of the questions raised in the above text. In carrying out this analysis, we will be examining an NO-NO 2 database recently recorded during the second phase of the NASA GTE/CITE 3 airborne field mission. The sampling region encompassed by these data included the subtropical/tropical North and South Atlantic Ocean, as shown in Figure 1.

OBSERVATIONAL DATABASE
The species NO, NO2, and NOy (where NOy = NO + NO 2 + NO 3 + N20 5 + HNO 3 + HO2NO 2 + PAN + RNOx)were measured simultaneously using the two-photon/laser-induced fluorescence (TP/LIF) technique [Bradshaw et al., 1985;Sandholm et al., 1990Sandholm et al., , 1992. This speetroseopically selective NO technique simultaneously determined ambient NO, NO produced from the photolysis of ambient NO 2, and NO produced from the reduction of ambient NOy compounds. The latter process involved the use of a 300øC gold catalytic surface with 0.3 55 CO as a reducing agent. The photolytic conversion of NO 2 to NO was achieved using a 1-kW filtered UV-arc continuumlight source having a passband of 350 nm <3, < 410 nm. The photolyric yield ranged from 3055 to 6055 and sample residence times ranged from 2 to 4.5 s. A porcelain-glass-coated common inlet was used to sample ambient air in an orientation perpendicular to the airstream. Constant volumetric flow (circa 250 lpm) and, consequently, sample residence time (circa 0.25) was maintained throughout this common sampling inlet manifold. Three separate LIF sampling cells were used: one each for measuring the LIF signal from ambient NO, photolytically produced NO from NO 2, and catalytically produced NO from NOy. Flow from the common inlet manifold was split at a position near (e.g., < 0.5 m) each detection system. Standard addition calibrations using NO and NO 2 were routinely carried out by injecting a known concentration of standards into the common ambient sampling manifold. All flows were measured using linear mass flow meters that had been cross compared to volume displacement standards. The accuracy of these instrument calibrations has been estimated at + 16 % for NO and q-1855 for NO2 and NOy at the 9555 confidence limit.
Limits of detection for a typical 3-min signal integration period were 2.5 parts per trillion by volume(pptv) for NO and 12 pptv for NO•, based on a signal-to-noise ratio of 2:1. The typical q-10% at 500 pptv, increasing to q-2055 at 200 pptv.
The The north/south tropical Atlantic CITE 3 database, as defined by missions 11-19, consisted of 231 independent NO-NO• measurements. Of the 231 total data points, 190 of these had (S/N) signal/noise ratios of > 2:1, and 120 had signal/noise (S/N) values >3:1. The S/N designation as used in this text will always mean that both NO and NO2 satisfy the indicated signal-tonoise ratio criteria. As will be discussed in greater detail below, our purpose in S/N filtering the NO-NO 2 database was to assess the possible influence of systematic errors (specifically offset errors) and to assess the statistical robustness of several data subsets of the total database. The authors note that in the case of offset errors, any attempt to salvage low S/N data by assigning weighting factors to them can be dangerous since, on average, low S/N data (reflecting low mixing ratios) are more prone to being influenced by offset errors than high SIN data. The assignment of a weighting factor to individual data points implies that in all cases the major error associated with the data is random in nature. Quite clearly, this is not always the case. (e.g., 10 NO-NO 2 data pairs) is unique in that it represents the only NO-NO 2 data recorded during the CITE 3 tropical Atlantic campaign in which moderate levels of NO and NO 2 were observed. "Moderate," in this case, means NO 2 mixing ratios of 90 to 450 pptv and NO of 20 to 150 pptv. In continental United States, these levels would most likely be classified as relatively clean rural air. In the tropical Atlantic, typical boundary layer mixing ratios for NO 2 and NO were 21 and 5 pptv, respectively. Thus for purposes of discussion in this text we have used the labelling "tropical-marine boundary layer (plume)" to describe these data.

Figures 2a and 2b indicate that based on a S/N filter level of
> 2:1, NO and NO 2 mixing ratios ranged from 4 to 48 pptv and 16 to 69 pptv, respectively. The median values for NO and NO 2 were 17 and 35 pptv, respectively. When filtered at the 3' 1 S/N level, the range in NO and NO 2 decreased to 6 to 32 pptv and 24 to 69 pptv, respectively. For the latter ease, the median value for NO was 17 pptv and that for NO 2 was 40 pptv, indicating that NO 2 had a greater sensitivity to S/N filtering than did NO .  Tables la and lb  that the chemistry of the NO-NO 2 system is correctly represented by equation (2), and (4) that the model has accurately estimated the concentrations of the critical radical species HO2, CH302, and RO2; the significance of the ratio Roe/R c is that the median value for R•./R c for a specified data set should be unity (see below for a more detailed discussion of this topic). Significant departure from unity thus signifies a potential problem in any one or all of the above areas.
The data presented in Figure 3 are for an S/N filter level of > 2:1. Only NO-NO 2 data at the > 2:1 S/N level have been used since only these data were deemed acceptable for a preliminary quantitative assessment of equation (2). Recall that the LIF limit-of-detection for NO and NO 2 was defined in terms ofS/N = 2:1. For example, when referenced to the continuous lognormal curve shown in the figure, the distribution is seen to have the general characteristics of a lognormal distribution. The fact that this type of distribution is found is not totally surprising since all values of R•./R c are necessarily confined to positive numbers, and the expected median value is unity. Statistically, a common distribution that arises from these restrictions is one that is lognormal. The continuous lognormal curve shown in Figure 3 was derived from the statistical characteristics of the logtransform distribution, e.g., the mean and the standard deviation.
Still another interesting qualitative feature of the R•./R c frequency distribution plot is the apparent multimode structure of this distribution. In this case there appears to be at least two separate modes and quite possibly three.
From a quantitative point of view the frequency versus Roe/R c plot also raises an interesting question as related to the comparison of model predictions with theory. For example, the mean R•./R c values for data filtered at S/N > 2:1 and _> 3:1 were 1.58 and 1.53, respectively; and the median values were 1.29 and 1.34, respectively. It may be recalled that the mean R•./R c value for the re-analyzed CITE 2 airborne data set [Cha•neides et al., 1990] was also approximately 1.5, and the corresponding median value has been estimated at 1.33. Thus at the "ensemble" data analysis level, the CITE 2 and CITE 3 data sets give quite similar results. For both data sets the "expected" median value for the ratio R•./R c (e.g., for the case of perfect conditions experimentally and theoretically) would be unity. This follows from the fact that under perfect conditions the median values for R e and R c would be equal. Therefore it is the median that defines a "reference point" for comparison of observations with theory [Hines and Montgomery, 1972]. Under these same perfect conditions the "expected mean", EM, for Roe/R c would be given by: where S 2 is the variance, as defined by the log-transform plot of

R•./R c [Hines and Montgomery, 1972].
In the case of the CITE 3 database the median value of R•./R c is seen to be significantly shitted relative to the expected value of 1.0, thus suggesting the presence of one or more problems in our understanding of the relevant photøehernieal processes related to the NO-NO2 system and/or in the quality of the input data to the model. In this context, it is quite significant that the CITE 3 database shows strong evidence of being composed of two or more independent distributions. The latter conclusion is further supported by the results from a scatter plot of the quantities (NO2)oexpt and (NO2)caic, as shown in Figure 4. In this plot we have attempted to optimize any inherent correlation between these two quantities by using only data filtered at the S/N -> 3:1 level. In addition, all high mixing ratio NO-NO2 plume data identified earlier in the text (10 points) were removed. Because of the noisy nature of these data, we also have displayed them in the form of a "bin" plot as shown in Figure 5.
In this case the X axis was divided into ten concentration bins where each bin was assigned 11 data points. The results from the latter plot make more apparent that there is a The results from these plots indicate the following: (1) the latitudinal R•/R c data show a very clear systematic difference between the northern and the southern hemisphere, with the northern hemispheric values being significantly higher than those in the southern hemisphere. Within the northern hemisphere there also appears to be two independent data groupings. (2) Based on a visual inspection, the latitudinal trends in Roe/R c appear to be positively correlated with NOy but are anticorrelated with the general trend in 03.
Both CO and NO x show no simple correlation with the Roe/R c trends. (3) The altitudinal trends in the southern hemisphere appear to be a strong function of the S/N filter level. For example, the _> 3:1 filtered data show Roe/R c values centered around 1.0 for high altitudes (z > 3.7 km), dropping to a little less than 1 for middle altitudes (2.2 -3 km) and then dropping to considerably less than 1 (e.g., 0.60) for altitudes corresponding to the trade wind inversion (1.3 -2.2 km).
By contrast, databased on a S/N range of 2:1 to 3:1 show RE/R C values centered around 1.0 for very low altitudes (boundary layer); but for the trade wind region, most RE/R c values are far greater than unity. The RE/R c distribution at still higher altitudes is similar to that for the 3' 1 S/N filtered data set. The critical variables CO and NOy show no simple trends relative to RE/R c and altitude. (The authors note that only in the southern hemisphere was there sufficient data as a function of altitude to look for possible trends.) Based on the trends in RE/R c as well as the general trends in the ancillary measurements (Figures 6a-6b, 7a-7d), we have concluded that the bulk of the CITE 3 NO-NO 2 data can be assigned to three major data subsets. These have been labeled here as free-tropospheric northern hemisphere (FTNH), freetropospheric tropical northern hemisphere (FTTNH), and freetropospheric southern hemisphere (FTSH). These three regional Still another interesting feature of the northern hemispheric data block is the fact that the trend in RE/P c values shows a major discontinuity just to the south of 15øN, increasing from an average value of 2 to a value of -3 (see Figure 6a). It was this trend, in fact, that suggested that the northern hemisphere database should be examined as two independent data sets. From meteorological data it now appears that one of the major factors involved in this discontinuity was a major change in meteorological conditions during transit flights 11A and 1 lB. As discussed by Shipham  possible restricted our analysis to those data having S/N > 3:1.
In the case of the FTTNH data set, only six data points were available at S/N > 3' 1; and thus no further analysis was carded out at this level.
As discussed previously in the text, the expected median value for the ratio Roe/R c (i.e., both theory and observations being error free) would be unity. Based on this criterion, it can be seen that for both northern hemispheric data subsets there is very substantial disagreement between model predictions and observations. In contrast to the northern hemisphere, the southern hemispheric data subset, FTSH, showed very good agreement between model predictions and observations (see Table 2). The median and mean values of Roe/R c for this data block were 0.97 and 1.01 q-0.04 (SDOM), respectively. The corresponding mean value for R•_ was 2.44. In sharp contrast to the meteorological picture presented for the northern hemisphere, the southern hemispheric picture was relatively simple for the altitude range of 2.5 to 5.5 km. The 5-day isentropic back trajectories indicate that the air mass sampled had originated in the eastern South Atlantic. Although these trajectories do appear to intersect the continent of Africa, the transit time to Brazil is estimated to be of the order of 5 to 9 days. Thus the general meteorological picture for the southern hemispheric component of CITE 3 suggests that a very well aged air mass was sampled. This picture also appears to be in concert with the midaltitude southern hemispheric NOy    Table 2 shows also the RF./R c values calculated from equation (1). These results reflect the degree to which "simple theory" can be used to explain the CITE 3 NO-NO 2 observations. In every case, the RE/R c values are seen to be significantly higher than those given by equation (2). In fact, with the exception of the FTSH data the difference between "simple" and "expanded" theories equalled or exceeded a factor of 1.5. Thus the substantial difference between the RF./R c results from "simple" versus "expanded" theory demonstrate the important role peroxy radicals played in defining the regional photochemical environment of the CITE 3 database. This point is further illustrated in Table 3 where the percentage of the total conversion of NO to NO 2 contributed by peroxy radicals is given. These results indicate that the peroxy radical contribution ranged from 17% to 48%, with the northern hemispheric data defining the middle to high end (e.g., 29% to 48%) and the southern hemispheric, FTSH, data the low end (e.g., 17%). The difference between hemispheres reflects the higher average mixing ratio of 0 3 for the FTSH versus the FTNH database (e.g., 50 versus 75 (ppbv)) as well as the lower average UV flux for the FTSH data block (e.g., J(OID) = 1.99 x 10 '5, FTSH, versus 3.78 x 10 '5, FTNH). For both regions it was generally found that the most important of the peroxy species was HO 2 (-50%), followed by CH30 2 (-27%) and RO 2 (-23%). Total peroxy radical, PO 2, mixing ratios for high Sun conditions ranged from 30 to 60 pptv both for the FTNH and the FTSH data sets.
In addition to the three major data blocks, two smaller data subsets also were defined. Each involved a uniquely different chemical environment.
These smaller databases involved the  In contrast to the TWISH database, we believe that the data block labeled TMBL(P) does merit further comment. In this ease, even though the database is weak in terms of numbers (i.e., only 10 data points), the highly elevated mixing ratios for NO and NO 2 produced data pairs having an average S/N ratio of nearly eight. Based on assumed hydrocarbon levels of 25 and 5 ppbv of equivalent C3H 8 (i.e., see Tables lb and 2), the median values for Roe/R c were calculated at 0.91 and 0.99, respectively; the corresponding mean values for this data subset were .94 + 0.09 and 1.04 :l: 0.10 (SDOM), respectively. Thus even though the data are very limited, one may draw the tentative conclusion that for "moderate" mixing ratios of NO x (100 to 500 pptv) tropicalmarine boundary layer chemistry is well represented by equation (2). An interesting aspect of the TMBL(P) data subset was the observation that the NOy mixing ratio during this data run ranged from a low of 6 ppbv to a high of 10 ppbv. In spite of these high levels of NOy, it appears that they resulted in no interference in the measurement of NO 2 (see discussion later in text).
For reader convenience we have summarized all data subset acronyms and their respective regional environmental descriptions in Table 4. Table 2, it can be seen that there are two data subsets in which model predictions and experimental observations are consistent and two where the agreement is quite poor. As discussed earlier in the text, there also was one data subsets (e.g., TWISH) that had so few data pairs at S/N values of •3:1 as to render the final results of very limited value. Data subsets showing a high level of agreement between theory and observation are those labeled here as TMBL(P) and FTSH, whereas significant disagreement is seen in the two northern hemispheric data sets, i.e., FTNH and FTTNH. To gain further insight concerning the large Roe/R c bias in the northern hemispheric data, three questions were explored:

Based on the highest S/N filtered data summarized in
(1) How does the magnitude of disagreement compare with the magnitude of the random and systematic errors associated with the

analysis? (2) What are the major chemical and physical differences between those databases showing poor agreement versus those in good agreement with model predictions? (3) Based on the findings from questions 1 and 2, what conclusions may be drawn concerning possible shortcomings in the CITE 3 measurements versus shortcomings in our model description of the NO-NO 2 photochemical system?
Given that the FTNH data set was the most robust data subset of the two data subsets showing a high RE/P c bias, it served as the principal focus of our investigation of questions 1-3. To proceed with this analysis, however, further reflection is required on defining the "reference point" for the ratio Roe/R c. Recall that earlier in the text for the case where the ratio of two lognormal Under the above conditions, the expected median is unity and the expected mean is also closely approximated by unity. As seen from Table 2, all data subsets but FTTNH satisfy or closely approximate the condition given above. Further evidence supporting this position can be found in the mathematical closeness of the median and the mean for the FTNH, FTSH, and TMBL(P) databases where the agreement is seen to be within 5 %.
Given, then, the statistical conditions described above, the analysis that follows will be based on the use of the arithmetic mean. For all data sets analyzed here, it is the standard deviation of the mean (SDOM) that will be used to define how well the mean is known. In this case, the mean RE/R c value is over 10 oremoved from that predicted by equation (2). To assess the relative contributions of R E and R c to the random error associated with RE/R c, a propagation of error analysis was performed on the basic equation R = RE/R c. In our analysis all 40 (S/N >3:1) NO-NO: FTNH data pairs were evaluated. These results showed that the relative error in R c (e.g., aRc/Rc) was 20% and that for R E was 35%. Thus when added in quadrature, the relative contribution of R c and R E to the total uncertainty in the ratio Roe/R c was 25 and 75 %, respectively. These results indicate that the uncertainties associated with the measurement of NO 2 and NO were the major sources of random error in the evaluation of the ratio RE/R c.

Concerning question (1) which addresses the issue of random and systematic errors, it can be seen from
In the context of question (1), the potential importance of systematic errors was also explored. Systematic errors can arise both due to errors in the measurements of individual physical and chemical variables used in the overall evaluation of RE/R c (e.g., NO2, NO, 0 3, UVZ, UVN, CO, H20, NMHC, T, and P) or due to errors in the many photochemical and gas kinetic rate constants used in modelling calculations. In the latter ease, it is important to recognize that it doesn't matter whether the error in a "J" or "k" coefficient has been caused by a random or systematic error in some experiment used to determine its value. Since the sign of the error does not change from one run to another, it always manifests itself as a systematic error in the model output and therefore in the value of R c.
In assessing the resultant systematic error from the "J" and "k" values used in our calculations, we have employed a Monte Carlo procedure similar to that presented by Thompson and Stewart [1991]. In our evaluation the estimated errors in all "J" and "k" values are those given by [Demore et al. 1992]. The errors were assigned a lognormal probability distribution for values that must be positive, and normal distributions were assigned to errors for values that could be either positive or negative. In our analysis an average temperature of 273 K was used for the FTNH data in randomly producing 1000 independent sets of "J" and "k" values.
Based on a representative data point from the FTNH environment, calculations were then carried out for each set of "J" and"k" values. The output from these calculations gave probability distributions for the systematic error due to "J" and "k" in the model results. For example, the probability distribution for RE/R c had a standard deviation of 0.64. While the true systematic error is unknown, an error the magnitude of one standard deviation does shi• the original FTNH mean value of 1.74 to 1.1 (e.g., 86% of the bias is removed). Although this might be viewed as an encouraging result, it raises the obvious difficulty that all other data subsets would have to be corrected by approximately the same percentage (e.g., 37 % of the initial RE/R c value). Quite clearly, the application of this magnitude of correction to the FTSH and TMBL(P) data would result in an unaeeeptably large negative bias in the Roe/R c mean values for these databases. This suggests that the actual magnitude of the systematic error in Roe/R c, associated with the J and k values, is most likely significantly smaller than that cited above. If so, a major component of the FTNH Roe/R c bias would remain.
Concerning possible systematic errors in the observational data, the situation here is more complex than that involving the J and k uncertainties. Part of this complexity arises from the fact that for airborne-observational field data a given systematic error may affect the data recorded all of the time and therefore be viewed as stationary throughout the data set, or it may appear only under very specific conditions in which case the sign of the error may be known but the magnitude would vary in time. Thus the latter type of error may apply to only one segment of an overall data set. As discussed in the text above, a stationary type systematic error cannot resolve the discrepancy between the FTNH and the FTSH data subsets.
To assess the possible effects of variable systematic errors, we have examined both the FTNH and FTSH data in terms of their respective critical chemical and physical parameters. We have defined a "critical" parameter in this case as one whose value would have a near 1' 1 linear effect on the value of Roe/R c. Thus given that it satisfied the criteria of being nonuniform in its impact on the FTNH and FTSH databases, a "critical" parameter would lead to a correction in the Roe/R c values for the FTNH data without a concomitant adjustment in the Roe/R c values of FTSH.
To identify these "critical" parameters, sensitivity calculations were performed on all four data subsets. The results are shown in Tables 5a and 5b. For these calculations, Roe/R c values were estimated for independent changes in each of the listed test variables (e.g., H20, CO, 03, NO, NO2, UV(zenith), and UV(nadir)) where each test variable was adjusted in value in stepwise increments which ranged up to a factor of q-2. The calculations were performed on a single representative run from each data subset that had an Roe/R c value close to the mean value for that database. For the hydrocarbon sensitivity tests, the original hydrocarbon level was decreased by a factor of 2.0 but increases in hydrocarbon levels were carried out by adding the surrogate hydrocarbon species toluene in 2.0 ppbv steps (see discussion below for details).
The results from these calculations indicate that the "critical" parameters in the evaluation of Roe/R c are 03, NO, NO2, UV(zenith), and (non-methane hydrocarbons) (NMHCs). As related to the FTNH and FTSH databases, the discussion that follows is focused on examining whether significant systematic errors may have been present in one of the above parameters for one data subset but not the other.
Concerning the "critical" species 03, we found no evidence in the data themselves nor anything recorded in flight books that suggests that there were any problems in the performance of this instrument. Similarly, an evaluation of the critical parameter UV(zenith) indicated that this parameter was an unlikely source of systematic error. Flight notes gave no indication of any failure in this instrument during the mission nor was there any indication of any significant shilt in the instrument's calibration factor as measured by the Eppley Company before and alter the CITE 3 mission. Recall however, that the major use of the UV(zenith) measurement in our analysis was to correct our two-stream model in the fact that the measurement of NO2, unlike NO, is not a direct measurement. As discussed earlier in the text, the NO 2 measurement involves the initial step of photo fragmentation (NO 2 -I-hv • NO -I-O), followed by the two-photon LIF detection of the NO photofragment. Thus to the extent that an unknown/unidentified NOy species (i.e., some form of organic nitrate) undergoes photolytic decomposition and produces NO 2 as a product, this NO 2 photofragment would lead to a signal indistinguishable from that derived from natural NO 2.
Alternatively, any NOy species that has a high rate of decomposition on surfaces or that could thermally decompose in the gas phase to form NO 2 (e.g., N2Os, HO2NO2, PAN ) also would have the potential for creating an interference problem. The in:portanee of the latter source increases as the sample residence time becomes long and/or there are significant differences in the sample chamber temperature relative to the ambient outside air temperature. As configured during CITE 3, the PF-LIF NO 2 instrument used a continuum light source that was filtered to restrict the photolysis wavelength from 350 to 410 nm. This relatively narrow bandwidth centered at long wavelengths most likely eliminated most NOy species from becoming a problem. In addition the sample residence times were kept very short (i.e., 2 to 4.5 s), and the sample inlet lines were maintained within a few degrees of the outside air temperature. Tests by Georgia Tech investigators using known amounts of HNO 3 and a limited number of alkylnitrates and nitroalkanes also suggest the absence of interferences from these classes of compounds [Sandholm et al., 1990[Sandholm et al., , 1992. On the other hand, when the sum of individually measured NOy compounds, as recorded in other airborne field programs, is compared against total measured NOy, very frequently a "shortfall" has been found in the budget. This suggests that there are still several NOy compounds that remain unidentified [Hubler et al., 1992;$ingh et al., 1993;Ridley, 1991;Sandholm et al., 1992Sandholm et al., , 1993. Of the known but yet unmeasured NOy species, the lablie compounds HO2NO 2 and N20 5 could perhaps present the most serious problem. These compounds, in addition to their thermal decomposition in the gas phase, could very well undergo rapid decomposition on surfaces. For both of these compounds NO2 would most likely be one of the decomposition products.
The issue of NOy interference in the measurement of NO2 is centrally important to this analysis since a comparison of the FTNH and FTSH NOy data (e.g., see Figure 6b) indicates a considerable difference in the NOy mixing ratios for these two data subsets. For example, the FTNH data show NOy mixing ratios ranging from 0.9 to 6.2 ppbv with an average value of 2.5 ppbv. By contrast, the FTSH data gave an average NOy mixing Still another consideration in our assessment of the NOy-NO 2 interference hypothesis was the possible importance of the ratio NOy/NO 2. For example, it may be argued that the larger this ratio the higher the probability that even a small percent decomposition of NOy would have a significant impact on the value of (NO2)Expt. In actual fact, all that is required to shift the observed FTNH Roe/R c average from 1.74 to 1.0 is an (NO2)lnterf ' level of 18 pptv. (Recall the reported average NO 2 mixing ratio for the FTNH data was 42 pptv.) This NO 2 interference level (i.e., 18 pptv) represents less than 1% of the estimated average FTNH NOy mixing ratio of 2500 pptv. In an effort to explore the latter possibility, a scatter plot was made of Roe/R c versus NOy/NO2; see Figure 11. The results from this binned data plot gave an r 2 value of 0.41. Although more indicative of an interference problem, this modest r 2 value does not convincingly show that (NO2)lnterf from NOy was the primary source of the FTNH Roe/R c bias. By contrast, a similar analysis of the FTSH data gave a still lower r 2 value of 0.34.
In a final attempt to gain further insight into the NOy interference question, we also examined the possible influence of the highly labile NOy species HO2NO 2, per nitric acid. The concentration of this species is defined by the thermal equilibrium (R7) NO 2 + HO 2 +M • HO2NO 2. As shown in Table 6, for the average temperatures reported for the FTNH and FTSH data sets, the estimated mixing ratio for HO2NO 2 is -2 pptv and therefore should have been of little consequence as an interference. However, within the la error bar given for the equilibrium constant for this species, its mixing ratio could have ranged from a high of 15 pptv to a low of 0.1 pptv. To assess the interference potential of HO2NO 2, all RE/P c values (i.e., S/N > 3) from the FTNH database were plotted versus temperature. The argument here would be if HO2NO 2 were a major source of (NO2)lnterf , the lower the temperature the more HO2NO 2 present and the larger the expected value of RE/P c. The results from this regression analysis showed no such correlation. In fact, the reverse trend was found, that is, the value of RE/P c was found to increase slightly with increasing temperature. The data set consists of all FTNH data pairs having S/N >_3:1.

Equally important here is the observation made earlier in the text
All "plume" data have been removed. samples (e.g., C2 • C4) showed that NMHC levels were relatively low, having an integrated reactivity with OH that ranged from 0.76 to 3.88 ppbv equivalent propane (C3H8). These low NMHC levels typically had a 1095 or less effect on the evaluated Roe/R c ratios for the FTNH data (see e.g., Table 3). Based on these results, the impact of NMHCs on the Roe/R c FTNH bias would appear to be quite small. This raises the question, however, whether the five NMHC samples taken during mission 11 were indeed representative of the actual northern hemisphere hydrocarbon environment over the designated flight track (e.g., Figure 1). Evidence suggesting that they were can be seen in the CO data. For example, the northern hemisphere middle-altitude (z >_ 3 kin) CO data averaged only 90 ppbv, having a maximum value of 115 ppbv. Quite clearly, these CO levels reflect those one would expect to find in a relatively clean environment, i.e., one having been little affected by significant combustion sources. An important aspect of the CITE 3 CO measurements, as related to the question of representative sampling, is that these measurements were recorded continuously using the well established TDL (tunable diode laser) technique. Thus it could be argued that the CO measurements tend to support the contention that the limited number of NMHC grab samples taken during the transit flight were representative of the actual hydrocarbon environment. Evidence suggesting that atmospheric conditions might have been chemically more complex during the northern hemisphere transit flight can be found in the CITE 3 NOy measurements and the meteorological data. As discussed earlier in the text, the 5 day back trajectories for the Wallops Island, Virginia, to Puerto Rico transit flight show that the air sampled between 3.6 and 4.8 km (aircraft altitude) had moved off the northeastern and central eastern U.S. coast turned abruptly south (the effects of Hurricane Gabriel) and then moved down through the western Atlantic toward the Caribbean. The estimated time that this air was over the western Atlantic before aircraft sampling, occurred ranges from 1/2 to 3 days (e.g., ranging from the first hour of flight to the end of mission 11A as Puerto Rico was approached). For most of the flight from Wallops Island, Virginia, to Puerto Rico, then the general classification of the air sampled would be "middle-altitude continental outflow." The back trajectories, in combination with synoptic data, also suggest that for 1-3 days before passing over the U.S. coast, the air parcel sampled had experienced extensive convective activity in combination with heavy rain showers [Shipha•n et al., this issue]. Thus the general overall picture that emerges is one which suggests that there had been extensive mixing of surface sources into midtropospheric air at 3 to 5 km.
The chemical evidence supporting the notion of a more complex middle-altitude chemical environment during the northern hemisphere transit flight involves the reported NOy measurements. For example, the high average mixing ratios for NOy, the high degree of variability, and the strong correlation between dew point and NOy levels (e.g., r 2 = 0.81) would seem to be consistent with significant ground level NO x being transported into the middle troposphere. If the latter were true, however, it also would seem to follow that one should have observed elevated levels of both NMHCs and CO.
A scenario that potentially resolves this apparent conflict would involve the hypothesis that the major source of NOy was not predominantly ground level combustion but rather lightning. The latter source appears to be compatible with the observed elevated, variable NOy levels cited above and is also consistent with the high correlation found between NOy and dew point. Provided that the additional assumption is made that the major component of the convectively transported ground level air was predominantly rural in its composition, it is also consistent with the CO and hydrocarbon observations. Based on the lightning hypothesis, then, the air sampled during the transit flight at 3 to 5 km would have had elevated levels of NOy, modest levels of CO, and the hydrocarbon mix would have been more characteristic of biogenic emissions. Unfortunately, as shown in Table lb, the latter type species were not analyzed in the CITE 3 grab samples. However, based on recent results summarized by Chameides et al. [ 1992] for the time period of September, surface isoprene levels as high as 0.7 ppbv could be expected. (The concentrations of propane (C3H8) and toluene equivalent in reactivity to this level of isoprene are 70 and 10 ppbv, respectively). These surface levels of isoprene would most likely have been diluted by factors of 4 to 5 due to convective transport to higher altitudes and also would have been further reduced in concentration due to their rapid reaction with OH radicals. Even so, these losses do not preclude the possibility that some mix of unreaeted parent hydrocarbon (i.e., isoprene) and reactive oxidation by-products may have provided a source of peroxy radicals that significantly exceeded that which was used in our initial calculations of the ratio R•/R c [Calvert and Madronich, 1987;Calvert, 1989, 1990]. In an effort to test this NMHC hypothesis, several hydrocarbon modelling simulations were carried out in which the integrated chemical effect of a reactive hydrocarbon mix was simulated using the surrogate hydrocarbon species toluene. In the first set of simulations, the surrogate species was added to gas mixtures corresponding to specific FTNH runs where the runs chosen had RE/R c values close to that of the mean for the entire data set, e.g., 1.74. The results from these simulations are those shown in Table 5b and reflect the hydrocarbon sensitivity tests cited earlier in the text. From Table 5b one can estimate that at a toluene mixing ratio of 3 ppbv, 80% of the FTNH Roe/R c bias is removed, reaching a value of 1.15. In fact, for toluene mixing ratios much greater than 4 ppbv, the rate of reduction in the value of Roe/R c decreases significantly due to the suppression of OH and a subsequent plateauing of the peroxy radical concentration. (Note that based on a diurnally-averaged OH level of 1 x 106/em 3 for middle-altitude conditions, toluene would have an estimated lifetime of approximately 2 days at 275 K.) At levels of 3 ppbv, the peroxy radical conversion of NO to NO 2 was estimated to be -50% of the total conversion rate of NO. By contrast, from Table 3 it can be seen that when using the background hydrocarbon mixture for the FTNH data, corresponding to our previously defined level 1 , peroxy radicals accounted for only 29% of the NO to NO 2 conversion rate (level 1 is approximately equivalent to 0.4 ppbv toluene). Overall, the above results suggest that from a theoretical point of view the addition of a moderately reactive surrogate hydrocarbon to the gas mixture of each sampling run does remove the major component of the R•./Rc bias in the FTNH data. However, these results raise yet a different question. If such enhanced hydrocarbon mixtures were present, what impact might they have on 03 (i.e., 03 tendency), and are these estimates in reasonable agreement with actual 03 observations? To address this question for the hydrocarbon conditions stated above, the 03 tendency was evaluated using equation (6):

P(O3) = [NO](k,•[HO2] + ks[RO2] ) -k•o[O(•D)][H2 ¸] -[O3](k•[HO2] + k•2[OH])
where the new reactions (R9) ---(R12) represent the major 03 destruction processes, e.g., involves a process that might best be described as "/'fiament transport." Newell et al. suggest that these filaments (i.e., rivers of unique air) may define a very significant transport process for H20 and possibly other trace constituents within the troposphere. A very interesting feature of this 20 ø -23.4øN sampling interval is the fact that what appears to be relatively clean air resulted in a calculated average value for R•./R c of 1.14. Much like the FTSH data, therefore, the observations between 23.4 • and 20øN latitude seem to reflect a quite different photochemical environment than that observed for the remainder o f the northern hemisphere. Of the 15 data pairs involved in this data subset, eight had S/N NO-NO 2 ratios of 3:1 or greater. Upon removing this small segment of data (e.g., those having S/N _>3:1) from the overall FTNH data subset resulted in a new mean value for RE/P c of 1.80, up from the original value of 1.74.

SUMMARY AND CONCLUSIONS
The CITE 3 NO-NO 2 database has provided a unique opportunity to look at some important aspects of current photochemical theory. Our analysis of these data, as related to evaluating the photochemical ratio R•./Rc, identified two major data subsets for focused analyses. The regional environmental setting for each data subset has been labeled here as freetropospheric (2.6-4.9 km) northern hemisphere (FTNH) and freetropospheric (2.4-5.5 km) southern hemisphere (FTSH). The air mass type for each of these cases was categorized as: middlealtitude middle-latitude continental outflow recently imprinted with ground level sources and lightning (FTNH); and middlealtitude tropical maritime air mixed with a small component of well-aged biomass emissions (FTSH). The results from our analysis of these two higher-quality data sets revealed the following key points: (1) Airborne NO-NO 2 data when used in quantitative assessments of photochemical theory need to be carefully screened in terms of the regional environmental setting in which they are recorded. In some cases the screening of the data in terms of S/N ratios may also be important. (2) The CITE 3 data set, when analyzed as per point 1, produced two reasonably high quality data sets that were found to give quite different results when evaluating the photochemical test ratio R•./R c. We have found that the consistency between the observations from the southern hemispheric data subset (i.e., FTSH) and model predictions and the significant inconsistency of the northern hemispheric data subset (i.e., FTNH) points toward a systematic error in the FTNH observational data rather than a shortcoming in our current understanding of fundamental photochemical processes, although the latter possibility cannot be totally ruled out. (3) The FTNH data, which had an Roe/R c bias of 0.74 +_ 0.07, was most likely influenced by (1) An NO2 interference from a yet unidentified NOy species; (2) The presence o f unidentified/unmeasured hydrocarbons, the integrated chemical reactivity of which would be equivalent to --2.7 ppbv of toluene; or (3) Some combination of (1) and (2). (4) To reduce the potential impact of problems of the type cited above, future airborne missions should endeavor to (1) make more comprehensive hydrocarbon measurements of C2-C•0 species that include both conventional hydrocarbons as well as those containing oxygen and nitrogen; (2) expand field measurements of NOy to include more of its many component forms (e.g.,