Uncertainties in climate assessment for the case of aviation NO

Abstract

Nitrogen oxides emitted from aircraft engines alter the chemistry of the atmosphere, perturbing the greenhouse gases methane (CH₄) and ozone (O₃). We quantify uncertainties in radiative forcing (RF) due to short-lived increases in O₃, long-lived decreases in CH₄ and O₃, and their net effect, using the ensemble of published models and a factor decomposition of each forcing. The decomposition captures major features of the ensemble, and also shows which processes drive the total uncertainty in several climate metrics. Aviation-specific factors drive most of the uncertainty for the short-lived O₃ and long-lived CH₄ RFs, but a nonaviation factor dominates for long-lived O₃. The model ensemble shows strong anticorrelation between the short-lived and long-lived RF perturbations (R² = 0.87). Uncertainty in the net RF is highly sensitive to this correlation. We reproduce the correlation and ensemble spread in one model, showing that processes controlling the background tropospheric abundance of nitrogen oxides are likely responsible for the modeling uncertainty in climate impacts from aviation.

Quantifying uncertainties in climate processes is a major priority for climate research, as exemplified by the widespread use of probability distributions and uncertainties in the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (1). To understand and attribute changes in the greenhouse gases and aerosols that force climate change, we rely on chemistry-transport models (CTMs), which require further characterization of their errors (e.g., 2–5). Aviation emissions are estimated to contribute 5% of anthropogenic climate forcing, with a nearly threefold uncertainty (6). This forcing occurs mainly through aviation-induced cloudiness and emissions of CO₂ and nitrogen oxides. Climate impacts of other aviation emissions—CO, SO₂, soot, hydrocarbons, and water vapor—are highly uncertain, but estimated to be much smaller (6). In this paper we quantify one type of climate forcing and its uncertainty, the impact of aviation emissions on the greenhouse gases ozone (O₃) and methane (CH₄).

Aircraft engines emit NO and NO₂ (NOx), which are indirect greenhouse gases through their chemical reactions impacting O₃ production and the CH₄ lifetime. As an O₃ precursor, NOx emissions increase the tropospheric column of O₃ and its radiative forcing. Aviation NOx emissions have a stronger impact on O₃ than surface emissions on a per molecule basis because they occur at high altitudes (7). Previous model-based estimates of the amount of O₃ generated by aviation NOx emissions vary by up to 100% and these differences have been attributed to a range of causes: e.g., the ratios of NO∶NO₂ and OH∶HO₂, background NOx levels (8), location and time of emissions (9–11), the amount of sunlight (12), and atmospheric mixing (13, 14).

Early work on climate forcing by aviation assumed that the impacts would be short-lived and limited to the northern latitudes containing the major flight corridors because of the short lifetime (months) of O₃. Subsequent studies, however, found that the short-lived, aviation-induced O₃ and CO perturbations propagate to the tropics where they boost OH levels and trigger a long-lived negative perturbation to global CH₄ abundance (15). An equally long-lived, global, and negative O₃ perturbation accompanies the CH₄ perturbation, and is based on numerous studies showing that the increase in mean tropospheric O₃ since preindustrial times is partially due to the increase in CH₄ (16, 17). The negative aviation radiative forcing (RF) due to long-lived CH₄ and O₃ perturbations nearly cancels the short-lived positive RF due to O₃ (12), which creates large uncertainty in the net forcing. Although most studies find that aviation NOx emissions cause net positive RF, some find negative forcing (8, 11, 18). This work aims to quantify the RF uncertainty using two complementary approaches. Part of the analysis is similar to that in earlier studies and IPCC assessments: We look at the ensemble of published model results and perform sensitivity analyses on poorly understood processes with a single CTM. We also develop another analysis here: We decompose the RF into components and factors, assess the possible errors in each, and then propagate the final uncertainty in RF. Taken together, these approaches point to atmospheric measurements that would reduce uncertainties in key processes involved in climate forcing.

RF Uncertainty from Model Ensembles

In this section we analyze aviation NOx RF from the ensemble of all published studies since the IPCC aviation assessment (15), which synthesized earlier work. Fig. 1 shows each RF estimate from the last 12 y, broken down into its short-lived and long-lived O₃ and CH₄ components. These estimates come from diverse models, including CTMs (7, 10–12, 18–21) and coupled chemistry-climate general circulation models (GCMs) (7, 15, 19–21). They are based on a wide range of scenarios for aviation and other emissions. Studies find little evidence for nonlinearity of the atmospheric response to aviation emissions over the range 0.4 to 1.3 Tg(N) a^-1 used in the ensemble here (7, 18), so we normalize all results to 1 Tg(N) a^-1 from aviation.

Fig. 1.

Steady-state RF (mW m^-2) from short-lived O₃, long-lived CH₄, and long-lived O₃ caused by aviation NOx emissions. Values are scaled to 1 Tg(N) a^-1. For papers reporting multiple models or scenarios, the key is O, Oslo-CTM2; L, LMDz-INCA; M, TM4; Q, ULAQ; T, p-TOMCAT; U, UCI; N2015, NASA-2015; N1992, NASA-1992. Derwent et al. (10) is corrected by Stevenson et al. (11). An asterisk (*) indicates studies that did not report long-lived O₃ RF. The means and standard deviations are given for each of the three components.

The RF values in Fig. 1 and throughout this work are the steady-state responses to aviation NOx emissions. Most studies calculate these by the difference between two steady-state simulations with differing aviation NOx emissions (7, 18–21). Three studies report 100-year integrated RF responses to pulse emissions of aviation NOx (10–12), which are comparable to the steady-state response because the integration time is much longer than the CH₄ perturbation lifetime (10–15 y). The latter studies use pulse durations between 1 mo and 1 y, which could conceivably increase the spread of RF estimates, but we find their results to be within the envelope of RF from steady-state simulations. Our ensemble statistics would not change by excluding them. We exclude one recent aviation study (22) because its results derive from the same model simulations as another included study (21) and another because of its different experimental design (23). Stated RFs are based on composition changes in the troposphere and do not include stratospheric response, although some studies also include the lower stratosphere, where aviation NOx also increases O₃ (e.g., 15) and the greenhouse effect is similar to the upper troposphere. The CH₄ RF here also excludes changes to stratospheric water vapor, which are estimated to increase RF cooling by about 15% (24). Transient CH₄ perturbations for any particular year are always less than predicted at steady state due to the decadal lifetime of CH₄ and the steadily increasing aviation emissions (25). We use steady-state CH₄ RF here to avoid making assumptions about historical emissions and perturbations, and also because it is directly proportional to the global warming potential (GWP).

None of the three RF estimates in Fig. 1 exhibit any trends over time of publication. Likewise, their spread has not changed over a decade of model development. We find no clear differences between GCMs and CTMs in any individual RF component or in the net RF, although one of the CTMs (model T in Fig. 1) has the largest component RFs in several studies, perhaps due to its lower CO abundance and larger ozone production efficiency (7, 18, 20). At the opposite end, Unger et al. (23) find much smaller RF from short-lived O₃ that is not statistically significant and is confounded by simultaneous changes in aviation aerosols and clouds. Treating all of the data in Fig. 1 as independent samples of a normal distribution, we estimate the steady-state RF effects from 1 Tg(N) a^-1 from aviation to be +27.3 ± 9.7 mW m^-2 due to short-lived O₃, -16.1 ± 5.6 mW m^-2 due to long-lived CH₄, and -6.6 ± 3.3 mW m^-2 due to long-lived O₃.

RF Uncertainty from Factor Decomposition

Past efforts to quantify the uncertainty in the climate impacts of aviation NOx have relied primarily on ensembles of model simulations, as in the prior section. In this section we present an alternative method for determining the uncertainty of aviation impacts based on decomposing the RF into its key factors. We use literature review and expert judgment to assess the uncertainty in each factor and propagate these to the uncertainty in RF. Many of the factors can be derived from the general climate forcing literature and do not require model simulations that are specific to aviation. A major advantage of this approach is that we can identify the dominant sources of uncertainty.

The steady-state, global mean RF from aviation NOx emissions (F) can be broken into components due to the short-lived O₃ response (F_short O3), and the long-lived CH₄ and O₃ responses (F_long CH4 and F_long O3, respectively):

[1]

These components can in turn be factorized into key terms summarizing the chemical interactions between NOx, O₃, and CH₄, including their feedbacks:

[2]

[3]

[4]

where d[O₃]_short/dE is the short-lived O₃ response to aviation NOx emissions (E) and dL_CH4/dE is the accompanying relative change in CH₄ lifetime, dF/d[O₃] and dF/d[CH₄] are the RF efficiencies of tropospheric O₃ and CH₄, [CH₄] = 1.78 ppm is the global mean CH₄ abundance, f_CH4 is the feedback factor, which prolongs the lifetime of CH₄ perturbations (16), and d[O₃]/d[CH₄] is the O₃ component of the long-lived CH₄ perturbation (26). Note that the product of the first three terms on the right-hand sides of Eqs. 3 and 4 equals the steady-state response of CH₄ abundance to aviation emissions (14).

Table 1 reviews the best estimates and one-sigma uncertainties (68% confidence interval) of each factor in Eqs. 2–4, including the values adopted in this work. Atmospheric measurements provide no direct constraints on these factors. Although the range of models does not encompass all potential uncertainties and errors, it does in this case provide a large ensemble.

Table 1.

Factors governing the climate impact of aviation NOx emissions

Factor, unit	Value	Uncertainty*	Source†
d[O3]short/dE	0.9	± 25%	(15)
DU(O3) [Tg(aviation N) a-1]-1	0.6	± 15%	(19)
	0.51	± 36%‡	(20)
	0.74		UCI
	0.60	± 0.15	Adopted
dLCH4/dE	−2.4	± 20%	(15)
%[Tg(aviation N) a-1]-1	−1.8	± 36%	(19)
	−1.55	± 38%	(20)
	−1.7		UCI
	−1.7	± 0.35	Adopted
fCH4	1.38	± 0.05	(16)
d ln(CH4 lifetime)/d ln[CH4]	1.30		(17)
	1.52		UCI
	1.40	± 0.10	Adopted
d[O3]/d[CH4]	3.7	± 0.8	(16)
DU(O3) ppm(CH4)-1	2.4		(17)
	3.3		UCI
	3.5	± 1.0	Adopted
dF/d[CH4]	370	± 7.2%	(27)
mW m-2 ppm(CH4)-1	370	± 7.2%	Adopted
dF/d[O3]	38§		(28)
mW m-2 DU(O3)-1	48§		(29)
	42§	± 12	(30)
	36	± 6	(7)
	36	± 8	Adopted

DU, Dobson unit.

^*Stated uncertainties are one standard deviation (68% confidence interval).

^†UCI values are from the UCI CTM used in this work. Adopted values are used throughout this analysis.

^‡Model standard deviation is < 25% without the single outlier model.

^§RF calculated for preindustrial to present changes in tropospheric O₃, not aviation-specific pattern.

Only two of the factors—d[O₃]_short/dE and dL_CH4/dE—require model simulations specific to aviation to derive their values and uncertainties, although dF/d[O₃] is also somewhat dependent on the aviation perturbation (see below). The values adopted here for d[O₃]_short/dE and dL_CH4/dE encompass recent estimates based on model ensembles (19, 20), but exclude values from IPCC (15) that were based on early CTMs, which had coarse resolution at aircraft altitudes, and were not supported by subsequent studies. The poorly understood sources of model differences in these terms are discussed below, but likely involve the range of NO∶NO₂ and OH∶HO₂ ratios and background NOx abundances across the models.

The remaining factors can be estimated from general literature on atmospheric composition and climate, without requiring model studies or calculations specific to aviation. The O₃ response to long-lived CH₄ changes, here d[O₃]/d[CH₄] = 3.5 ± 1.0 DU(O₃) [ppm(CH₄)]^-1, can be calculated from models (17, this work) and is qualitatively supported by numerous studies showing that the increase in mean tropospheric O₃ since the preindustrial era is partially due to the increase in CH₄ (16). Our estimated range for the CH₄ feedback factor (f_CH4 = 1.4 ± 0.1) includes values from models with widely varying chemical mechanisms (16, 17, this work). The final terms involve mapping the changes in greenhouse gas composition to RF. For CH₄ perturbations, which are globally uniform, the RF efficiency, dF/d[CH₄] = 370 mW m^-2 [ppm(CH₄)]^-1 ± 7.2%, is relatively well-known and nearly constant across all models (27). For O₃ perturbations, the RF is less well-known because the perturbation patterns themselves vary significantly between models. The RF efficiency may also differ between the short-lived and long-lived O₃ responses because they have different spatial and temporal distributions, but we adopt one value here, dF/d[O₃] = 36 ± 8 mW m^-2 [DU(O₃)]^-1, because the range derived from aviation-specific model tests (7) overlaps that from studies of the preindustrial to present-day changes in O₃ RF (28–30).

Table 2 shows the values and uncertainties of all RF components obtained by propagating the adopted values in Table 1 through Eqs. 1–4. For simplicity, the uncertainty calculations here assume that all factors are uncorrelated, which is reasonable for the individual O₃ and CH₄ RF components, but not for their sum, as we discuss in the next section. Each one of the RF estimates derived from the factor decomposition agrees with the model ensemble estimate within its 68% confidence intervals. Furthermore, neglecting net RF for the moment, the uncertainties of each RF component derived from the factor decomposition are comparable to the standard deviations of the model ensemble. This level of agreement illustrates the usefulness of the factor decomposition method, which might be extended to RF of aviation contrails by considering factors such as flight paths, likelihood of ice supersaturation, and rates of ice sedimentation and evaporation, among others, but such an analysis is beyond the scope of this work.

Table 2.

Steady-state RF (mW m^-2) from aviation NOx emissions, 1 Tg(N) a^-1

RF term	Factor decomposition*	Model ensemble†
Short-lived O3	+21.6 ± 7.2	+27.3 ± 9.7
Long-lived CH4	−15.7 ± 3.6	−16.1 ± 5.6
Long-lived O3	−5.3 ± 2.2	−6.6 ± 3.3
Net	+0.6 ± 8.3‡	+4.5 ± 4.5

^*Calculated from Eqs. 1–4 with factor values given in Table 1, assuming that all factors are uncorrelated. Stated uncertainties are one standard deviation (68% confidence interval).

^†Calculated from models shown in Figs. 1 and 2. Stated uncertainties are one standard deviation.

^‡Accounting for correlations due to all common factors in Eqs. 2–4 reduces the uncertainty to ± 8.1 mW m^-2.

An additional benefit of the factor decomposition, which is not available from the model ensemble approach, is that we can easily determine which of the terms contributes most to uncertainty in RF. For the short-lived O₃ response, uncertainties in d[O₃]_short/dE and dF/d[O₃] contribute roughly equally to the uncertainty in F_short O3. For the long-lived CH₄ response, however, the uncertainty in dL_CH4/dE causes about 80% of the uncertainty in F_long CH4. In both of these cases, the dominant uncertainty originates in factors that require aviation-specific studies. In contrast, for the long-lived O₃ response, almost 50% of the variance in F_long O3 originates in the d[O₃]/d[CH₄] term, which reflects uncertainty in the global response of O₃ to CH₄ perturbations and is not a unique response to aviation emissions. The dL_CH4/dE and dF/d[O₃] uncertainties explain most of the remaining variance of long-lived O₃ RF.

Correlation of RF Components and its Causes

Fig. 2 shows the ensemble of net RF across the models. For three early studies that did not calculate F_long O3, we fill in values by scaling the reported F_long CH4:

[5]

The net RF from aviation NOx for the model ensemble is +4.5 ± 4.5 mW m^-2 for emissions of 1 Tg(N) a^-1. Although 17 of the 21 net RF estimates are positive, the 68% confidence interval includes zero. This range is similar to those in recent studies that use a subset of the models analyzed here [e.g., +4.3 ± 3.4 (20), -0.3 to +6.0 (31)].

Fig. 2.

Steady-state net RF (mW m^-2) caused by aviation NOx emissions. Values are scaled to 1 Tg(N) a^-1. See Fig. 1 caption for the key to multimodel papers. An asterisk (*) indicates that the long-lived O₃ RF was obtained using Eq. 5. The mean and standard deviation is given.

The net RF derived from the factor decomposition is +0.6 mW m^-2. If we naively assume no correlation among the RF components, then its one-sigma range is +0.6 ± 8.3 mW m^-2, which encompasses the ensemble-based estimate and likewise spans zero. However, this assumption of no correlation gives an uncertainty in the net RF that is nearly twice as large as the spread in models, despite the similarity of standard deviations for each RF component found above. This discrepancy persists if we account for correlations due to common factors in Eqs. 2–4 (uncertainty reduces to ± 8.1 mW m^-2). The ensemble-derived standard deviation of net RF is also much smaller than would be expected if the variance of each of the ensemble-derived component RFs were independent (expected ± 11.6 mW m^-2), implying that there are important correlations between the components.

Fig. 3 shows that short-lived O₃ RF and long-lived CH₄ RF are strongly correlated in the model ensemble (R² = 0.79), which has previously been assumed but not demonstrated (15). In the absence of quantitative data, earlier work assumed 100% anticorrelation between the O₃ and CH₄ RF components, but recognized that this assumption would give confidence intervals for net aviation NOx RF that are too small (22). For example, if we assumed 100% anticorrelation, instead of the actual 79%, between the positive and negative RF components from the model ensemble, the inferred uncertainty in net RF would be ± 0.8 mW m^-2, which is one-fifth of the actual ensemble variation. Assuming the same with the factor decomposition, we would expect net RF uncertainty of ± 1.4 mW m^-2, which is one-third of the ensemble variation. Using a subset of the model ensemble studied here, Myhre et al. (7) found that the largest net aviation NOx RF occurred in models with the largest ratio of initial O₃ column change to fractional CH₄ lifetime change and the full ensemble also supports this conclusion.

Fig. 3.

Steady-state RF (mW m^-2) from short-lived O₃ versus long-lived CH₄ caused by aviation NOx emissions. Values are scaled to 1 Tg(N) a^-1. Model data (black diamonds) and their mean (gray circle) are from Fig. 1. UCI CTM sensitivity studies show change relative to the mean: lightning NOx emissions [4 Tg(N) a^-1] reduced 50% (L-NOx/2), surface NOx emissions (45 Tg(N) a^-1) reduced 50% (S-NOx/2), surface CO emissions [983 Tg(CO) a^-1] reduced 50% (S-CO/2), convective fluxes reduced 50% (convection/2), kinetics favoring O₃ production (kinetics), and a previous model version. Long-lived CH₄ RF here excludes the impact on stratospheric water vapor.

To assess the causes of correlation between the RF components, we test the sensitivity of the University of California, Irvine (UCI) CTM to processes and parameters that likely reflect important differences among the models in the ensemble, similar to an earlier study of global O₃ and CH₄ budgets (32). Hoor et al. (20) describe the base CTM configuration (version 5.6). RF changes in the sensitivity tests are overlaid on Fig. 3 based on factor changes given in Table S1. Large and similar changes in both short-lived O₃ and long-lived CH₄ RF components come from halving either convective fluxes or surface NOx emissions [originally 45 Tg(N) a^-1]. Both increase the aviation RF due to short-lived O₃ and decrease the (negative) RF due to long-lived CH₄ by 3 to 8 mW m^-2 each. Halving the lightning NOx source [originally 4 Tg(N) a^-1] similarly shifts both aviation RF components in the same direction as the first perturbations but by less than 2 mW m^-2, possibly because most lightning occurs in the tropics far from the major flight routes. Considering the export efficiency of surface NOx to the free troposphere is 5–20% (33, 34), the changes to d[O₃]_short/dE, dL_CH4/dE and RF per unit change in reactive nitrogen supply to the free troposphere agree within a factor of 3 in these sensitivity tests. In all of these tests the RF responses in the UCI CTM lie along the axis of variation for the model ensemble, indicating that varying treatments of convection, surface emissions, and lightning NOx are plausible causes of the ensemble spread and anticorrelation of aviation RFs. Prior studies show that O₃ and CH₄ RF responses to aviation NOx emissions are sensitive to the spatial variation in background NOx within a single model (8). We suggest that variation in mean tropospheric NOx between models can explain the spread of O₃ and CH₄ RFs in the ensemble.

Changes in aviation RF with temporal development of the UCI CTM (previous version 5.4) also lie along the major axis of ensemble variation, although it is unclear which changes to transport, emissions, wet deposition, and photolysis are responsible. In another sensitivity test we halve the surface CO emissions [originally 983 Tg(CO) a^-1]. Short-lived O₃ RF drops by 8 mW m^-2 in response, because CO is an O₃ precursor, but the long-lived CH₄ RF changes very little, possibly because of the offsetting decrease in OH consumption by CO. To test the effect of kinetic uncertainties, we adjust rate coefficients for gas-phase reactions to the upper limit of their uncertainties (35) in favor of O₃ production from NOx (+15% for HO₂ + NO → OH + NO₂, -10% for NO + O₃ → NO₂ + O₂, -15% for HO₂ + O₃ → OH + 2O₂). This adjustment induces small changes in both RF components. In both of these sensitivity tests the slopes of O₃ vs. CH₄ RF responses depart from the slope of the model ensemble, which suggests that variations in kinetics and CO emissions between models are not major causes of the spread in reported aviation RFs. In a final sensitivity test (not plotted), we use 2004 meteorology instead of 2005. The RF change was negligibly small for both components, indicating that interannual variability has little effect on aviation RF.

Conclusions

RF of aviation NOx emissions is often evaluated through a model ensemble approach where the standard deviation of the ensemble is used as a measure of the uncertainty. From published results, we show that one important metric, steady-state RF, has not changed much in value or uncertainty with model and emission developments over the last decade. Overall, published work suggests that the steady-state, net RF is +4.5 ± 4.5 mW m^-2 for aviation NOx emissions of 1 Tg(N) a^-1, which is consistent with recent studies using a subset of the models analyzed here. Other important RFs from aviation include CO₂, contrails, and induced cirrus. Current CO₂ forcing is about +28 mW m^-2 for 2005 (22), but not in steady state. Contrail and cirrus RFs are typically estimated to be positive [+33 mW m^-2 (22)] although some recent studies find larger positive and negative cirrus effects [-140 to +120 mW m^-2 (36, 37)]. These steady-state contrail and cirrus RF estimates should be increased by about 15% to compare with the 1 Tg(N) a^-1 values used here.

The GWP allows for direct comparison of aviation emissions of NOx with those of CO₂. From the model ensemble here, the net RF from 1 Tg(N) aviation emission pulse integrated over 100 y is +4.5 mW a m^-2 and that of 1 Tg of CO₂ is 0.087 mW a m^-2 (derived from ref. 27). Thus, the ensemble-derived GWP of aviation NOx (as N) is 52 ± 52. If we had included the additional cooling attributed to CH₄-driven changes in stratospheric water vapor (7), then this number would be closer to zero, but the uncertainty would increase. Similar 100-year GWP ranges were derived by Fuglestvedt et al., (31) (-2 to 71) and Myhre et al., (7) (-21 to 67), although the latter included the stratospheric water vapor effect. Given aviation CO₂∶NOx emission ratios of 800∶1 by mass of N (7, 22), the 100-year integrated RF from aviation NOx is about 7% ± 7% of that from aviation CO₂.

Using a complementary approach, we factor the aviation NOx RF into its key terms and assess the best value and uncertainty of those terms, many of which reflect basic knowledge of atmospheric chemistry and RF and are not specific to aviation. The three component RF values and their uncertainties propagated from the factor decomposition, assuming no correlations among the factors, agree with the ensemble mean values and standard deviations. This second method identifies those terms largely responsible for the overall uncertainty and thus can be used to guide research priorities. For example, aviation-specific terms drive most of the uncertainty for the short-lived O₃ and long-lived CH₄ RFs—d[O₃]_short/dE, dF/d[O₃], and dL_CH4/dE—whereas a nonaviation term dominates for long-lived O₃, d[O₃]/d[CH₄]. The sum of the three components (i.e., net RF) derived from the factorization has twice the uncertainty of the model ensemble, when we assume that the factors are uncorrelated. Conversely, if we assume 100% anticorrelation of the positive and negative RF components, the uncertainty in net RF is three to five times smaller than the model ensemble. The short-lived O₃ and long-lived CH₄ RF components are strongly anticorrelated (R² = 0.87) in the model ensemble, but this anticorrelation cannot be identified with the factorization. Thus, the two approaches are synergistic.

Sensitivity studies with the UCI CTM identify processes that control background NOx in the free troposphere as the likely cause of both the spread in model results and the anticorrelation between O₃ and CH₄ RFs. Simulations with less background tropospheric NOx calculated greater impacts from aviation NOx. This trend is consistent with the well-known decrease in O₃ production as NOx increases (e.g., 38, 39) although most aviation studies find no clear evidence of chemical nonlinearities. Improved assessments of the current and future climate impacts from aviation NOx depend on more accurate simulations of tropospheric composition, including advances in modeling convection, lightning, and surface emissions, as well as better characterization of the NOx distribution. Comparisons with NOx measurements in the free troposphere should identify which models match the observations and if these predict similar aviation impacts. However, current NOx climatologies (40) are assembled from campaigns of opportunity and do not have the necessary statistics or coverage to adequately test the models. An observational NOx climatology that includes HOx (OH + HO₂) sources and samples a wide range of latitudes and photochemical regimes in both hemispheres at seasonal intervals would aid process studies in global atmospheric chemistry and their application to changing emissions and climate.