Global free tropospheric NO 2 Abundances Derived using a Cloud Slicing Technique applied to Satellite Observations from the Aura Ozone Monitoring Instrument (OMI)

. We derive free-tropospheric NO 2 volume mixing ratios (VMRs) and stratospheric column amounts of NO 2 by applying a cloud slicing technique to data from the Ozone Monitoring Instrument (OMI) on the Aura satellite. In the cloud-slicing approach, the slope of the above-cloud NO 2 column versus the cloud scene pressure is proportional to the NO 2 VMR. In this work, we use a sample of nearby OMI pixel data from a single orbit for the linear ﬁt. The OMI data include cloud 5 scene pressures from the rotational-Raman algorithm and above-cloud NO 2 vertical column density (VCD) (deﬁned as the NO 2 column from the cloud scene pressure to the top-of-the-atmosphere) from a differential optical absorption spectroscopy (DOAS) algorithm. Estimates of stratospheric column NO 2 are obtained by extrapolating the linear ﬁts to the tropopause. We compare OMI-derived NO 2 VMRs with in situ aircraft proﬁles measured during the NASA Intercontinental Chemical Transport 10 Experiment Phase B (INTEX-B) campaign in 2006. The agreement is generally within the estimated uncertainties when appropriate data screening is applied. We then derive a global seasonal climatology of free-tropospheric NO 2 VMR in cloudy conditions. Enhanced NO 2 in the free troposphere commonly appears near polluted urban locations where NO 2 produced in the boundary layer may be transported vertically out of the boundary layer and then horizontally away from the 15 source. Signatures of lightning NO 2 are also shown throughout low and middle latitude regions in summer months. A proﬁle analysis of our cloud slicing data indicates signatures of uplifted and transported anthropogenic NO 2 in the middle troposphere as well as lightning-generated NO 2 in the upper troposphere. Comparison of the climatology with simulations from the Global Modeling Initiative (GMI) for cloudy conditions (cloud optical thicknesses > 10) shows similarities in the 20 spatial patterns of continental pollution outﬂow. However, there are also some differences in the seasonal variation of free-tropospheric NO 2 VMRs near highly populated regions and in areas affected by lightning-generated NO x . Stratospheric column


Introduction
Tropospheric nitrogen dioxide (NO 2 ) is mainly produced by fossil fuel combustion, biomass burning, and soil emission near the Earth's surface and by lightning and aircraft emissions in middle and upper troposphere. NO 2 is an important tropospheric constituent, because it is both a pollutant and climate agent. It has adverse effects on human health (Brook et al., 2007) and is one of six criteria 30 pollutants designated by the US Environmental Protection Agency (EPA). It is contributes to the formation of ozone, another EPA criteria pollutant. NO 2 also has both direct and indirect radiative effects. The direct effect results from NO 2 absorption of incoming sunlight in the ultraviolet (UV) and visible (VIS) spectral range (e.g., Solomon et al., 2007;Vasilkov et al., 2009). Because NO 2 is an ozone precursor and affects tropospheric concentrations of methane, it also has indirect short-and 35 long-wave radiative effects (e.g. Fuglestvedt et al., 2008;Wild et al., 2001;Shindell et al., 2009). NO 2 has distinct absorption features in the UV/VIS (primarily at blue wavelengths) that can be remotely sensed by satellite spectrometers using Differential Optical Absorption Spectroscopy (DOAS) techniques. For example, tropospheric vertical column densities (VCDs) of NO 2 have been estimated using spectral radiance measurements from the Global Ozone Monitoring Experi-40 ment (GOME) , SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) , the Ozone Monitoring Instrument (OMI) Bucsela et al., 2006Bucsela et al., , 2008, and the Second Global Ozone Monitoring Experiment (GOME-2) (Munro et al., 2006). The retrieved tropospheric columns of NO 2 have been evaluated with aircraft, ground-based, and balloon measurements. For example, OMI-derived 45 VCDs show moderately good agreement with aircraft measurements from the NASA Intercontinental Chemical Transport Experiment-A (INTEX-A) and -B (INTEX-B) Experiment Boersma et al., 2008Boersma et al., , 2011, ground-based direct-sun DOAS measurements (Herman et al., 2009), and multi-axis DOAS measurements Hains et al., 2010).
With their global coverage, satellite tropospheric column estimates have provided important infor-50 mation related to tropospheric NO x chemistry and transport. Satellite retrievals show decreases of NO 2 tropospheric columns over the United States in recent years (Russell et al., 2012;Duncan et al., 2013) and Europe (Castellanos and Boersma, 2012). These reductions result from emission controls and the economic recession. Reductions in NO 2 were also observed over Beijing and the surrounding areas during the 2008 olympic and paralympic games (Witte et al., 2009). Lamsal et al. (2013) 55 showed that OMI-derived surface NO 2 concentrations are highly correlated with urban population, but that the NO 2 to population relationship is geographically dependent. Satellite measurements of tropospheric NO 2 columns have also been utilized to study sources and long range transport of NO x in conjunction with chemical transport models (e.g., Martin et al., 2003Martin et al., , 2006Zhang et al., 2007;Beirle et al., 2004Beirle et al., , 2011Jaeglé et al., 2005;Frost et al., 2006;Boersma et al., 2008;Lin et al., al., 2008;Lamsal et al., 2010).
There have been only a few studies that have utilized cloudy satellite NO 2 observations, and they have primarily focused on lightning-generated NO x (e.g., Boersma et al., 2005). Cloudy data are 65 typically discarded in most studies that use satellite-derived tropospheric NO 2 columns, because clouds screen the near-surface from observation. However, the screening property of clouds can be exploited to provide unique estimates of NO 2 concentrations in the free troposphere using cloudslicing techniques. It is otherwise difficult to separate the boundary layer portion of the NO 2 column from the free-tropospheric contribution. Cloud slicing can also be used to estimate stratospheric 70 NO 2 burdens. Ziemke et al. (2001Ziemke et al. ( , 2003Ziemke et al. ( , 2005Ziemke et al. ( , 2009) pioneered cloud slicing approaches to estimate free tropospheric O 3 concentrations as well as stratospheric column amounts of O 3 . The ozone derived from cloud slicing has been validated by extensive comparisons with ozonesondes (Ziemke et al., 2003) and Microwave Limb Sounder (MLS) data (Ziemke et al., 2009). Ziemke et al. (2010) and Ziemke and Chandra (2010) subsequently derived tropospheric and stratospheric ozone clima-75 tologies, and Ziemke et al. (2010) developed the ozone El Niño-Southern Oscillation (ENSO) index that has been compared with chemistry-climate simulations (Oman et al., 2013).
Measurements of NO 2 in the free-troposphere are sparse. Aircraft in situ measurements, lidar observations, and balloon-sonde soundings have been confined mainly to field campaigns that are limited in spatial and temporal extents. UV/VIS limb soundings provide vertical profiles of NO 2 , 80 but the measurements are limited to the stratosphere (Bovensmann et al., 1999).
Cloud-slicing of NO 2 from satellite measurements can potentially provide additional information about spatial and temporal variations in free tropospheric NO 2 concentrations. Model studies show that lightning NO x production contributes to free tropospheric NO 2 abundances, but magnitudes and distributions are still largely unknown; in particular, vertical distributions of lightning NO x are 85 dependent upon the characteristics of the convection parameterizations in the models (Choi et al., 2005(Choi et al., , 2008Allen et al., 2012;Martini et al., 2011). The NO 2 lifetime in the free troposphere (up to a week or more) allows for intercontinental transport of uplifted anthropogenic and lightninggenerated NO 2 (e.g., Li et al., 2004;Wang et al., 2006;Zhang et al.,, 2008;Walker et al., 2010).
While this transport has been simulated, global NO 2 observations in the free troposphere have not 90 been available for extensive evaluation. In addition, knowledge of the distributions of NO 2 in the free troposphere is important for calculations of its anthropogenic radiative forcing (e.g. Fuglestvedt et al., 2008;Wild et al., 2001;Shindell et al., 2009).
In this study, we use OMI to infer free tropospheric NO 2 VMRs and stratospheric column amounts of NO 2 . To derive these quantities, we use the OMI-inferred above-cloud NO 2 columns and cloud 95 parameters from highly cloudy scenes. We evaluate the derived OMI NO 2 VMRs with available aircraft data from the NASA INTEX-B campaign. We derive a global seasonal climatology of free tropospheric NO 2 VMRs from OMI. For reference, we show an example of a comparison with NO 2 fields simulated by a chemical-transport model, the Global Modeling Initiative (GMI). We also con-struct coarse profiles for several regions with sufficient cloud pressure variability. Finally, we infer 100 seasonal, zonal-mean stratospheric column amounts of NO 2 and compare them with independent estimates including simulations from GMI.
2 Data description 2.1 Space-based measurements from OMI OMI is a UV/VIS grating spectrometer that flies aboard the NASA Aura spacecraft (Levelt et al., 105 2006). Aura is in a sun-synchronous orbit with a local equator crossing time of 13:35 ± 0:05 (ascending node). OMI provides daily global coverage with a nadir pixel size of approximately 13×24 km 2 and a swath width of about 2600 km. It has separate channels for UV and VIS observations. The OMI spectral resolutions in the VIS and UV channels are 0.63 and 0.45 nm, respectively. An obstruction outside the instrument (known as the "row anomaly") has reduced the swath coverage 110 starting in May 2008. In order to avoid the row anomaly, we focus on OMI data obtained from 2005-2007.

OMI cloud scene pressure
OMI has two independent cloud retrieval algorithms. They are described in detail by Stammes et al. (2007). Here, we provide a brief explanation of these algorithms. One algorithm uses the 115 collision-induced O 2 -O 2 absorption band near 477 nm in the VIS channel; its official product name is OMCLDO2 (Acarreta et al., 2004;Sneep et al., 2008). The other makes use of the filling-in effect of rotational Raman scattering (RRS) at wavelengths from 345 to 354 nm in the UV-2 channel (OMCLDRR) (Joiner and Vasilkov, 2006;Vasilkov et al., 2008).
Both algorithms use the Mixed Lambertian Equivalent Reflectivity (MLER) model that accurately 120 reproduces the observed Rayleigh scattering or atmospheric absorption in a cloudy scene (Koelemeijer and Stammes, 1999;Ahmad et al., 2004). The MLER model utilizes the independent pixel approximation; it treats a measured cloudy pixel radiance (I m ) as a weighted sum of two independent subpixels: clear (I clr ) and cloudy (I cld ). The clear and cloudy subpixels are weighted by an effective cloud fraction (f c ), i.e., where P terrain is the terrain pressure and P c is the cloud optical centroid pressure; P c can be considered as a reflectance-weighted pressure located inside a cloud (Vasilkov et al., 2008;Joiner et al., 2012). This is distinct from the cloud-top pressure derived from thermal infrared measurements. To model I cld and I clr , clouds and the Earth's surface are treated as Lambertian reflectors (i.e., through 130 which no light is transmitted). For the clear-sky contribution, the surface LER is taken from a precomputed climatology that varies in space and time. The Lambertian clouds are treated as having a fixed albedo of 0.8. In scenes containing transmissive clouds with an overall LER < 0.8, f c < 1; the clear subpixel contribution (first term in the right-hand side of Eq. 1) accounts for light transmitted through the cloud. We also note that f c is different from the geometric cloud fraction as it is designed 135 to account for cloud transmission within the context of the MLER model. We have found that f c is practically spectrally invariant over the US/VIS wavelengths considered here. In the OMCLDRR algorithm, f c is retrieved by inverting Eq. 1 at a wavelength unaffected by RRS. Then P c is retrieved to be consistent with the observed amount of RRS filling-in.
We also make use of a wavelength-dependent quantity known as the cloud radiance fraction (f r ), 140 defined as the fraction of radiance contributed by clouds (and aerosol). Within the context of the MLER model, f r is computed as Cloud optical centroid pressures from OMCLDO2 and OMCLDRR are very similar, particularly for pixels with high values of f c and f r (Joiner et al., 2012). However, there are some subtle dif-145 ferences, particularly over the Pacific where there is a high incidence of multi-layer clouds. As a result, cloud slicing NO 2 VMRs derived with the two cloud products exhibit some differences in spatial patterns, particularly over equatorial pacific and Gulf of Mexico. In this work, we use P c from OMCLDRR. For reference, we show sample results that use OMCLDO2 P c in Appendix D.

150
NO 2 slant column densities (SCD) are retrieved from solar backscattered radiances in the VIS channel with a spectral fitting window of 405-465 nm. These data are provided in the OMNO2A product . Fitting errors of NO 2 SCDs range from 0.3-1 × 10 15 cm −2 . There is evidence that NO 2 SCDs are positively biased by ∼ 25% (Krotkov, 2013); therefore our estimates from cloud slicing will be biased by the same amount.

155
Here, we divide the OMI NO 2 SCD by the geometric air mass factor (AMF geometric ) to obtain estimates of NO 2 VCDs in highly cloudy conditions. AMF geometric is given by where SZA and VZA are the solar and view zenith angles, respectively. AMF geometric is appropriate for use in an atmosphere where the effects of Rayleigh scattering are relatively small. This is 160 generally the case for highly cloudy observations at NO 2 wavelengths at moderate SZAs.
It is useful at this point to introduce the concept of cloud scene pressure (P scene ) given by The derived NO 2 VCD in a cloudy pixel can be interpreted as the total column from P scene to the topof-the-atmosphere (i.e., the total column above P scene ). Eq. 4 is derived by assuming that the NO 2 165 profile is vertically uniform between P terrain and P c (Joiner et al., 2009). Because this condition will not be met for NO 2 in highly polluted regions, here we use only pixels where f r > 0.9. For these pixels, the below-cloud contribution to the observed VCD (i.e., from the second term on the right hand side of Eq. 4) is small and P scene P c . Like P c , P scene is located below the physical cloud top altitude. Henceforth we refer to the derived NO 2 VCD in a cloudy scene (NO 2 VCD = NO 2 SCD / 170 AMF geometric ) as the above-cloud NO 2 VCD.

OMNO2B estimates of Stratospheric and Tropospheric NO 2 VCDs
Stratospheric NO 2 VCDs are estimated and reported in the OMNO2B product (Bucsela et al., 2013).
We use these estimates as an independent check on our derived stratospheric NO 2 VCDs from cloud slicing. The OMNO2B procedure for estimating stratospheric VCDs is explained in detail by Buc-175 sela et al. (2013). Here we provide a brief explanation of the procedure. First, an initial VCD is obtained by dividing the NO 2 SCD (Sect. 2.1.2) by a stratospheric air mass factor (i.e., the air mass factor is calculated using radiative transfer, assuming that all NO 2 is contained in the stratosphere).
Then, the stratospheric VCD is estimated for two cases: (1) In clean areas (with small amounts of tropospheric NO 2 ), stratospheric VCDs are obtained by subtracting GMI estimates of the tropo-180 spheric column from the initial VCD. Spatial smoothing is then applied to the resulting geographic field; (2) Where there is substantial tropospheric NO 2 pollution, the stratospheric VCDs are estimated using spatial interpolation from the surrounding clean regions. Tropospheric NO 2 VCDs are then estimated by taking the differences between the total and stratospheric SCDs and converting them to VCDs using appropriate stratospheric and tropospheric AMFs. intervals (Thornton et al., 2000;Perring et al., 2010;Bucsela et al., 2008). At 1 Hz, the mixing ratio observations have precisions ranging from ±23 pptv at 1000 hPa to ±46 pptv at 200 hPa at a signal to noise ratio of 2.

GMI model simulation
We use GMI chemical transport model simulations for comparison with our NO 2 cloud slicing results. A detailed model description is provided in Duncan et al. (2007) and Strahan et al. (2007).
Here, we provide a brief explanation of the model. The model is driven by Goddard Earth Observing System 5 (GEOS-5) meteorological fields (Rienecker et al., 2011). The GMI spatial resolution is 200 2°latitude× 2.5°longitude. The GMI vertical extent is from the surface to 0.01 hPa, with 72 levels; vertical resolution ranges from ∼150 m in the boundary layer to ∼1 km in the free troposphere and lower stratosphere. Model outputs are sampled at the local time of the Aura overpass.
The GMI chemistry combines stratospheric chemical mechanisms (Douglass et al., 2004) with detailed tropospheric O 3 -NO x -hydrocarbon chemistry that has its origins in the Harvard GEOS-

205
Chem model (Bey et al., 2001). In addition to chemistry, the model includes various emissions sources, aerosol microphysics, deposition, radiation, advection, and other important chemical and physical processes including lightning NO x production (Allen et al., 2010).
In this study, we extract GMI NO 2 concentrations/burdens for four different sets of conditions and vertical ranges (three tropospheric and one stratospheric): (1)

Cloud slicing technique
The cloud slicing technique takes advantage of optically thick clouds to estimate a VMR of a target trace gas in the free troposphere between the clouds (Ziemke et al., 2001(Ziemke et al., , 2003. We infer NO 2 VMRs using the slope derived from linearly fitting the collocated OMI above-cloud column NO 2 220 to cloud scene pressures. Figure 1 illustrates a simple example of this technique (not to scale). We require at least two nearby above-cloud NO 2 VCDs for different cloud scene pressures as in Fig. 1-(a). The two OMI measurements are shown in a pressure-VCD coordinate plane in Fig. 1- (b). NO 2 VCD (VCD NO 2 ) between the two pressure levels P1 and P2 (P1 < P2) can be derived by integrating the NO 2 VMR (VMR NO 2 ) over pressure from P1 to P2, i.e., where R air is the gas constant, k B is the Boltzmann constant, and g is the gravitational acceleration.
Assuming a constant mixing ratio over the range P1 to P2 in Eq. 5, the mean NO 2 VMR in this pressure interval is given by

230
From this relationship, the NO 2 VMR in the pressure range of OMI cloud measurements is proportional to the fitted slope of NO 2 VCD versus cloud scene pressure, as shown in Fig. 1-(c). The confidence interval of NO 2 VMR also can be derived from the linear fit if more than two observations are available. In Fig. 1-(c), we show the pressure range of the NO 2 VMR (vertical error bar) as well as the confidence interval (horizontal error bar).

235
In addition, a stratospheric column NO 2 can be derived by extrapolating the linear fit to the tropopause as shown in Fig. 1 -(d). This estimate is based on the assumption that the mixing ratio of NO 2 remains constant from the lowest cloud pressure to the tropopause. By assuming a uniform free tropospheric NO 2 VMR, we limit the number of retrieved parameters to 2 (slope and y-intercept, related to free-tropospheric VMR and stratospheric VCD, respectively). This simplifies 240 the retrieval and its error analysis.
While the cloud slicing technique derives the free tropospheric NO 2 VMR without the need for a prescribed stratospheric column or other a priori information, it relies on several assumptions and conditions. The method works well only with a relatively large number of nearby cloudy OMI pixels that have a sufficient variation in cloud pressure. We also note that the derived NO 2 VMR 245 information is based on the assumption that NO 2 is vertically and horizontally well mixed in the given pressure range and spatial extent of the OMI pixel collections. In addition, we assume that the stratospheric column remains constant during the time period and over the area of the OMI pixel sample. Finally, the absolute magnitudes of the derived tropospheric mixing ratios and stratospheric columns are only as accurate as the above-cloud NO 2 VCDs. Errors in the derived cloud scene 250 pressures may contribute additional uncertainty. It should also be noted that the NO 2 VMRs are derived in highly cloudy conditions. These conditions may not be representative of the general all-sky atmosphere.
In order to ensure that appropriate data are used for cloud-slicing, we apply rigorous data filtering criteria. This results in the use of approximately 10-15 % of the available pixel data depending on 255 season and geolocation. The data selection criteria are summarized in Table 1 and discussed in detail in Appendix A1.
Although we show a case of two adjacent OMI measurements in Fig. 1      Overall, this comparison, even with its intrinsic limitations, provides some confidence in the ability to estimate NO 2 mixing ratios with OMI cloud slicing.
For comparison between OMCLDRR and OMCLDO2 results, a scattergram using OMI VMRs magnitudes and scatter as compared with OMCLDRR. When we exclude the high standard error cases, OMCLDO2 data result in slightly higher scatter and lower correlation versus INTEX-B.

Global Seasonal Climatology of free tropospheric NO 2 VMR
We construct a seasonal climatology of OMI free tropospheric NO 2 as explained in Sect. 3 and 345 Appendix A2. In analyzing the global climatology, we focus on spatial and temporal variations of the NO 2 VMR rather than its absolute magnitude. In this section, we examine aspects of the OMI free tropospheric NO 2 climatology in the context of anthropogenic and lightning contributions. We also show GMI free tropospheric NO 2 VMRs for comparison.

Anthropogenic contributions
In the northern hemisphere (NH) winter (Dec-Feb), the primary source of free tropospheric NO 2 appears to be anthropogenic emissions; high free tropospheric VMRs are seen over densely populated 390 regions and the lightning contribution is expected to be negligible during these months (top right panel of Fig. 4). Over most of the highly populated areas of North America, southeast (SE) Asia, and Europe, free tropospheric NO 2 VMRs are higher in winter (Dec-Feb) as compared with summer (Jun-Aug). It is well known that boundary layer NO 2 VMRs are generally higher in winter as compared with summer owing to a longer chemical lifetime in winter; the OMI-derived tropospheric 395 columns (the first row of Fig. C3 in Appendix C), that are dominated by boundary layer pollution in heavily populated areas, also reflect higher values in winter than in summer. In contrast to NO 2 VMRs from OMI, the NO 2 VMRs from GMI are higher in summer as compared with winter over southeast Asia (the second row of Fig. 4 for cloudy conditions, and Fig. C2 for all-sky conditions), while the tropospheric column NO 2 from GMI is higher in winter in this region (the second row of Overall, OMI NO 2 VMRs have lower values in the SH during the austral winter as compared with the NH. This is also shown in the GMI output. It should be noted that there are not many large population centers in the SH, particularly at high latitudes, nor as much NO x contribution from 405 aircraft at high latitudes in the SH as compared with the NH. However, it should also be noted that cloud slicing data are not available around many of the major population centers in the SH (e.g., Johannesburg, South Africa and Sao Paulo, Brazil) owing to a lack of optically thick clouds and/or cloud pressure variation.
Regarding transport of anthropogenic NO 2 , we focus on winter months when lightning NO 2 con-410 tributions are likely to be small. The OMI cloud slicing NO 2 climatology shows a spatial patterns consistent with pollution outflow from North America and Asia. For example, the persistent Asian northeasterly outflow of NO 2 via the Bering Sea resembles that of CO (e.g., Liang et al., 2004), a tracer of incomplete combustion emissions. The spatial extents of continental outflows are different for the free tropospheric VMRs and tropospheric columns. This might be explained by extended 415 transport at higher altitudes where the NO 2 lifetime is longer.

Lightning contributions
A band of enhanced NO 2 appears extensively during the summer in the both hemispheres (∼0-30°a nd possibly higher latitudes in the NH). The low cloud scene pressures (shown in the fifth row of Fig. C1 in Appendix C) in these regions are indicative of frequent convection. In particular, extensive 420 enhancements in summertime NO 2 VMRs over NH tropical and subtropical oceans, are similar to modeled lightning NO x enhancements in previous studies (e.g., Choi et al., 2008;Allen et al., 2012;Martini et al., 2011;Walker et al., 2010). This suggests that lightning is a major source of free tropospheric NO 2 in tropical and subtropical regions in summer. Because the SH is far less polluted than the NH, potential NO 2 enhancements due to lightning are more apparent there. Finally, we note 425 that these extensive NO 2 enhancements indicated by cloud slicing during summer over oceans are not as apparent in the OMI tropospheric columns.
While the locations of these apparent lightning-enhancements of NO 2 are similar in summer in both GMI and OMI data sets, there are a few key differences to note. For example, the seasonality of the NO 2 enhancements over oceans shown by OMI is not as strongly reflected in the GMI out-430 put. In addition, there appears to be a much stronger land/ocean contrast in GMI than in the OMI climatology.
For comparison, we also show maps of free-tropospheric NO 2 climatology obtained with OM-CLDO2 cloud data in Fig. D2 of Appendix D. The OMCLDO2 climatology shows very similar spatial and temporal patterns as compared with that derived using OMCLDRR data presented here 435 with slightly lower VMRs in general. However, the OMCLDO2 climatology does not show a strong signature of lightning-enhanced NO 2 over the tropical North Pacific in Jun-Aug as is shown in the OMCLDRR climatology. This is discussed in more detail in Appendix D.

Profile analysis
We examine the pressure dependence of the derived VMRs over large regions (to reduce random 440 errors) in order to provide a rough vertical distribution of free tropospheric NO 2 . We highlight two types of areas: (1) East Asia and its outflow region to focus on anthropogenic contributions, and (2) tropical portions of the NH and SH to examine potential lightning contributions. East Asia; we attempt to avoid boundary layer contamination in order to preserve the assumption of uniform NO 2 VMRs over the observed cloud pressure range. We obtain a profile down to 850 hPa in the outflow region because there is little boundary layer pollution in that area.The profile of East Asia clearly indicates the presence of uplifted anthropogenic NO 2 in the middle troposphere of 600-800 hPa. In the outflow region, the NO 2 VMRs are higher at P 700 hPa as compared with those 455 at P> 800 hPa. This suggests that there is not a significant surface source NO 2 , and that uplifted anthropogenic NO 2 is transported at around ∼700 hPa or above in this region. Figure 6 shows variations in the derived NO 2 profiles in tropical regions of the NH and SH. Here, we examine two latitudinal bands with enhanced summertime NO 2 based on the spatial distributions shown in Fig. 4. Again, owing to the large number of samples, the standard errors are relatively small 460 (∼5 pptv). In summer, the NO 2 VMRs increase with altitude in both hemispheres. The profile shapes suggest that NO 2 sources, presumably lightning, are located primarily in the upper troposphere in these regions. This is consistent with aircraft measurements (e.g., Huntrieser et al., 2009) and modeling studies (e.g., Allen et al., 2010Allen et al., , 2012Martini et al., 2011) of lightning-generated NO x . In contrast, NO 2 VMR profiles are more uniform in winter, possibly owing to less frequent lightning 465 activity associated with convection in the shifting Inter-Tropical Convergence Zone (ITCZ). We note that the winter baseline NO 2 VMR is higher in NH by approximately a factor of two possibly due to more pollution sources in NH. In contrast, the summertime profiles of NO 2 are very similar in the NH and SH.
Overall, our analysis indicates a capability of the cloud slicing technique to retrieve NO 2 profile 470 information when provided with a relatively large sample size. Our profile results are consistent with an anthropogenic source for the enhanced NO 2 in middle to high latitudes off the coasts of highly populated areas. They also indicate a lightning source in the summer over tropical areas, primarily located in the upper troposphere.

475
We generated estimates of stratospheric NO 2 columns as described in Sect. VMRs do not always exhibit good agreement. Small-scale temporal and spatial variability, poor 500 collocation, and fairly large OMI measurement uncertainties contribute to these discrepancies.
We generated global seasonal maps of free tropospheric NO 2 VMRs as well as free tropospheric NO 2 vertical profiles over selected regions. With appropriate data filtering over a three year time period, we obtain a sufficient number of cloudy OMI measurements to cover most of the Earth.
Confidence intervals for individual cloud slicing VMRs are fairly large; however, averaging over nine 505 months (3 months × 3 years) reduces random errors and provides a reasonable estimate of the mean values. The free-tropospheric NO 2 VMR climatology shows distinct spatial and seasonal patterns; these patterns differ from those of OMI-estimated tropospheric NO 2 columns. The combination of mapped and profile analyses indicates that spatial patterns of the OMI-derived free tropospheric NO 2 are consistent with (1) uplifted anthropogenic NO 2 over densely populated regions; (2) continental 510 outflow of anthropogenic NO 2 ; and (3) lightning-generated NO x , particularly in summer months at low to middle latitudes with a source located primarily in the upper troposphere. Anthropogenic sources appear to dominate in the winter hemisphere, especially in the northern hemisphere at high latitudes near heavily populated regions, while lightning contributions dominate over ocean at low to middle latitudes in summer in both hemispheres.

515
GMI model simulations suggest that NO 2 VMRs vary with cloud conditions by altering the photochemistry. Spatial patterns of continental outflow show general agreement between the OMI cloud slicing climatology and GMI simulations for cloudy conditions. However, some differences, particularly with respect to lightning-generated NO x , were noted.
We also provided estimates of NO 2 stratospheric columns from the cloud slicing technique. These estimates agree well with those from the OMNO2B algorithm that are based on a completely independent technique (NO 2 columns over clean regions). The two OMI stratospheric NO 2 estimates display similar seasonal and latitudinal zonal mean variations. These variations are also consistent with those produced in GMI simulations. The excellent agreement between these stratospheric column NO 2 estimates provides a closure validation of the free tropospheric OMI cloud slicing results.

525
Our overall analysis shows that the cloud slicing technique can provide valuable information on the free tropospheric distribution of NO 2 that is distinct from the derived tropospheric total columns.
In particular, we expect to apply this technique to future geostationary missions including the NASA Earth Ventures Instrument (EVI) 1 selected mission Tropospheric Emissions: Monitoring of Pollution (TEMPO) over the North America (Chance et al., 2013) and the Korean Geostationary Environ-530 ment Monitoring Spectrometer (GEMS) over the Asia-Pacific region (Kim, 2012). These missions should provide excellent cloud slicing results; they will provide improved sampling (with higher spatial and temporal resolutions) as compared with OMI.

Appendix A
Additional details in applying the cloud slicing technique 535

A1 Data Filtering Criteria
We apply the following checks to ensure that only high quality data are used in our analysis. With these checks, approximately 10-15 % of OMI pixels are retained, depending on season and geolocation: (1) we use only pixels with f r > 0.9 to remove OMI pixels with an insufficient cloud shielding of the boundary layer; (2) we remove data with aerosol indices > 1.0, because absorbing aerosols are 540 known to produce biases in the retrieved cloud properties (Vasilkov et al., 2008); (3) we exclude data with solar zenith angles (SZA) > 80°; the use of the geometrical AMFs may not be appropriate at higher SZAs owing to higher amounts of Rayleigh scattering; (4) we exclude data affected by snow and ice because UV/VIS cloud measurements cannot differentiate between snow/ice and clouds; In the presence of snow/ice, we cannot be assured of boundary layer cloud shielding. We use a flag 545 for snow-and ice-covered pixels based on the Near-real-time SSM/I EASE-grid daily global Ice and snow concentration and Snow Extent (NISE) data set (Nolin et al., 1998) provided in OMCLDRR product.
We also apply checks to ensure sufficient cloud variability; we only use collections with at least 30 OMI pixels, a cloud pressure standard deviation > 35 hPa, and a cloud pressure range > 200 550 hPa. Finally, we employ outlier checks to remove data that fall outside the range expected from our assumptions including a uniform mixing ratio over the appropriate pressure range and homogeneous stratospheric column over the corresponding area; we empirically selected a threshold of 2σ from the linear fit for this check.
A2 Application of cloud slicing to seasonal climatology 555 In order to create a global seasonal climatology of free-tropospheric NO 2 VMRs, we average individual retrievals in three month segments (one for each season) using data collected over 3 years (2005)(2006)(2007). We grid the data at a spatial resolution of 6°latitude × 8°longitude.
In Fig. A1, we show two examples of how the NO 2 VMRs are calculated for a single grid box.
For these examples, we use only one month in summer (June) and winter (January). The grid box 560 encompasses New York City, NY, USA. In order to remove pixels affected by substantial vertical gradients in the NO 2 VMR, we use only cloudy data with P scene < a lower boundary (P lower , gray lines) where the mean NO 2 vertical profile is relatively well mixed according GMI; specifically, P lower is pressure above which the absolute magnitude of vertical gradient of monthly-mean NO 2 VMR < 0.33 pptv/hPa. Note that P lower varies with season (as shown in Fig. A1) and geolocation 565 (not shown). For reference, we also show GMI daily and monthly mean profiles.
Using an OMI pixel collection from a single orbit, we calculate the free tropospheric NO 2 VMR (small black dots), the confidence interval (horizontal bars), and the pressure range (vertical bars).
Then, we average the derived single-orbit NO 2 VMRs (weighted inversely by the square of the confidence intervals) to obtain a single representative NO 2 VMR for the given time period (large 570 black dots).
In Fig. A1, we have shown data from one month for simplicity. To construct a seasonal climatology, we use the same spatial grid but a larger temporal window (3 months×3 years) to reduce the sampling biases and random noise. For quality control of the climatology, we show data only where the NO 2 VMR standard error of the mean < 50% for NO 2 VMR > 20 pptv or NO 2 VMR standard 575 error of the mean < 10 pptv for NO 2 VMR ≤ 20 pptv. With these criteria, there are some areas with no OMI-derived NO 2 VMRs. These are mainly areas with little variability in cloud pressure or regions covered with ice/snow. A similar approach is used to obtain gridded values of the stratospheric NO 2 column.

Appendix B
580

Additional case studies of OMI and INTEX-B comparisons
We show additional comparisons in which OMI and INTEX-B NO 2 VMR display poor agreement.
These discrepancies are presumably caused by small-scale spatial and temporal variations in NO 2 VMRs, different cloud conditions that might alter the NO x photochemistry, and/or poor collocations. Figure B1 shows a case with discrepancies likely due to the differences in the locations, times, and 585 the spatial scales of the measurements. The DC-8 profile was taken over a small area near Houston  Figure B2 shows an example of small scale spatial variations in NO 2 profiles as seen by the aircraft measurements. The second column of Fig. B2 shows two DC-8 NO 2 profiles that were taken on the same day at nearby locations. The first column shows the two corresponding OMI 595 pixel collections closest to the DC-8 profiles. In order to differentiate the two cases, the first row uses dark blue for tne DC-8 profile and light blue for OMI pixels, and the second row uses red for the DC-8 profile and pink for OMI pixels. Since the two DC-8 profiles encompass many of the same OMI pixels, the shared pixels are marked with purple on the map (top right This variability may be due to actual variability in the NO 2 profile over the course of a day and/or errors in the OMI measurements. Figure B3 shows a case of OMI cloud slicing VMR variation between orbits for one DC-8 NO 2 profile. The first and second panels of Fig. B3 show two OMI pixel collections taken from two adjacent orbits on the same day. They correspond to one DC-8 610 profile taken over the Pacific north of Hawaii. Even though the OMI pixel collections cover a similar area and time, the resulting NO 2 VMRs differ by ∼30 pptv. This variability may be due to a small scale feature such as a transported pollution plume, altered photochemistry due to the different solar illuminations or cloud conditions, and/or measurement uncertainties in the OMI data, although the differences appear to be outside the expected OMI uncertainties.

615
Appendix C Auxiliary data to interpret cloud slicing NO2 VMR Here, we show auxiliary data that is helpful for quality assurance and interpretation of the NO 2 VMR climatology. The first row of Fig. C1 shows the gridded numbers of OMI pixel collections that are used to derive the seasonal free tropospheric NO 2 climatology. The maps show a sufficiently large 620 number of collections (> 60) for many areas of interest. Large numbers of collections are available over the frontal storm track regions of the North Atlantic, North Pacific and Southern ocean as well as the intertropical convergence zone (ITCZ). In addition, there are large numbers of orbits at high latitudes (> 60°), because these regions can have more than one overpass (orbit) per day. However, some relatively cloud free areas (e.g., the Sahara) as well as oceanic regions, in areas of subsidence with little cloud pressure variability, have smaller numbers of collections (<20).
The second row of Fig. C1 shows the weighted root mean square (RMS) of 95% confidence intervals of NO 2 VMRs. As discussed above, the confidence interval is a measure of the fitting VMRs. This is a measure of how much the individually fitted NO 2 VMRs vary in each grid box.
Similar to the confidence interval, the standard deviations are large in areas of high NO 2 VMRs (major urban areas and continental plumes) and areas with small clouds amounts and/or small cloud variability (deserts and oceans near 20°N latitude). In addition, high standard deviations are present near ∼60°S in Sep.-Nov., possibly owing to stratospheric variability and/or larger errors at high solar 640 zenith angles.
The fourth row of Fig. C1 shows maps of the standard error of the mean for the gridded NO 2 VMR climatology (i.e., the standard deviation divided by square root of the number of measurements). The standard errors provide an estimate of uncertainty for the spatial and temporal variations shown in the climatology (in the absence of a constant bias). We use this quantity for quality control as described 645 in Sect. 4.2.
The fifth row of Fig. C1 shows maps of the OMCLDRR cloud scene pressure for the gridded NO 2 VMR climatology. Owing to significant light penetration inside clouds, the lowest mean cloud pressures are around 450 hPa, well below the typical cloud top pressures. The cloud pressures also vary with season.
650 Figure C2 shows seasonal mean GMI free tropospheric NO 2 VMRs for all-sky conditions. While the maps of all-sky VMR show similar patterns as compared with those of cloudy conditions, all-sky NO 2 VMRs are generally lower over urban regions and higher over oceans than cloudy NO 2 VMRs. Figure C3 shows tropospheric column NO 2 from OMI (upper row) and GMI (bottom row). OMI and GMI tropospheric columns NO 2 agree very well, showing higher columns in winter and lower 655 columns in summer over major urban areas. This seasonal variation is also shown in the OMI climatology of free tropospheric NO 2 VMR as presented in Sect. 4.2.1.

Appendix D OMCLDO2 sample results
While we used OMCLDRR cloud parameters for analysis in the main text, here we show results 660 obtained when using cloud parameters from the OMCLDO2 product. Similar to Fig. 3, Figure D1 shows a scattergram of INTEX-B and OMI cloud slicing NO 2 VMRs, but using OMCLDO2 cloud data.  Joiner et al. (2010) showed that there is a high frequency of multi-layer clouds in the 685 NH Pacific. The two cloud algorithms may behave differently in these complex conditions as Raman scattering has a linear response with cloud pressure, while oxygen dimer absorption has a pressuresquared dependence.
Acknowledgements. This material is based upon work supported by the National Aeronautics and Space Administration under agreement NNH10ZDA001N-AURA issued through the Science Mission Directorate for the 690 Aura Science Team managed by Kenneth Jucks. We thank the the OMI data processing team and algorithm developers, particularly F. Boersma and P. Veefkind, the GMI data processing team, particularly S. Strahan, and