Nitrogen oxides in the global upper troposphere: interpreting cloud-sliced NO₂ observations from the OMI satellite instrument

Nitrogen oxides (${\mathrm{NO}}_x\equiv \mathrm{NO}+{\mathrm{NO}}_{\mathrm{2}}$) in the upper troposphere (UT) have a large impact on global tropospheric ozone and OH (the main atmospheric oxidant). New cloud-sliced observations of UT NO₂ at 450–280 hPa (∼6–9 km) from the Ozone Monitoring Instrument (OMI) produced by NASA and the Royal Netherlands Meteorological Institute (KNMI) provide global coverage to test our understanding of the factors controlling UT NO_x. We find that these products offer useful information when averaged over coarse scales ($\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$, seasonal), and that the NASA product is more consistent with aircraft observations of UT NO₂. Correlation with Lightning Imaging Sensor (LIS) and Optical Transient Detector (OTD) satellite observations of lightning flash frequencies suggests that lightning is the dominant source of NO_x to the upper troposphere except for extratropical latitudes in winter. The NO₂ background in the absence of lightning is 10–20 pptv. We infer a global mean NO_x yield of 280±80 moles per lightning flash, with no significant difference between the tropics and midlatitudes, and a global lightning NO_x source of 5.9±1.7 Tg N a⁻¹. There is indication that the NO_x yield per flash increases with lightning flash footprint and with flash energy.

1 Introduction

Nitrogen oxides (${\mathrm{NO}}_x\equiv \mathrm{NO}+{\mathrm{NO}}_{\mathrm{2}}$) in the upper troposphere (UT) have profound effects on the oxidizing capacity of the atmosphere and on climate, but the factors controlling their concentrations are poorly understood. NO_x in the UT impacts climate by efficiently producing ozone where it is a potent greenhouse gas (Dahlmann et al., 2011; Worden et al., 2011; Rap et al., 2015) and by increasing the concentration of OH (the main tropospheric oxidant) (Murray et al., 2012, 2014). Primary NO_x sources in the UT include lightning, aircraft, convective injection, and downwelling from the stratosphere (Ehhalt et al., 1992; Jaeglé et al., 1998; Bertram et al., 2007). NO_x cycles chemically with reservoir species including nitric acid (HNO₃), pernitric acid (HNO₄), dinitrogen pentoxide (N₂O₅), peroxyacetyl nitrate (PAN), and other organic nitrates, thus defining the NO_y chemical family (NO_y≡NO_x + reservoirs). Effective loss of NO_x from the UT is through subsidence of NO_y to lower altitudes where deposition of HNO₃ provides the ultimate sink. The residence time of NO_y in the UT is 10–20 days (Prather and Jacob, 1997). The lifetime of NO_x against conversion to short-lived reservoirs varies from ∼3 h in the convective outflow of thunderstorms to 0.5–1.5 days in background air (Nault et al., 2016). Chemical recycling from these reservoirs maintains relatively high UT NO_x background concentrations (Bradshaw et al., 2000; Baehr et al., 2003; Nault et al., 2016).

Representation of lightning NO_x in chemical transport models (CTMs) is particularly uncertain. Physically based parameterizations relating lightning frequency to deep convective cloud tops, convective mass flux, convective precipitation, or high-cloud ice content have poor predictive capability (Tost et al., 2007; Allen et al., 2010; Murray et al., 2012; Finney et al., 2014), limiting our ability to estimate the response of lightning NO_x to future climate (Finney et al., 2016, 2018). An alternative is to prescribe flash densities from space-based observations and static NO_x production rates per flash (Sauvage et al., 2007; Allen et al., 2010; Murray et al., 2012). NO_x production efficiencies per flash in the literature vary from <10 to 5000 moles nitrogen per flash (mol N fl⁻¹) (Schumann and Huntrieser, 2007; Murray, 2016). Global CTMs typically use 100–500 mol N fl⁻¹, sometimes assuming higher production rates at midlatitudes than in the tropics (Hudman et al., 2007; Ott et al., 2010), and a global lightning NO_x source of 3–7 Tg N a⁻¹ to match observations of tropospheric ozone and NO_y species (Sauvage et al., 2007).

Our understanding of UT NO_x has so far been evaluated with observations from aircraft campaigns (Drummond et al., 1988; Jacob et al., 1996; Crawford et al., 1997; Jaeglé et al., 1998; Bradshaw et al., 2000; Hudman et al., 2007; Stratmann et al., 2016). There are also long-term NO_x measurements from instruments on board commercial aircraft dating back to the 1990s, but these are mostly over the north Atlantic and the NO₂ measurements have low precision and interference from thermally unstable NO_x reservoir compounds (Brunner et al., 2001). A number of studies have used satellite observations of tropospheric NO₂ columns from solar backscatter to infer lightning NO_x emissions (Beirle et al., 2010; Pickering et al., 2016; Bucsela et al., 2010), including in combination with global models (Boersma et al., 2005; Martin et al., 2007; Miyazaki et al., 2014). These studies estimate global lightning NO_x emission of 1 to 8 Tg N a⁻¹.

New cloud-sliced satellite products of tropospheric NO₂ mixing ratios at 280–450 hPa (∼6–9 km) offer additional vertical resolution by retrieving partial NO₂ columns above clouds and exploiting differences in heights of neighboring clouds to calculate NO₂ mixing ratios (Choi et al., 2014; Belmonte-Rivas et al., 2015). There are two new products of seasonal mean UT NO₂ mixing ratios retrieved from Ozone Monitoring Instrument (OMI) partial NO₂ columns by research groups at the Royal Netherlands Meteorological Institute (KNMI) and NASA. The KNMI product has been evaluated against UT NO₂ from the Tracer Model version 4 (TM4) CTM. Large regional differences between OMI and TM4 are attributed to model deficiencies in lightning NO_x and uplift of anthropogenic pollution (Belmonte-Rivas et al., 2015). The NASA UT product is new to this work and follows a similar retrieval approach to the midtropospheric (900–650 hPa) product of Choi et al. (2014). That product was evaluated with aircraft observations of NO₂ and interpreted with the Global Modeling Initiative (GMI) CTM (Choi et al., 2014). Choi et al. (2014) identified large discrepancies between modeled and observed NO₂ seasonality over regions influenced by pollution and lightning.

Here, we compare the two UT NO₂ products, obtained with distinct retrieval methods, and use aircraft observations of NO₂ from multiple NASA DC8 aircraft campaigns to arbitrate and evaluate the information that can be derived from the satellite datasets. We go on to test current understanding of UT NO_x and the implications for lightning emissions using the GEOS-Chem CTM.

2 OMI observations of upper troposphere NO₂

OMI is on board the NASA Aura satellite launched into Sun-synchronous orbit in July 2004. It has an overpass time of about 13:30 LT, a swath width of 2600 km, and a horizontal resolution of 13 km × 24 km at nadir (Levelt et al., 2006). Columns of NO₂ along the instrument viewing path (slant columns) are obtained by spectral fitting of solar backscattered radiation in the 405–465 nm window (Boersma et al., 2011). Standard products of total and tropospheric column NO₂ are screened for cloudy scenes using a cloud radiance fraction threshold of 0.5. Partial columns of NO₂ above cloudy scenes can be used to estimate vertically resolved NO₂ mixing ratios, as was first demonstrated with satellite observations of ozone (Ziemke et al., 2001). This approach, so-called cloud slicing, assumes a uniform trace gas concentration between two horizontally nearby clouds at different altitudes, so that the UT NO₂ mixing ratio is proportional to the slope of the partial columns versus the corresponding cloud pressures at the optical center of the cloud. Two products of seasonal mean UT NO₂ have been retrieved from OMI following distinct retrieval steps detailed below: a product from KNMI at 330–450 hPa for 2006 (Belmonte-Rivas et al., 2015) and from NASA at 280–450 hPa for 2005–2007 following an approach similar to that used to retrieve midtropospheric NO₂ (Choi et al., 2014). In what follows, we distinguish between the two OMI NO₂ products as KNMI and NASA.

The KNMI product uses DOMINO v2.0 slant columns (Boersma et al., 2011) and cloud fractions and altitudes from the O₂–O₂ absorption cloud product, OMCLDO2 (Acarreta et al., 2004) over partially to very cloudy scenes (cloud radiance fraction >0.5). Contamination due to NO₂ from below (up to 66 % over polluted land masses) is estimated using the TM4 model and removed. Stratospheric NO₂ from an assimilated product (Belmonte-Rivas et al., 2014) is also removed. An air mass factor (AMF) (detailed in Boersma et al., 2004) that accounts for viewing geometry, surface albedo, light attenuation by gases along the viewing path, and sensitivity to NO₂ vertical distribution is applied to the resultant partial slant columns to convert to vertical columns. Additional data filtering removes scenes with solar zenith angle (SZA) ≥ 70^∘ and surface albedo ≥30 %. Resultant daily vertical partial columns are aggregated on consistent pressure and horizontal ($\mathrm{1}{}^{\circ }\times \mathrm{1}{}^{\circ }$) grids and used to determine seasonal mean UT NO₂ mixing ratios for grid squares with at least 30 measurements. UT NO₂ centered at 380 hPa (range 330–450 hPa) is estimated as the difference between partial tropospheric columns retrieved above two neighboring clouds with cloud pressures in the ranges 330–450 and 380–500 hPa, respectively. Biases from sampling cloudy scenes, such as the effect of clouds on photochemistry, are corrected using TM4. These are small (typically <20 %) in the UT (Belmonte-Rivas et al., 2015).

The NASA UT NO₂ product for 2005–2007, centered at 350 hPa (∼280–450 hPa), uses updated version 3 slant columns (OMNO2 v3.0) (Krotkov et al., 2017) that correct for a positive bias in the DOMINO v2.0 product with improved spectral fitting (Marchenko et al., 2015; van Geffen et al., 2015). Partial columns from the cloud height to the top of the atmosphere are retrieved for individual OMI pixels above very cloudy scenes (cloud radiance fraction >0.7) to minimize contamination from below. Cloud fraction and height are from the OMCLDO2 product (Acarreta et al., 2004). The AMF accounts for viewing path geometry and light scattering by clouds with uniform scatter that is optically thick and geometrically thin (near-Lambertian clouds). Data filtering is applied to remove scenes with SZA > 80^∘, snow/ice cover, and severe aerosol pollution that could be misclassified as clouds. Daily UT NO₂ is estimated for neighboring partial columns with sufficient cloud variability (cloud pressure distance >160 hPa) and well-mixed NO₂ (NO₂ vertical gradient <0.33 pptv hPa⁻¹ diagnosed with the GMI CTM). The stratospheric column is assumed uniform above neighboring clouds and thus is removed when differencing two nearby partial columns. Daily values of UT NO₂ are gridded to obtain seasonal means at $\mathrm{5}{}^{\circ }\times \mathrm{8}{}^{\circ }$ (latitude × longitude) for scenes with at least 50 measurements. Gaussian weighting is applied to assign higher weighting to UT NO₂ closest to 350 hPa. Choi et al. (2014) used a similar approach to retrieve midtropospheric NO₂ except that cloud fraction and height were from the rotational Raman scattering product (OMCLDRR), and successful retrieval required a stricter cloud radiance fraction of 0.9, a minimum of 30 measurements, and a wider minimum cloud pressure distance of 200 hPa. A shift in cloud radiance fraction threshold from 0.9 (Choi et al., 2014) to 0.7 (this work) only introduces a small (<5 %) difference in the retrieved partial columns due to contamination from below, as estimated by Pickering et al. (2016) for OMI scenes over the Gulf of Mexico.

Figure 1 compares seasonal mean UT NO₂ from the two satellite products in December–February and June–August. KNMI NO₂ is gridded to the NASA coarse grid. Data for March–May and September–November are in the Supplement (Fig. S1). KNMI NO₂ has greater coverage than the NASA product due to a lower cloud fraction threshold in the retrieval. The two products exhibit very different spatial features. Spatial correlation between the two products (Pearson's correlation coefficient between coincident grid squares) is R=0.41 in December–February and R=0.38 in June–August. There is marginal improvement in the correlation with further spatial averaging. At $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$, we find R=0.50 in December–February and R=0.45 in June–August. The correlation only increases substantially in September–November from R=0.49 at $\mathrm{5}{}^{\circ }\times \mathrm{8}{}^{\circ }$ (Fig. S1) to R=0.66 at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$. KNMI is systematically lower than NASA in all seasons for coincident grid squares, varying from 16 % lower in June–August to 48 % lower in December–February at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$.

Figure 1 Upper troposphere (UT) NO₂ from the OMI satellite instrument. Seasonal mean UT NO₂ from KNMI in 2006 at 330–450 hPa (a, b) is compared to NASA in 2005–2007 at 280–450 hPa (c, d). Data are at $\mathrm{5}{}^{\circ }\times \mathrm{8}{}^{\circ }$ horizontal resolution for December–February (a, c) and June–August (b, d). Grey areas indicate no data and, for NASA, scenes with fewer than 50 measurements.

Contamination of UT NO₂ from below the cloud may still be present in the datasets despite attempts to correct for this using the TM4 model in the case of KNMI and by only considering very cloudy scenes in the case of NASA. These include a large enhancement in KNMI NO₂ (>90 pptv) over southern Africa in June–August when there is intense biomass burning, and the NO₂ hotspot over northeast China in all seasons in both products (Figs. 1, S1). Belmonte-Rivas et al. (2015) caution that the contamination correction in the KNMI product relies on accurate simulation of NO₂ vertical distribution.

Figure 2 NASA DC8 upper troposphere NO₂ over North America in spring–summer (March–August). Observations are from the thermal-dissociation laser-induced fluorescence (TD-LIF) instrument at 450–280 hPa, 11:00–16:00 LT, and without stratospheric influence. Campaigns include the Intercontinental Chemical Transport Experiment – North America Phase A (INTEX-A) in June–August 2004 (Singh et al., 2006), Phase B (INTEX-B) in March–May 2006 (Singh et al., 2009), Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS) in March–April and June–July 2008 (Jacob et al., 2010), Deep Convective Clouds and Chemistry (DC3) in May–June 2012 (Barth et al., 2015), and Studies of Emissions, Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys (SEAC⁴RS) in August 2013 (Toon et al., 2016).

3 Evaluation of OMI upper troposphere NO₂ with aircraft observations

We evaluate the OMI UT NO₂ products with observations from NASA DC8 aircraft campaigns over North America and Greenland in spring–summer, for which dense coverage is available (Fig. 2). These include the Intercontinental Chemical Transport Experiment – North America Phase A (INTEX-A), Phase B (INTEX-B), Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS), Deep Convective Clouds and Chemistry (DC3), and Studies of Emissions, Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys (SEAC⁴RS) campaigns. Only INTEX-B is in the same year as the OMI products but we consider interannual variability to be a small source of error. All NO₂ measurements are from thermal-dissociation laser-induced fluorescence (TD-LIF) instruments (Day et al., 2002). These are susceptible to interference from decomposition of thermally unstable reservoir compounds including methyl peroxy nitrate (CH₃O₂NO₂) and HNO₄ (Browne et al., 2011). Publicly available DC3 and SEAC⁴RS TD-LIF NO₂ are already corrected for this interference. We apply a correction for the other campaigns using the relationship between temperature and percentage interference from Browne et al. (2011). Observed mean ambient air temperature in the UT during INTEX-A was 246 K, corresponding to 20 % interference. That for INTEX-B was 241 K (30 % interference) and 236 K for ARCTAS (38 % interference).

Figure 3 Evaluation of OMI and GEOS-Chem upper troposphere NO₂ with aircraft observations. Individual points are Pearson's correlation coefficients between gridded March–August mean UT NO₂ measured from the aircraft and OMI KNMI in 2006 (blue), OMI NASA in 2005–2007 (red), and GEOS-Chem in 2006 (green) for grid averaging domains of $\mathrm{5}{}^{\circ }\times \mathrm{8}{}^{\circ }$ (latitude × longitude), $\mathrm{10}{}^{\circ }\times \mathrm{16}{}^{\circ }$, $\mathrm{15}{}^{\circ }\times \mathrm{24}{}^{\circ }$, $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$, and $\mathrm{25}{}^{\circ }\times \mathrm{40}{}^{\circ }$. Inset values are the number of points at each resolution. The domain sampled is shown in Fig. 2.

Figure 4 March–August upper troposphere NO₂ over North America. All data are at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$. Background colors in the different panels show concentrations from KNMI, NASA, and GEOS-Chem (GC). Circles show the aircraft observations (same in all panels). Aircraft observations are for 11:00–16:00 LT. The model is sampled in the satellite overpass time window (12:00–15:00 LT). Model and aircraft data are at 280–450 hPa and screened for stratospheric influence using ozone ∕ CO >1.25 mol mol⁻¹. Inset boxes show reduced major axis (RMA) regression statistics and mean NO₂ for coincident grid squares. Grey grid squares indicate no coincident observations.

There are also NO₂ observations from the recent NASA ATom campaign and from the In-service Aircraft for a Global Observing System (IAGOS) commercial aircraft campaign (Berkes et al., 2018). These use chemiluminescence instruments that are also susceptible to interference. Chemiluminescence and TD-LIF NO₂ are consistent during the SEAC⁴RS campaign for the altitude range considered in this work (6–9 km) (Travis et al., 2016), but the interference from chemiluminescence is challenging to quantify due to dependence on the operator and instrument design that varies across campaigns (Reed et al., 2016).

Figure 2 shows the sampling extent of TD-LIF UT NO₂ over North America and Greenland in spring–summer at 450–280 hPa around the satellite overpass (11:00–16:00 LT) for scenes not influenced by the stratosphere (diagnosed with collocated ozone ∕ CO >1.25 mol mol⁻¹; Hudman et al., 2007). Concentrations of UT NO₂ exceed 80 pptv over the eastern US due to lightning NO_x emissions and convective transport of boundary layer pollution, and are <30 pptv over the rest of the domain.

Figure 3 shows the spatial correlation between the March–August mean gridded aircraft data and the OMI UT NO₂ from the KNMI and NASA products as a function of horizontal resolution. There is no significant spatial correlation between the OMI products and aircraft NO₂ at $\mathrm{5}{}^{\circ }\times \mathrm{8}{}^{\circ }$ (R<0.1) and $\mathrm{10}{}^{\circ }\times \mathrm{16}{}^{\circ }$ (R<0.2). The correlation improves with further spatial averaging, peaking at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ (R=0.56 for KNMI, R=0.64 for NASA). The satellite products are also spatially consistent at this resolution (R=0.89), but KNMI is 43 % lower than NASA.

Figure 4 compares the spatial distribution of OMI and aircraft UT NO₂ at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ over North America. Domain mean KNMI UT NO₂ is 38 % lower than the aircraft observations, compared to 2.2 % higher for NASA UT NO₂. Both products exhibit less variability (reduced major axis, RMA, regression slopes <1) and high bias in background NO₂ compared to the aircraft observations (positive RMA intercepts of 5.9±1.4 pptv for KNMI and 9.2±2.7 pptv for NASA). We proceed with the NASA UT NO₂ product at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$, as correlation peaks at this resolution and the NASA product is more consistent with domain mean aircraft UT NO₂ than the KNMI product.

4 Constraints on upper tropospheric NO_x

The NASA product provides near-global coverage of UT NO₂ to assess current understanding of regional UT NO_x sources and dynamics by comparing to UT NO₂ from the GEOS-Chem CTM (version 10-01; http://wiki.seas.harvard.edu/geos-chem/index.php/GEOS-Chem_v10-01, last access: 28 November 2017) driven by NASA Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2) reanalysis meteorology. The model horizontal resolution is $\mathrm{2}{}^{\circ }\times \mathrm{2.5}{}^{\circ }$ and the output is regridded to $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ for comparison with OMI. GEOS-Chem is sampled under all-sky conditions in the satellite overpass window (12:00–15:00 LT). We find that the effect on NO₂ of sampling the model under cloudy conditions is small. Isolating NO₂ under very cloudy conditions using MERRA-2 cloud fractions decreases modeled UT NO₂ by no more than 5 pptv in the tropics/subtropics and less at higher latitudes. We use output from the model for 2006 following a 1-year spin-up for chemical initialization. Interannual variability in UT NO₂, determined as the difference between modeled 2006 and multi-year mean (2005–2007) UT NO₂, is small (<4 pptv) everywhere except central Africa year round (4–12 pptv), the Arctic north of 60^∘ N (up to 25 pptv), and the Middle East in June–August and northern India in March–May (both 10–20 pptv). Recent evaluation of model NO₂ with observed vertical profiles from the SEAC⁴RS aircraft campaign shows no significant bias in the 6–9 km range of the OMI product (Travis et al., 2016; Silvern et al., 2018).

Local GEOS-Chem emissions of NO_x in the UT include aircraft and lightning. Aircraft emissions from the Aviation Emissions Inventory Code (AEIC) inventory (Stettler et al., 2011) total 0.82 Tg N in 2006, which is much less than lightning in the same year (6.5 Tg N). Transport from the stratosphere is simulated using a climatology of NO_y species concentrations from the GMI model above the tropopause (Murray et al., 2012) and is very small (0.4 Tg N a⁻¹ as total NO_y). Lightning in the model is estimated using the parameterization implemented by Murray et al. (2012). This includes an initial estimate of lightning flashes using the Price and Rind (1992, 1993, 1994) relationship between cloud-top height and lightning flashes. These are then scaled to the same annual global flash frequency (46 fl s⁻¹) and regional distribution as the climatology from the combined Lightning Imaging Sensor (LIS) and Optical Transient Detector (OTD) high-resolution monthly climatology (LIS/OTD HRMC) (Cecil et al., 2014). The standard GEOS-Chem model has higher NO_x yields per flash at northern midlatitudes (north of 35^∘ N) than in the tropics (500 mol N fl⁻¹ versus 260 mol N fl⁻¹), but we find that this overestimates observed OMI UT NO₂ by 10–20 pptv (20 %–40 %) at northern midlatitudes in summer when the lightning source is dominant. Here, we address this overestimate by assuming a NO_x yield of 260 mol N fl⁻¹ everywhere. This decreases global lightning NO_x emissions by 15 % from 6.5 to 5.5 Tg N a⁻¹. The lightning parameterization in GEOS-Chem does not distinguish lightning NO_x production from flashes within or between clouds (intra- or inter-cloud) or from the cloud to the Earth's surface (cloud to ground).

Figure 3 shows the spatial correlation between the model and aircraft observations. The model is more consistent with the aircraft observations than OMI at fine spatial resolution. Like OMI, GEOS-Chem correlation with the aircraft observations improves with spatial averaging, peaking at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ (R=0.75). Figure 4 also shows comparison of March–August GEOS-Chem UT NO₂ with the aircraft observations at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$. The model is sampled over the same pressure range as NASA (280–450 hPa) around the OMI overpass (12:00–15:00 LT) and is filtered for stratospheric influence using model ozone ∕ CO >1.25 mol mol⁻¹. Domain average UT NO₂ from the model is 19% lower than the aircraft measurements, and the model also overestimates background UT NO₂ (intercept of 7.5±1.0 pptv) and underestimates the variability (slope of 0.45±0.09).

Figure 5 compares seasonal mean OMI and GEOS-Chem UT NO₂ in December–February and June–August. The other seasons are shown in the Supplement (Fig. S2). Formation of PAN, HNO₄, and CH₃O₂NO₂ accounts for over 75 % of NO_x loss in the model in all seasons. Lower concentrations of UT NO₂ in the Northern Hemisphere winter compared to summer in the model is mostly because lightning activity is at a minimum. The model underestimates UT NO₂ in the northern midlatitudes in winter by 20–40 pptv, suggesting misrepresentation of another process in the model, such as excessive NO_x loss by N₂O₅ hydrolysis in aerosols (Kenagy et al., 2018). The particularly large bias over polluted regions in winter could also be due to contamination of the UT NO₂ retrievals by enhanced boundary layer NO₂.

Figure 5 Observed and modeled upper troposphere NO₂. The figure shows NASA OMI seasonal mean UT NO₂ for 2005–2007 (a, b) and corresponding GEOS-Chem model values (c, d). The model is sampled at 280–450 hPa during the satellite overpass (12:00–15:00 LT) and filtered for stratospheric influence. Data are at $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ horizontal resolution for December–February (a, c) and June–August (b, d). Grey grid squares in the top panels indicate no OMI data.

Figure 6 Log–log relationship between upper troposphere NO₂ and lightning flash frequencies, and linear regression fits between the two. Individual points are coincident seasonal mean UT NO₂ from OMI (a) and GEOS-Chem (b) versus seasonal mean LIS/OTD lightning flash climatologies for coincident $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ grid squares in the northern midlatitudes (>30^∘ N; blue) and tropics (<30^∘ N; orange). Northern midlatitude points exclude December–February, which show poor correlation with lightning flashes (see text for details). Lines and legends show reduced major axis linear regression fits to the data with corresponding Pearson's correlation coefficients. The regression lines plot as curves on the log–log scale, highlighting the NO₂ background at low lightning flash rates and the correlation of NO₂ with lightning at high flash rates.

Figure 7 Relationship between OMI and GEOS-Chem derived lightning NO_x production rates and LIS lightning properties: energy (radiance), duration, and footprint area. Individual points are seasonal mean $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ grid squares at 40^∘ N–40^∘ S.

Figure 6 shows the log–log relationship between seasonal mean LIS/OTD lightning flash climatology and seasonal mean UT NO₂ from OMI and GEOS-Chem, and the corresponding reduced major axis linear regression fits. Data are divided into northern midlatitudes and tropics. We exclude the contaminated observations over northeast China and the wintertime northern midlatitude grid squares that show no correlation with lightning flashes (R<0.1). Results from multi-model sensitivity studies indicate that UT NO_x in winter is predominantly from surface sources, with a smaller contribution from extratropical lightning (Grewe et al., 2001). Background concentration of UT NO₂ (intercepts in Fig. 6) from non-lightning sources is 10–20 pptv and is 3–5 pptv higher in the northern midlatitudes than in the tropics. The slopes for the linear fits to lightning flash frequency are consistent between the OMI observations and GEOS-Chem, and show similar slopes for northern midlatitudes and the tropics. Fitting the ratio between OMI observations and GEOS-Chem on the $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ grid implies a NO_x yield per flash of 280±80 mol N fl⁻¹ with no significant difference between midlatitudes and the tropics, and no significant difference with the GEOS-Chem prior estimate of 260 mol N fl⁻¹. Our prior estimate of global lightning source was 5.5 Tg N a⁻¹, and the improved estimate is 5.9±1.7 Tg N a⁻¹.

Properties of lightning flashes, including energy, duration, and footprint area, have been retrieved from the OTD and LIS sensors (Beirle et al., 2014). The flash footprint area is the spatial extent of lightning detection events contributing to the flash (collection of local events) diagnosed by the satellite data. Figure 7 shows the relationship between OMI and GEOS-Chem derived lightning NO_x production rates and LIS lightning properties from Beirle et al. (2014). The strongest correlation is with lightning extent (R=0.50), followed by energy (R=0.40). The correlation with flash duration is weak (R=0.25). The relationships in Fig. 7 suggest a dependence of lightning NO_x production rates on lightning flash energy of 510±80 mol N (J m⁻² sr⁻¹ µm${}^{-\mathrm{1}}{)}^{-\mathrm{1}}$ and on flash footprint area of 2.0±0.3 mol N  km⁻², possibly offering guidance for relating NO_x yields to physical properties in global models rather than the current approach of assigning static values.

5 Conclusions

Measurements of NO_x in the UT have mainly been from aircraft campaigns that are limited in space and time. Two new cloud-slicing UT NO₂ products from OMI produced by KNMI and NASA offer the potential to address uncertainties in our understanding of UT NO_x sources. Here, we intercompared these products, evaluated them with aircraft observations, and used them together with GEOS-Chem model simulations to demonstrate a dominance of lightning as a source of UT NO_x.

The KNMI and NASA UT NO₂ products use very different retrieval methods. Seasonal mean concentrations from the two products show weak global correlation at the $\mathrm{5}{}^{\circ }\times \mathrm{8}{}^{\circ }$ (latitude × longitude) resolution of the NASA retrieval, with some improvement when the data are further averaged to $\mathrm{20}{}^{\circ }\times \mathrm{32}{}^{\circ }$ (R=0.5–0.7). At that resolution, they show correlation with in situ aircraft observations of UT NO₂ over North America for different years (R=0.56–0.64). The KNMI product is biased low by 38 % relative to the aircraft observations, while the NASA product has no significant bias.

We find from the relationship of OMI UT NO₂ with LIS/OTD flash rates that most NO_x in the upper troposphere is from lightning, except in the midlatitudes in winter. The background NO₂ concentration in the absence of lightning is 10–20 pptv. The relationship suggests no difference in NO_x yields per flash between the midlatitudes and the tropics, in contrast to the higher yields at midlatitudes often assumed in models. We derive a global mean lightning NO_x production rate per flash of 280±80 mol N fl⁻¹, from which we infer a best estimate for the global lightning NO_x emission of 5.9±1.7 Tg N a⁻¹.