This study explores the upper tropospheric anticyclone above eastern North America and its influence on the summertime buildup and decay of lightning NOx (LNOx) and thunderstorm outflow. LNOx transport is simulated with a particle dispersion model that releases a LNOx tracer from the locations of millions of cloud-to-ground lightning flashes during May–September 2004 and 2006. On average, upper tropospheric zonal flow in May transitions to a closed anticyclone above northern Mexico and the southern United States in July that strengthens in August and rapidly decays in September. Concentrations of the LNOx tracer reach a maximum above the southern United States and Gulf of Mexico in July and August. Fourteen study sites across North America exhibit high day-to-day variability of the LNOx tracer in the upper troposphere during summer, with the sites most heavily influenced by the North American summer monsoon having the greatest background concentrations. During late spring and September the western sites have low concentrations with little variability. In general, the west coast sites plus Barbados have the most aged thunderstorm outflow, while the east coast sites have the least aged outflow. More than 80% of summertime upper tropospheric NOx above the eastern United States is produced by lightning. To produce the best available observation-based view of upper troposphere NOx above North America, measurements from six aircraft campaigns are combined in a single composite plot. The modeled upper tropospheric NOx matches the general continental-scale distribution of NOx in the composite plot, supporting the dominant role of LNOx in the simulations.
 Tropospheric ozone is a central trace gas that influences the oxidizing capacity of the troposphere, and is also an important greenhouse gas, especially in the upper troposphere where ozone changes have a greater impact on surface temperatures [Forster and Shine, 1997; Intergovernmental Panel on Climate Change, 2007]. A recent study by Stevenson et al.  examined the tropospheric ozone budget as calculated by an ensemble of 26 chemical transport model simulations. Approximately one third of the tropospheric ozone burden is due to anthropogenic activities, resulting in widespread ozone increases since the late 1800s [Marenco et al., 1994; Staehelin et al., 1994; Lamarque et al., 2005; Oltmans et al., 2006]. Lightning NOx (LNOx) emissions are believed to be a major natural influence on upper tropospheric ozone production, especially over the tropics and in the midlatitudes during summer, however large uncertainties still remain regarding the annual LNOx budget, with current estimates in the range 5 ± 3 Tg a−1 (see the recent literature review by Schumann and Huntrieser ). While satellites have provided reliable climatological estimates of the total number of lightning flashes across the globe annually [Christian et al., 2003], uncertainty arises from the quantity of NOx produced per flash owing to high variability in the intensity and spatial extent of each flash [Schumann and Huntrieser, 2007].
 A major LNOx producing region of the Earth that is receiving particular attention is North America due to its high LNOx emissions during summer [Biazar and McNider, 1995; Flatøy and Hov, 1997] and its unique transport patterns that allow LNOx to accumulate in the upper troposphere. Aircraft-based LNOx observations began above North America with the PRE-STORM experiment in June, 1985 over Oklahoma and Kansas [Dickerson et al., 1987]. During July and August, 1989, Ridley et al.  measured NOx above New Mexico and found large enhancements in the upper troposphere due to lightning emissions within thunderstorms produced by the North American summer monsoon. Extensive sampling of NOx between eastern North America and Europe occurred during the NOXAR and POLINAT 2 experiments in the summers of 1995 and 1997, by an instrumented SwissAir commercial airliner. Analyses of these data sets [Brunner et al., 1998, 2001; Jeker et al., 2000], along with NOx measurements from the 1998 STREAM experiment above southeastern Canada [Lange et al., 2001] all concluded that lightning associated with summertime convection caused NOx enhancements in air masses above eastern North America as well as in similar air masses advected into the western North Atlantic Ocean. The STERAO experiment above Colorado in summer 1996 led to new insights into thunderstorm structure and LNOx production [Dye et al., 2000; DeCaria et al., 2005]. Strong lightning NO enhancements have also been observed above Florida by aircraft [Ridley et al., 1994], and lightning NO2 enhancements have been observed above the Gulf of Mexico by the polar orbiting GOME instrument [Beirle et al., 2006]. Most recently, measurements by the NASA DC-8 during the summertime 2004 ICARTT experiment expanded our knowledge of the NOx distribution across eastern North America, with lightning once again shown to play a dominant role in the upper troposphere [Cooper et al., 2006; Bertram et al., 2007; Hudman et al., 2007].
 Current research is also focusing on the photochemical production of ozone in the free troposphere above North America due to LNOx emissions, ranging from the scale of individual storms [Decaria et al., 2005] to the regional and continental scale. A modeling study by Zhang et al.  indicated that even though North American anthropogenic NOx emissions are far greater than North American LNOx emissions, LNOx should dominate ozone production in the upper troposphere above North America during summer owing to its direct injection into the upper troposphere. Li et al.  used a chemical transport model to suggest that a widespread upper tropospheric ozone enhancement should exist above the southern United States during summer, the result of convective lofting of surface ozone, and in situ production by anthropogenic NOx, LNOx, and in part by HOx radicals produced from convectively lifted formaldehyde that originates from biogenic isoprene. The ozone could be produced over many days because of the trapping of the ozone precursors in the semipermanent upper tropospheric anticyclone located above eastern North America during summer. While only five ozonesondes from Huntsville, Alabama were available to verify the simulation they did show enhanced ozone in the upper troposphere and the authors found that in order to make their model match the observations they had to increase the model LNOx emissions by a factor of 4.
 The scarcity of free tropospheric ozone measurements above summertime North America was rectified by the 1 July to 15 August 2004 INTEX Ozonesonde Network Study (IONS) experiment that provided ozonesondes from multiple sites across eastern North America plus one upwind site at Trinidad Head, California [Thompson et al., 2007]. Cooper et al.  combined the IONS ozonesondes with ozone profiles from MOZAIC commercial aircraft at several U.S. and Canadian airports, and ozone profiles from a lidar in southern California, to produce the largest set of free-tropospheric ozone profiles ever gathered across North America in a single season. Once the model-calculated stratospheric ozone contribution had been removed from all tropospheric profiles the data did reveal an upper tropospheric ozone enhancement above the southern USA, in agreement with the previous modeling studies [Zhang et al., 2003; Li et al., 2005]. On average the upper troposphere of eastern North America contained an additional 16 ppbv of ozone compared to the west coast, with a maximum enhancement of 24 ppbv above Houston, Texas. A detailed study of NOx sources indicated that lightning produced 78–95% of the NOx in the upper troposphere above eastern North America, and that 69–84% of the ozone enhancement was due to in situ ozone production from LNOx with the remainder due to transport of ozone from the surface or in situ ozone production from other sources of NOx. A follow-up IONs experiment was conducted in August 2006 with improved sampling across western North America [Cooper et al., 2007]. The data showed that the upper tropospheric ozone maximum recurs each year, its location and strength influenced by the strength and location of the summertime upper tropospheric anticyclone that traps convectively lofted ozone, ozone precursors and LNOx above the southeastern United States. The North American summer monsoon that flows northward along the Rocky Mountains is embedded within the western side of the anticyclone and also marks the westernmost extent of the ozone maximum. Simulations by the ECHAM5/MESSy1 atmospheric chemistry general circulation model indicate LNOx emissions led to the production of 25–30 ppbv of ozone at 250 hPa above the southern United States during the study period. A recent analysis of ozone measured by four commercial and one research aircraft throughout the upper troposphere of the United States during the late 1970s provides further evidence that upper tropospheric ozone above the western United States is less than ozone above the northeastern United States and Mexico [Schnadt Poberaj et al., 2007].
 Previous work on summertime LNOx emissions has focused on just 1- or 2-month periods for a particular year, coincident with aircraft measurement campaigns, without providing a broader context for the observations. The purpose of this study is to describe and quantify the buildup and decay of LNOx and aged thunderstorm outflow above North America over the spring and summer months in 2004 and 2006. These results show when and where deep convection has the greatest potential for influencing the chemistry of the upper troposphere above North America and the North Atlantic Ocean. The findings will also serve as a guide for the planning of future aircraft missions aimed at quantifying the impact of deep convection on the NOx, HOx and ozone budgets in the same regions. The selected May–September time period is significant because it covers the transition from late spring weather conditions, that are dominated by midlatitude cyclones and frontal-driven convection, to late summer conditions that are dominated by the North American summer monsoon, the upper tropospheric anticyclone and air mass thunderstorms. Furthermore, the influence of the collapse of the summer monsoon flow on LNOx emissions is explored. We contrast this process between the years 2004 and 2006 with weaker and stronger than average upper tropospheric anticyclones, respectively. To simulate the transport as accurately as possible we employ a Lagrangian particle dispersion model with high-resolution global-scale winds and LNOx tracers emitted from the exact times and locations of all cloud-to-ground lightning flashes detected above the United States, northern Mexico and the surrounding waters. Finally, we compare the modeled LNOx in the upper troposphere to anthropogenic sources of NOx and to a “climatology” of measured NOx mixing ratios above summertime North America.
2.1. FLEXPART LNOx Simulation
 The FLEXPART Lagrangian particle dispersion model [Stohl et al., 1998, 2005] was used to simulate the transport of LNOx emitted from thunderstorms. The model calculates the trajectories of a multitude of particles and was driven by global ECMWF wind fields with a temporal resolution of 3 h (analyses at 0000, 0600, 1200, and 1800 UTC; 3-h forecasts at 0300, 0900, 1500, and 2100 UTC), horizontal resolution of 1° × 1°, and 60 (90) vertical levels in 2004 (2006). Particles are transported both by the resolved winds and parameterized subgrid motions. To account for convection, FLEXPART uses the parameterization scheme of Emanuel and Živković-Rothman , which is implemented at each 15-min model time step, and is intended to describe all types of convection. Verified by Forster et al. , the scheme produces convective precipitation amounts that are in relatively good agreement with observations both in the tropics and extratropics. It also improves the agreement of FLEXPART simulations with airborne tracer measurements compared to simulations without the convection parameterization.
 The LNOx emissions in this study are not produced by a grid-scale model parameterization scheme. Instead, LNOx production was simulated with FLEXPART above North America and the surrounding waters according to the exact times and locations of each cloud-to-ground (CG) lightning flash observed by lightning detection networks. CG flashes were detected over the continental USA by the National Lightning Detection Network (NLDN), with a detection efficiency (DE) better than 90% (M. J. Grogan, Report on the 2002–2003 US NLDN system-wide upgrade, 2004, http://www.vaisala.com/businessareas/measurementsystems/thunderstorm/knowledgecenter/aboutnldn). For regions between 25°N and 50°N, but outside of the continental United States, CG flashes were detected with the experimental long-range lightning detection network (LRLDN). DE is 60–80% at 60°N and 40–60% at 21°N and increases with proximity to the USA (K. Cummins, personal communication, Vaisala-Thunderstorm, 2006). For regions between the equator and 25°N, CG and IC flashes were quantified according to a seasonally and diurnally varying climatology based on 5 years of data from the polar orbiting Lightning Imaging Sensor (LIS) and Optical Transient Detector (OTD) instruments [Christian et al., 2003].
 To account for the dropoff in detection efficiency with distance from shore all NOx emissions from LRLDN flashes were scaled by the inverse of the DE. Furthermore we scaled the NOx emissions per CG flash to account for the intracloud (IC) flashes not detected by the NLDN and LRLDN. Monthly gridded IC:CG ratios calculated for the continental United States [Boccippio et al., 2001] typically range between 1 and 10 for individual 0.5° × 0.5° grid cells and were applied to the NLDN data. Little is known about the IC:CG ratio above the waters surrounding North America, so the annual average ratio of 2.9 above the United States (as determined by Boccippio et al. ) was applied to the LRLDN data.
 We selected a LNOx emission rate of 3.5 kg N flash−1 based on the most recent review of all available LNOx emission estimates [Schumann and Huntrieser, 2007]. Following the recommendations of Ridley et al.  we treat the NOx production per IC flash as equal to the production from a CG flash. For each day during May–September 2004 and 2006 (plus a 20-day spin-up time in April), a single FLEXPART trajectory particle was released at the exact time and location of the millions of CG lightning flashes detected by the NLDN and LRLDN. Additional trajectory particles were released from the region of the lightning climatology. Representing a passive LNOx tracer the particles were allowed to advect for a transport time of 20 days, after which they were removed from the simulation. Computer memory limitations only permitted the release of one FLEXPART particle per flash. In total the simulation represented the transport of LNOx emissions from 320 million IC and CG flashes in 2004 (using 144 million trajectory particles) and from 279 million IC and CG flashes in 2006 (using 134 million trajectory particles).
 To account for the fact that LNOx emissions occur mainly in the upper troposphere but over a wide range of altitudes [Pickering et al., 1998; DeCaria et al., 2005], the particles were released at random between the tropopause and 6 km above sea level according to a normal distribution which was allowed to slide up and down with the height of the tropopause. No particles were released below 6 km and the mode of the release height was between 10 and 11 km. Upon release the particles were permitted to move vertically according to the model vertical winds and the convection scheme. This vertical particle release distribution is based upon the altitudes at which NOx is measured in and around North American thunderstorms. Ridley et al.  summarize vertical NO profiles from three different aircraft studies of North American thunderstorms, with maximum NO values occurring between 10 and 12 km, 8–13 km and 12–14 km above New Mexico, Colorado and Florida, respectively. During ICARTT the DC-8 tropospheric survey flights show maximum NOx values in the 10- to 12-km range. In contrast the North American NO or NOx profiles reviewed by Ridley et al.  and Hudman et al.  all show an average NOx minimum in the 4–6 km range, with no evidence to support a strong influence from LNOx on the midtropospheric NOx distribution. As discussed below the FLEXPART LNOx tracer reproduces the general shape of the DC-8 NOx profiles, indicating that LNOx emissions below 6 km do not make a major contribution to the overall tropospheric NOx distribution of the midtroposphere.
2.2. FLEXPART Simulation of Other NOx Tracers
 FLEXPART was also used to simulate the transport of aircraft NOx emissions above North America using a global inventory of aircraft emissions created by combining the civil traffic inventory for 1999 of Sutkus et al.  with an estimate for 1999 military, charter, and general aviation obtained by extrapolating the earlier inventory work of Mortlock and Van Alstyne . Emissions in these inventories were calculated using great circle routes and provided on a 1° × 1° by 1 km altitude grid for each month of 1999. Actual civil aircraft routes vary from day to day because of air traffic congestion, prevailing winds or to avoid turbulence. The use of idealized great circle routes as opposed to actual flight routes can result in some errors but this has been found to be relatively small (∼5%) [Forster et al., 2003]. While aircraft NOx emissions are small compared to surface anthropogenic NOx emissions their direct injection into the free troposphere makes them an important source of NOx in the upper troposphere, totaling about 15% of global LNOx emissions on an annual basis [Schumann and Huntrieser, 2007].
 The transport of North American anthropogenic NOx emissions was simulated with FLEXPART, on the basis of the point, on-road, nonroad and area sources from the U.S. EPA National Emissions Inventory, base year 1999, with spatial partitioning of area type sources at 4 km resolution. This database covers all sources in the United States, sources in Canada south of 52°N, and sources in Mexico north of 24°N [Frost et al., 2006]. Point sources representing eastern USA power plants were updated to account for the 50% reduction in power plant emissions that occurred between 1999 and 2003 [Frost et al., 2006]. Emissions for all other regions of North America were taken from the EDGAR 3.2 Fast Track 2000 data set, which estimates year 2000 emissions using the EDGAR 3.2 estimates for 1995 and trend analyses for the individual countries. EDGAR uncertainty estimates are roughly 50% or greater [Olivier and Berdowski, 2001].
 Finally, the transport of NOx from the lower stratosphere to the mid and upper troposphere via stratospheric intrusions was also calculated with FLEXPART, which simulates these transport conditions very well [James et al., 2003; Cooper et al., 2005]. Stratospheric NOx transport was simulated above North America (−140°W to −40°W) by filling the lower stratosphere with a NOx tracer (the average NOx mixing ratio was set to 220 ppbv, which is typical for summertime [Hegglin et al., 2006]), and then allowing the model winds to advect the tracer into the troposphere as dictated by stratosphere-troposphere exchange processes.
 To simulate the photochemical decay of the lightning and aircraft/anthropogenic NOx emissions an e-folding lifetime of 4 days was imposed on the tracers for some aspects of the analysis. The 4-day lifetime was taken from the analysis by Bertram et al.  of NOx measurements in the upper troposphere above North America during summer 2004. This lifetime is based on an initial LNOx mixing ratio of 800 pptv, and was derived for 10 km, the altitude with the greatest fraction of convectively influenced air masses. To account for the rapid oxidation of NOx in the atmospheric boundary layer and the fact that only a small percentage of surface NOx emissions ever reach the free troposphere, any FLEXPART particle that was emitted from the surface and subsequently reached an altitude greater than 2 km was tagged. These tagged particles were considered to have escaped the atmospheric boundary layer and their NOx values were reduced by 86% as determined by Hudman et al.  for summertime conditions above North America.
2.3. In Situ NOx Measurements
 The sum of the FLEXPART lightning, anthropogenic surface, aircraft and stratospheric NOx tracers is compared to in situ NOx measurements from six separate field experiments. These experiments are briefly summarized below, and information on the measurement techniques can be found in the accompanying references.
 The ELCHEM experiment occurred during July and August 1989 and involved 12 flights of the NCAR Sabreliner jet aircraft above New Mexico when the region was dominated by either synoptic high pressure or moist southerly flow associated with the North American Monsoon [Ridley et al., 1994]. The data set includes in situ NO and NO2 measurements.
 The second and third experiments utilized an instrumented commercial Swissair B-747 that measured NOx on hundreds of flights between Europe and the United States/Asia during the NOXAR experiment (May 1995 to May 1996) and the POLINAT 2 experiment (August–September 1997) [Brunner et al., 1998, 2001]. The present study focuses on the daytime portions of the 29 flights between Europe and the eastern United States during June–August 1995 and the 12 flights between Europe and the eastern United States during August 1997.
 The fourth data set comes from the Cirrus Regional Study of Tropical Anvils and Cirrus Layers–Florida Area Cirrus Experiment (CRYSTAL-FACE) above Florida, the Gulf of Mexico and the Caribbean Sea during June–July 2002 [Ridley et al., 2004]. NO values are from in situ measurements from the NASA WB-57 aircraft. NO2 was calculated from in situ O3 and NO measurements assuming photostationary equilibrium under clear sky conditions using methodology similar to that used by Brunner et al. . NO2 was about 11% of NOx in the upper troposphere. When the WB-57 sampled thunderstorm outflow within cirrus clouds solar radiation could be reduced by as much as 50%, violating the clear-sky assumption. Conversely when sampling air above cirrus cloud, solar radiation could be enhanced owing to cloud albedo. These uncertainties affect the jNO2 values used in the calculation of NO2, leading to a NOx uncertainty of ±12%.
 The fifth data set comes from the 1996 Stratosphere-Troposphere Experiments: Radiation, Aerosols and Ozone (STERAO) above the vicinity of eastern Colorado. In situ NO measurements were made from the University of North Dakota Citation aircraft [Stith et al., 1999]. NO2 was calculated using the same methodology as for the CRYSTAL-FACE data.
 The sixth and final data set comes from the 18 flights of the NASA DC-8 jet aircraft during the 1 July to 15 August 2004 ICARTT (International Consortium for Atmospheric Research on Transport and Transformation) study and includes in situ NO and NO2 measurements [Bertram et al., 2007]. These flights covered much of the eastern USA, southeastern Canada, the adjacent regions of the North Atlantic Ocean, and also included one survey flight of the western United States.
 LNOx has a strong impact on ozone in the upper troposphere and many of the results from this study will be discussed in relation to the IONS and MOZAIC ozone profiles collected above North America during the 2004 and 2006 IONS summertime experiments [Cooper et al., 2006, 2007], as summarized in section 1. In the present study we focus particular attention on the behavior of LNOx emissions and transport above the 14 sites that launched daily ozonesondes across North America during August 2006. There are three reasons for this decision: (1) these 14 sites form the basis of our understanding of upper tropospheric ozone mixing ratios above North America, but up until now there has been little understanding of the seasonal and year-to-year variability of LNOx above these sites; (2) these sites were carefully chosen to cover the major inflow and outflow regions of the North American atmosphere and are therefore representative of North America's major transport pathways; and (3) now that infrastructure is in place these sites are likely to be utilized during future ozonesonde intensives and knowledge of LNOx in these regions is important for the planning of future studies.
 The 14 site locations are shown in Figure 1, with the names and abbreviations of the sites listed in Table 1. Figure 1 also shows the median August 2006 filtered tropospheric ozone (FTO3) mixing ratios at 10–11 km as measured by daily ozonesondes at the 14 study sites (as reported by Cooper et al. ). FTO3 is the quantity of ozone present in the troposphere after the model calculated contribution from the stratosphere is removed. The result is the amount of ozone that is produced within the troposphere, without confounding influence from recent stratospheric ozone intrusions (the methodology is described by Cooper et al. ).
Table 1. Names and Abbreviation of the 14 Ozonesonde Sites Shown in Figure 1
Barbados, Caribbean Sea
Boulder, Colorado, USA
Bratt's Lake, Saskatchewan, Canada
Huntsville, Alabama, USA
Kelowna, British Columbia, Canada
Mid-Atlantic, USA (includes sondes from Beltsville, Maryland and Wallops Island, Virginia (170 km apart)
Mexico City, Mexico
Narragansett, Rhode Island, USA
Ontario, Canada (includes sondes from Egbert and Walsingham, 190 km apart)
Sable Island, Canada, North Atlantic Ocean
Socorro, New Mexico, USA
Table Mountain, California, USA
Texas Coast, Houston and nearby waters, USA
Trinidad Head, California, USA
Figure 1 shows a clear upper tropospheric ozone maximum above the southeastern United States, its location and strength closely related to the location and strength of the upper tropospheric anticyclone [Cooper et al., 2007]. The anticyclone establishes itself above the southern United States during summer in conjunction with the southerly flow of the North American summer monsoon that flows into the western side of the anticyclone. The results presented below show that the seasonal buildup and decay of this anticyclone governs the summertime buildup and decay of LNOx and thunderstorm outflow above the southern United States.
Figure 2a shows a 20-year climatology (1987–2006) of the 250-hPa geopotential height and wind barbs for May through September (from the NCEP/NCAR Reanalysis data set [Kalnay et al., 1996]). On average, the upper troposphere exhibits zonal flow in May, with a ridge building above southern Mexico in June. By July the ridge has formed a closed circulation anticyclone above northern Mexico and the southern USA that strengthens in August, followed by a rapid decrease of the height, horizontal extent and latitude of the anticyclone in September. This pattern occurs in both 2004 (Figure 2b) and 2006 (Figure 2c). However, the geopotential height of the center of the anticyclone was lower than average in July 2004 and shifted farther south than average in both July and August 2004. In contrast the anticyclone was stronger than average and positioned farther north in both July and August 2006, which led to long residence times in the upper troposphere and very high ozone mixing ratios above Huntsville (HUN), Alabama (Figure 1), as described by Cooper et al. .
Figures 3 and 4 show the quantity of FLEXPART LNOx tracers above North America during May–September of 2004 and 2006, respectively. Figures 3a and 4a show the average monthly column values of the 1-h LNOx tracer indicating the LNOx emission regions. LNOx emission regions were generally similar in both years with the Sierra Madre Mountains of northwestern Mexico becoming prominent in July and August, the region with the greatest summertime lightning frequency in North America [Murphy and Holle, 2005]. It might be expected that because 2006 had a stronger anticyclone it would have conditions that would produce more lightning than 2004. However, LNOx emissions above North America were greater in 2004 than 2006 from May through August, with seasonal peak emissions of 0.30 Tg N in July 2004 and 0.24 Tg N in August 2006. This is largely due to LNOx emissions being much stronger above the waters adjacent to the coast of the southeastern United States during summer 2004, especially the northern coast of the Gulf of Mexico. The greater lightning frequency above these regions was caused by frequent cold front passages in 2004 that stalled above the warm coastal waters.
Figures 3b and 4b show the average monthly column of the 20-day LNOx tracer with a 4-day e-folding lifetime imposed. This particular tracer indicates the regions where LNOx would most likely be found in the upper troposphere above North America. In both years the stronger zonal flow in May and June quickly advects the LNOx to the east coast of North America and out into the western North Atlantic Ocean. In July and August, LNOx emissions increase and the monsoon flow and the anticyclone trap the LNOx above southern North America, but both LNOx emissions and the anticyclone weaken in September resulting in much lower column LNOx values. The June LNOx values are much greater above the southeastern United States in 2004 than 2006 owing to 40% greater LNOx emissions and a stronger anticyclonic circulation in the upper troposphere above this region. The July anticyclone was weaker in 2004 than 2006 but the 30% greater LNOx emissions in 2004 led to greater column LNOx values above the southeastern United States and Gulf of Mexico. LNOx emissions were similar in August 2004 and 2006 but the stronger and more northerly anticyclone in 2006 led to greater LNOx values above the southern United States.
Figures 3c and 4c depict the location of a 20-day LNOx tracer. Given the approximate 4-day NOx lifetime in the upper troposphere very little LNOx would remain after 20 days. However, this tracer can be used to indicate the average location of aged thunderstorm outflow in terms of both the mass of air detrained from thunderstorm anvils, and in terms of photochemically produced compounds like ozone and HNO3. The distribution of the 20-day tracer is similar to that of the 4-day e-folding lifetime tracer but with greater column values downwind of North America. These plots indicate that in summer a plume of aged thunderstorm outflow extends, on average, from North America to the west coasts of northern Africa and southern Europe.
 The day-to-day variability of the 20-day LNOx tracer is investigated for each of the 14 IONS-06 ozonesonde sites in Figures 5 and 6 for 2004 and 2006, respectively. Figures 5 and 6 (top) give the daily average column values of the tracer above the 5 sites that are predominantly upwind of the major North American lightning emission regions (see Figure 1). These sites show very little variation from May though September with the exception of Table Mountain in southwestern California. During July and August the LNOx tracer that accumulates in the anticyclone occasionally escapes to the tropical and subtropical eastern Pacific Ocean where it can then be transported back to North America at subtropical latitudes by westerly winds advecting across southern California.
Figures 5 and 6 (middle) show that the 20-day LNOx tracer above the 5 sites in central North America have a much greater influence from thunderstorm outflow especially in June, July and August, as expected given their proximity to the major LNOx emission regions. During 2004 the sites experienced greater highs and lows of the 20-day tracer due to a combination of stronger LNOx emissions and more variable transport due to the frequent frontal passages. In contrast, July 2006 had much lower LNOx emissions while August had a more persistent upper level anticyclone that maintained greater minimum values of the LNOx tracer. This transport pattern kept air masses with aged thunderstorm outflow circulating above the southern United States and led to greater ozone mixing ratios in the upper troposphere above the southeastern United States in August 2006 than in 2004 [Cooper et al., 2007]. In both years the column values of the 20-day LNOx tracer decreased abruptly in September owing to a decrease in LNOx emissions and the weakening and southward migration of the anticyclone.
 Finally, Figures 5 and 6 (bottom) show the 20-day LNOx tracer above the four sites in eastern Canada and the eastern United States, downwind of the anticyclone. The column LNOx tracer values follow the same general pattern as the central North America sites, but with reduced values and greater variability. The variability is due to both localized LNOx emissions associated with frontal passages and with episodic export of aged thunderstorm outflow from the anticyclone.
Figure 7 combines the results from Figures 3, 4, 5 and 6 with additional information on the age of the 20-day LNOx tracer above each site. Overall the upwind sites have the most aged tracer. Typically much more than 50% of the tracer is older than 4 days, reflecting the small local LNOx emissions and the episodic impact of air masses with aged thunderstorm outflow. The relative age of the 20-day tracer is less above the central North America sites which have a combination of high local NOx emissions as well as aged thunderstorm outflow. The tracer age is even less above the eastern North America sites owing to a decreased influence from air masses with aged thunderstorm outflow.
4.1. Challenges of Model Verification
 The results presented so far describe the quantity of LNOx above North America and downwind regions in terms of column values of either a passive 20-day LNOx tracer, or a LNOx tracer with a 4-day e-folding lifetime. These types of tracers are well suited for describing the relative distribution of LNOx or aged thunderstorm outflow in the atmosphere as well as the daily, monthly or interannual variability. The fact that the transport of the tracers was simulated using high-resolution global winds with LNOx emissions from the exact times and locations of detected CG flashes builds confidence in the tracers. A more critical test of the accuracy of the tracers is to compare them to actual NOx measurements in the upper troposphere. We conduct such a comparison below, but before proceeding further we first review the challenges of modeling NOx in the upper troposphere.
 The first challenge is that at the present time there exists no data set of in situ measurements that fully characterizes NOx in the upper troposphere above all of North America in a single summertime period. The two best data sets in terms of spatial coverage are (1) the NOXAR/POLINAT2 experiments that provided 41 summertime flights along the east coast of North America and midlatitude to high-latitude regions of the North Atlantic Ocean during 1995 and 1997, and (2) the 18 flights of the NASA DC-8 jet aircraft during the July–August 2004 ICARTT experiment [Singh et al., 2006] with good coverage above the eastern United States, southeast Canada and the western North Atlantic. However, these data sets provide little or no data above western North America, Mexico or the Gulf of Mexico.
 The second challenge is the correct quantification of LNOx emissions. Current estimates of the mass of NOx emitted per flash vary from 0.5 to 9.3 kg N flash−1 (see the most current review by Schumann and Huntrieser ) with the value used in this study close to the middle of the range. The IC:CG ratio is also highly variable and largely unknown above the oceans. Finally, the injection height of the LNOx emissions is also not well characterized with current North America estimates based on the in situ sampling and cloud-resolved modeling of just a few thunderstorms [Pickering et al., 1998; DeCaria et al., 2005; Ridley et al., 2004].
 Other sources of upper tropospheric NOx must also be quantified. These include: surface emissions from soils, biomass burning and anthropogenic sources, as well as free tropospheric emissions from aircraft and injection of NOx from the stratosphere. Surface NOx emissions require accurate simulation of their transport from the boundary layer to the upper troposphere. To verify the deep convective transport of surface emissions to the upper troposphere, we produced animations (not shown) of the FLEXPART anthropogenic surface NOx tracer in the upper troposphere overlaid onto GOES water vapor imagery during August 2006. The NOx tracer only appeared in the upper troposphere in conjunction with deep convective cells or warm conveyor belts, providing qualitative verification that the model transports the tracer to the upper troposphere under the correct conditions.
 Finally, our modeled LNOx mixing ratios are largely determined by the imposed NOx lifetime in the upper troposphere which varies with altitude, season and location, as reviewed by Schumann and Huntrieser . The 4-day e-folding lifetime for NOx in the upper troposphere is based on the best available data and modeling from the ICARTT experiment during summer 2004 [Bertram et al., 2007; Ren et al., 2008], but it is an average value applied in the present study to all altitudes of the upper troposphere and to all photochemical situations.
4.2. Comparison Between Modeled and Observed NOx During ICARTT
 We begin by first comparing modeled NOx values to the NASA DC-8 NOx measurements above North America during July–August 2004. The comparison is limited to times when the aircraft was in the troposphere. The modeled NOx includes emissions from lightning, the stratosphere and anthropogenic emissions from the surface and aircraft. Emissions from soils and biomass burning are much less than anthropogenic surface emissions, with soils accounting for about 15% of summertime surface NOx emissions from the United States [Jaeglé et al., 2005]. As less then 15% of soil and biomass burning NOx emissions are exported to the free troposphere [Hudman et al., 2007], their mixing ratios are very small in the upper troposphere compared to summertime LNOx mixing ratios [Cooper et al., 2006]. Therefore, soil and biomass burning NOx were not included in the comparison. Figure 8 shows that the modeled NOx vertical profile has a similar shape to the measured profile. Both have a relative maximum in the boundary layer where mixing ratios are enhanced by anthropogenic emissions, low values in the midtroposphere and the greatest overall values in the upper troposphere. The ratios of the mass of modeled to measured NOx are 1.3, 1.0, 1.5 and 1.7 in the 0- to 3-, 3- to 6-, 6- to 9-, and 9- to 12-km layers, respectively. Agreement between the model and observations is quite good from the surface up to 6 km. Above 6 km the model overestimates the mass of NOx in the upper troposphere by 50–70%, largely driven by roughly 25% of the cases which have very high modeled NOx values, as shown in Figure 8. The results become more complicated between 9 and 12 km where the NOx mass is overpredicted by 70% but the median modeled NOx values are less than the measured median values. In fact the whole range of modeled NOx values becomes much broader than the measurements. Some of the underprediction of the NOx values in the 10th and 25th percentile range is likely due to omission of aged NOx from natural and anthropogenic emissions upwind of North America.
 As previous studies have shown, NOx in the summertime North American upper troposphere is dominated by lightning emissions, therefore the overestimation of NOx for very high values (90th percentile) is due to too strong of an influence from modeled LNOx emissions. We can only speculate that perhaps the model produces too much LNOx during strong storms, or maybe the model does not adequately disperse the LNOx tracer. It is also possible that in some cases the DC-8 may have flown near to but did not penetrate the most concentrated LNOx plumes in order to avoid dangerous conditions. Another source of error is the constant NOx lifetime of 4 days. We conducted some sensitivity tests of the NOx lifetime in the upper troposphere allowing it to vary with altitude (shorter NOx lifetime at lower altitudes and longer NOx lifetime at higher altitudes) and mixing ratio (shorter lifetime at greater NOx mixing ratios, and longer lifetime at smaller NOx mixing ratios). Decreasing the lifetime at greater NOx mixing ratios did lead to an improvement in the agreement between the modeled and measured 90th percentile but only by about 10%. The sensitivity tests resulted in little change between the modeled and measured median values. Overall the 4-day lifetime appears to be robust for estimating median NOx values, and we therefore use median values for the rest of the comparisons below.
 Additional insight into the overestimation of NOx is gained by comparing our FLEXPART simulation of the ICARTT NOx measurements to the GEOS-CHEM simulation by Hudman et al. . The GEOS-CHEM simulation releases 7 kg N flash−1 at the times and locations of lightning flashes specified by the model's convection scheme, whereas the FLEXPART simulation releases 3.5 kg N flash−1 at the times and locations of the NLDN and LRLDN detected lightning flashes. Both models underestimate the median DC-8 NOx mixing ratios in the upper troposphere. But whereas FLEXPART overestimates the higher NOx values, GEOS-CHEM underestimates them, despite releasing twice as much NOx per flash. Both models have similar numbers of flashes above the United States, but GEOS-CHEM greatly underestimates the number of flashes above the coastal waters and the Sierra Madre Mountains. FLEXPART emits LNOx above the coastal waters and the Sierra Madre Mountains according to the CG flashes detected by the LRLDN, so for these regions FLEXPART should be accurate in terms of the location and time of the emissions. However, the FLEXPART simulation assumes 2.9 IC flashes per CG flash above these regions, which may be an overestimate, as little is known about the IC:CG ratio in these regions. An overestimate of LNOx emissions in this region of upper tropospheric recirculation may explain FLEXPART's very high 75th and 90th LNOx percentiles. Better quantification of the number of IC and CG flashes above Mexico, the Gulf of Mexico and the coastal North Atlantic Ocean is required to reconcile the LNOx emissions of the two model approaches.
4.3. Upper Tropospheric NOx Climatology
 An important science question to answer is: What is the typical summertime distribution of NOx across the North American upper troposphere? So far no single measurement campaign has been conducted to answer this question. At present all we can do is produce a limited “climatology” by building a composite plot of upper tropospheric in situ NOx measurements from several aircraft campaigns over the past 20 years. Such a plot is shown in Figure 9a, which uses all tropospheric NOx measurements (averaged into 2-min samples) between 8.5 and 15 km during June, July and August from the ELCHEM (1989), STERAO (1996), NOXAR (1995), POLINAT 2 (1997), CRYSTAL-FACE (2002), and ICARTT (2004) experiments. Each 1° × 1° grid cell shows the median value of all NOx measurements in that grid cell, with smoothing over a 4° × 4° range. These data sets contain NOx measurements from a wide range of air masses but can't be considered to represent true average NOx conditions above North America. For example, the NOXAR and POLINAT2 data sets are biased toward afternoon conditions when daytime trans-Atlantic commercial flights are most common above eastern North America. The ICARTT DC-8 flight tracks were not planned to randomly sample the atmosphere above North America, but were targeted toward a variety of air masses that were influenced by pollution, biomass burning plumes, convective thunderstorm outflow and relatively clean background conditions. In general the NOXAR/POLINAT2 and ICARTT flights avoided regions of active convection and therefore did not sample fresh and highly concentrated LNOx plumes. In contrast the ELCHEM, STERAO, and CRYSTAL-FACE experiments deliberately targeted regions of active convection and sampled both background air upwind of the storms and fresh LNOx plumes downwind of the storms. While these data sets have biases in their sampling procedures we can still use the data to provide a broad estimate of the upper tropospheric NOx distribution across North America as each experiment sampled a wide range of air masses, and the use of median values in Figure 9a helps to reduce the influence of extreme events.
 Overall the composite NOx plot (Figure 9a) indicates that the greatest upper tropospheric mixing ratios are located above the southeastern United States and the westernmost North Atlantic Ocean, with decreasing values toward the northeastern United States, southeast Canada and high-latitude regions of the North Atlantic. Data across much of the western United States are relatively sparse but indicate mixing ratios are less than above the southeastern United States. The measurements from the STERAO and ELCHEM experiments are of special interest because they provide dense sampling above eastern Colorado and New Mexico. Even though these data sets were biased toward sampling active convection the median NOx values are only half the values measured above the eastern United States by aircraft that avoided active convection. This comparison highlights the enhanced background NOx values that are routinely encountered above the eastern and southern United States. The high NOx values above south central Canada are from a single flight of the DC-8 on July 15, 2004, that sampled a region of strong thunderstorm outflow produced by a line of storms that marched from west to east across the Great Plains of southern Canada and the northern USA over the previous 48 h. Such high NOx values are likely to occur episodically in this region and are not a persistent feature.
Figures 9b and 9c show median modeled NOx values (lightning, stratospheric, aircraft and surface anthropogenic emissions) for the same seasonal and vertical range as in Figure 9a, during 2004 and 2006. Given that Figures 9b and 9c are for specific years, while Figure 9a is a composite of several years, a direct comparison between the modeled and measured values is not meaningful. But assuming that Figure 9a indicates typical NOx mixing ratios in the summertime upper troposphere we can at least verify whether or not the modeled values are in the correct range, and then use the modeled values to draw conclusions about regions where measurements are not available.
 For summer 2004 (Figure 9b) the modeled NOx has very high (>0.8 ppbv) average values above the southeastern USA and the adjacent portions of the Gulf of Mexico and western North Atlantic Ocean. There are a few grid cells in Figure 9a showing median measurement values at or above 0.8 ppbv, especially above Florida, but in general the measured values above the southeastern United States are less than 0.8 ppbv. For this region the model over predicts NOx, which is likely related to strong LNOx emissions along the coastal areas of the southeastern United States that had high levels of lightning activity during summer 2004 (Figure 3). The model reproduces well the decrease in NOx from the southeastern United States toward Canada and the high-latitude North Atlantic Ocean. The modeled NOx values for 2006 not only have the same general distribution as the measured data but have much better quantitative agreement above the southeastern United States. Overall the model output from both years captures the main features of the upper tropospheric NOx distribution and are typically well within a factor of 2 of the measured values.
 As mentioned earlier the NOXAR/POLINAT2 and ICARTT NOx data that cover most of the measurement region in Figure 9a are biased away from air masses containing active convection and fresh LNOx emissions. In contrast, the FLEXPART output in Figures 9b and 9c contains both aged and fresh LNOx emissions. To make a more accurate comparison we show in Figure 9d just the measurements from the NOXAR/POLINAT2 and ICARTT experiments and in Figures 9e and 9f we show the 2004 and 2006 FLEXPART output with all LNOx emissions less than 12 h old removed. Removal of fresh LNOx emissions reduces the modeled NOx above Mexico, the Gulf of Mexico and the southeastern United States by about 30%. In general the 2004 modeled NOx above the southeastern United States comes into better agreement with the measurements, while the 2006 modeled values above the southeastern United States are slightly less than the measured values. But overall the main conclusion does not change, the modeled NOx maximum above the southeastern United States and coastal waters is supported by the measurements.
 Given the reasonable agreement between the model and measurements we infer from the model output that the most concentrated portion of the upper tropospheric NOx enhancement extends westward toward the Sierra Madre Mountains of northwestern Mexico, southward to southern Mexico and Cuba, and eastward as far as Bermuda. Figures 3 and 4 indicate that lesser quantities of aged LNOx extend across the North Atlantic Ocean toward North Africa and Europe. This plume generally corresponds to the average NO2 plume detected by the satellite-based SCIAMACHY instrument during May–October, 2004 [Martin et al., 2006]. Model simulations indicate that the SCIAMACHY plume is primarily in the upper troposphere and has a major contribution from lightning emissions, although this plume is shifted farther northward than the LNOx plume calculated by FLEXPART.
Zhang et al.  calculated NOx mixing ratios at 250 hPa above North America during July and found that LNOx contributes to more than 80% of the NOx above the eastern and southern United States. Similar calculations for the present study in the 8.5–12 km range show that in both 2004 and 2006, LNOx accounts for more than 70% of upper tropospheric NOx above most of the United States, Central America and the surrounding waters, and exceeds 90% across northern Mexico, the Gulf of Mexico and Florida (Figures 10a and 10b). The earlier comparison between the total mass of the FLEXPART NOx tracer to the DC-8 measurements showed an overestimate in the upper troposphere. If we assume most of this error is due to the LNOx portion of the NOx tracer and then reduce the LNOx tracer by 40% to bring the mass of the modeled NOx tracer more in line with the DC-8 measurements, we find that LNOx still accounts for 70–90% of NOx above the eastern and southern USA (Figures 10c and 10d). The dominance of LNOx in the upper troposphere appears to be a very robust result. Even if we reduce the IC:CG flash ratio by a factor of 2, and reduce the quantity of NOx emitted per flash by a factor of 2 (i.e., reduce total LNOx emissions by a factor of 4), we find that LNOx still accounts for 50–85% (50–80%) of the upper tropospheric NOx above the eastern and southern United States in 2004 (2006).
 The major conclusions from this study are as follows.
 1. The seasonal strengthening and decay of the anticyclone in the upper troposphere above eastern North America governs the summertime buildup and decay of LNOx and thunderstorm outflow above the same region.
 2. On average the upper troposphere exhibits zonal flow in May, with a ridge building above southern Mexico in June. By July the ridge forms a closed circulation anticyclone above northern Mexico and the southern United States that strengthens in August, followed by a rapid decrease of the height, horizontal extent and latitude of the anticyclone in September.
 3. Concentrations of both the 4-day lifetime LNOx tracer and the 20-day LNOx tracer reach a maximum above the southern United States and Gulf of Mexico in July and August.
 4. All 14 study sites across North America exhibit strong daily variability in the concentrations of the LNOx tracer in the upper troposphere during summer, with the sites most heavily influenced by the North American summer monsoon having the greatest background concentrations. During late spring and September the west coast and Rocky Mountain sites have low concentrations of the LNOx tracer with little variability.
 5. In general the west coast sites plus Barbados have the most aged thunderstorm outflow, while the east coast sites have the least aged outflow.
 6. More than 80% of the NOx in the summertime upper troposphere above the eastern and southern USA is produced by lightning.
 7. The FLEXPART NOx tracer distribution in the 2004 and 2006 summertime upper troposphere matches the general continental-scale distribution of NOx as measured by several aircraft campaigns. Comparison of the mixing ratios between the modeled and measured NOx shows good agreement above the northeastern United States, southeastern Canada and the western North Atlantic, with the modeled values slightly greater over the southeastern and mid-Atlantic states in 2006, and substantially greater in 2004.
 8. Until routine survey flights of atmospheric composition can be implemented across much of North America, Figure 9a will have to serve as our best observation-based estimate of upper tropospheric NOx during summer.
 This study has provided simulations of LNOx emissions and transport above North America that are about as accurate as can currently be achieved. The LNOx tracer was emitted from the exact times and locations of all CG flashes detected above the United States, northern Mexico and the surrounding waters. The simulation was conducted with a Lagrangian particle dispersion model that does not suffer from artificial grid-scale diffusion of tracers as occurs in Eulerian models, and the model was driven by high-resolution global wind fields. The estimate of the quantity of NOx emitted per flash was taken from the most recent synthesis of all published values, and the estimate of the NOx lifetime in the upper troposphere was derived from the best available in situ measurements above summertime North America. However, uncertainty arises from: (1) the fact that the LNOx emitted per flash and the NOx lifetime are average values in a highly variable environment; (2) the ratio of IC:CG flashes is also highly variable and largely unknown above the oceans; (3) the release altitude of the LNOx tracer is parameterized and not explicit; (4) the tracer is released in a highly convective environment with the sub-grid-scale vertical motions in the model described by a parameterized convection scheme.
 There are several requirements to improve the simulation of LNOx in the upper troposphere above North America. The first would be the detection of the actual locations of all IC flashes in addition to the detection of CG flashes across the continent and surrounding waters. This may be possible by 2015 when NOAA-NASA will launch the first of the GOES-R geostationary satellite series. These satellites will contain the Geostationary Lightning Mapper instrument that will monitor all lightning flashes occurring anytime and anywhere in the Western Hemisphere. The second requirement is extensive measurements of NOx and HOx across North America in air masses with both fresh and aged thunderstorm outflow. A full characterization of the spatial distribution of NOx across the continent is necessary for model comparison and the HOx measurements will allow the calculation of the NOx lifetime under various conditions across the continent and at many altitudes. These data can also be used to better constrain the quantity of NOx produced per flash especially if lightning mapping arrays are utilized. The third would be to release the LNOx from the actual flash locations in a chemical transport model that is run on a grid scale fine enough to explicitly simulate deep convection on the continental scale.
 We thank Stuart McKeen and Greg Frost at the University of Colorado/NOAA ESRL for making the North American NOx emission inventory available to us, and Dennis Boccippio at NASA for providing the gridded IC:CG ratios. The 1989 ELCHEM Sabreliner NOx measurements and 1996 STERAO Citation NO measurements were kindly provided by Brian Ridley. We also thank the flight crew of the NASA WB-57, Brian Ridley (NCAR; NO data), Erik Richard (University of Colorado; O3 data) and Karen Rosenlof (NOAA ESRL; aircraft position data) for the CRYSTAL-FACE data. NCEP reanalysis data were provided by the NOAA/ESRL Physical Sciences Division, Boulder Colorado from their Web site at http://www.cdc.noaa.gov/. NLDN and LRLDN data were collected by Vaisala-Thunderstorm and supplied to us by the Global Hydrology Resource Center (GHRC) at NASA Marshall Space Flight Center (MSFC). The LIS/OTD 2.5° low-resolution lightning climatologies (v0.1 gridded satellite data) are preliminary data sets produced by the NASA LIS/OTD Science Team (Principal Investigator, H. J. Christian, NASA/MSFC) available from GHRC (http://ghrc.msfc.nasa.gov). EDGAR (http://www.mnp.nl/edgar) is a product of the National Institute for Public Health and the Netherlands Organisation for Applied Scientific Research and is part of the Global Emissions Inventory Activity of IGBP/IGAC.