Lightning is a large and variable source of nitrogen oxides (NOx ≡ NO + NO2) to the upper troposphere. Precise estimates of lightning NOx (LNOx) production rates are needed to constrain tropospheric oxidation chemistry; however, controls over LNOx variability are poorly understood. Here, we describe an observational analysis of variability in LNO2 with lightning type by exploiting U.S. regional differences in lightning characteristics in the Southeast, South Central, and North Central United States. We use satellite NO2 measurements from the Ozone Monitoring Instrument with Berkeley High Resolution vertical column densities, a combined lightning data set derived from the Earth Networks Total Lightning Network and National Lightning Detection NetworkTM measurements, and hourly winds from the European Centre for Medium-Range Weather Forecasts climate reanalysis data set (ERA5) over May–August 2014–2015. We find evidence that cloud-to-ground (CG) strokes produce a factor of 9–11 more NO2 than intracloud (IC) strokes for storms with stroke rates of at least 2,800 strokes·cell−1·hr−1. We show that regional differences in LNO2 production rates are generally consistent with regional patterns CG and IC stroke frequency and stroke current density. A comparison of stroke-based and flash-based CG/IC LNO2 estimates suggests that CG LNO2 is potentially underestimated when derived with flash data due to the operational definition of CG lightning. We find that differences in peak current explain a large portion of CG/IC LNO2 variability, but that other factors must also be important, including minimum stroke rate. Because IC and CG strokes produce NOx in distinct areas of the atmosphere, we test the sensitivity of our results against the atmospheric NO2 vertical distribution assumed in the a priori profiles; we show that the relative CG to IC LNO2 was generally insensitive to the assumed NO2 vertical distribution.
- Lightning NO2 production rates in the United States vary with regional differences in average current per stroke
- Cloud-to-ground (CG) strokes produce more LNO2 than intracloud (IC) strokes in the United States
- LNO2 derived from flash rather than stroke data may obscure CG and IC NO2 production per stroke differences
In the upper troposphere (UT), nitrogen oxides (NOx ≡ NO + NO2) largely control the atmospheric oxidation capacity and the chemical production of ozone (O3), a short-lived greenhouse gas and air pollutant (Labrador et al., 2004; Liaskos et al., 2015; Murray et al., 2013; Murray et al., 2014; Tost, 2017). Lightning is the dominant source of NOx to the UT at 2- to 8-Tg N·year−1 (Schumann & Huntrieser, 2007), contributing as much as 70% of UT NOx in the tropics, subtropics, and summertime midlatitudes (Allen et al., 2010; Jourdain & Hauglustaine, 2001; Lawrence et al., 1995; Sauvage et al., 2007; Zhang et al., 2003). The mechanism by which lightning produces NO is generally known; however, it has proven difficult to precisely derive lightning NOx production rates (LNOx) by correlating measurements of lightning frequency with space-based NO2 observations in the vicinity of storms (Beirle et al., 2010; Pickering et al., 2016; Schumann & Huntrieser, 2007). Reported LNOx production per flash spans three orders of magnitude (Beirle et al., 2010; Bucsela et al., 2010; Cooray et al., 2009; Cummings et al., 2013; Finney et al., 2016; Huntrieser et al., 2008; Huntrieser et al., 2009; Huntrieser et al., 2011; Jourdain et al., 2010; Koshak, 2014; Koshak et al., 2014; Laughner & Cohen, 2017; Liaskos et al., 2015; Martini et al., 2011; Miyazaki et al., 2014; Murray, 2016; Murray et al., 2013; Nault et al., 2017; Ott et al., 2010; Pollack et al., 2016) and, averaged over the globe, is subject to a factor of 4 uncertainty (Schumann & Huntrieser, 2007). The wide range in LNOx estimates is thought to be caused by multiple factors such as challenges associated with correlating measured NO2 and/or NO with lightning frequency in the dynamic atmosphere; NOx and lightning measurement uncertainties and variable detection efficiencies; differences in employed instrumentation and lack of standardized LNOx metrics; influences of nonlightning UT NOx sources, such as aircraft or the convection of surface emissions; and the natural variability in many lightning physical properties that affect LNOx production, including lightning type, peak current, and flash length (Wang et al., 1998; Huntrieser et al., 2007; Ripoll et al., 2014a; Ripoll et al., 2014b; Bruning & Thomas, 2015; Koshak, Solakiewicz, et al., 2015).
Lightning nomenclature consists of two fundamental definitions: strokes and flashes. A stroke (sometimes referred to as return stroke in the case of lightning connecting to the ground or K-change for an intracloud stroke) is defined as an impulsive surge of current occurring during a lightning discharge, with this surge thought to cause NO production. A flash consists of one or more strokes connected through a complex structure of hot channels and (cold) streamer coronas. Lightning monitoring networks measure both strokes and flashes and define flashes as one or more strokes sufficiently close in time and space as determined by the specific spatial and temporal parameters of the stroke clustering. LNOx studies have been conducted using a variety of approaches, including with satellite observations, theoretical models, laboratory experimentation, and aircraft sampling, and a challenge in the literature is that results are reported in units of per flash or per stroke and as LNO, LNO2, or LNOx depending on what nitrogen oxide measurements are available. To our knowledge, LNOx analyses utilizing Lightning Location System (LLS) measurements have to date based LNOx estimates on flash data, rather than strokes.
Lightning strokes are categorized as cloud to ground (CG) or intracloud (IC). By definition, CG flashes are stroke clusters with one or more CG strokes that may also include one or more IC strokes; IC flashes are stroke clusters with no CG strokes. CG strokes discharge between the lower-cloud charge region (~3–5 km above ground level [agl]) and the ground. IC strokes discharge between intracloud charge regions, typically consisting of a midlevel negative charge region (~5 km agl) and an upper-level positive charge region (~10 km agl) (Stolzenburg et al., 1998). Total and relative CG and IC stroke frequency is a direct function of the charge structure of thunderstorms, with higher-altitude charge regions favoring IC strokes and producing few to no CG strokes (Mansell et al., 2002; Mansell et al., 2010; Rust et al., 2005). The altitudes of the charge regions are related to cloud liquid water depletion, with slower liquid depletion rates leading to higher-altitude in-cloud charge regions (Bruning et al., 2014). Because of this, the ratio of CG to IC strokes varies regionally with climate and geography (Koshak, Cummins, et al., 2015; Medici et al., 2017; references therein). While IC strokes are more common, accounting for ~70% of all lightning, CG strokes tend to exhibit higher peak currents and typically have longer channel lengths and greater cloud-top optical area extent than IC strokes (Koshak, 2010; Koshak, 2014; Rakov & Uman, 2003). As a result, individual CG strokes are theorized to produce 10 times more NOx per stroke than individual IC strokes (Price et al., 1997).
Variability in lighting type, that is, CG versus IC stroke prevalence, could contribute a portion of the estimated LNOx uncertainty (Huntrieser et al., 2008; Schumann & Huntrieser, 2007), but observational evidence is inconclusive. Some studies indicate greater LNOx for CG strokes and flashes, but others suggest comparable CG and IC LNOx production rates. For example, Koshak et al. (2014) found that CG flashes produced approximately 10 times more LNOx than IC flashes (mean values of 484- and 35-mol NOx·flash−1, respectively) in Northern Alabama August thunderstorms (2005–2009). These estimates were obtained using the NASA Lightning Nitrogen Oxides Model (LNOM), which ingested three-dimensional lightning channel mapping data from the North Alabama Lightning Mapping Array (LMA), peak current information from the National Lightning Detection NetworkTM (NLDN), and laboratory-derived LNOx production parameterizations and theoretical considerations. In an expanded analysis over 9 years (2004–2012) in the same region, Koshak (2014) computed CG and IC production rates of 604- and 38-mol NOx·flash−1, respectively. Carey et al. (2016) widened the application of the LNOM to include a suite of radar measurements of storm environment, such as storm updraft volume and graupel mass and volume, to calculate CG/IC LNOx of ~8 (919/116-mol NOx·flash−1).
By contrast, there are reports that CG and IC strokes/flashes produce similar amounts of NOx. From a purely electrostatic analysis of the charge neutralized by a lightning flash, Gallardo and Cooray (1996) concluded that CG and IC flashes dissipate similar amounts of energy, which, assuming LNOx scales linearly with energy per flash, produced comparable LNOx. Cooray et al. (2009) predicted that CG and IC flashes both produced 3.3-mol NO per coulomb of charge neutralized in a theoretical analysis. Barthe and Barth (2008) found low LNOx sensitivity to CG/IC flash variability, in this case using vertical winds in convective cells to parameterize LNOx in a cloud-resolving model (~125-mol NO·flash−1). DeCaria et al. (2005) and Ott et al. (2010) also found comparable CG and IC LNOx production (500-mol NO·flash−1), determined using cloud-resolved model simulations of several thunderstorms using various CG and IC LNOx parameterizations, constraining them with in-cloud NOx distributions measured by aircraft. The NASA Global Modeling Initiative chemical transport model assumed equivalent per flash CG and IC LNOx but implemented geographic differences between midlatitude (500-mol NO·flash−1) and tropical (250-mol NO·flash−1) lightning, in agreement with empirical studies of midlatitude (480-mol NO·flash−1) (Allen et al., 2010; Jourdain et al., 2010) and tropical (250-mol NO·flash−1) lightning (Bucsela et al., 2010). At the same time, Marais et al. (2018) did not observe differences between midlatitude and tropical LNOx, deriving LNOx of 280 ± 80-mol NOx per total (CG + IC) flash using satellite-based lightning measurements that do not distinguish between CG and IC.
In light of uncertainties in CG versus IC LNOx production rates, here we exploit regional differences in relative CG and IC stroke frequency across the United States to investigate variability in the ratio of mean LNO2 produced by CG strokes to that produced by IC strokes (CG/IC LNO2). We combine observations of NO2 from the satellite-based Ozone Monitoring Instrument (OMI) Berkeley High Resolution (BEHR) retrieval and lightning data from a combined data product using the Earth Networks Total Lightning Network (ENTLN) and NLDN across the contiguous United States (CONUS) and in three geographic areas, the Southeast, South Central, and North Central United States, where there are pronounced differences in CG/IC strokes (Figure 1, Table 1). We focus on storms with short spatial and temporal delays between stroke occurrence and NO2 observation and NO2 measurements with high cloud radiative fraction (CRF) conditions, using a methodology developed by Pickering et al. (2016). Second, we compare CG and IC LNOx estimates when derived from stroke versus flash frequency data and describe LNOx variability as a function of stroke peak current data, as opposed to total or type-speciated stroke number. Finally, we discuss uncertainties and test the sensitivity of our results against different OMI BEHR a priori NO2 profiles.
|U.S. region||CONUS||Southeast||South Central||North Central|
|Mean daily stroke number (x 103 strokes)|
|Mean daily current density (x 103 A·stroke−1)|
|Total||5.7 ± 1.9||5.0 ± 2.0||5.7 ± 1.6||5.2 ± 1.8|
|CG||21.0 ± 4.5||20.9 ± 4.1||20.5 ± 4.0||17.9 ± 4.8|
|IC||4.5 ± 1.3||3.6 ± 1.2||4.6 ± 1.1||4.6 ± 1.4|
|Mean daily current (x 106 A)|
- Note. Daily current density uncertainties are 95% confidence intervals.
The OMI onboard the Aura satellite observes the total atmospheric NO2 column density with global coverage obtained every 2 days since 2008 (due to the OMI row anomaly), passing over the United States at ~1:50 p.m. local time (LT). OMI detects NO2 by differential optical absorption spectroscopy and is a nadir-viewing imaging spectrometer that measures backscattered solar radiation over the wavelength range 270–500 nm with a spectral resolution of approximately 0.5 nm. We use tropospheric vertical column densities from the BEHR product Version 3-0A, which is a regional retrieval focused on North America that incorporates a number of high-resolution a priori inputs (Laughner et al., 2018; Russell et al., 2011). The BEHR V3-0A air mass factor (AMF) is based on high spatial resolution albedo measurements (0.05° × 0.05°) from the Moderate Resolution Imaging Spectroradiometer 6 MODIS BRF product and monthly NO2 and temperature profiles simulated by WRF-Chem at 12-km spatial resolution. Modeled NO2 profiles include a source for lightning NO2 and boundary layer pollution lofted to the UT by convection, and the profiles were updated to year 2012 emissions (Laughner et al., 2018). We used the “visible-only” product, defined as vertical columns in which the below-cloud NO2 contributions have not been inferred, regridded to 0.05° × 0.05°.
To limit the impact of anthropogenic pollution on lightning production estimates, we first removed NO2 columns greater than 3 × 1015 molecules·cm−2, which have been shown to be typical of urban NO2 concentrations (Beirle et al., 2010). A factor of 2 increase or decrease in this cutoff did not alter our conclusions. Second, we only considered NO2 columns with a CRF greater than 0.9. Pickering et al. (2016) demonstrated that this CRF threshold effectively filtered out most below-cloud NO2. While these two selection criteria increased the sensitivity of our analysis by reducing noise from below-cloud surface NOx sources, they potentially biased the results low because a portion of lightning-generated NO2 should be located beneath the upper portion of clouds. OMI provides cloud optical centroid pressures, which are the atmospheric pressures to which NO2 is detected, and these pressures are often deeper than the tops of most thunderstorm clouds at 400–600 hPa (Vasilkov et al., 2008).
We created a combined stroke data product using observations from the ENTLN and NLDN. The networks continuously measure lightning stroke occurrence times, geographic locations, type (CG versus IC), polarity, and peak currents. Both networks detect radio frequency radiation emitted by lightning using magnetic direction finding (NLDN) and time of arrival (NLDN and ENTLN) (Rakov, 2013). For the joint product, we merged the ENTLN and NLDN data sets following a methodology similar to Bitzer and Burchfield (2016), ultimately keeping a set of unique strokes and flashes. In brief, we first included all events detected by either network, then tested for duplicates (i.e., events recorded by both networks) determined as any match within the specified time and distance constraints of 1 ms and 25 km. The time constraint was limited by the NLDN 1-ms time resolution. In the case of a match, the duplicate event was counted as a single unique event, and data from ENTLN were incorporated into the final merge. If either network reported a stroke as CG, then that unique stroke was categorized as CG.
ENTLN and NLDN CG stroke detection efficiencies are generally considered to be >90% within the CONUS. High detection efficiencies result from CG strokes being easier to detect at low sensor density (Liu & Heckman, 2011) because they are on average more powerful, larger, and longer in duration. A localized study in Florida reported 97% CG detection by ENTLN using high-speed video confirmation that channels connected to ground indicate (Zhu et al., 2017). Due to high CG detection efficiencies across both networks, we use CG stroke counts as is (no detection efficiency correction applied). Zhu et al. (2017) empirically determined peak current estimation errors of ±15% by comparing ENTLN measurements to observations of triggered—CG strokes, which we included as an additional source of uncertainty in current-based results.
IC stroke detection efficiencies are lower and more variable with network sensor density than CG detection efficiencies. Therefore, to evaluate regional variability in CG/IC LNOx, we first needed to consider the impacts of spatially variable IC detection efficiencies. To do this, we compared the ENTLN and NLDN combined product (ENTLN + NLDN) with observations from the Lightning Imaging Sensor (LIS) onboard the Tropical Rainfall Measuring Mission (TRMM) satellite. TRMM-LIS has a detection efficiency for total lightning flashes (TRMM-LIS does not distinguish between CG and IC) in North America of ~80% (Bitzer et al., 2016; Boccippio et al., 2002). Because most lightning flashes and strokes are IC (e.g., 94% of strokes in CONUS in May–August 2014; Table 1), LIS group number (similar to stroke number) approximately equals IC stroke number. To determine stroke/group matches between ENTLN + NLDN and TRMM-LIS data sets, we employed the same selection criteria as Bitzer et al. (2016): ±10 ms and 20 km. The resulting total stroke detection efficiency map is shown in Figure 2. Average IC stroke detection efficiencies in the CONUS (up to 38° latitude), Southeast, and South Central United States regions were 45%, 63%, and 44%, respectively, and in general agreement with past estimates for North America (Rudlosky, 2015; Thompson et al., 2014). The average IC flash detection efficiency in CONUS was 88%. The North Central United States was out of view of TRMM-LIS; therefore, we compared IC stroke observations from ENTLN (not the combined product) for each region to those from the International Space Station (ISS) LIS over the time period March–December 2017. ISS-LIS is the same lightning sensor as TRMM-LIS but onboard the ISS, which has an orbit that extends up to ~55° latitude and covers the globe in ~1.5 hr. Detection efficiencies in the North Central region were approximately equal to the South Central United States using ISS-LIS; therefore, we use the detection efficiency value of 44% derived using TRMM-LIS for the North Central United States in this study.
ENTLN flash criteria were 0.7 s and 10 km (Liu & Heckman, 2011), and NLDN flash criteria were 1 s and 10 km (Cummins et al., 1998). In May–August 2014, across CONUS, the average number of strokes per flash, known as the multiplicity, was 6.3 for CG + IC, 8.7 for CG flashes, and 5.9 for IC (with IC strokes corrected for detection efficiency). In the case that multiplicity is high and the relative number of CG strokes in the cluster is low, a CG flash, as operationally defined for LLS data sets, would appear to produce a similar amount of NOx as an IC flash, assuming a similar number of strokes. Multiplicity varied geographically across CONUS, with higher strokes per flash in the U.S. Mid-Atlantic and Florida (Figure 3). High multiplicity areas were better sampled by the LLS. Regional detection efficiency-corrected multiplicities (May–August 2014) were as follows: Southeast, 7.3 (CG + IC), 10.7 (CG), and 6.8 (IC); South Central, 8.1 (CG + IC), 12.2 (CG), and 7.7 (IC); and North Central, 5.9 (CG + IC), 9.5 (CG), and 5.6 (IC).
To account for displacement of LNOx downwind of lightning before OMI sampling, we used hourly winds from the European Centre for Medium-Range Weather Forecasts (ECMWF) climate reanalysis data set (ERA5) (Dee et al., 2011). ERA5 has 37 pressure levels and a horizontal resolution of 31 km2.
3 Deriving LNO2
LNOx is produced in the form of NO, while OMI detects NO2. NO is converted to NO2 through reactions with O3 and peroxy radicals, and in the presence of sunlight, NO2 is photolyzed back to NO, regenerating O3. This chemistry reaches steady state within minutes during the daytime and NOx partitions to an NO-to-NO2 ratio as a function of the atmospheric oxidant concentration, temperature, and solar radiation (Leighton, 1961). While O3 abundance in the UT near storms usually ranges from 60–100 ppb, O3 mixing ratios at storm cloud edges have been reported as high as 150–400 ppb, with elevated O3 filaments extending approximately a hundred kilometers downwind of storms (Pan et al., 2014). Enhanced UT O3 is caused in part by the overshooting of thunderclouds into the stratosphere, which pull high concentrations of O3 into the troposphere (Pan et al., 2014), new O3 chemical production from precursors lifted to the UT in convective updrafts (Brunner et al., 1998; Huntrieser et al., 2016b; Pickering et al., 1990), and potentially direct O3 formation from cold ionization (electrical corona discharge) around the hot plasma channel (Bozem et al., 2014; Kotsakis et al., 2017; Minschwaner et al., 2008). Using aircraft observations from the Deep Convection Clouds and Chemistry (DC3) experiment, Pan et al. (2014) found evidence of stratospherically influenced elevated O3 concentrations at cloud edges on more than half of the flight segments that sampled of tropopause-reaching storms. All else being equal, a change in O3 from 80 to 160 ppb would alter NO/NO2 by more than 40%, adding uncertainty to LNOx (or LNO) estimates derived from NO2 observations. This estimate was computed at 10 km agl following data presented for the Southeast U.S. region during the SEAC4RS mission (Silvern et al., 2018): jNO2 = 0.014 s−1 (NO2 photolysis rate), kNO + O3 = 3 × 10−12exp(−1,500/T) (NO + O3 rate expression; Burkholder et al., 2015), T = 240 K, and with 75% of the NO to NO2 conversion occurring via reaction with O3. Because of the temperature dependence of kNO + O3, modeled NO/NO2 ratios increase sharply with altitude (Burkholder et al., 2015); however, aircraft observations indicate that NO/NO2 ratios are lower than predicted in the UT. On average, over the Southeast United States during SEAC4RS, NO/NO2 ratios above 8 km equaled 1.4 mol·mol−1, which were less than half those modeled by the GEOS-Chem chemical transport modeled and sampled along the flight track (Travis et al., 2016). The cause of this discrepancy is uncertain, with Silvern et al. (2018) suggesting a combination of positive interferences in aircraft NO2 measurements from known and unknown thermally unstable organic nitrates (Nault et al., 2015; Silvern et al., 2018) and errors in jNO2 and kNO + O3 at low temperatures. As a result, we opted not to assume an NO/NO2 ratio but instead reported our results as LNO2 rather than LNOx. To facilitate comparison with previously published LNOx estimates, when required, we use the observed NO/NO2 ratio 1.4 mol·mol−1 or NOx/NO2 ratio 2.4 mol·mol−1 (Silvern et al., 2018).
To compute LNO2, we adapted the method described in Pickering et al. (2016), which is outlined here as follows. First, stroke measurements from the combined ENTLN + NLDN data set were summed over a 1° × 1° grid for the local time period of 11:26 a.m. LT (2.4 hr before the OMI overpass) to 1:50 p.m. LT (OMI overpass time), and the NO2(0) data were averaged to this same grid. Correlations between NO2(0) and strokes were limited to those co-occurring in the same grid cell in order to reduce LNO2 variability caused by atmospheric mixing and chemistry on longer timescales. To distinguish the LNOx signal from UT background NO2 (NO2,background) “noise,” Pickering et al. (2016) used a flash frequency threshold of 1,000 flashes·cell−1·hr−1. However, Bucsela et al. (2019) recently demonstrated that LNOx production efficiency varied with storm flash rate such that lower flash rates yielded higher LNOx, suggesting that LNOx estimates were highly sensitive to the application of a minimum flash number. That said, without a flash frequency threshold, there was no guarantee that grids used in the analysis were even electrically active, and, under this condition, the R2 of the NO2(0) and strokes correlation across CONUS was ~0. Here, we opted to apply a minimum hourly flash rate of 1,000 following Pickering et al. (2016), thereby only including grids with lighting in the calculation. Because the average cell residence time was 2.4 hr, our flash frequency threshold was therefore 2,400 flashes·cell−1. During our study period, flashes produced 2.8 times more LNO2 than strokes (calculated without a minimum flash or stroke threshold), so the equivalent minimum stroke number per cell was 6,720. To test the sensitivity of our results to this stroke rate threshold, we repeated our full analysis at minimum stroke rates spanning 1–10,000 strokes·cell−1·hr−1 (Figure 4a). In agreement with Bucsela et al. (2019), we observed a decreasing trend in LNO2 with increasing stroke frequency that was accompanied by an increasing trend in R2. For example, increasing the minimum stroke rate to 5,000 strokes·cell−1·hr−1 decreased total LNO2 by 13% and increased R2 to 0.6 (from 0.54). At a higher minimum stroke rate of 10,000 strokes·cell−1·hr−1, total LNO2 decreased by 31%, and R2 increased to 0.66. As a consequence of focusing on higher stroke rate storms, our absolute LNO2 production efficiencies yielded lower LNO2 per flash compared to those that would be derived using a lower threshold and should be interpreted in the context of the LNOx dependence on stroke frequency (Bucsela et al., 2019). Critical to this work, we also observed a dependence on minimum stroke rate for the CG/IC LNO2, which was largely caused by variability in CG LNO2 (Figure 4), which we report in section 4.1 and discuss in section 4.3.
4 Results and Discussion
4.1 Total (CG + IC), CG, and IC LNO2 by Region
Correlations between NO2(0) and stroke number for CONUS and the three U.S. regions are shown in Figure 6. The resulting LNO2 (mol NO2·stroke−1) is given in Table 2 for all strokes (CG + IC) and for CG and IC separately. The highest total (CG + IC) LNO2 production rates were observed in the South Central United States, the region with the most frequent lightning strokes and where strokes have the largest mean stroke current density (Table 1). The Southeast United States experienced similar average LNO2 compared to CONUS but experienced the lowest stroke frequency and stroke current density. The North Central region had the lowest total LNO2 production rate, but stroke number and daily stroke current densities were comparable to the Southeast United States. The significance and statistical characteristics of LNO2 correlations are presented in Tables 3 and 4, respectively. Regional LNO2 (CG + IC) differences were significant at the 5% level, with the exception of the Southeast and South Central United States, where differences were statistically significant at the 10% level (Table 3). Total LNO2 in the Southeast United States was statistically indistinguishable from CONUS, while the South Central and North Central United States were significantly different at the 10% level. The linear fits resulted in y-intercepts of ~10% or less of the LNO2 estimates (Table 4), indicating small contributions from nonlightning sources in all regions. Correlation coefficients for CONUS and in the South Central and Southeast United States were R2 0.5–0.6. The poorest correlation was observed in the North Central United States (R2 = 0.39), which was potentially caused in part by the smaller number of data points in the fitting (N) and uncertainty in the detection efficiency correction because of the lack TRMM-LIS overlap.
|LNO2 (mol NO2·stroke−1)|
|U.S. region||CONUS||Southeast||South Central||North Central|
|Total||1.6 ± 0.1||1.6 ± 0.2||1.9 ± 0.2||1.2 ± 0.2|
|CG||10.7 ± 2.5||9.7 ± 2.9||10.9 ± 4.4||8.7 ± 11.5|
|IC||1.1 ± 0.1||0.9 ± 0.3||1.2 ± 0.4||0.9 ± 0.5|
- Note. LNO2 was calculated for grids with a minimum stroke rate of 2,800 strokes·cell−1·hr−1. Uncertainties are 95% confidence intervals.
|p values for slope differences|
|U.S. region||CONUS||Southeast||South Central||North Central|
|U.S. region||y-intercept (x 106)||N||R2 (CG + IC)||R2 (equation 2)|
|CONUS||0.17 ± 0.03||174||0.54||0.57|
|Southeast||0.07 ± 0.02||58||0.63||0.67|
|South Central||0.15 ± 0.03||78||0.50||0.55|
|North Central||0.14 ± 0.03||44||0.39||0.39|
- Note. Results include grids with a minimum stroke rate of 2,800 strokes·cell−1·hr−1.
Solving equation 2, we calculated greater CG than IC LNO2 production rates (mol NO2·stroke−1) (Table 2). For CONUS, we found ~9.7 times higher LNO2 by CG compared to IC strokes. By region, we computed over 11 times greater CG than IC LNO2 in the Southeast, ~9 times greater in the South Central, and ~10 times greater than IC LNO2 in the North Central United States. R2 values only slightly improved by conducting the linear fit with equation 2, except in the North Central United States where there was no change, suggesting that factors other than stroke type influenced the variance. Regional variability in CG versus IC LNO2 corresponded to CG to IC differences in mean daily current densities (Table 1), with the highest ratio of CG to IC current densities in the Southeast and the lowest in the North Central United States (CG/IC rather than absolute CG current densities). In most regions, small ratios of CG to IC LNO2 were accompanied by smaller ratios of CG to IC daily current densities, and vice versa, so differences in CG/IC LNO2 were compensated for by relative current density differences; however, the North Central region was an outlier. Regional CG/IC differences were not explained by regional variability in IC detection efficiencies. Uncounted IC strokes, or undercorrected detection efficiencies, would have led to false enhancements in CG/IC LNO2; instead, we observed both the largest CG/IC and highest IC detection efficiencies in the Southeast United States. Absolute values of CG/IC LNO2 were comparable to ratios reported by other researchers. Koshak et al. (2014) and Carey et al. (2016) used the LNOM and data from the LMA to estimate CG/IC LNOx of 7.9–13.8 flashes over Alabama, located between the South Central and Southeast United States. While their studies were performed using a small number of thunderstorms, as opposed to the several months of storms in this study, the LMA has a near perfect detection efficiency within its field of view of ~200-km radius (Thomas et al., 2000; Thomas et al., 2004), providing flash extent information to account for differences in flash length. Wang et al. (1998) also found a positive correlation between NO production and current in laboratory sparks, although the correlation was nonlinear.
While the results above were calculated with a minimum stroke rate of 2,800 strokes·cell−1·hr−1, we also tested the sensitivity of CG/IC LNO2 to this threshold (Figure 4). We observe a decreasing trend in CG/IC LNO2 from 23 when the threshold is 1 stroke·cell−1 down to 1.6 when it is 10,000 strokes·cell−1·hr−1. That said, even at minimum stroke rate as high as 10,000 strokes·cell−1·hr−1, which only included 386 grids over the study window compared to 2,246 for a minimum stroke rate of 1 stroke·cell−1, CG LNO2 was still at least 50% greater than IC LNO2 (CG/IC LNO2 = 1.6). The trend in CG/IC was driven by decreasing CG LNO2 with increasing stroke rates, with IC LNO2 comparatively steady over the minimum stroke rate range. This suggests that observations of the total LNO2 dependence on storm stroke rate were caused in large part by changes CG LNOx production efficiency. This variability with stroke rate may also provide at least a partial explanation for differences in CG/IC determined in past observations studies, as past studies have used inconsistent filtering for storm flash frequency.
4.2 Influence of Lightning Data: Stroke, Flash, and Current
To investigate how stroke-based LNO2 production rates differed from results derived using flash data, we compared stroke-based results to flash-based absolute and CG/IC LNO2 across CONUS (Figure 7, Table 5). On average, there were 6.3 strokes per flash; however, the total (CG + IC) LNO2 per flash was only 2.8 times larger than when calculated per stroke. This implies that the use of flash data may underestimate per-flash total LNO2 by ~55% and per-flash CG LNO2 by ~70%. We computed that CG/IC LNO2 equaled 3.4 using flash data (at a minimum flash rate of 1,000 flashes·cell−1·hr−1), which was considerably lower than the ratio of 9.7 determined using stroke data and consistent with our hypothesized biases based on the CG flash operational definition.
|Lightning data||Stroke (mol NO2·stroke−1)||Flash (mol NO2·flash−1)||Peak current (mol NO2·stroke−1)|
|Total||1.6 ± 0.1||6.6 ± 4||1.6 ± 0.1|
|CG||10.7 ± 2.5||18.5 ± 7.9||4.7 ± 2.7|
|IC||1.1 ± 0.2||5.4 ± 0.9||1.3 ± 0.2|
- Note. Current has been converted to units mol NO2·stroke−1 by dividing by the average current per stroke. Uncertainties are slope 95% confidence intervals.
Because NO is presumably produced by current along a lightning channel, comparison of OMI NO2 with measured lightning current should provide a more direct correlation and may reveal errors inherent in quantifying LNO2 from stroke or flash frequency observations. We calculated LNO2 using ENTLN peak current (mol NO2·A−1) for CONUS by least squares linear regression and equation 2 (Figure 7). To facilitate comparison with stroke-based LNO2 and literature flash-based estimates, we multiplied the current-based LNO2 by the mean current per stroke for each lightning type: 5.6 kA (CG + IC), 20.9 kA (CG), and 4.5 kA (IC). Current-based total and IC LNO2 were within 10% of stroke-based estimates but lower than flash-based LNO2 results. Current-based CG LNO2 was 56% lower than calculated using stroke data LNO2 but as much as 76% lower than flash-based LNO2.
If LNO2 production rates for CG versus IC lightning were only a function of peak current, we would expect CG/IC ≈ 1 in units mol NO2·A−1/NO2·A−1; however, current-derived CG/IC equaled 3.6. The difference in stroke-based CG/IC and current-based CG/IC (Table 5) indicates that peak current may account for up to ~60% of CG/IC LNO2 differences but that other factors, to which CG strokes are more sensitive than IC strokes, are also important. In addition to higher peak currents, CG strokes tend to be larger and longer (Koshak, 2014), which may contribute to differences not explained by peak current alone. Huntrieser et al. (2008) found that flash size played a key role in LNOx production, while peak current was minor, even though the analysis used a model predicting a positive correlation between spark current and NOx production (Wang et al., 1998). To reduce detection efficiency effects, Huntrieser et al. (2008) only included lightning strokes larger than 10 kA in both subtropical and tropical storms. However, tropical storms tend to have smaller flashes and by extension lower peak current flashes. It is therefore possible that selecting flashes with currents larger than 10 kA introduced a selection bias toward tropical results over subtropical results and obscured the effect of peak current on LNOx production.
4.3 Comparisons and Uncertainties
To compare our results to previous work, we converted the stroke-based LNO2 (units mol NO2·stroke−1) to LNOx (mol NOx·flash−1) by assuming NOx/NO2 ratio 2.4 mol·mol−1 and 6.3 strokes·flash−1. For CONUS, total (CG + IC) LNOx was 24.2-mol NOx·flash−1, which was lower than previous midlatitude results that range 33- to 500-mol NOx·flash−1 (e.g., Beirle et al., 2010; Ott et al., 2010; Schumann & Huntrieser, 2007). Recently, Nault et al. (2017) proposed that CONUS LNOx should be increased to 665-mol NO·flash−1, contrary to the low production rate inferred here. There were three key differences between Nault et al. (2017) and this study: the NO2 measurement timing and source (satellite-based versus aircraft observations); lightning discharge data type (strokes versus flashes); and inference method of lightning production rates (direct calculation versus model-observation comparison). While the focus of our study was not to improve LNOx accuracy but instead on identifying differences in CG/IC NOx emissions, we attributed the lower estimates derived here at least in part to inherent differences between different LLS detection sensitivities (discussed below), coarse OMI spatiotemporal resolution compared to aircraft data, statistical rather than case study-based analysis, and the decision to limit NO2 data to high CRF pixels. In addition, Allen et al. (2019) used the North Alabama LMA to correct the World Wide Lightning Location Network (WWLLN) detection efficiency and TRMM-LIS for a second-order, region-specific correction for a study in the tropics, finding tropical flashes produced 170 ± 100-mol NOx·flash−1. Because recent work has demonstrated that LNOx production rates vary with flash rate, such that LNOx decreases with increasing storm flash frequency (Bucsela et al., 2019), LNOx comparisons between studies are complicated as results often do not report storm flash rates. With no flash frequency threshold, Bucsela et al. (2019) found an LNOx rate of 180 ± 100-mol NOx·flash−1. Under the same conditions (minimum flash rate = 1), we computed LNOx equal to 36-mol NOx·flash−1. The higher observed LNOx in Allen et al. (2019) and Bucsela et al. (2019) may be a combined result of differences in WWLLN detection efficiency correction, which is larger than for ENTLN, as well as the stroke versus flash biases discussed earlier in this paper. Finally, Gordillo-Vázquez et al. (2019) performed a global modeling study using various parameterizations for lightning to determine that LNOx was on average between 310–441 mol·flash−1; however, oceanic lightning is known to have larger peak currents than continental lightning (Nag & Cummins, 2017), which should result in higher LNOx.
Because LLS systems, such as ENTLN, NLDN, and WWLLN, are more sensitive to CG than IC strokes (Murphy et al., 2014; Rudlosky & Shea, 2013; Zhu et al., 2016), the measured current per stroke (or flash) is biased high even after correcting for detection efficiency. The ENTLN detects both CG and IC lightning with varying detection efficiencies across CONUS (Liu & Heckman, 2011; Rudlosky, 2015; this study). While past analyses using NLDN and WWLLN have applied corrections for CG and IC detection efficiencies, these corrections do not account for the temporal and spatial variability in detection efficiency inherent in all lightning data sets. This is especially true for IC strokes and flashes, as the detection efficiency corrections are large and therefore highly uncertain. Bucsela et al. (2010) used TRMM-LIS measurements to upscale CG/IC observations from the Costa Rica Lightning Detection Network, a system with similar sensors to the NLDN. Bucsela et al. (2010) ultimately estimated LNOx of 100–250 mol·flash−1, greater than both our stroke-based (77% greater) and flash-based (84% greater) results. While our method is otherwise similar, Pickering et al. (2016) estimated total LNOx over the Gulf of Mexico using WWLLN data, with total (CG + IC) corrected using LIS, to be 80 ± 45-mol NOx·flash−1, which is ~70% larger than our stroke-based results and 80% greater than our flash-based results.
LNO2 uncertainties were added in quadrature, treating individual errors as uncorrelated (Table 6): NO2 slant column measurements, ±15%; unobserved NO2 below the OMI optical centroid pressure, ±15%; stratospheric vertical column subtraction, ±40% (Pickering et al., 2016); and wind-related variability in the stroke counting window (t), ±15%. We estimated ENTLN detection efficiency errors as ±16% for total and IC strokes, which was the standard deviation of detection efficiency estimates over CONUS relative to TRMM-LIS. Due to high detection efficiencies for CG strokes, we estimated uncertainties of ±5%. The y-intercepts (~105 mol) implied small contributions from nonlightning sources. Intercept values were larger than in Pickering et al. (2016), which were stated to be near 0; however, their region of study was the Gulf of Mexico, where surface NO2 emissions were minimal. Cloud-resolving models and aircraft studies over rural areas found that 83–90% of NOx within the anvil of a storm was produced by lightning (DeCaria et al., 2000; DeCaria et al., 2005); as a result, we considered errors in the tropospheric background NO2 column to be ±20% (Bucsela et al., 2019). We used ENTLN current errors of ±15% (Zhu et al., 2017), determined for negative polarity CG strokes. We know of no estimates of IC current error and therefore applied ±15% to all current measurements. Combined uncertainties were ±50% (CG) and ±52% (IC) LNO2 using stroke frequency data and ±52% (CG) and ±54% (IC) LNO2 using current data. Since most strokes were IC, we used IC uncertainty estimates for total (CG + IC) LNO2.
|Strokes and Flashes|
|Slant column measurementd||5%||5%|
|Optical centroid pressurea,d||15%||15%|
|Stratospheric NO2 removalb||40%||40%|
|Stroke/flash counting due to wind||15%||15%|
Restricting the analysis to high CRF pixels may have also contributed to LNO2 estimates that were biased low and influenced the relative CG to IC results, as approximately 30% of LNOx is expected below the cloud (Pickering et al., 2016). To investigate this, we performed four sensitivity tests varying the a priori NO2 profile and combining these custom profiles with BEHR scattering weights to derive AMFs using equation (3) in Laughner et al. (2018). New AMFs were applied to OMI slant columns to compute vertical column densities, which were then used to recalculate LNO2. Because of slight differences in how the AMF was applied here from the BEHR retrieval, we included a baseline run with BEHR a priori NO2 profiles. Lightning NO2 profiles were represented by a vertical Gaussian distribution for which we varied altitude of the peak (hPa), peak magnitude (ppb), and peak width (hPa) inferred from the NO and NOx aircraft measurements in thunderstorm outflows during the DC3 campaign reported in Huntrieser et al. (2016a). The sensitivity test parameters were as follows: peak altitude, 200 and 500 hPa; magnitude, 0.1 and 0.9 ppb NO2; and width, 60 and 120 hPa. We found that the relative CG to IC LNO2 was generally insensitive to the NO2 vertical distribution, with case study results varying by less than 5% (Table 7). The largest effects on CG/IC were observed by increasing the width of the NO2 distribution, effectively adding more LNO2 throughout the cloud. The greatest total LNO2 was determined when the majority of NO2 was narrowly distributed (0.1 ppb and 60 hPa) high in the atmosphere (200 hPa) or the original magnitude (0.9 ppb and 60 hPa) low in the atmosphere (500 hPa). Changing the NO2 distribution peak height had a stronger effect (15%) on CG LNO2 than changing the peak magnitude (8%). Overall, this analysis implied that the NO2 profile did not have a major impact in the total LNO2, with less than 15% difference between all test cases. Increasing the source height from 500 to 200 hPa produced a 13% difference, less than our 30% estimated errors due to our use of high CRF data. Furthermore, these results suggested that relative CG to IC LNO2 was not a result of the NO2 profile used by BEHR. Because the DC3 campaign focused on anvil sampling, not on sampling entire clouds, the vertical extent of LNOx in active storm clouds could be greater than used here (60 and 120 hPa); however, the effect of widening the distribution would likely be too small to alter our conclusions.
|LNO2 (mol NO2·stroke−1)|
|Parameters||Test 1||Test 2||Test 3||Test 4||BEHR a priori|
|Total||1.5 ± 0.1||1.3 ± 0.1||1.5 ± 0.1||1.3 ± 0.1||1.5 ± 0.1|
|CG||9.8 ± 2.2||9.0 ± 1.9||10.4 ± 2.1||8.5 ± 1.8||10.2 ± 2.4|
|IC||1.0 ± 0.2||0.9 ± 0.1||1.0 ± 0.1||0.9 ± 0.1||1.1 ± 0.2|
- Note. Uncertainties are 95% confidence intervals.
The theoretical advantage of the BEHR profiles was that the high spatial resolution of the profiles is theoretically better able to capture the vertical distribution of NO2 resulting from lightning than retrievals that use coarser profiles (the subproduct with daily profiles would theoretically better capture the temporal variation in vertical distribution due to lightning but did not cover the time period of interest for this study). However, the profile sensitivity results suggested that the variability in profile shape unique to lightning had smaller effect on LNO2 estimates than did errors from other sources. Here, we used the “visible-only” AMFs, which did not estimate the below-cloud NO2. Given that the error due to obscured below-cloud NO2 is 30%, a hybrid approach using daily BEHR profiles to estimate the below-cloud component could be a significant improvement. To date, validation of the BEHR a priori profiles has focused on the day-to-day variability in urban areas (Laughner et al., 2019) or the average performance in rural areas with significant lightning influence (Laughner et al., 2019; Laughner & Cohen, 2017). Recent work by Zhu et al. (2019) has improved the average accuracy of the simulation from which BEHR a priori profiles are derived, but validating the ability of those profiles to reproduce the day-to-day variability of lightning NO2 in the profiles remains to be done.
We conducted an observational analysis using OMI BEHR NO2 columns and a combined ENTLN + NLDN lightning data product to investigate variability in per-stroke LNO2 as a function of lightning type. For CONUS, we found that total (CG + IC) LNO2 equaled 1.6 ± 0.1-mol NO2·stroke−1, which was lower than typically reported in the literature. We computed LNO2 production rates for three U.S. geographical regions, the Southeast, South Central, and North Central United States, with distinct lightning characteristics, relative prevalence of CG and IC lightning strokes, and stroke current densities. We showed that regional LNO2 differences were generally explained by variability in these lightning properties, with average current per stroke being an especially strong predictor of LNO2 variability. We separately estimated the CG and IC LNO2, finding a factor of 9–11 difference in CG/IC for all regions. We also find that this factor is significantly dependent on the storm stroke rate, with CG/IC LNO2 decreasing with increasing stroke frequency, which may provide partial explanation as to why some past observational studies have found similar CG and IC NO2 production efficiencies. Because of how CG and IC flashes are operationally defined, we demonstrated that LNO2 estimates derived from flash data underestimated CG/IC compared to stroke-based results. Due to the larger uncertainties in current measurements, it was unclear whether the stroke or current data were more accurate. Finally, we studied the sensitivity of our results to various BEHR NO2 a priori profile shapes, finding that CG/IC was generally insensitive to changes in the magnitude, height, and width of a Gaussian NO2 distribution derived from empirical aircraft measurements.
The authors thank Dr. Scott Rudlosky for his helpful comments and insight. BEHR NO2 V3 vertical columns are available for download at https://behr.cchem.berkeley.edu/download-behr-data/. ENTLN measurements were obtained freely by request from Earth Networks (https://www.earthnetworks.com). The authors acknowledge Vaisala, Inc., for providing the NLDN data set. The ERA5 is provided by ECMWF (http://apps.ecmwf.int/data-catalogues/era5/). ECMWF data contain modified Copernicus Climate Change Service Information 2014–2015.
- 2010). Impact of lightning NO emissions on North American photochemistry as determined using the Global Modeling Initiative (GMI) model. Journal of Geophysical Research-Atmospheres, 115(D22), D22301. https://doi.org/10.1029/2010JD014062
- 2019). Lightning NOx production in the tropics as determined using OMI NO2 retrievals and WWLLN stroke data. Journal of Geophysical Research-Atmospheres, 124(23), 13498–13518. https://doi.org/10.1029/2018JD029824
- 2008). Evaluation of a new lightning-produced NOx parameterization for cloud resolving models and its associated uncertainties. Atmospheric Chemistry and Physics, 8(16), 4691–4710. https://doi.org/10.5194/acp-8-4691-2008
- 2010). Direct satellite observation of lightning-produced NOx. Atmospheric Chemistry and Physics, 10(22), 10965–10986. https://doi.org/10.5194/acp-10-10965-2010
- 2016). Bayesian techniques to analyze and merge lightning locating system data. Geophysical Research Letters, 43(24), 12605–12613. https://doi.org/10.1002/2016GL071951
- 2016). A Bayesian approach to assess the performance of lightning detection systems. Journal of Atmospheric and Oceanic Technology, 33(3), 563–578. https://doi.org/10.1175/JTECH-D-15-0032.1
- 2002). Performance assessment of the optical transient detector and Lightning Imaging Sensor. Part I: Predicted diurnal variability. Journal of Atmospheric and Oceanic Technology, 19(9), 1318–1332. https://doi.org/10.1175/1520-0426(2002)019
- 2014). Influence of corona discharge on the ozone budget in the tropical free troposphere: A case study of deep convection during GABRIEL. Atmospheric Chemistry and Physics, 14(17), 8917–8931. https://doi.org/10.5194/acp-14-8917-2014
- 2015). Lightning channel length and flash energy determined from moments of the flash area distribution. Journal of Geophysical Research-Atmospheres, 120(17), 8925–8940. https://doi.org/10.1002/2015JD023766
- 2014). Continuous variability in thunderstorm primary electrification and an evaluation of inverted-polarity terminology. Atmospheric Research, 135-136, 274–284. https://doi.org/10.1016/j.atmosres.2012.10.009
- 1998). Large-scale nitrogen oxide plumes in the tropopause region and implications for ozone. Science, 282(5392), 1305–1309. https://doi.org/10.1126/science.282.5392.1305
- 2019). Midlatitude lightning NOx production efficiency inferred from OMI and WWLLN data. Journal of Geophysical Research-Atmospheres, 124(23), 13475–13497. https://doi.org/10.1029/2019JD030561
- 2010). Lightning-generated NOx seen by the Ozone Monitoring Instrument during NASA's Tropical Composition, Cloud and Climate Coupling Experiment (TC4). Journal of Atmospheric Research: Atmospheres, 115(D10), D00J10. https://doi.org/10.1029/2009JD013118
- 2015). Chemical kinetics and photochemical data for use in atmospheric studies: Evaluation number 18. Pasadena, CA: Jet Propulsion Laboratory, National Aeronautics and Space Administration.
- 2016). The kinematic and microphysical control of lightning rate, extent, and NOx production. Journal of Geophysical Research-Atmospheres, 121(13), 7975–7989. https://doi.org/10.1002/2015JD024703
- 2009). On the NOx production by laboratory electrical discharges and lightning. Journal of Atmospheric and Solar-Terrestrial Physics, 71(17-18), 1877–1889. https://doi.org/10.1016/j.jastp.2009.07.009
- 2013). Cloud-resolving chemistry simulation of a Hector thunderstorm. Atmospheric Chemistry and Physics, 13(5), 2757–2777. https://doi.org/10.5194/acp-13-2757-2013
- 1998). The U.S. National Lightning Detection Network and applications of cloud-to-ground lightning data by electric power utilities. IEEE Transactions on Electromagnetic Compatibility, 40(4), 465–480. https://doi.org/10.1109/15.736207
- 2005). Lightning-generated NOx and its impact on tropospheric ozone production: A three-dimensional modeling study of a Stratosphere-Troposphere Experiment: Radiation, Aerosols, and Ozone (STERAO-A) thunderstorm. Journal of Geophysical Research-Atmospheres, 110(D14), D14303. https://doi.org/10.1029/2004JD005556
- 2000). A cloud-scale model study of lightning-generated NOx in an individual thunderstorm during STERAO-A. Journal of Geophysical Research-Atmospheres, 105(D9), 11,601–11,616. https://doi.org/10.1029/2000JD900033
- 2011). The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quarterly Journal of the Royal Meteorological Society, 137(656), 553–597. https://doi.org/10.1002/qj.828
- 2019). Polyfitn, https://www.mathworks.com/matlabcentral/fileexchange/34765-polyfitn, MATLAB Central File Exchange. Access date: December 29, 2019.
- 2016). Response of lightning NOx emissions and ozone production to climate change: Insights from the Atmospheric Chemistry and Climate Model Intercomparison Project. Geophysical Research Letters, 43(10), 5492–5500. https://doi.org/10.1002/2016GL068825
- 1996). Could cloud-to-cloud discharges be as effective as cloud-to-ground discharges in producing NOx? Tellus Series B: Chemical and Physical Meteorology, 48(5), 641–651. https://doi.org/10.1034/j.1600-0889.1996.t01-4-00003.x
- 2019). Comparison of six lightning parameterizations in CAM5 and the impact on global atmospheric chemistry. Earth and Space Science, 6(12), 2317–2346. https://doi.org/10.1029/2019EA000873
- 2016b). On the origin of pronounced O3 gradients in the thunderstorm outflow region during DC3. Journal of Geophysical Research-Atmospheres, 121(11), 6600–6637. https://doi.org/10.1002/2015JD024279
- 2016a). Injection of lightning-produced NOx, water vapor, wildfire emissions, and stratospheric air to the UT/LS as observed from DC3 measurements. Journal of Geophysical Research-Atmospheres, 121(11), 6638–6668. https://doi.org/10.1002/2015JD024273
- 2009). NOx production by lightning in Hector: First airborne measurements during SCOUT-O3/ACTIVE. Atmospheric Chemistry and Physics, 9(21), 8377–8412. https://doi.org/10.5194/acp-9-8377-2009
- 2011). Mesoscale convective systems observed during AMMA and their impact on the NOx and O3 budget over West Africa. Atmospheric Chemistry and Physics, 11(6), 2503–2536. https://doi.org/10.5194/acp-11-2503-2011
- 2007). Lightning-produced NOx over Brazil during TROCCINOX: Airborne measurements in tropical and subtropical thunderstorms and the importance of mesoscale convective systems. Atmospheric Chemistry and Physics, European Geosciences Union, 7(11), 2987–3013.
- 2008). Lightning activity in Brazilian thunderstorms during TROCCINOX: Implications for NOx production. Atmospheric Chemistry and Physics, 8(4), 921–953. https://doi.org/10.5194/acp-8-921-2008
- 2001). The global distribution of lightning NOx simulated on-line in a general circulation model. Physics and Chemistry of the Earth, Part C: Solar, Terrestrial & Planetary Science, 26(8), 585–591. https://doi.org/10.1016/S1464-1917(01)00051-4
- 2010). Lightning NOx emissions over the USA constrained by TES ozone observations and the GEOS-Chem model. Atmospheric Chemistry and Physics, 10(1), 107–119. https://doi.org/10.5194/acp-10-107-2010
- 2014). The NASA Lightning Nitrogen Oxides Model (LNOM): Application to air quality modeling. Atmospheric Research, 135-136, 363–369. https://doi.org/10.1016/j.atmosres.2012.12.015
- 2010). Optical characteristics of OTD flashes and the implications for flash-type discrimination. Journal of Atmospheric and Oceanic Technology, 27(11), 1822–1838. https://doi.org/10.1175/2010JTECHA1405.1
- 2014). Global lightning nitrogen oxides production. In V. Cooray (Ed.), The lightning flash ( 2nd ed., chap. 19, p. 928). London: Institution of Engineering and Technology.
- 2015). Variability of CONUS lightning in 2003–12 and associated impacts. Journal of Applied Meteorology and Climatology, 54(1), 15–41. https://doi.org/10.1175/JAMC-D-14-0072.1
- 2015). A return stroke NOx production model. Journal of the Atmospheric Sciences, 72(2), 943–954. https://doi.org/10.1175/JAS-D-14-0121.1
- 2017). Ozone production by corona discharges during a convective event in DISCOVER-AQ Houston. Atmospheric Environment, 161, 13–17. https://doi.org/10.1016/j.atmosenv.2017.04.018
- 2017). The Version 3 OMI NO2 standard product. Atmospheric Measurement Techniques, 10(9), 3133–3149. https://doi.org/10.5194/amt-10-3133-2017
- 2004). Strong sensitivity of the global mean OH concentration and the tropospheric oxidizing efficiency to the source of NOx from lightning. Geophysical Research Letters, 31(6). https://doi.org/10.1029/2003GL019229
- 2017). Quantification of the effect of modeled lightning NO2 on UV-visible air mass factors. Atmospheric Measurement Techniques, 10(11), 4403–4419. https://doi.org/10.5194/amt-10-4403-2017
- 2018). The Berkeley High Resolution tropospheric NO2 product. Earth System Science Data, 10(4), 2069–2095. https://doi.org/10.5194/essd-10-2069-2018
- 2019). Evaluation of Version 3.0B of the BEHR OMI NO2 product. Atmospheric Measurement Techniques, 12(1), 129–146. https://doi.org/10.5194/amt-12-129-2019
- 1995). Lightning and atmospheric chemistry: The rate of atmospheric NO production. In Handbook of atmospheric electrodynamics, (Vol. 1, pp. 189–202). Boca Raton, FL: CRC Press.
- 1961). Photochemistry of air pollution. San Diego, CA: Academic Press.
- 2015). Sensitivity of tropical tropospheric composition to lightning NOx production as determined by replay simulations with GEOS-5. Journal of Geophysical Research-Atmospheres, 120(16), 8512–8534. https://doi.org/10.1002/2014JD022987
- 2011). The application of total lightning detection and cell tracking for severe weather prediction. In 91st American Meteorological Society Annual Meeting, (pp. 1-10). Seattle, WA.
- 2002). Simulated three-dimensional branched lightning in a numerical thunderstorm model. Journal of Geophysical Research-Atmospheres, 107(D9), ACL 2-1–ACL 2-12. https://doi.org/10.1029/2000JD000244
- 2010). Simulated electrification of a small thunderstorm with two-moment bulk microphysics. Journal of the Atmospheric Sciences, 67(1), 171–194. https://doi.org/10.1175/2009JAS296
- 2018). Nitrogen oxides in the global upper troposphere: Interpreting cloud-sliced NO2 observations from the OMI satellite instrument. Atmospheric Chemistry and Physics, 18(23), 17017–17027. https://doi.org/10.5194/acp-18-17017-2018
- 2011). The impact of North American anthropogenic emissions and lightning on long-range transport of trace gases and their export from the continent during summers 2002 and 2004. Journal of Geophysical Research-Atmospheres, 116(D7), D07305. https://doi.org/10.1029/2010JD014305
- 2017). The intracloud lightning fraction in the contiguous United States. Monthly Weather Review, 145(11), 4481–4499. https://doi.org/10.1175/MWR-D-16-0426.1
- 2008). Observation of enhanced ozone in an electrically active storm over Socorro, NM: Implications for ozone production from corona discharges. Journal of Geophysical Research-Atmospheres, 113(D17), D17208. http://doi.org/10.1029/2007JD009500
- 2014). Global lightning NOx production estimated by an assimilation of multiple satellite data sets. Atmospheric Chemistry and Physics, 14(7), 3277–3305. https://doi.org/10.5194/acp-14-3277-2014
- 2014). Enhanced cloud lightning performance of the US National Lightning Detection Network following the 2013 upgrade. Paper presented at 23rd International Lightning Detection Conference & 5th International Lightning Meteorology Conference, Tucson, Arizona.
- 2016). Lightning NOx and impacts on air quality. Current Pollution Reports, 2(2), 115–133. https://doi.org/10.1007/s40726-016-0038-0
- 2013). Interannual variability in tropical tropospheric ozone and OH: The role of lightning. Journal of Geophysical Research-Atmospheres, 118(19), 11,468–11,480. https://doi.org/10.1002/jgrd.50857
- 2014). Factors controlling variability in the oxidative capacity of the troposphere since the Last Glacial Maximum. Atmospheric Chemistry and Physics, 14(7), 3589–3622. https://doi.org/10.5194/acp-14-3589-2014
- 2017). Negative first stroke leader characteristics in cloud-to-ground lightning over land and ocean. Geophysical Research Letters, 44(4), 1973–1980. https://doi.org/10.1002/2016GL072270
- 2015). Measurements of CH3O2NO2 in the upper troposphere. Atmospheric Measurement Techniques, 8(2), 987–997. https://doi.org/10.5194/amt-8-987-2015
- 2017). Lightning NOx emissions: Reconciling measured and modeled estimates with updated NOx chemistry. Geophysical Research Letters, 44(18), 9479–9488. https://doi.org/10.1002/2017GL074436
- 2010). Production of lightning NOx and its vertical distribution calculated from three-dimensional cloud-scale chemical transport model simulations. Journal of Geophysical Research-Atmospheres, 115(D4), D04301. https://doi.org/10.1029/2009JD011880
- 2014). Thunderstorms enhance tropospheric ozone by wrapping and shedding stratospheric air. Geophysical Research Letters, 41(22), 7785–7790. https://doi.org/10.1002/2014GL061921
- 2016). Estimates of lightning NOx production based on OMI NO2 observations over the Gulf of Mexico. Journal of Geophysical Research-Atmospheres, 121(14), 8668–8691. https://doi.org/10.1002/2015JD024179
- 1990). Model calculations of tropospheric ozone production potential following observed convective events. Journal of Geophysical Research-Atmospheres, 95(D9), 14049–14062. https://doi.org/10.1029/JD095iD09p14049
- 2016). Airborne quantification of upper tropospheric NOx production from lightning in deep convective storms over the United States Great Plains. Journal of Geophysical Research-Atmospheres, 121(4), 2002–2028. https://doi.org/10.1002/2015JD023941
- 1997). NOx from lightning: 2. Constraints from the global atmospheric electric circuit. Journal of Geophysical Research-Atmospheres, 102(D5), 5943–5951. https://doi.org/10.1029/96JD02551
- 2013). Electromagnetic methods of lightning detection. Surveys in Geophysics, 34(6), 731–753. https://doi.org/10.1007/s10712-013-9251-1
- 2003). Lightning: Physics and effects. New York: Cambridge University Press.
- 2014a). On the dynamics of hot air plasmas related to lightning discharges: 1. Gas dynamics. Journal of Geophysical Research-Atmospheres, 119(15), 9196–9217. https://doi.org/10.1002/2013JD020068
- 2014b). On the dynamics of hot air plasmas related to lightning discharges: 2. Electrodynamics. Journal of Geophysical Research-Atmospheres, 119(15), 9218–9235. https://doi.org/10.1002/2013JD020068
- 2015). Evaluating ENTLN performance relative to TRMM/LIS. Journal of Operational Meteorology, 3(2), 11–20. https://doi.org/10.15191/nwajom.2015.0302
- 2013). Evaluating WWLLN performance relative to TRMM/LIS. Geophysical Research Letters, 40(10), 2344–2348. https://doi.org/10.1002/grl.50428
- 2011). A high spatial resolution retrieval of NO2 column densities from OMI: Method and evaluation. Atmospheric Chemistry and Physics, 11(16), 8543–8554. http://doi.org/10.5194/acp-11-8543-2011
- 2005). Inverted-polarity electrical structures in thunderstorms in the Severe Thunderstorm Electrification and Precipitation Study (STEPS). Atmospheric Research, 76(1-4), 247–271. https://doi.org/10.1016/j.atmosres.2004.11.029
- 2007). Remote sensed and in situ constraints on processes affecting tropical tropospheric ozone. Atmospheric Chemistry and Physics, 7(3), 815–838. https://doi.org/10.5194/acp-7-815-2007
- 2007). The global lightning-induced nitrogen oxides source. Atmospheric Chemistry and Physics, 7(14), 3823–3907. https://doi.org/10.5194/acp-7-3823-2007
- 2018). Observed NO/NO2 ratios in the upper troposphere imply errors in NO-NO2-O3 cycling kinetics or an unaccounted NOx reservoir. Geophysical Research Letters, 45(9), 4466–4474. https://doi.org/10.1029/2018GL077728
- 1998). Electrical structure in thunderstorm convective regions: 1. Mesoscale convective systems. Journal of Geophysical Research-Atmospheres, 103(D12), 14,059–14,078. https://doi.org/10.1029/97JD03546
- 2000). Comparison of ground-based 3-dimensional lightning mapping observations with satellite-based LIS observations in Oklahoma. Geophysical Research Letters, 27(12), 1703–1706. https://doi.org/10.1029/1999GL010845
- 2004). Accuracy of the lightning mapping array. Journal of Geophysical Research-Atmospheres, 109(D14), D14207. https://doi.org/10.1029/2004JD004549
- 2014). A comparison of two ground-based lightning detection networks against the satellite-based Lightning Imaging Sensor (LIS). Journal of Atmospheric and Oceanic Technology, 31(10), 2191–2205. https://doi.org/10.1175/JTECH-D-13-00186.1
- 2017). Chemistry-climate interactions of aerosol nitrate from lightning. Atmospheric Chemistry and Physics, 17(2), 1125–1142. https://doi.org/10.5194/acp-17-1125-2017
- 2016). Why do models overestimate surface ozone in the Southeast United States? Atmospheric Chemistry and Physics, 16(21), 13,561–13,577. https://doi.org/10.5194/acp-16-13561-2016
- 2008). Evaluation of the OMI cloud pressures derived from rotational Raman scattering by comparisons with other satellite data and radiative transfer simulations. Journal of Geophysical Research: Atmospheres, 113, D15S19. https://doi.org/10.1029/2007JD008689
- 1998). Nitric oxide production by simulated lightning: Dependence on current, energy, and pressure. Journal of Geophysical Research-Atmospheres, 103(D15), 19,149–19,159. https://doi.org/10.1029/98JD01356
- 2003). Impacts of anthropogenic and natural NOx sources over the U.S. on tropospheric chemistry. Proceedings of the National Academy of Sciences of the United States, 100(4), 1505–1509. https://doi.org/10.107/pnas.252763799, https://doi.org/10.1073/pnas.252763799
- 2019). Lightning NO2 simulation over the contiguous U.S. and its effects on satellite NO2 retrievals. Atmospheric Chemistry and Physics. https://doi.org/10.5194/acp-2019-216 in review
- 2016). A study of National Lightning Detection Network responses to natural lightning based on ground truth data acquired at LOG with emphasis on cloud discharge activity. Journal of Geophysical Research-Atmospheres, 121(24), 14,651–14,660. https://doi.org/10.1002/2016JD025574
- 2017). Evaluation of ENTLN performance characteristics based on the based on the ground truth natural and rocket-triggered lightning data acquired in Florida. Journal of Geophysical Research-Atmospheres, 122(18), 9858–9866. https://doi.org/10.1002/2017JD027270