Introduction

Due to its adverse impact on human health1 and ecosystem well being,2 tropospheric ozone (O3) has been at the center of air quality research during the past decades.3,4,5 Ground-level O3 is either produced through complex photochemical reactions involving nitrogen oxides (NOx) and volatile organic compounds (VOCs),6 with both anthropogenic and natural origins, or transported from other locations.7 As air quality regulations are tightened and the contribution of anthropogenic sources to tropospheric O3 levels is reduced in US and other economically-developed regions, the natural sources becomes more increasingly important.8,9,10,11,12

Lightning generates relatively large but uncertain quantities of NOx and thus strongly impacts the composition of trace gases in the troposphere.13,14,15 Globally, lightning is estimated to generate 2–8 Tg N per year16 which is considerably smaller than present-day global contributions from anthropogenic (~20.5 Tg N per year) and biomass burning (~5.5 Tg N per year) sources.17 Despite the strong influence on troposphere burdens of O3 and OH, impacts of lightning NOx emissions on the surface-level air composition and chemistry have been suggested by past studies to be generally small, and thus except for some applications with global models, regional air quality assessments have historically not included lightning NOx emissions.16 Uncertainties in lightning NOx emissions are influenced by limitations in the description of space and time variability in lightning frequency and intensity, NOx production rates from lightning flashes, and the vertical distribution and transport of lightning NOx after its production. Since lightning activity is often associated with deep convection in the atmosphere, the lightning flash rate can be parameterized using convection schemes.18,19 However, convection schemes in meteorological models generally show low skill in producing the convective precipitation and the lightning distribution.20,21 Though many studies report enhancement in surface-level NOx levels following thunderstorm activity, the impacts on surface-level O3 are ambiguous and typically complicated by the local photochemical conditions such as the availability of other precursors and sunlight.22,23,24,25 Regional modeling calculations have shown up to 10 ppb enhancements in simulated instantaneous O3 mixing ratios8,13,23,25 due to addition of lightning NOx.

Heath et al.26 demonstrated that assimilation of high-quality lightning data from the national lightning detection network (NLDN) in the weather research and forecast (WRF) model considerably improved the simulation of summertime convective activity and rainfall over the continental United States. In this study, we build upon this prior work26,27 by implementing lightning NOx production in the Community Multiscale Air Quality (CMAQ)28,29,30,31 model based on the NLDN lightning flash data32,33 and perform CMAQ simulations driven by the lightning-assimilated WRF outputs with (hereafter denoted NLDN) and without (denoted Base) the NLDN-lightning NOx production.

Results

The model performance for O3 predictions, in terms of a particular implementation of lightning NOx production, depends on correct NOx emissions from other sources in the Base model. Uncertainties in specification of precise activity patterns of mobile and industrial sources lead to uncertainties in space and time variability in estimated emissions from these sources that then translate to biases and errors in model predicted O3 mixing ratios. The addition of lightning NOx emissions can then result in further deterioration of model performance for predicted O3 at many locations such as in the eastern U.S. as illustrated in Fig. 1a, which presents the change in bias in simulated surface daily maximum 8 h (DM8HR) O3 between the NLDN and Base simulations. However, significant impact of lightning NOx emissions on improving skill of model predictions of surface O3 in the Mountain West States (MWS: Arizona, New Mexico, Nevada, Utah, Colorado, and Wyoming) is illustrated in Fig. 1a, and further emphasized in Fig. 1b. During the summer of 2011 in the MWS region there were over 800 measured DM8HR O3 exceedances above 70 ppb as summarized in Table 1. Even though both model simulations with and without lightning NOx underestimated the observed values, the simulation incorporating impacts of lightning NOx reduced the mean bias across the measurement locations (Fig. 1b) significantly during periods with lightning activity (depicted in Fig. 1c). Significant impacts of lightning NOx on day to day variations in domain-mean surface DM8HR O3 over the MWS region is also noted in Fig. 1d, which illustrates that lightning NOx emissions increase region-average DM8HR O3 by up to 6 ppb and helps reduce the model low-bias. Recent modeling analysis34 in the southeast U.S. has speculated that NOx motor vehicle emissions in U.S. National Emission Inventory (NEI) may be overestimated. However, the role of uncertainties associated with other modeled processes as well as the inventory base year used in the analysis, on the magnitude of this overestimation, are not readily apparent (e.g., Simon et al.10). The MWS region which is the focus of our analysis is a sparsely populated area and consequently anthropogenic emissions in the region are relatively small. Thus, the relative impacts of uncertainties in anthropogenic NOx on the total NOx burden in the region is also comparatively smaller relative to other regions in the U.S. In recent complementary analyses,31 we have comprehensively compared model predicted NOx mixing ratios with observations for different regions, which also indicated that the model has the best performance over MWS for NOx predictions. Thus, the impact of uncertainties in characterizing NOx emissions from other sectors on the inferred impacts of lightning NOx emissions in the MWS region are expected to be negligible.

Fig. 1: Impact of lightning events on O3 production during summer 2011.
figure 1

a Mean bias reduction over continental US between simulations with and without lightning NOx. b Same as a but focus on the MWS. c The average daily lightning flashes during July 2011. d Time series of observed and modeled DM8HR O3 mixing ratios. e Time series of the mean bias difference between the two simulations and the daily mean lightning flashes.

Full size image
Table 1 Summary of the observed and modeled DM8HR O3 mixing ratios in MWS.
Full size table

To further delineate the impact of lightning activity on daily variations in surface O3, we examined the correlation between lightning activity and the change in model DM8HR O3 bias (Fig. 1e) (see the Supplementary Information (SI) and Supplementary Fig. 5 for detailed calculation of the lightning flashes; here the grid cell increments of ten are used). The time series of DM8HR O3 mixing ratios (Fig. 1d) and mean bias difference (Fig. 1e) clearly indicate that the model simulation with lightning NOx (NLDN) performed much better episodically in matching the observed values than the simulation without lightning NOx (Base). Days with highest lightning activity also corresponded to days when the largest improvements in DM8HR O3 mean bias were realized through the incorporation of lightning NOx emissions, strongly suggestive of the large role of lightning NOx emissions in modulating surface-level O3 mixing ratios in this region. The comparison of the daily average mean bias in DM8HR O3 over the AQS sites in the MWS for both the Base and the NLDN cases presented in Supplementary Fig. 3 further illustrates the improvement in model skill in capturing variations in surface DM8HR O3 in the region due to inclusion of lightning NOx emissions.

The relative importance of lightning as a source of reactive nitrogen in the troposphere over the MWS region is further highlighted in Fig. 2 which compares the strength of this source with other sources traditionally considered in estimating O3 in the region. Figure 2a–c show the spatial distributions of NOx emissions from anthropogenic, soil, and lightning sources over the MWS region for July 2011. Figure 2d presents the relative contributions of these three source categories to daily NOx emissions, while Fig. 2e quantifies the daily fraction of total NOx emissions that can be attributed to lightning. Note, for Fig. 2d, e, we limit the analysis to portions of the MWS region east of the 115°W longitude to minimize the influence of urban centers. As indicated in Fig. 2e, the lightning produced NOx constitutes 15–50% of daily NOx emissions in these remote regions and on average ~30% during July 2011, resulting in the significant impact on surface O3 illustrated earlier. The release of lightning NOx in such magnitudes into the atmosphere in this region could also potentially modulate atmospheric nitrogen deposition amounts in the region, and further emphasizes the need for air quality models to include lightning NOx emissions to accurately simulate both atmospheric chemistry and deposition in the MWS region.

Fig. 2: NOX emissions from different sources in time and space during July, 2011.
figure 2

a Anthropogenic NOX emissions. b Soil NOX emissions. c Lightning NOX emissions. d Daily total emissions excluding those from the area west of −115° longitude. e The daily lightning NOx ratio to the total NOX emissions.

Full size image

The significant impacts of lightning on air quality in the MWS region are however not unique to the summer of 2011. To leverage observations from the 2014 DISCOVER-AQ (Deriving Information on Surface Conditions from COlumn and VERtically Resolved Observations Relevant to Air Quality) field campaign, as a case study, WRF-CMAQ model simulations with (NLDN) and without (Base) lightning NOx emissions were performed for a period encompassing the field study and model results were compared with measurements at the Boulder Atmospheric Observatory (BAO) site. Maximum surface-level O3 mixing ratios were observed on July 28th, and the evolution of vertical O3 structure inferred from the lidar measurements during 26–29 July 2014 is shown in Fig. 3a. In these illustrations, the observed lidar O3 profiles are aggregated to the model vertical grid and temporally averaged to hourly values. Figure 3b shows the corresponding Base model (without lightning NOx) simulated values and Fig. 3c shows the corresponding simulated values when lightning NOx is included. Significant discrepancies are noted between the simulated and observed evolution of vertical O3 distributions on these days, with the model generally underestimating the development of high O3. Also, noted in the observed lidar profiles on most days, is the existence of localized high O3 aloft, suggestive of in situ production rather than intrusion of stratospheric air from higher altitudes. Comparison of Fig. 3a, b, c indicates that the simulated evolution of vertical O3 distribution and values with lightning NOx (Fig. 3c) matched the observations more closely than the case without lightning emissions. In general, the Base model shows overestimation on July 26th and underestimation on July 28th and 29th, while the simulation with lightning NOx improved the predicted O3 values for both these cases, i.e., decreased values on July 26th and increased values on July 28th and 29th. The difference between the two model cases is illustrated in Fig. 3d. The evolution of peak O3 over this period is further depicted in Fig. 4a, b that compare model and observed vertical profiles of daily maximum O3 mixing ratios for both model cases. From July 26 to 29, the observed peak values increased by over 20 ppb at the surface (from below 70 ppb to above 90 ppb), while the simulated values without lightning NOx decreased by 10 ppb (from 80 to 70 ppb); the simulations changed from more than 10 ppb over-prediction on July 26th to more than 20 ppb under-prediction on July 29th. However, when lightning NOx is included (Fig. 4b), the simulated surface daily maximum O3 mixing ratios are much closer to the observed values (compared to the Base case, the simulated O3 values decreased from 87 to 72 ppb on July 26th and increased from 76 to 89 ppb on July 29th, respectively). Lightning flash data for this period (see Supplementary Fig. 10) indicate periods of active lightning strikes in the region, especially on July 28th and 29th, further suggesting the contribution of lightning NOx to the observed aloft O3 and subsequent contribution to surface enhancements. These changes in model simulated O3 mixing ratios during this period in response to the injection of lightning NOx emissions further indicate the non-linear relationship between O3 and its precursors. Depending on the location and timing of its injection, lightning NOx emissions can either produce or destroy O3. In proximity of the lightning location, O3 is chemically consumed due to titration by the freshly emitted NO. In contrast, when NOx from lightning is transported to locations downwind of the lightning flash, in the presence of sunlight it reacts with VOCs in the surrounding air to produce O3. Note that the uniform distribution of simulated O3 values under about 2 km shown in Fig. 4 and Supplementary Fig. 11 is due to the fact that maximum O3 values typically occur during early afternoon hours when the planetary boundary layer (PBL) is well mixed.

Fig. 3: Lidar measured and simulated O3 mixing ratios at BAO site.
figure 3

a Lidar measurements, b base simulations, c simulations with NLDN, d simulation difference (NLDN – Base). The legend on the left is for a–c and the legend at the right bottom is for d.

Full size image
Fig. 4: Lidar measured and simulated O3 mixing ratios at BAO site.
figure 4

a Maximum O3 mixing ratios across the layers on each day over the episode (Lidar and Base), and b The same as a, but for Lidar and NLDN. Note that solid lines represent lidar measurements on different days while the circles are the model values.

Full size image

The importance of lightning NOx as a modulator of tropospheric chemistry in the MWS region is further illustrated in Fig. 5 which presents the observed daily mean lightning flashes across the MWS region for the month of July for the years 2011 through 2016 (similar patterns are also observed for August as illustrated in the SI). Figure 6 shows the estimated NOx emissions released due to these lightning flash rates using the NOx production algorithm implemented in CMAQ.30,31 Similar intensity of lightning activity occurs every summer (Fig. 5) resulting in comparable amounts of lightning NOx emissions released into the atmosphere in the MWS region suggesting that summertime O3 levels in the region could be significantly modulated by this natural source.

Fig. 5
figure 5

Daily mean lightning flashes over MWS during July from 2011 to 2016.

Full size image
Fig. 6
figure 6

NOx emissions from lightning in July from 2011 to 2016 over MWS.

Full size image

Discussion

With the most up-to-date lightning NOx production schemes and the vertical distribution algorithm in CMAQ as described and evaluated in recent publications,30,31 we examined the importance of lightning NOx as a modulator of tropospheric chemistry in the U.S. Mountain West States. During summer, lightning NOx emissions provide a continuous NOx input to the airshed over the remote MWS region with average monthly contributions to the total NOx budget of 30%. As a result, significant impact on O3 air quality is both observed and confirmed through analysis of detailed model simulations with and without lightning NOx emissions and their detailed comparisons with available routine network and specialized field campaign observations of surface and aloft O3 in the region.

As revealed by the detailed model calculations for summer 2011, lightning NOx can contribute up to 21 ppb (or 43%) of the hourly values of surface-level O3 in the region. Given the comparable level of lightning activity in other years, similar levels of contribution of lightning to surface O3 across the region can be expected. More importantly, as technological advances and regulatory measures reduce anthropogenic NOx emissions in the region, the relative contributions of lightning NOx to O3 levels in the region may increase. Consequently, accurate quantification of lightning NOx emissions, its contribution to background O3 levels, and its role in modulating air quality in the region will become increasingly important.

Methods

CMAQ simulations are performed for the April–September 2011 period over a domain encompassing the contiguous United States and discretized with 12 km horizontal grid spacing. The Base model configuration and simulation details are the same as those detailed in Kang et al.31 with the CB6 chemical mechanism and emissions input data from the 2011 NEI for 2011 simulations. For simulations during the year 2014, the emissions input data are still based on 2011 NEI with 2014 year-specific emissions for EGU (electric generating units), fire, and biogenic sectors from the three states (Colorado, Utah, and Wyoming) constituting the MWS study region. NOx emissions from electric generating units are based on Continuous Emissions Monitoring and thus are relatively well constrained. The NEI utilizes the MOVES (https://www.epa.gov/moves) modeling system which incorporates detailed information of motor vehicle fleets, miles traveled, fuel information, and the latest emission factors to estimate emissions from this sector. Emissions from wildfires are based on year-specific daily fire-activity from the Hazard Mapping System fire detections and the SMARTFIRE system (http://www.getbluesky.org/smartfire/docs/Raffuse 2007.pdf). The lightning-induced NOx is calculated based on the NLDN observed cloud-to-ground (CG) lightning flashes that are gridded to the model grid as hourly total flashes, and intra-cloud (IC) flashes based on the climatological IC-to-CG ratio. The vertical distribution of lightning NOx release follows Kang et al.30 The lightning NOx production rate over land-based grid cells is set to 350 moles/stroke (this value is about the middle of the range suggested by recent studies13 that have explicitly reported lightning-NOx yields using observational constraints). To evaluate the performance of the two model cases for 2011, we compare simulated hourly and DM8HR O3 mixing ratios with corresponding measured values at sites from the air quality system network. To further examine the impact of lightning NOx emissions on atmospheric composition and chemistry we conducted model simulations for the period of the 2014 DISCOVER-AQ field campaign in Colorado and analyzed model time-height O3 distributions in conjunction with those inferred from lidar measurements taken by the National Oceanic and Atmospheric Administration’s Earth System Research Laboratory35,36 at BAO.