Summertime impact of convective transport and lightning NOx production over North America: modeling dependence on meteorological simulations

Global-scale chemical transport model simulations indicate lightning NOx dominates upper tropospheric O3 production above Eastern North America during summertime but vary in their estimates. To improve our understanding, a regional-scale model (REAM) with higher resolution is applied. To examine the uncertainties in modeling the impact of convective transport and lightning NO x production on upper tropospheric chemical tracer distributions, REAM simulations of chemical tracers are driven by two meteorological models, WRF and MM5, with different cumulus convective parameterizations. The model simulations are evaluated using INTEX-A aircraft measurements and satellite measurements of NO 2 columns and cloud top pressure, and we find that mid and upper tropospheric trace gas concentrations are affected strongly by convection and lightning NOx production. WRF with the KF-eta convection scheme simulates larger convective updraft mass fluxes below 150 hPa than MM5 with the Grell scheme. The inclusion of the entrainment and detrainment processes leads to more outflow in the mid troposphere in WRF than MM5. The ratio of C2H6/C3H8 is found to be a sensitive parameter to convective outflow; the simulation by WRF-REAM is in closer agreement with INTEX-A measurements than MM5-REAM, implying that convective mass fluxes by WRF are more realistic. WRF also simulates lower cloud top heights (10– 12 km) than MM5 (up to 16 km), and hence smaller amounts of estimated (intra-cloud) lightning NO x and lower emission altitudes. WRF simulated cloud top heights are in better agreement with GOES satellite measurements than MM5. Simulated lightning NOx production difference (due primarCorrespondence to: C. Zhao (chun.zhao@eas.gatech.edu) ily to cloud top height difference) is mostly above 12 km. At 8–12 km, the models simulate a contribution of 60–75% of NOx and up to 20 ppbv of O3 from lightning, although the decrease of lightning NO x effect from the Southeast to Northeast and eastern Canada is overestimated. The model differences and biases found in this study reflect some major uncertainties of upper tropospheric NO x and O3 simulations driven by those in meteorological simulations and lightning parameterizations.


Introduction
Tropospheric distributions of trace gases are driven in part by meteorological conditions.Convection and associated lightning NO x production are two important meteorological processes affecting the production and distribution of tropospheric chemical tracers (e.g., Wang et al. 2001;Doherty et al. 2005;Hudman et al., 2007;Choi et al., 2005Choi et al., , 2008a)).Convection redistributes trace gases vertically and significantly affects atmospheric chemical and transport processes during long-range transport (e.g., Wang et al., 2000Wang et al., , 2001;;Doherty et al., 2005;Hess, 2005;Folkins et al., 2006;Kiley et al., 2006;Hudman et al., 2007).Li et al. (2005) and Choi et al. (2008b) showed the importance of convection in ventilating air pollutants from the continental boundary layer of the United States (US) and providing a conduit for US pollution to the Western North Atlantic Ocean.
Simulations of convective transport have large uncertainties.Several studies found substantial divergences among Chemical Transport Model (CTM) simulations arising from the difference in various cumulus parameterizations and underlying meteorological fields (e.g., Prather and Jacob, 1997;Prather et al., 2001;Collins et al., 2002;Doherty et al., Published by Copernicus Publications on behalf of the European Geosciences Union. 2005).To properly evaluate model simulations of convective transport and lightning NO x production, extensive atmospheric measurements are needed.One such dataset is the Intercontinental Chemical Transport Experiment -North America (INTEX-A) collected during summer (3 July to 15 August) 2004 over North America (Singh, et al., 2006), in which a large number of cases for active convection and large amounts of lightning NO x production were measured (e.g., Hudman et al., 2007;Bertram et al., 2007).
Lightning is a major source of NO x (NO 2 +NO) in the upper troposphere.NO x is thought to be produced during the return stroke stage of a cloud-to-ground flash and the leader stage of an intra-cloud flash but there remains a great deal of uncertainty in the mechanism of NO x production in lightning flashes (e.g., Schumann and Huntrieser, 2007).The lightning flash rate is often parameterized as functions of meteorological variables such as convective updraft mass fluxes (UMF), convective available potential energy (CAPE), convective cloud top height, and precipitation rate (e.g., Price et al., 1993;Allen et al., 2000;Choi et al., 2005Choi et al., , 2008a)).Lightning NO x significantly enhances tropospheric NO 2 columns, in particular, over the ocean, where NO 2 columns are more sensitive to lightning NO x production due to less impact of surface NO x emissions (e.g., Choi et al. 2005Choi et al. , 2008a;;Martin et al., 2006;Bertram et al., 2007).Cooper et al. (2009) presented a summary of many related observational and modeling studies over the US and suggested that lightning contributes to more than 80% of summertime NO x in the upper troposphere in the region.It also increases the concentrations of O 3 and PAN in the free troposphere (e.g., Labrador et al., 2004;Cooper et al., 2006;Hudman et al., 2007).Hudman et al. (2007) found that lightning enhanced O 3 concentrations by 10-17 ppbv and PAN by 30% in the upper troposphere based on the INTEX-A measurements over Eastern North America and the Western North Atlantic Ocean during summer 2004 using the GEOS-CHEM model.Recent satellite measurements including NO 2 columns from the SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) were used to show lightning enhanced NO 2 over the North Atlantic Ocean, and to constrain the global lightning NO x emissions in the range of 4-8 Tg N/yr (Martin et al., 2006(Martin et al., , 2007)).
Previous analyses with global-scale models have shown difficulties in accurately quantifying the lightning-induced NO x and O 3 production in the upper troposphere above North America during summer, as summarized by Cooper et al. (2009), with more recent estimates by Pfister et al. (2008) and Hudman et al. (2009).Simulated convective transport of tracers and lightning NO x production are sensitive to underlying meteorological fields.To study the sensitivities in simulating their impact on trace gas simulations, we use a Regional chEmical trAnsport Model (REAM) with 70×70 km 2 resolution driven by two meteorological models with different convection schemes, the Weather Research and Forecasting (WRF) model (v3.0,Skamarock et al., 2005) with the KF-eta scheme (Kain, 2003) and the Fifth-Generation NCAR/Penn State Mesoscale Model (MM5) (v3.6.1,Grell et al., 1995) with the Grell scheme (Grell et al., 1993).When compared to the convective transport and lightning NO x features measured during INTEX-A, the model difference between WRF-REAM and MM-5 REAM is attributed to the underlying meteorological fields, particularly the convection related variables.
Our analysis proceeds as follows.In Sect.2, we describe the REAM model and the measurements used in the study.The convective impact on tropospheric tracers is analyzed in Sect.3. The lightning impact is examined in Sect. 4. Conclusions are given in Sect. 5.

Model description
The REAM model driven by MM5 assimilated meteorological fields (MM5-REAM) was described by Choi et al. (2008a).Previously, this model was applied to investigate a number of tropospheric chemistry and transport problems at northern mid latitudes (Choi et al., 2005(Choi et al., , 2008a, b;, b;Jing et al., 2006;Wang et al., 2006;Gillus et al., 2008) and in the polar regions (Zeng et al., 2003(Zeng et al., , 2006;;Wang et al., 2007).In this work, the REAM model is developed to use the WRF assimilated meteorological fields (WRF-REAM).Large changes are apparent in the free tropospheric chemical distributions when WRF fields are used in place of MM5.
The model has a horizontal resolution of 70 km with 23 vertical layers below 10 hPa.Meteorological fields are assimilated using either MM5 or WRF constrained by the NCEP reanalysis products (NNRP).The horizontal domain of MM5 or WRF has 5 extra grids beyond that of REAM on each side to minimize potential transport anomalies near the boundary.Most meteorological fields are archived every 30 min except those related to convective transport and lightning parameterizations (e.g., cloud top height, cloud base height, convective mass fluxes, and convection available potential energy CAPE), which are archived every 5 min.Chemical initial and boundary conditions for chemical tracers in REAM are obtained from the global simulation for the same period using the GEOS-CHEM model driven by GEOS-4 assimilated meteorological fields (Bey et al., 2001).Anthropogenic and biogenic emission algorithms and inventories are adapted from the GEOS-CHEM model (Choi et al., 2005(Choi et al., , 2008a Turquety et al. (2007).The default inventories in GEOS-Chem for ethane and propane are used since the ethane and propane emission ratios in the VISTAS inventory appear to be problematic, similar to the problem previously found by Wang et al. (1998).Sub-grid convective transport in WRF-REAM and MM5-REAM is developed to be consistent with the KF-eta and Grell schemes implemented in WRF and MM5, respectively.The KF-eta scheme in WRF is developed based on the KF scheme (Kain and Fritsch, 1993).It utilizes a simple cloud model with moist updrafts and downdrafts, including the effects of detrainment and entrainment.Shallow convection is allowed for any updraft that does not reach minimum cloud depth for precipitating clouds; this minimum depth varies as a function of cloud-base temperature (Kain, 2003).The Grell scheme in MM5 is based on the rate of destabilization or quasi-equilibrium, a simple single-cloud scheme with updraft and downdraft fluxes and compensating motion determining the heating/moistening profile (Grell et al., 1993).
A newer and apparently quite different version of the Grell scheme (Grell and Devenyi, 2002) is available in the WRF model.Hence, the results shown in this study do not apply to the Grell scheme in WRF.We did not use the newer Grell (deep convection) scheme in WRF because there is no shallow convection scheme that can be paired with the Grell scheme in WRF, which is not the case in MM5.Shallow convection can be quite effective in ventilating pollutants from the boundary layer (e.g., Choi et al., 2005).A second reason is that MM5 with the Grell scheme has been widely used in previous regional chemical and transport modeling studies.Both KF-eta and (MM5) Grell convective schemes simulate moist updrafts and downdrafts.One notable difference is that the KF-eta scheme includes cloud entrainment and detrainment during convection but the Grell scheme does not.This difference is reflected in simulating the North America outflow of the pollutants and evaluated with INTEX-A measurements in this study.
The cloud-to-ground lightning flash rate is parameterized as a function of convective mass fluxes and CAPE on the basis of the observed cloud-to-ground lightning flashes by the National Lightning Detection Network (NLDN) in summer 2004 as described by Choi et al. (2005).The parameterization ensures the dynamic consistency between simulated lightning NO x production and simualted convection events.The IC/CG flash ratio is calculated following Wang et al. (1998).It is assumed that IC and CG flashes have the same energy (Ott et al., 2003;Choi et al., 2005).Lightning NO x is distributed vertically following the mid-latitude profile by Pickering et al. (1998).We set a NO x production rate of 250 moles NO per flash in this study through trial and error analysis such that model simulations are consistent with in situ and satellite observations.This production rate happens to agree with the value suggested by Schumann and Huntrieser (2007).

Aircraft observations
The Intercontinental Chemical Transport Experiment -North America Phase A (INTEX-A) was aimed at understanding the transport and transformation of gases and aerosols on transcontinental and intercontinental scales above Eastern North America (Singh et al., 2006).In this study, the INTEX-A measurements of C 2 H 6 , C 3 H 8 , HNO 3 , NO, NO 2 , and O 3 from the NASA DC-8 aircraft are used.C 2 H 6 and C 3 H 8 were measured with 1 pptv detection limit and 2-10% nominal accuracy (Simpson et al., 2000).HNO 3 was measured with 5-10 pptv detection limit and 10-15% nominal accuracy (Talbot et al., 1999;Crounse et al., 2006).NO was measured with a precision of 50 pptv with 1-min time integration (Ren et al., 2008).NO 2 was measured with 1 pptv detection limit and 10% nominal accuracy (Thornton et al., 2000).O 3 was measured with 1 ppbv detection limit and 5% nominal accuracy (Avery et al., 2001).All instruments on the DC-8 are described in detail by Singh et al. (2006).One-minute merge data of NO, NO 2 , HNO 3 , and O 3 and original data of C 2 H 6 and C 3 H 8 from DC-8 from July 1 st to 14 August 2004 are used (http://www-air.larc.nasa.gov/cgi-bin/arcstat).Some compounds were measured by two different techniques such as HNO 3 .When both measurements are available, the average values are used.When we compare model simulations with measurements, the model output is sampled at the time and locations of in situ measurements.

Tropospheric NO 2 columns
The SCIAMACHY instrument onboard the ENVISAT satellite has a spatial resolution of 30×60 km 2 and a 6-day global coverage.Tropospheric columns of NO 2 retrieved from SCIAMACHY and its uncertainties are calculated by Martin et al. (2006).The retrieval uncertainties are due to spectral fitting, the spectral artifact from the diffuser plate, removal of the stratospheric column, and air mass factor calculation.Measurements with cloud fraction greater than 30% are excluded in order to reduce the impact of clouds on the satellite retrievals.A more detailed description regarding the tropospheric NO 2 columns from SCIAMACHY and its validation with INTEX-A measurements can be found in Martin et al. (2006).

Cloud top pressure
The operational data collection phase of the International Satellite Cloud Climatology Project (ISCCP) began in July weather satellites (Rossow and Schiffer, 1991).The measurements of cloud top pressure over North America provided by the ISCCP DX dataset with 3-hourly 30 km sampled pixels processed from the images of GOES-10 and GOES-12 satellites are used to evaluate the model simulated cumulus cloud top heights.The measurements with the cloud top pressure larger than 500 hPa are excluded in the study to filter out the low cloud information.

Dependence of convective transport on cumulus parameterization
Figure 1a shows the spatial distributions of the mean updraft mass fluxes of deep convection at three pressure levels (800, 500, and 300 hPa) from WRF and MM5 simulations with KF-eta and Grell convection schemes, respectively, for July and August 2004.WRF and MM5 simulate generally similar spatial distributions of mass fluxes with strong convection events over the Western and Southeastern US, Mexico, and the Western North Atlantic Ocean.One clear difference is that the updraft fluxes at 500 and 800 hPa are much higher in WRF than MM5.The mass fluxes in WRF are not as spatially concentrated over the Western-Central US and are greater over the Southeast than MM5. Figure 1b shows the vertical profiles of mass fluxes from the two models including the entrainment and detrainment fluxes only from WRF averaged over North America (domain shown in Fig. 1a).

Convective impact on export of pollutants
In the REAM model, convective transport lifts pollutants from the boundary layer into the free troposphere.As a result, concentrations increase at higher altitudes and decrease at lower altitudes.In model simulations, the change of concentrations as a function of altitude reflects the strength of convective transport.Here we use C 3 H 8 as an example.Figure 2 shows the relative changes of C 3 H 8 driven by convection at the surface and four pressure levels (800, 500, 300, and 150 hPa) for July and August 2004 in the two models.Both models show decreases of C 3 H 8 at the surface and 800 hPa.At 500 hPa, convective transport increases C 3 H 8 in WRF-REAM particularly over the Southeast because of entrainment and detrainment and updraft flux convergence.MM5-REAM, in contrast, shows a general convection-driven decrease.At higher altitudes, both models show increasing concentrations due to convection.However, the largest increase is at 300 hPa in WRF-REAM but at 150 hPa in MM5-REAM.The maximum outflow altitude is higher in MM5-REAM because the convective top is higher in MM5 (Fig. 1b).
The difference of the simulated C 2 H 6 concentrations between the two models is within 10% (Appendix A).WRF-REAM simulated C 3 H 8 concentrations are 10-30% higher than MM5-REAM in the free troposphere (3-8 km), in better agreement with the INTEX-A observations (Appendix A).To minimize the effects of emission uncertainties and the large vertical gradient of C 2 H 6 and C 3 H 8 in this analysis, we investigate the convective effect on C 2 H 6 /C 3 H 8 ratios (Wang and Zeng, 2004).The chemical lifetime of C 3 H 8 (2 weeks) is shorter than C 2 H 6 (>1 month).Long-range transport of C 3 H 8 is less efficient and we expect to see a larger convective transport effect on C 3 H 8 than C 2 H 6 .
We compare the median profiles of C 2 H 6 /C 3 H 8 in both models with the INTEX-A measurements over the outflow region of the Western North Atlantic Ocean in Fig. 3.The measurements and corresponding simulated results are averaged into 1-km vertical bins.There are >50 measurements for each 1-km vertical bin.We also show the sensitivity results when convective transport is turned off in the models.  of 4-5 in the boundary layer.The observed ratio reaches a maximum of 9 at 3 km and gradually decreases to 4-5 at 11 km.Generally speaking, the ratio of C 2 H 6 /C 3 H 8 increases in the troposphere as a result of differential chemical aging and atmospheric mixing (Wang and Zeng, 2004).Therefore, the ratio of C 2 H 6 /C 3 H 8 tends to increase from the boundary layer to the free troposphere.The observed decrease of C 2 H 6 /C 3 H 8 ratio reflects the effect of convective transport, which mixes upper tropospheric (high C 2 H 6 /C 3 H 8 ratio) air masses with low C 2 H 6 /C 3 H 8 ratio air masses lifted from the boundary layer into the free troposphere.We note that the amount of mixing is determined by flux vertical convergence, not by the direct fluxes shown in Fig. 1a.The measurement variability is larger in the lower troposphere, reflecting a mixture of fresh continental air with low C 2 H 6 /C 3 H 8 ratios and aged marine air with high C 2 H 6 /C 3 H 8 ratios over the Western North Atlantic Ocean.Among the model simulations, both standard models reproduce the general profiles of the observed C 2 H 6 /C 3 H 8 ratios; the profile from WRF-REAM is in closer agreement with the measurements.MM5-REAM median profile is at the upper bound of the measurements at 4-9 km.More telling of the model difference is in the sensitivity simulations.Without convective transport, the simulated median C 2 H 6 /C 3 H 8 ratios in WRF-REAM would be a factor 2-3 too high compared to the measurements.In MM5-REAM, the effect of convective transport is evident only in the upper troposphere (above 7 km) as indicated in Fig. 2. The lack of convective mixing in MM5-REAM results in large overestimates of the C 2 H 6 /C 3 H 8 ratios in the free troposphere at 3-9 km.The convective effect in MM5-REAM becomes larger than WRF-REAM above 11 km.There is no direct in situ observation to evaluate the model performance above 11 km.What we will show in Sect.4.1 is that the convective cloud top is overestimated in MM5 compared with GOES satellite observations, particularly over the Western North Atlantic Ocean.WRF simulations are in closer agreement with the observations.
We also examine the effect of convective scavenging of soluble HNO 3 .We assume that HNO 3 is removed in convective updrafts in the model (e.g., Wang et al., 2001).This wet scavenging pathway effectively removes HNO 3 lifted from the boundary layer.However, HNO 3 produced from lightning NO x is not scavenged in this process.With entrainment (such as in WRF-REAM), background HNO 3 entrained into cumulus clouds is also removed.Without entrainment scavenging, upper tropospheric HNO 3 concentrations can be high from lightning NO x .In general, simulated HNO 3 concentrations are lower in WRF-REAM than MM5-REAM and are in better agreement with the INTEX-A measurements although both model simulated median HNO 3 profiles are within the standard deviations of the measurements (Appendix A).WRF simulates larger convective mass fluxes than MM5 and also includes entrainment fluxes (Fig. 1b).Both factors contribute to larger wet scavenging in WRF-REAM.

Cumulus cloud top and lightning NO x production
We compare model simulated tropospheric NO 2 columns with SCIAMACHY measurements during INTEX-A period (1 July-15 August) (Martin et al., 2006) to illustrate the difference of lightning NO x production between the two models (Fig. 4).The temporal resolution of SCIAMACHY is low, covering the globe every 6 days.After filtering out measurements with cloud fractions >30%, there are only about 2 days of measurements per model grid over most regions of the Eastern US during INTEX-A period; on average, there are 3 days of measurements over the US.Therefore, the comparison here is qualitative in nature.Some of the overestimates in the models can be traced back to simulated lightning influence during one of the measurement days.WRF-REAM and MM5-REAM simulations are very similar when lightning NO x is excluded.When including lightning NO x , WRF-REAM simulated NO 2 columns are lower than MM5-REAM and are closer to the limited observations.The spatial correlation is also higher in WRF-REAM (R=0.73)than MM5-REAM (R=0.58).Lightning   NO x concentrations are lower in WRF-REAM than MM5-REAM.For example, NO 2 columns above 12 km are mainly due to lightning NO x .They are much lower in WRF-REAM than in MM5-REAM (Fig. 4).Over the Western North Atlantic Ocean, NO 2 columns above 12 km account for 10% of the total columns in WRF-REAM but ∼50% in MM5-REAM.Specifying a lower NO x production rate per flash in MM5-REAM than WRF-REAM can correct the high bias in MM5-REAM.However, the correction will also lead to large underestimations in MM5-REAM compared to INTEX-A aircraft measurements (to be discussed in the next section).
The large difference in simulated lightning NO x production between WRF-REAM and MM5-REAM is due mainly to the difference in the simulated cumulus cloud top heights.The simulated vertical distribution of lightning NO x in both models follows the mid-latitude profile by Pickering et al. (1998).Figure 5 shows the vertical distributions of lightning NO x production in the two models averaged over the INTEX-A regions.MM5-REAM simulates the lightning NO x maximum at ∼15 km much higher than that in WRF-REAM at ∼12 km.It is important to note that even though MM5-REAM simulates much more total lightning NO x than WRF-REAM, the two models simulate similar lightning NO x production at 2-12 km, which will explain why they simulate similar lightning impact on the upper tropospheric (8-12 km) NO 2 and O 3 concentrations shown in the next section.Our lightning NO x parameterization is based on the observed cloud-to-ground (CG) flash rates from the NLDN network (Choi et al., 2005(Choi et al., , 2008a)).The intra-cloud (IC) lightning flash rates are estimated in the model as a function of the freezing altitude and cumulus cloud top height (Wang et al., 1998).A higher cloud top height generally leads to higher lightning NO x production.We therefore evaluate model simulated cumulus cloud top heights with the measurements by GOES-10 and GOES-12 satellites from the DX cloud dataset of the ISCCP (Rossow and Schiffer, 1991) in Fig. 6.Clearly, the problem is in MM5 results, where cloud top pressures are underestimated over most regions of the Gulf of Mexico, the southeastern US and the Western North Atlantic Ocean.An exception is over Southern Florida, where MM5 simulated cloud top heights are more consistent with the observations than WRF, which may indicate that the entrainment and detrainment are overestimated in WRF over that region.The overestimates of cloud top heights lead to higher IC/CG flash ratios and overestimates of lightning NO x production in these regions (Fig. 4).The average IC/CG flash ratio over the US from WRF-REAM is 5, much lower than that of 7 from MM5-REAM during the INTEX-A period.It also becomes apparent that lightning NO x in MM5-REAM is injected too high in altitude (Fig. 5).Convection in WRF with the KF-eta scheme extends to a lower altitude of 10-12 km, rather than up to 16 km in MM5 with the Grell scheme.Satellite measurements of NO 2 (indirectly) and cloud top pressure (directly) indicate that cloud top height simulated by WRF is more realistic.

Effect of lightning NO x during INTEX-A
The large model difference in lightning NO x is not necessarily reflected in the comparison with aircraft NO x measurements because the flight ceiling of the DC-8 is 12 km.Figure 7 shows the comparisons of upper tropospheric NO x at 8-12 km along the DC-8 flight tracks.The difference between WRF-REAM and MM5-REAM is not as significant as we found in Fig. 4-6 because of the similar lightning NO x emissions from the two models at 2-12 km (Fig. 5).Upper tropospheric NO x in both models are driven by lightning, which increases NO x mixing ratios by a factor of up to 5 (∼500 pptv).Both models simulate larger lightning impact over the South Eastern US than over the Northeast and East-ern Canada.Measurements indicate that the model underestimates lightning NO x production in the latter regions.
Figure 8 shows the comparison of the latitudinal distribution of upper tropospheric NO x (8-12 km) over Eastern North America (25 • N-55 • N and <90 • W).Generally, both models significantly underestimate lightning NO x over the regions north than 35 • N. MM5-REAM overestimates NO x concentrations over the southeastern US due in part to the subsidence of large amounts of lightning NO x above 12 km (Fig. 5).Both models simulate that NO x concentrations decrease by a factor of >5 from the Southeast to the Northeast and Eastern Canada, much larger than a factor of 2 or less in the measurements.Similar large biases in the simulated south-to-north decrease of lightning NO x over Eastern North America can also be found in previous studies (e.g., Li et al., 2005;Hudman et al., 2007;Cooper et al., 2006 and2009).Figure 1a shows that convective mass fluxes in the upper troposphere in both WRF and MM5 are generally low over the Northeast.Measurements by the NLDN network also show low CG flashes there.Therefore, the model underestimate may reflect that the lightning parameterization should be formulated differently over the northern regions from southern regions of Eastern North America.
Lightning NO x is a major source of O 3 in the upper troposphere and significantly affects the budget of tropospheric O 3 .Hudman et al. (2007Hudman et al. ( , 2009) ) found lightning can increase upper troposphere O 3 concentrations by 10-17 ppbv and Cooper et al. (2006) found an increase of 11-13 ppbv on average and suggested a maximum of 24 ppbv based on the box model analysis during INTEX-A over the Eastern US.We find, here, that O 3 concentrations are increased by up to ∼20 ppbv (Fig. 7) over the region and the average O 3 enhancement is ∼10 ppbv over the region.The results are in line with previous studies.Despite the difference in the underlying meteorological fields, simulated O 3 concentrations and their sensitivities to lightning NO x are similar between WRF-REAM and MM5-REAM since lightning Atmos.Chem.Phys., 9, [4315][4316][4317][4318][4319][4320][4321][4322][4323][4324][4325][4326][4327]2009 www.atmos-chem-phys.net/9/4315/2009/Tropospheric O 3 production from surface emissions of NO x and volatile organic compounds (VOCs) and transport from the stratosphere also make significant contributions to upper tropospheric O 3 (Choi et al., 2008a).

Relative contributions of surface and lightning emissions to tropospheric NO x
The relative importance of the different odd nitrogen sources in the troposphere, particularly lightning NO x , over the US has been investigated in previous studies (e.g., Cooper et al., 2009, and Cooper et al. (2009), although they found some evidence for model overestimation of lightning NO x over the Southeast and our simulations have a clear low bias compared to INTEX-A measurements over the Northeast and Eastern Canada (Fig. 7).Above 12 km, the two models clearly diverge.WRF-REAM and MM5-REAM calculate, ∼50% and ∼90% of the NO x are due to lightning, respectively.The NO x concentrations at 12-15 km from the MM5-REAM simulation are more than double those from the WRF-REAM simulation due to lightning.The divergence between WRF-REAM and MM5-REAM above 12 km reflects the lightning NO x vertical profiles in Fig. 4. The NO x mixing ratios due to surface emissions in the MM5-REAM simulation are ∼50% greater than those from the WRF-REAM simulation because of the absence of dilution from entrainment and detrainment and the higher cloud top height in MM5 simulation.
We also estimate the source contributions to total reactive nitrogen (NO y ) at 8-12 km.Both models suggest contributions of ∼40 and 10% to NO y from lightning and surface emissions over North America at 8-12 km, respectively.Previously, Allen et al. (2000) estimated that 13% and 16% are due to lightning and surface emissions over North America for October-November 1997 during the SONEX Experiment, respectively.More intensive summertime lightning is likely the reason for a larger lightning impact in our results.

Conclusions
REAM driven by two meteorological models, WRF (WRF-REAM) and MM5 (MM5-REAM) with different convective schemes, is used to evaluate the model sensitivities in convective transport and lightning NO x production to meteorological simulations.When compared to the convective transport and lightning NO x features measured during INTEX-A, we find that simulated convective transport and lightning NO x production are very sensitive to the difference of the underlying meteorological fields particularly the variables directly affected by the cumulus convection scheme.
WRF with the KF-eta scheme simulates larger updrafts from the lower troposphere, resulting in significantly more outflow at 3-9 km than MM5 with the Grell scheme.A sensitivity chemical indicator affected by this outflow is the C 2 H 6 /C 3 H 8 ratio.While WRF-REAM shows large decreases (up to a factor of 2) of the C 2 H 6 /C 3 H 8 ratio at 3-9 km due to convective outflow, the change is relatively small in MM5-REAM.In comparison, the two model results are in agreement in the boundary layer and 10-11 km.WRF-REAM simulations are clearly in closer agreement with the INTEX-A observations.Larger mass fluxes as well as entrainment and detrainment in WRF-REAM also lead to more scavenging of soluble HNO 3 in the free troposphere than MM5-REAM.The simulated median profile of HNO 3 by WRF-REAM is in closer agreement with the measurements than MM5-REAM, although the observed variation is larger than the model difference.
WRF with the KF-eta scheme simulates lower convective cloud top heights than MM5 with the Grell scheme.The cloud top height directly affects the model estimates of intra-cloud lightning production.Consequently, WRF-REAM simulates less lightning NO x than MM5-REAM and the maximum lightning NO x altitude of 12 km in WRF-REAM is lower than 15 km in MM5-REAM.Measurements of tropospheric NO 2 columns from SCIAMACHY provide a qualitative comparison, which suggests that WRF-REAM is closer to the observations, although the lower temporal resolution and cloud presence over convective regions greatly reduced the number of valid measurements.Evaluation using the ISCCP cloud top height measurements from GOES satellites clearly demonstrated that MM5 simulated convective cloud tops are too high over the southeastern US and the Western North Atlantic Ocean.
We note that the large model difference in lightning NO x production occurs mostly above 12 km, where no in situ measurements were available from INTEX-A.For future field missions targeting the effect of lightning NO x and convective transport, observations above 12 km are needed.
Despite the large differences discussed previously, the two models show similar agreement with upper tropospheric in situ NO x measurements.Over the observation regions of INTEX-A, the two models show consistent results for the effect of lightning NO x in the upper troposphere (8-12 km): (1) lightning enhances upper tropospheric NO x concentrations by up to a factor of >5 (∼500 pptv) and NO 2 columns by a factor of >1.5 over the ocean; (2) lightning and surface emissions over North America contribute to NO x (NO y ) at 8-12 km by 60-75% (40%) and ∼10% (10%), respectively; and (3) lightning NO x increases O 3 concentrations by up to 20 ppbv with an average of 10 ppbv.These results are generally consistent with previous studies conducted with coarser resolution global models.
A major model bias is that the decrease of lightning NO x effect (at 8-12 km) from the Southeast to the Northeast and Eastern Canada is significantly overestimated.Inspections of previous modeling results show similar biases.This model bias results from lesser convective activities simulated by both MM5 and WRF and lesser cloud-to-ground lightning flash rates in the observations.The bias indicates a need for a different lightning parameterization for the Southeast from the Northeast and eastern Canada.

Appendix A REAM model evaluations with INTEX-A measurements
The evaluation here largely follows that by Hudman et al. (2007).Figure A1 compares the simulated and observed vertical distributions of C 2 H 6 , C 3 H 8 , NO x , and HNO 3 concentrations during INTEX-A.The measurements and corresponding model results are averaged into 1-km vertical bins.There are >200 measurements for each 1-km vertical bin.The model successfully reproduces the observed concentrations of the C 2 H 6 and C 3 H 8 in the free troposphere.The difference of simulated C 2 H 6 between WRF-REAM and MM5-REAM is within 10%.WRF-REAM simulates 10-30% higher C 3 H 8 concentrations than MM5-REAM in the free troposphere (3-8 km).Both models overestimate the lower tropospheric C 2 H 6 and C 3 H 8 concentrations, likely resulting from the uncertainties of their emissions.The comparison over the outflow region of the Western Atlantic Ocean is similar.REAM simulated vertical NO x profiles are similar to the GEOS-CHEM result shown by Hudman et al. (2007).The observed NO x profile by Hudman et al. (2007) is lower than shown here or that by Cooper et al. (2009); the reason is unclear.The observed C-shape profile is simulated by REAM.The upper tropospheric NO x is underestimated.Figure 7 and 8 shows that most of the underestimation is over the Northeast and Eastern Canada.Increasing the lightning NO x production rate per flash in the model would lead to an overestimation over the Southeast and cause a large bias compared to satellite observed tropospheric NO 2 columns (Fig. 4).WRF-REAM and MM5-REAM simulated NO x profiles from the surface to 12 km are similar.
HNO 3 is generally well simulated by both WRF-REAM and MM5-REAM in the free troposphere but overestimated in the boundary layer.WRF-REAM simulated 15-35% less HNO 3 concentrations in the free troposphere than MM5-REAM, in closer agreement to the measurements.
Fig. 1a.Mean deep convective updraft mass fluxes from WRF and MM5 simulations for July and August 2004.
Figure 1b.Vertical profiles of mean mass fluxes of deep convection from WRF and M M5 simulations, and the average entrainment and detrainment fluxes from the WRF simulation for July and August 2004 over North America (shown in fig.1a).Positive (negative) fluxes are updrafts (downdrafts).

Fig. 1b .
Fig. 1b.Vertical profiles of mean mass fluxes of deep convection from WRF and MM5 simulations, and the average entrainment and detrainment fluxes from the WRF simulation for July and August 2004 over North America (shown in Fig. 1a).Positive (negative) fluxes are updrafts (downdrafts).

Figure 2 .Fig. 2 .
Fig. 2. Percentage changes of C 3 H 8 in the standard model simulations from the model simulations without convective transport for July and August 2004 at the surface, and 150, 300, 500, and 800 hPa.Results for WRF-REAM and MM5-REAM are shown.

Fig. 3 .
Fig. 3. Observed and simulated vertical profiles of median C 2 H 6 /C 3 H 8 ratios in the outflow regions over the Western North Atlantic Ocean.There are >50 measurements for each 1-km interval.Black squares show the observed means at 1-km interval.The horizontal bars show the standard deviations."std" denotes the standard simulation."w/o conv" denotes the simulation where convective transport is turned off.
Figure 4. Tropospheric NO2 columns derived from SCIAMACHY measurements [Martin et al., 2006] and simulated by WRF-REAM and MM5-REAM during the INTEX-A period (July 1 st to August 15 th , 2004).Tropospheric NO 2 columns from the standard simulation and a sensitivity simulation without lightning NO x are shown.Also shown are the tropospheric columns above 12 km in the standard simulation.Only the measurements with cloud fractions < 30% and the corresponding simulation results are used.White areas indicate that no measurement data are available.

Fig. 4 .Figure 5 .
Fig. 4. Tropospheric NO 2 columns derived from SCIAMACHY measurements (Martin et al., 2006) and simulated by WRF-REAM and MM5-REAM during the INTEX-A period (1 July to 15 August 2004).Tropospheric NO 2 columns from the standard simulation and a sensitivity simulation without lightning NO x are shown.Also shown are the tropospheric columns above 12 km in the standard simulation.Only the measurements with cloud fractions <30% and the corresponding simulation results are used.White areas indicate that no measurement data are available.

Fig. 5 .
Fig. 5. Mean lightning NO x production rate profiles in WRF-REAM and MM5-REAM for 1 July-15 August 2004 averaged over the INTEX-A region.

Fig. 6 .
Fig. 6.Mean cumulus cloud top pressures measured by GOE-10 and GOE-12 satellites and simulated by WRF and MM5 for 1 July-15 August 2004.Measurement data >500 hPa (and corresponding model results) are excluded to filter out the low cloud information.

Figure 7 .Fig. 7 .
Fig. 7. Observed and simulated upper tropospheric NOx and O 3 concentrations along DC-8 flight tracks at 8-12 km during the INTEX-A experiment.Results from the standard simulations and sensitivity simulations without lightning NO x using WRF-REAM and MM5-REAM are shown.The impacts of lightning (rightmost column) are estimated by subtracting the sensitivity results from the standard model results.

Figure 8 .
Figure 8. Observed and simulated latitudinal distributions of median upper tropospheric NOx (8-12 km) over eastern North America (25 o N-55 o N and <90 o W).There are >20 measurements for each 2 o latitude band.Black squares show the observed means and the vertical bars show the standard deviations.

Fig. 8 .
Fig. 8. Observed and simulated latitudinal distributions of median upper tropospheric NO x (8-12 km) over Eastern North America (25 • N-55 • N and <90 • W).There are >20 measurements for each 2 • latitude band.Black squares show the observed means and the vertical bars show the standard deviations.

Figure A1 .
Figure A1.Observed and simulated vertical profiles of median C 2 H 6 , C 3 H 8 , NO x , and 6 HNO 3 over the INTEX-A region on July 1 st -August 15 th , 2004.Black squares show the 7 observed means and the horizontal bars show the standard deviations.There are >200 8 data points for each 1-km interval.9 10 11 12 13 14 15 16 17 18 Fig.A1 Observed and simulated vertical profiles of median C 2 H 6 , C 3 H 8 , NO x , and HNO 3 over the INTEX-A region on 1 July-15 August, 2004.Black squares show the observed means and the horizontal bars show the standard deviations.There are >200 data points for each 1-km interval.
).One exception is that the emissions of NO x , CO, and (≥C 4 alkanes) over the US are prepared by Sparse Matrix Operator Kernel Emissions (SMOKE) model (http: //cf.unc.edu/cep/empd/products/smoke/index.cfm) for 2004 projected from the Visibility Improvement State and Tribal Association of the Southeast (VISTAS) 2002 emission inventory, since we found that these emissions are more consistent with INTEX-A measurements than the default inventories The observations show the lowest median C 2 H 6 /C 3 H 8 ratio www.atmos-chem-phys.net/9/4315/2009/Atmos.Chem.Phys., 9, 4315-4327, 2009 Mean cumulus cloud top pressures measured by GOE-10 and GOE-12 13 satellites and simulated by WRF and MM5 for July 1 st -August 15 th , 2004.Measurement 14 data > 500 hPa (and corresponding model results) are excluded to filter out the low cloud www.atmos-chem-phys.net/9/4315/2009/Atmos.Chem.Phys., 9, 4315-4327, 2009 the references therein).We use WRF-REAM and MM5-REAM to estimate the lightning and www.atmos-chem-phys.net/9/4315/2009/Atmos.Chem.Phys., 9, 4315-4327, 2009 surface NO x contributions over the INTEX-A regions (covering the US and Western North Atlantic Ocean) from 1 July-15 August.In our simulations, WRF-REAM and MM5-REAM show similar results up to 12 km.Lightning contribution to NO x increases from ∼10% in the boundary layer to 60-75% at 8-12 km.In contrast, the surface emission contributions decrease from 80% in the boundary layer to ∼10% at 8-12 km.Our estimation of lighting contribution is smaller than that of 80-95% by