Comprehensive evaluations of diurnal NO2 measurements during DISCOVER-AQ 2011: effects of resolution-dependent representation of NOx emissions

Abstract

Nitrogen oxides (${\mathrm{NO}}_{\mathit{\text{x}}}$ $=$ $\mathrm{NO}$ $+$ ${\mathrm{NO}}_{\mathrm{2}}$) play a crucial role in the formation of ozone and secondary inorganic and organic aerosols, thus affecting human health, global radiation budget, and climate. The diurnal and spatial variations in ${\mathrm{NO}}_{\mathrm{2}}$ are functions of emissions, advection, deposition, vertical mixing, and chemistry. Their observations, therefore, provide useful constraints in our understanding of these factors. We employ a Regional chEmical and trAnsport model (REAM) to analyze the observed temporal (diurnal cycles) and spatial distributions of ${\mathrm{NO}}_{\mathrm{2}}$ concentrations and tropospheric vertical column densities (TVCDs) using aircraft in situ measurements and surface EPA Air Quality System (AQS) observations as well as the measurements of TVCDs by satellite instruments (OMI: the Ozone Monitoring Instrument; GOME-2A: Global Ozone Monitoring Experiment – 2A), ground-based Pandora, and the Airborne Compact Atmospheric Mapper (ACAM) instrument in July 2011 during the DISCOVER-AQ campaign over the Baltimore–Washington region. The model simulations at 36 and 4 $\mathrm{km}$ resolutions are in reasonably good agreement with the regional mean temporospatial ${\mathrm{NO}}_{\mathrm{2}}$ observations in the daytime. However, we find significant overestimations (underestimations) of model-simulated ${\mathrm{NO}}_{\mathrm{2}}$ (${\mathrm{O}}_{\mathrm{3}}$) surface concentrations during nighttime, which can be mitigated by enhancing nocturnal vertical mixing in the model. Another discrepancy is that Pandora-measured ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs show much less variation in the late afternoon than simulated in the model. The higher-resolution 4 $\mathrm{km}$ simulations tend to show larger biases compared to the observations due largely to the larger spatial variations in ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions in the model when the model spatial resolution is increased from 36 to 4 $\mathrm{km}$. OMI, GOME-2A, and the high-resolution aircraft ACAM observations show a more dispersed distribution of ${\mathrm{NO}}_{\mathrm{2}}$ vertical column densities (VCDs) and lower VCDs in urban regions than corresponding 36 and 4 $\mathrm{km}$ model simulations, likely reflecting the spatial distribution bias of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions in the National Emissions Inventory (NEI) 2011.

1 Introduction

Nitrogen oxides (${\mathrm{NO}}_{\mathit{\text{x}}}$ $=$ $\mathrm{NO}$ $+$ ${\mathrm{NO}}_{\mathrm{2}}$) are among the most important trace gases in the atmosphere due to their crucial role in the formation of ozone (${\mathrm{O}}_{\mathrm{3}}$) and secondary aerosols and their role in the chemical transformation of other atmospheric species, such as carbon monoxide (CO) and volatile organic compounds (VOCs) (Cheng et al., 2017, 2018; Fisher et al., 2016; Li et al., 2019; Liu et al., 2012; Ng et al., 2017; Peng et al., 2016; Zhang and Wang, 2016). ${\mathrm{NO}}_{\mathit{\text{x}}}$ is emitted by both anthropogenic activities and natural sources. Anthropogenic sources account for about 77 % of the global ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions, and fossil fuel combustion and industrial processes are the primary anthropogenic sources, which contribute to about 75 % of the anthropogenic emissions (Seinfeld and Pandis, 2016). Other important anthropogenic sources include agriculture and biomass and biofuel burning. Soils and lightning are two major natural sources. Most ${\mathrm{NO}}_{\mathit{\text{x}}}$ is emitted as NO, which is then oxidized to ${\mathrm{NO}}_{\mathrm{2}}$ by oxidants, such as ${\mathrm{O}}_{\mathrm{3}}$, the hydroperoxyl radical (${\mathrm{HO}}_{\mathrm{2}}$), and organic peroxy radicals (${\mathrm{RO}}_{\mathrm{2}}$).

The diurnal variations in ${\mathrm{NO}}_{\mathrm{2}}$ controlled by physical and chemical processes reflect the temporal patterns of these underlying controlling factors, such as ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions, chemistry, deposition, advection, diffusion, and convection. Therefore, the observations of ${\mathrm{NO}}_{\mathrm{2}}$ diurnal cycles can be used to evaluate our understanding of ${\mathrm{NO}}_{\mathit{\text{x}}}$-related emission, chemistry, and physical processes (Frey et al., 2013; Jones et al., 2000; Judd et al., 2018). For example, Brown et al. (2004) analyzed the diurnal patterns of surface NO, ${\mathrm{NO}}_{\mathrm{2}}$, ${\mathrm{NO}}_{\mathrm{3}}$, ${\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}$, ${\mathrm{HNO}}_{\mathrm{3}}$, OH, and ${\mathrm{O}}_{\mathrm{3}}$ concentrations along the east coast of the United States (US) during the New England Air Quality Study (NEAQS) campaign in the summer of 2002 and found that the predominant nighttime sink of ${\mathrm{NO}}_{\mathit{\text{x}}}$ through the hydrolysis of ${\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}$ had an efficiency on par with daytime photochemical loss over the ocean surface off the New England coast. Van Stratum et al. (2012) investigated the contribution of boundary layer dynamics to chemistry evolution during the DOMINO (Diel Oxidant Mechanisms in relation to Nitrogen Oxides) campaign in 2008 in Spain and found that entrainment and boundary layer growth in daytime influenced mixed-layer NO and ${\mathrm{NO}}_{\mathrm{2}}$ diurnal cycles on the same order of chemical transformations. David and Nair (2011) found that the diurnal patterns of surface NO, ${\mathrm{NO}}_{\mathrm{2}}$, and ${\mathrm{O}}_{\mathrm{3}}$ concentrations at a tropical coastal station in India from November 2007 to May 2009 were closely associated with sea breeze and land breeze, which affected the availability of ${\mathrm{NO}}_{\mathit{\text{x}}}$ through transport. They also thought that monsoon-associated synoptic wind patterns could strongly influence the magnitudes of NO, ${\mathrm{NO}}_{\mathrm{2}}$, and ${\mathrm{O}}_{\mathrm{3}}$ diurnal cycles. The monsoon effect on surface NO, ${\mathrm{NO}}_{\mathrm{2}}$, and ${\mathrm{O}}_{\mathrm{3}}$ diurnal cycles was also observed in China by Tu et al. (2007) on the basis of continuous measurements of NO, ${\mathrm{NO}}_{\mathrm{2}}$, and ${\mathrm{O}}_{\mathrm{3}}$ at an urban site in Nanjing from January 2000–February 2003.

In addition to surface ${\mathrm{NO}}_{\mathrm{2}}$ diurnal cycles, the daily variations in ${\mathrm{NO}}_{\mathrm{2}}$ vertical column densities (VCDs) were also investigated in previous studies. For example, Boersma et al. (2008) compared ${\mathrm{NO}}_{\mathrm{2}}$ tropospheric VCDs (TVCDs) retrieved from OMI (the Ozone Monitoring Instrument) and SCIAMACHY (SCanning Imaging Absorption SpectroMeter for Atmospheric CHartography) in August 2006 around the world. They found that the diurnal patterns of different types of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions could strongly affect the ${\mathrm{NO}}_{\mathrm{2}}$ TVCD variations between OMI and SCIAMACHY and that intense afternoon fire activity resulted in an increase in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs from 10:00 to 13:30 LT (local time) over tropical biomass burning regions. Boersma et al. (2009) further investigated the ${\mathrm{NO}}_{\mathrm{2}}$ TVCD change from SCIAMACHY to OMI in different seasons of 2006 in Israeli cities and found that there was a slight increase in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs from SCIAMACHY to OMI in winter due to increased ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions from 10:00 to 13:30 LT and a sufficiently weak photochemical sink and that the TVCDs from OMI were lower than SCIAMACHY in summer due to a strong photochemical sink of ${\mathrm{NO}}_{\mathit{\text{x}}}$.

Figure 1

The locations of surface and P-3B aircraft observations during the DISCOVER-AQ 2011 campaign. We mark the 36 $\mathrm{km}$ REAM grid cells with red lines and the 4 $\mathrm{km}$ REAM grid cells with black lines. Gray shading denotes land surface in the nested 4 $\mathrm{km}$ WRF domain, while the white area denotes ocean or water surface. Blue dots denote surface ${\mathrm{O}}_{\mathrm{3}}$ observation sites. Cross marks denote surface ${\mathrm{NO}}_{\mathrm{2}}$ observation sites, and their colors denote different measurement instruments: green for the Thermo Electron 42C-Y ${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzer, dark orchid for the Ecotech Model 9841/9843 T-${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzers, black for the Thermo Model 42C ${\mathrm{NO}}_{\mathit{\text{x}}}$ analyzer, and chocolate for the Teledyne API model 200eup photolytic ${\mathrm{NO}}_{\mathit{\text{x}}}$ analyzer. Circles denote Pandora sites, and the cyan circle denotes a Pandora site (USNA) on a ship. Black squares denote the inland P-3B aircraft spiral locations.

[Figure omitted. See PDF]

All of the above research, however, exploited only ${\mathrm{NO}}_{\mathrm{2}}$ surface or satellite VCD measurements. Due to the availability of ground-based ${\mathrm{NO}}_{\mathrm{2}}$ VCD observations, some recent studies tried to investigate the diurnal relationships between ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations and ${\mathrm{NO}}_{\mathrm{2}}$ VCDs (Kollonige et al., 2018; Thompson et al., 2019). For example, Zhao et al. (2019) converted Pandora direct-sun and zenith-sky ${\mathrm{NO}}_{\mathrm{2}}$ VCDs to ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations using concentration-to-partial-column ratios and found that the derived concentrations captured the observed ${\mathrm{NO}}_{\mathrm{2}}$ surface diurnal and seasonal variations well. Knepp et al. (2015) related the daytime variations in ${\mathrm{NO}}_{\mathrm{2}}$ TVCD measurements by ground-based Pandora instruments to the variations in coincident ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations using a planetary boundary layer height (PBLH) factor over the periods July 2011–October 2011 at the NASA Langley Research Center in Hampton, Virginia, and July 2011 at the Padonia and Edgewood sites in Maryland for the DISCOVER-AQ experiment, showing the importance of boundary layer vertical mixing on ${\mathrm{NO}}_{\mathrm{2}}$ vertical distributions and the ability of ${\mathrm{NO}}_{\mathrm{2}}$ VCD measurements to infer hourly boundary layer ${\mathrm{NO}}_{\mathrm{2}}$ variations. DISCOVER-AQ, the Deriving Information on Surface conditions from Column and Vertically Resolved Observations Relevant to Air Quality experiment (https://discover-aq.larc.nasa.gov/, last access: 6 April 2019), was designed to better understand the relationship between boundary layer pollutants and satellite observations (Flynn et al., 2014; Reed et al., 2015). Figure 1 shows the sampling locations of the summer DISCOVER-AQ 2011 campaign in the Baltimore–Washington metropolitan region. In this campaign, the NASA P-3B aircraft flew spirals over six air quality monitoring sites (Aldino – rural and suburban, Edgewood – coastal and urban, Beltsville – suburban, Essex – coastal and urban, Fairhill – rural, and Padonia – suburban) (Table S1 in the Supplement) and the Chesapeake Bay (Cheng et al., 2017; Lamsal et al., 2014) and measured 245 ${\mathrm{NO}}_{\mathrm{2}}$ profiles in 14 flight days in July (Zhang et al., 2016). During the same period, the NASA UC-12 aircraft flew across the Baltimore–Washington region at an altitude of about 8 $\mathrm{km}$ above sea level (a.s.l.), using the Airborne Compact Atmospheric Mapper (ACAM) to map the distributions of ${\mathrm{NO}}_{\mathrm{2}}$ VCDs below the aircraft (Lamsal et al., 2017). Furthermore, ground-based instruments were deployed to measure ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations, ${\mathrm{NO}}_{\mathrm{2}}$ VCDs, and other physical properties of the atmosphere (Anderson et al., 2014; Reed et al., 2015; Sawamura et al., 2014). Satellite OMI and GOME-2A (Global Ozone Monitoring Experiment – 2A) instruments provided ${\mathrm{NO}}_{\mathrm{2}}$ TVCD measurements over the campaign region at 13:30 and 09:30 LT, respectively. These concurrent measurements of ${\mathrm{NO}}_{\mathrm{2}}$ VCDs, surface ${\mathrm{NO}}_{\mathrm{2}}$, and vertically resolved distributions of ${\mathrm{NO}}_{\mathrm{2}}$ during the DISCOVER-AQ 2011 campaign, therefore, provide a comprehensive dataset to evaluate ${\mathrm{NO}}_{\mathrm{2}}$ diurnal and spatial variabilities and processes affecting ${\mathrm{NO}}_{\mathrm{2}}$ concentrations.

Section 2 describes the measurement datasets in detail. The Regional chEmistry and trAnsport Model (REAM), also described in Sect. 2, is applied to simulate the ${\mathrm{NO}}_{\mathrm{2}}$ observations during the DISCOVER-AQ campaign in July 2011. The evaluations of the simulated diurnal cycles of surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations, ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles, and ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs are discussed in Sect. 3 through comparisons with observations. In Sect. 3, we also investigate the differences between ${\mathrm{NO}}_{\mathrm{2}}$ diurnal cycles on weekdays and weekends and their implications for ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission characteristics. To corroborate our evaluation of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions based on ${\mathrm{NO}}_{\mathrm{2}}$ diurnal cycles, we further compare observed ${\mathrm{NO}}_{\mathit{\text{y}}}$ (reactive nitrogen compounds) concentrations with REAM simulation results in Sect. 3. Moreover, we assess the resolution dependence of REAM simulation results in light of the observations and discuss the potential distribution biases of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions by comparing the 36 and 4 $\mathrm{km}$ REAM simulation results with OMI, GOME-2A, and high-resolution ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs. Finally, we summarize the study in Sect. 4.

2 Datasets and model description

2.1 REAM

REAM has been widely applied in many studies (Cheng et al., 2017; Choi et al., 2008; Li et al., 2019; R. Zhang et al., 2018; Y. Zhang et al., 2016; Zhao et al., 2009). The model has a horizontal resolution of 36 $\mathrm{km}$ and 30 vertical layers in the troposphere. Meteorology fields are from a Weather Research and Forecasting (WRF; version 3.6) model simulation with a horizontal resolution of 36 $\mathrm{km}$. We summarize the physics parameterization schemes of the WRF simulation in Table S2. The WRF simulation is initialized and constrained by the NCEP coupled forecast system model version 2 (CFSv2) products (http://rda.ucar.edu/datasets/ds094.0/, last access: 10 March 2015) (Saha et al., 2011). The chemistry mechanism in REAM is based on GEOS-Chem v11.01 with updated aerosol uptake of isoprene nitrates (Fisher et al., 2016) and revised treatment of wet scavenging processes (Luo et al., 2019). A 2${}^{\circ }$ $\times$ 2.5${}^{\circ }$ GEOS-Chem simulation provides the chemical boundary and initial conditions.

Figure 2

Distributions of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions for the (a) 36 $\mathrm{km}$ and (b) 4 $\mathrm{km}$ REAM simulations around the DISCOVER-AQ 2011 region. Here ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions refer to the mean values ($\mathrm{molecules}{\mathrm{km}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$) in 1 week (Monday–Sunday).

[Figure omitted. See PDF]

Figure 3

Relative diurnal profiles of weekday and weekend ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions ($\mathrm{molecules}{\mathrm{km}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$) in the DISCOVER-AQ 2011 region (the 36 and 4 $\mathrm{km}$ grid cells over the 11 inland Pandora sites shown in Fig. 1) for the 36 and 4 $\mathrm{km}$ REAM. All the profiles are scaled by the 4 $\mathrm{km}$ weekday emission average value ($\mathrm{molecules}{\mathrm{km}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$).

[Figure omitted. See PDF]

Biogenic VOC emissions in REAM are from MEGAN v2.10 (Guenther et al., 2012). Anthropogenic emissions on weekdays are from the National Emission Inventory 2011 (NEI2011) (EPA, 2014) from the Pacific Northwest National Laboratory (PNNL), which has an initial resolution of 4 $\mathrm{km}$ and is regridded to REAM 36 $\mathrm{km}$ grid cells (Fig. 2). Weekday emission diurnal profiles are from NEI2011. The weekday-to-weekend emission ratios and weekend emission diurnal profiles are based on previous studies (Beirle et al., 2003; Boersma et al., 2009; Choi et al., 2012; de Foy, 2018; DenBleyker et al., 2012; Herman et al., 2009; Judd et al., 2018; Kaynak et al., 2009; Kim et al., 2016). These studies suggested that weekend ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions were 20 %–50 % lower than weekday emissions, and the weekend ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission diurnal cycles were different from weekdays; therefore, we specify a weekend-to-weekday ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission ratio of $\mathrm{2}/\mathrm{3}$ in this study. The resulting diurnal variations in weekday and weekend ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions over the DISCOVER-AQ 2011 region are shown in Fig. 3. The diurnal emission variation is lower on weekends than on weekdays.

To understand the effects of model resolutions on the temporospatial distributions of ${\mathrm{NO}}_{\mathrm{2}}$, we also conduct a REAM simulation with a horizontal resolution of 4 $\mathrm{km}$ during the DISCOVER-AQ campaign. A 36 $\mathrm{km}$ REAM simulation (discussed in Sect. 3.2) provides the chemical initial and hourly boundary conditions. Meteorology fields are from a nested WRF simulation (36, 12, 4 $\mathrm{km}$) with cumulus parameterization turned off in the 4 $\mathrm{km}$ domain (Table S2). Figure 1 shows a comparison of the 4 and 36 $\mathrm{km}$ REAM grid cells with DISCOVER-AQ observations, and Fig. 2 shows a comparison of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission distributions between the 4 and 36 $\mathrm{km}$ REAM simulations. The comparison of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission diurnal variations over the DISCOVER-AQ 2011 region between the 4 and 36 $\mathrm{km}$ REAM is shown in Fig. 3.

2.2

${\mathrm{NO}}_{\mathrm{2}}$ TVCD measurements by OMI and GOME-2A

The OMI instrument onboard the sun-synchronous NASA EOS Aura satellite with an Equator-crossing time of around 13:30 LT was developed by the Finnish Meteorological Institute and the Netherlands Agency for Aerospace Programs to measure solar backscattering radiation in the visible and ultraviolet bands (Levelt et al., 2006; Russell et al., 2012). The radiance measurements are used to derive trace gas concentrations in the atmosphere, such as ${\mathrm{O}}_{\mathrm{3}}$, ${\mathrm{NO}}_{\mathrm{2}}$, HCHO, and ${\mathrm{SO}}_{\mathrm{2}}$ (Levelt et al., 2006). OMI has a nadir resolution of 13 $\mathrm{km}$ $\times$ 24 $\mathrm{km}$ and provides daily global coverage (Levelt et al., 2006).

Two widely used archives of OMI ${\mathrm{NO}}_{\mathrm{2}}$ VCD products are available, NASA OMNO2 (v4.0) (https://disc.gsfc.nasa.gov/datasets/OMNO2_003/summary, last access: 26 September 2020) and KNMI DOMINO (v2.0) (https://www.temis.nl/airpollution/no2.php, last access: 14 January 2015). Although both use Differential Optical Absorption Spectroscopy (DOAS) algorithms to derive ${\mathrm{NO}}_{\mathrm{2}}$ slant column densities, they have differences in spectral fitting, stratospheric and tropospheric ${\mathrm{NO}}_{\mathrm{2}}$ slant column density (SCD) separation, a priori ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles, air mass factor (AMF) calculation, etc. (Boersma et al., 2011; Bucsela et al., 2013; Chance, 2002; Krotkov et al., 2017; Lamsal et al., 2021; Marchenko et al., 2015; Oetjen et al., 2013; van der A et al., 2010; Van Geffen et al., 2015). Both OMNO2 and DOMINO have been extensively evaluated with field measurements and models (Boersma et al., 2009, 2011; Choi et al., 2020; Hains et al., 2010; Huijnen et al., 2010; Ionov et al., 2008; Irie et al., 2008; Lamsal et al., 2014, 2021; Oetjen et al., 2013). The estimated uncertainty in the DOMINO TVCD product includes an absolute component of 1.0 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}$ and a relative AMF component of 25 % (Boersma et al., 2011), while the uncertainty in the OMNO2 TVCD product ranges from $\sim$ 30 % under clear-sky conditions to $\sim$ 60 % under cloudy conditions (Lamsal et al., 2014; Oetjen et al., 2013; Tong et al., 2015). In order to reduce uncertainties in this study, we only use TVCD data with effective cloud fractions $<$ 0.2, solar zenith angle (SZA) $<$ 80${}^{\circ }$, and albedo $\le$ 0.3. Both positive and negative TVCDs are considered in the calculation. The data affected by row anomaly are excluded (Boersma et al., 2018; R. Zhang et al., 2018).

For AMF calculation, DOMINO used daily TM4 model results with a resolution of 3${}^{\circ }$ $\times$ 2${}^{\circ }$ as a priori ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles (Boersma et al., 2007, 2011), while OMNO2 v4.0 used monthly mean values from the Global Modeling Initiative (GMI) model with a resolution of 1${}^{\circ }$ $\times$ 1.25${}^{\circ }$. The relatively coarse horizontal resolution of the a priori ${\mathrm{NO}}_{\mathrm{2}}$ profiles in the retrievals can introduce uncertainties in the spatial and temporal characteristics of ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs at satellite pixel scales. For comparison purposes, we also use 36 $\mathrm{km}$ REAM simulation results as the a priori ${\mathrm{NO}}_{\mathrm{2}}$ profiles to compute the AMFs and ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs with the DOMINO algorithm. The 36 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ data are first regridded to OMI pixels to calculate the corresponding tropospheric AMFs, which are then applied to compute OMI ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs by dividing the tropospheric SCDs from the DOMINO product by our updated AMFs.

The GOME-2 instrument onboard the polar-orbiting MetOp-A satellite (now referred to as GOME-2A) is an improved version of GOME-1 launched in 1995 and has an overpass time of 09:30 LT and a spatial resolution of 80 $\mathrm{km}$ $\times$ 40 $\mathrm{km}$ (Munro et al., 2006; Peters et al., 2012). GOME-2A measures backscattered solar radiation in the range from 240 to 790 $\mathrm{nm}$, which is used for VCD retrievals of trace gases, such as ${\mathrm{O}}_{\mathrm{3}}$, ${\mathrm{NO}}_{\mathrm{2}}$, BrO, and ${\mathrm{SO}}_{\mathrm{2}}$ (Munro et al., 2006). We use the KNMI TM4NO2A v2.3 GOME-2A ${\mathrm{NO}}_{\mathrm{2}}$ VCD product archived on http://www.temis.nl/airpollution/no2col/no2colgome2_v2.php (last access: 22 January 2015) (Boersma et al., 2007, 2011). GOME-2A-derived ${\mathrm{NO}}_{\mathrm{2}}$ VCDs have been validated with SCIAMACHY and MAX-DOAS measurements (Irie et al., 2012; Peters et al., 2012; Richter et al., 2011). As in the case of OMI, we use the same criteria to filter the ${\mathrm{NO}}_{\mathrm{2}}$ TVCD data and recalculate the tropospheric AMF values and GOME-2A TVCDs using the daily 36 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ profiles (09:00–10:00 LT).

2.3

Pandora ground-based ${\mathrm{NO}}_{\mathrm{2}}$ VCD measurements

Pandora is a small direct sun spectrometer which measures sun and sky radiance from 270 to 530 $\mathrm{nm}$ with a 0.5 $\mathrm{nm}$ resolution and a 1.6${}^{\circ }$ field of view (FOV) for the retrieval of the total VCDs of ${\mathrm{NO}}_{\mathrm{2}}$ with a precision of about 5.4 $\times$ 10${}^{\mathrm{14}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}$ (2.7 $\times$ 10${}^{\mathrm{14}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}$ for ${\mathrm{NO}}_{\mathrm{2}}$ SCD) and a nominal accuracy of 2.7 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}$ under clear-sky conditions (Herman et al., 2009; Lamsal et al., 2014; Zhao et al., 2020). There were 12 Pandora sites operating in the DISCOVER-AQ campaign (Fig. 1). Six of them are the same as the P-3B aircraft spiral locations (Aldino, Edgewood, Beltsville, Essex, Fairhill, and Padonia) (Table S1 and Fig. 1). The other six sites are Naval Academy (Annapolis, Maryland) (USNA – ocean), University of Maryland College Park (UMCP – urban), University of Maryland Baltimore County (UMBC – urban), Smithsonian Environmental Research Center (SERC – rural and coastal), Oldtown in Baltimore (Oldtown – urban), and Goddard Space Flight Center (GSFC – urban and suburban) (Table S1 and Fig. 1). In this study, we exclude the USNA site as its measurements were conducted on a ship (“Pandora(w)” in Fig. 1), and there were no other surface observations in the corresponding REAM grid cell. Including the data from the USNA site has a negligible effect on the comparisons of observed and simulated ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs. In our analysis, we ignore Pandora measurements with SZA $>$ 80${}^{\circ }$ (Fig. S1 in the Supplement) and exclude the data when fewer than three valid measurements are available within an hour to reduce the uncertainties in the hourly averages due to the significant variations in Pandora observations (Fig. S2).

Since Pandora measures total ${\mathrm{NO}}_{\mathrm{2}}$ VCDs, we need to subtract stratosphere ${\mathrm{NO}}_{\mathrm{2}}$ VCDs from the total VCDs to compute TVCDs. As shown in Fig. S3, stratosphere ${\mathrm{NO}}_{\mathrm{2}}$ VCDs show a clear diurnal cycle with an increase during daytime due in part to the photolysis of reactive nitrogen reservoirs such as ${\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}$ and ${\mathrm{HNO}}_{\mathrm{3}}$ (Brohede et al., 2007; Dirksen et al., 2011; Peters et al., 2012; Sen et al., 1998; Spinei et al., 2014), which is consistent with the significant increase in stratospheric ${\mathrm{NO}}_{\mathrm{2}}$ VCDs from GOME-2A to OMI. In this study, we use the GMI model-simulated stratospheric ${\mathrm{NO}}_{\mathrm{2}}$ VCDs in Fig. S3 to calculate the Pandora ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs. The small discrepancies between the GMI stratospheric ${\mathrm{NO}}_{\mathrm{2}}$ VCDs and satellite products do not change the pattern of Pandora ${\mathrm{NO}}_{\mathrm{2}}$ TVCD diurnal variations or affect the conclusions in this study.

2.4

ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCD measurements

The ACAM instrument onboard the UC-12 aircraft consists of two thermally stabilized spectrometers in the ultraviolet, visible, and near-infrared range. The spectrometer in the ultraviolet and visible band (304–520 $\mathrm{nm}$) with a resolution of 0.8 $\mathrm{nm}$ and a sampling of 0.105 $\mathrm{nm}$ can be used to detect ${\mathrm{NO}}_{\mathrm{2}}$ in the atmosphere. The native ground resolution of UC-12 ACAM ${\mathrm{NO}}_{\mathrm{2}}$ measurements is 0.5 $\mathrm{km}$ $\times$ 0.75 $\mathrm{km}$ at a flight altitude of about 8 $\mathrm{km}$ a.s.l. and a nominal ground speed of 100 $\mathrm{m}{\mathrm{s}}^{-\mathrm{1}}$ during the DISCOVER-AQ 2011 campaign (Lamsal et al., 2017), thus providing high-resolution ${\mathrm{NO}}_{\mathrm{2}}$ VCDs below the aircraft.

In this study, we mainly use the ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCD product described by Lamsal et al. (2017), which applied a pair-average co-adding scheme to produce ${\mathrm{NO}}_{\mathrm{2}}$ VCDs at a ground resolution of about 1.5 $\mathrm{km}$ (cross-track) $\times$ 1.1 $\mathrm{km}$ (along-track) to reduce noise impacts. In their retrieval of ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs, they first used the DOAS fitting method to generate differential ${\mathrm{NO}}_{\mathrm{2}}$ SCDs relative to the SCDs at an unpolluted reference location. Then they computed above- and below-aircraft AMFs at both sampling and reference locations based on the vector linearized discrete ordinate radiative transfer code (VLIDORT) (Spurr, 2008). In the computation of AMFs, the a priori ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles were from a combination of high-resolution (4 $\mathrm{km}$) CMAQ (the Community Multiscale Air Quality Modeling System) model outputs in the boundary layer and GMI simulation (2${}^{\circ }$ $\times$ 2.5${}^{\circ }$) results elsewhere in the atmosphere. Finally, the below-aircraft ${\mathrm{NO}}_{\mathrm{2}}$ VCDs at the sampling locations were generated by dividing below-aircraft ${\mathrm{NO}}_{\mathrm{2}}$ SCDs at the sampling locations by the corresponding below-aircraft AMFs. The below-aircraft ${\mathrm{NO}}_{\mathrm{2}}$ SCDs were the differences between the total and above-aircraft ${\mathrm{NO}}_{\mathrm{2}}$ SCDs. The total ${\mathrm{NO}}_{\mathrm{2}}$ SCDs were the sum of DOAS-fitting-generated differential ${\mathrm{NO}}_{\mathrm{2}}$ SCDs and ${\mathrm{NO}}_{\mathrm{2}}$ SCDs at the reference location, and the above-aircraft ${\mathrm{NO}}_{\mathrm{2}}$ SCDs were derived based on above-aircraft AMFs, GMI ${\mathrm{NO}}_{\mathrm{2}}$ profiles, and OMNO2 stratospheric ${\mathrm{NO}}_{\mathrm{2}}$ VCDs (Lamsal et al., 2017). The ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCD product had been evaluated via comparisons with other independent observations during the DISCOVER-AQ 2011 campaign, such as P-3B aircraft, Pandora, and OMNO2, and the uncertainty in individual below-aircraft ${\mathrm{NO}}_{\mathrm{2}}$ VCD is about 30 % (Lamsal et al., 2017). To keep the consistency of ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs, we exclude ${\mathrm{NO}}_{\mathrm{2}}$ VCDs measured at altitudes $<\mathrm{8}$ $\mathrm{km}$ a.s.l., which accounts for about 6.8 % of the total available ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCD data. We regrid the $\mathrm{1.5}\mathrm{km}\times \mathrm{1.1}\mathrm{km}$ ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs to the 4 $\mathrm{km}$ REAM grid cells (Fig. 1), which are then used to evaluate the distribution of ${\mathrm{NO}}_{\mathrm{2}}$ VCDs in the 4 $\mathrm{km}$ REAM simulation. As a supplement in Sect. 3.7, we also assess the 4 $\mathrm{km}$ REAM simulation by using the UC-12 ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs produced by the Smithsonian Astrophysical Observatory (SAO) algorithms, archived on https://www-air.larc.nasa.gov/cgi-bin/ArcView/discover-aq.dc-2011?UC12=1#LIU.XIONG/ (last access: December 31, 2019) (Liu et al., 2015a, b). This product is an early version of the SAO algorithm used to produce the Geostationary Trace gas and Aerosol Sensor Optimization (GeoTASO) and the GEOstationary Coastal and Air Pollution Events (GEO-CAPE) Airborne Simulator (GCAS) airborne observations in later airborne campaigns (Nowlan et al., 2016, 2018).

2.5

Surface ${\mathrm{NO}}_{\mathrm{2}}$ and ${\mathrm{O}}_{\mathrm{3}}$ measurements

The measurement of ${\mathrm{NO}}_{\mathit{\text{x}}}$ is based on the chemiluminescence of electronically excited ${\mathrm{NO}}_{\mathrm{2}}^{\ast }$, produced from the reaction of NO with ${\mathrm{O}}_{\mathrm{3}}$, and the strength of the chemiluminescence from the decay of ${\mathrm{NO}}_{\mathrm{2}}^{\ast }$ to ${\mathrm{NO}}_{\mathrm{2}}$ is proportional to the number of NO molecules present (Reed et al., 2016). ${\mathrm{NO}}_{\mathrm{2}}$ concentrations can be measured with this method by converting ${\mathrm{NO}}_{\mathrm{2}}$ to NO first through catalytic reactions (typically on the surface of heated molybdenum oxide (${\mathrm{MoO}}_{\mathit{\text{x}}}$) substrate) or photolytic processes (Lamsal et al., 2015; Reed et al., 2016). However, for the catalytic method, reactive nitrogen compounds other than ${\mathrm{NO}}_{\mathit{\text{x}}}$ (${\mathrm{NO}}_{\mathit{\text{z}}}$), such as ${\mathrm{HNO}}_{\mathrm{3}}$, peroxyacetyl nitrate (PAN), and other organic nitrates, can also be reduced to NO on the heated surface, thus causing an overestimation of ${\mathrm{NO}}_{\mathrm{2}}$. The magnitude of the overestimation depends on the concentrations and the reduction efficiencies of interference species, both of which are uncertain. The photolytic approach, which employs broadband photolysis of ambient ${\mathrm{NO}}_{\mathrm{2}}$, offers more accurate ${\mathrm{NO}}_{\mathrm{2}}$ measurements (Lamsal et al., 2015).

Figure 4

Hourly ratios of ${\mathrm{NO}}_{\mathrm{2}}$ measurements from the Teledyne API model 200 eup photolytic ${\mathrm{NO}}_{\mathit{\text{x}}}$ analyzer to ${\mathrm{NO}}_{\mathrm{2}}$ from coincident catalytic instruments for 2011 July. “CY42” denotes the ratios of photolytic ${\mathrm{NO}}_{\mathrm{2}}$ to ${\mathrm{NO}}_{\mathrm{2}}$ from the Thermo Electron 42C-Y ${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzer in Edgewood, “C42” denotes the ratios of photolytic ${\mathrm{NO}}_{\mathrm{2}}$ to ${\mathrm{NO}}_{\mathrm{2}}$ from the Thermo Model 42C ${\mathrm{NO}}_{\mathit{\text{x}}}$ analyzer in Padonia, and “ECO” denotes the ratios of photolytic ${\mathrm{NO}}_{\mathrm{2}}$ to ${\mathrm{NO}}_{\mathrm{2}}$ from the Ecotech Model 9841 T-${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzer in Padonia. “ECO” ratios are also used to scale ${\mathrm{NO}}_{\mathrm{2}}$ measurements from the Ecotech Model 9843 T-${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzer.

[Figure omitted. See PDF]

There were 11 ${\mathrm{NO}}_{\mathit{\text{x}}}$ monitoring sites operating in the DISCOVER-AQ region during the campaign (Fig. 1), including those from the EPA Air Quality System (AQS) monitoring network and those deployed for the DISCOVER-AQ campaign. Nine of them measured ${\mathrm{NO}}_{\mathrm{2}}$ concentrations by a catalytic converter. The other two sites (Edgewood and Padonia) had ${\mathrm{NO}}_{\mathrm{2}}$ measurements from both catalytic and photolytic methods. Different stationary catalytic instruments were used during the campaign: Thermo Electron 42C-Y ${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzer, Thermo Model 42C ${\mathrm{NO}}_{\mathit{\text{x}}}$ analyzer, Thermo Model 42I-Y ${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzer, and Ecotech Model 9843 and 9841 T-${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzers. In addition, a mobile platform – NATIVE (Nittany Atmospheric Trailer and Integrated Validation Experiment) with a Thermo Electron 42C-Y ${\mathrm{NO}}_{\mathit{\text{y}}}$ analyzer installed – was also deployed at the Edgewood site. The photolytic measurements of ${\mathrm{NO}}_{\mathrm{2}}$ in Edgewood and Padonia were from Teledyne API model 200eup photolytic ${\mathrm{NO}}_{\mathit{\text{x}}}$ analyzers. We scale catalytic ${\mathrm{NO}}_{\mathrm{2}}$ measurements using the diurnal ratios of ${\mathrm{NO}}_{\mathrm{2}}$ photolytic measurements to ${\mathrm{NO}}_{\mathrm{2}}$ from the corresponding catalytic analyzers (Fig. 4). Figure 4 shows the lowest photolytic–catalytic ratio in the afternoon, which reflects the production of nitrates and other reactive nitrogen compounds from ${\mathrm{NO}}_{\mathit{\text{x}}}$ in the daytime. When photolytic measurements are available, we only use the photolytic observations in this study; otherwise, we use the scaled catalytic measurements.

Nineteen surface ${\mathrm{O}}_{\mathrm{3}}$ monitoring sites were operating in the DISCOVER-AQ region during the campaign (Fig. 1). They measured ${\mathrm{O}}_{\mathrm{3}}$ concentrations by using a federal equivalent method (FEM) based on the UV absorption of ${\mathrm{O}}_{\mathrm{3}}$ (https://cfpub.epa.gov/si/si_public_file_download.cfm?p_download_id=520887&Lab=NERL, last access: 12 July 2021) with an uncertainty of 5 $\mathrm{ppb}$.

2.6

Aircraft measurements of ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles

In this study, we mainly use the ${\mathrm{NO}}_{\mathrm{2}}$ concentrations measured by the National Center for Atmospheric Research (NCAR) four-channel chemiluminescence instrument (P-CL) onboard the P-3B aircraft for the evaluation of REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles. The instrument has an ${\mathrm{NO}}_{\mathrm{2}}$ measurement uncertainty of 10 %–15 % and a 1 $\mathrm{s}$, 1$\mathit{\sigma }$ detection limit of 30 $\mathrm{pptv}$.

${\mathrm{NO}}_{\mathrm{2}}$ measurements from aircraft spirals provide us with ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles. Figure 1 shows the locations of the aircraft spirals during the DISCOVER-AQ campaign, except for the Chesapeake Bay spirals over the ocean. There were only six spirals available over the Chesapeake Bay, which have ignorable impacts on the following analyses. Therefore, we do not use them in this study. The remaining 239 spirals in the daytime for July 2011 are used to compute the average profiles of ${\mathrm{NO}}_{\mathrm{2}}$ for the six inland sites (Fig. 1).

The aircraft measurements were generally sampled from a height of about 300 $\mathrm{m}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ (above ground level) in the boundary layer to 3.63 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ in the free troposphere. We bin these measurements to REAM vertical levels. In order to make up the missing observations between the surface and 300 $\mathrm{m}$, we apply quadratic polynomial regressions by using aircraft data below 1 $\mathrm{km}$ and coincident ${\mathrm{NO}}_{\mathrm{2}}$ surface measurements.

Table 1

Comparison of the concentrations of ${\mathrm{NO}}_{\mathit{\text{y}}}$ and its components between REAM and P-3B aircraft measurements during the DISCOVER-AQ campaign.

			${\mathrm{NO}}_{\mathit{\text{y}}}$ ($\mathrm{ppb}$${}^{\mathrm{1}}$)	NO ($\mathrm{ppb}$)	${\mathrm{NO}}_{\mathrm{2}}\mathrm{\_}\text{NCAR}$ ($\mathrm{ppb}$)	${\mathrm{NO}}_{\mathrm{2}}\mathrm{\_}\text{LIF}$ ($\mathrm{ppb}$${}^{\mathrm{2}}$)	$\sum \text{PNs}$ ($\mathrm{ppb}$)	$\sum \text{ANs}$ ($\mathrm{ppb}$)	${\mathrm{HNO}}_{\mathrm{3}}$ ($\mathrm{ppb}$)	Derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ ($\mathrm{ppb}$${}^{\mathrm{3}}$)
36 $\mathrm{km}$${}^{\mathrm{4}}$	Weekday${}^{\mathrm{5}}$	P-3B	2.51 $\pm$ 2.09	0.18 $\pm$ 0.29	0.85 $\pm$ 1.13	0.68 $\pm$ 0.95	0.70 $\pm$ 0.58	0.31 $\pm$ 0.23	1.15 $\pm$ 0.73	2.86 $\pm$ 2.26
		REAM	3.64 $\pm$ 3.13	0.18 $\pm$ 0.30	0.74 $\pm$ 1.04	0.68 $\pm$ 0.89	0.54 $\pm$ 0.45	0.10 $\pm$ 0.09	1.80 $\pm$ 1.61	3.10 $\pm$ 2.70
		$R^{\mathrm{2}}$	0.33	0.35	0.38	0.34	0.37	0.38	0.24	0.41
	Weekend	P-3B	3.00 $\pm$ 2.18	0.15 $\pm$ 0.20	0.71 $\pm$ 0.80	0.63 $\pm$ 0.72	0.91 $\pm$ 0.53	0.36 $\pm$ 0.21	1.15 $\pm$ 0.79	2.96 $\pm$ 2.15
		REAM	3.78 $\pm$ 2.20	0.15 $\pm$ 0.17	0.54 $\pm$ 0.59	0.53 $\pm$ 0.58	0.53 $\pm$ 0.29	0.09 $\pm$ 0.06	2.31 $\pm$ 1.38	3.43 $\pm$ 2.26
		$R^{\mathrm{2}}$	0.29	0.28	0.41	0.45	0.27	0.39	0.50	0.51
4 $\mathrm{km}$	Weekday	P-3B	2.51 $\pm$ 2.15	0.19 $\pm$ 0.30	0.86 $\pm$ 1.27	0.68 $\pm$ 0.98	0.70 $\pm$ 0.59	0.31 $\pm$ 0.22	1.17 $\pm$ 0.74	2.90 $\pm$ 2.27
		REAM	3.81 $\pm$ 3.81	0.19 $\pm$ 0.35	0.79 $\pm$ 1.31	0.76 $\pm$ 1.20	0.46 $\pm$ 0.51	0.08 $\pm$ 0.10	2.03 $\pm$ 1.91	3.31 $\pm$ 3.28
		$R^{\mathrm{2}}$	0.28	0.22	0.26	0.32	0.37	0.29	0.38	0.47
	Weekend	P-3B	2.96 $\pm$ 2.13	0.14 $\pm$ 0.18	0.69 $\pm$ 0.74	0.63 $\pm$ 0.71	0.91 $\pm$ 0.51	0.35 $\pm$ 0.21	1.15 $\pm$ 0.80	2.94 $\pm$ 2.09
		REAM	4.36 $\pm$ 3.66	0.25 $\pm$ 0.40	0.85 $\pm$ 1.28	0.81 $\pm$ 1.23	0.41 $\pm$ 0.29	0.08 $\pm$ 0.08	2.54 $\pm$ 1.99	3.72 $\pm$ 3.52
		$R^{\mathrm{2}}$	0.21	0.15	0.19	0.18	0.16	0.23	0.38	0.37

${}^{\mathrm{1}}$ For P-3B, the concentrations of ${\mathrm{NO}}_{\mathit{\text{y}}}$, NO, and ${\mathrm{NO}}_{\mathrm{2}}\mathrm{\_}\text{NCAR}$ were measured by using the NCAR four-channel chemiluminescence instrument. The measurement uncertainties are 10 %, 10 %–15 %, and 10 % for NO, ${\mathrm{NO}}_{\mathrm{2}}$, and ${\mathrm{NO}}_{\mathit{\text{y}}}$, respectively. The 1 $\mathrm{s}$, 1$\mathit{\sigma }$ detection limits are 20, 30, and 20 $\mathrm{pptv}$ for NO, ${\mathrm{NO}}_{\mathrm{2}}$, and ${\mathrm{NO}}_{\mathit{\text{y}}}$, respectively (https://discover-aq.larc.nasa.gov/wp-content/uploads/sites/140/2020/05/Weinheimer20101005_DISCOVERAQ_AJW.pdf, last access: 28 June 2019). For REAM, ${\mathrm{NO}}_{\mathit{\text{y}}}$ is the sum of NO, ${\mathrm{NO}}_{\mathrm{2}}$, total peroxyacyl nitrates ($\sum \text{PNs}$), total alkyl nitrates ($\sum \text{ANs}$) (including alkyl nitrates and hydroxyalkyl nitrates), ${\mathrm{HNO}}_{\mathrm{3}}$, HONO, 2 $\times$ ${\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}$, ${\mathrm{HNO}}_{\mathrm{4}}$, first-generation C5 carbonyl nitrate (nighttime isoprene nitrate ISN1: ${\mathrm{C}}_{\mathrm{5}}{\mathrm{H}}_{\mathrm{8}}{\mathrm{NO}}_{\mathrm{4}}$), 2 $\times$ C5 dihydroxy dinitrate (DHDN: ${\mathrm{C}}_{\mathrm{5}}{\mathrm{H}}_{\mathrm{10}}{\mathrm{O}}_{\mathrm{8}}{\mathrm{N}}_{\mathrm{2}}$), methyl peroxy nitrate (MPN: ${\mathrm{CH}}_{\mathrm{3}}{\mathrm{O}}_{\mathrm{2}}{\mathrm{NO}}_{\mathrm{2}}$), propanone nitrate (PROPNN: ${\mathrm{CH}}_{\mathrm{3}}\mathrm{C}(=\mathrm{O}){\mathrm{CH}}_{\mathrm{2}}{\mathrm{ONO}}_{\mathrm{2}}$), nitrate from methyl vinyl ketone (MVKN: ${\mathrm{HOCH}}_{\mathrm{2}}\mathrm{CH}\left({\mathrm{ONO}}_{\mathrm{2}}\right)\mathrm{C}(=\mathrm{O}){\mathrm{CH}}_{\mathrm{3}}$), nitrate from methacrolein (MARCN: ${\mathrm{HOCH}}_{\mathrm{2}}\mathrm{C}\left({\mathrm{ONO}}_{\mathrm{2}}\right)\left({\mathrm{CH}}_{\mathrm{3}}\right)\mathrm{CHO}$), and ethanol nitrate (ETHLN: ${\mathrm{CHOCH}}_{\mathrm{2}}{\mathrm{ONO}}_{\mathrm{2}}$). ${}^{\mathrm{2}}$ For P-3B, the concentrations of ${\mathrm{NO}}_{\mathrm{2}}\mathrm{\_}\text{LIF}$, $\sum \text{PNs}$, $\sum \text{ANs}$, and ${\mathrm{HNO}}_{\mathrm{3}}$ were measured by applying the thermal-dissociation laser-induced fluorescence (TD-LIF) technique. The accuracy of TD-LIF measurements of ${\mathrm{NO}}_{\mathrm{2}}$, $\sum \text{PNs}$, $\sum \text{ANs}$, and ${\mathrm{HNO}}_{\mathrm{3}}$ is better than 15 %, and the detection limit for the sum of ${\mathrm{NO}}_{\mathrm{2}}$, $\sum \text{PNs}$, $\sum \text{ANs}$, and ${\mathrm{HNO}}_{\mathrm{3}}$ is $\sim$ 10 $\mathrm{ppt}$ 10 ${\mathrm{s}}^{-\mathrm{1}}$ (Day et al., 2002). ${}^{\mathrm{3}}$ To compare ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations from TD-LIF measurements with those from REAM, we calculate derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ as the sum of NO, ${\mathrm{NO}}_{\mathrm{2}}\mathrm{\_}\text{LIF}$, $\sum \text{PNs}$, $\sum \text{ANs}$, and ${\mathrm{HNO}}_{\mathrm{3}}$. Only when the concentrations of all the five species are available at the same hour in the same grid cell can we calculate derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ at the given hour in the given grid cell. Therefore, in Table 1, the averaged derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ values are not exactly equal to the sum of averaged NO, ${\mathrm{NO}}_{\mathrm{2}}\mathrm{\_}\text{LIF}$, $\sum \text{PNs}$, $\sum \text{ANs}$, and ${\mathrm{HNO}}_{\mathrm{3}}$ concentrations that only depend on the availability of a single species. In addition, the measurement times and frequencies between ${\mathrm{NO}}_{\mathit{\text{y}}}$ and derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ are not the same. A comparison between these two types of data needs coincident sampling, as described in the main text. ${}^{\mathrm{4}}$ Mean ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions over the six P-3B spiral sites are close (relative difference $<$ 4 %) between the 36 and 4 $\mathrm{km}$ REAM (Table S1). ${}^{\mathrm{5}}$ Due to different sampling times and locations between weekdays and weekends, we do not recommend a direct comparison between weekday and weekend values here.

In addition to using ${\mathrm{NO}}_{\mathrm{2}}$ concentrations from the NCAR four-channel instrument to evaluate REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles, we also use P-3B NO, ${\mathrm{NO}}_{\mathrm{2}}$, and ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations measured by the NCAR four-channel instrument and ${\mathrm{NO}}_{\mathrm{2}}$, total peroxyacyl nitrates ($\sum \text{PNs}$), total alkyl nitrates ($\sum \text{ANs}$) (including alkyl nitrates and hydroxyalkyl nitrates), and ${\mathrm{HNO}}_{\mathrm{3}}$ concentrations measured by the thermal-dissociation laser-induced fluorescence (TD-LIF) technique (Day et al., 2002; Thornton et al., 2000; Wooldridge et al., 2010) to evaluate the concentrations of ${\mathrm{NO}}_{\mathit{\text{y}}}$ from REAM (Table 1). All these P-3B measurements are vertically binned to REAM grid cells for comparisons with REAM results. In addition, below the P-3B spirals, four ${\mathrm{NO}}_{\mathit{\text{y}}}$ observation sites at Padonia, Edgewood, Beltsville, and Aldino were operating to provide continuous hourly ${\mathrm{NO}}_{\mathit{\text{y}}}$ surface concentrations during the campaign, which we also use to evaluate REAM-simulated ${\mathrm{NO}}_{\mathit{\text{y}}}$ surface concentrations in this study. We summarize the information of available observations at the 11 inland Pandora sites in Table S1.

3 Results and discussion

3.1 Evaluation of WRF-simulated meteorological fields

We evaluate the performances of the 36 $\mathrm{km}$ and nested 4 $\mathrm{km}$ WRF simulations using temperature, potential temperature, relative humidity (RH), and wind measurements from the P-3B spirals (Fig. 1) and precipitation data from the NCEP (National Centers for Environmental Prediction) Stage IV precipitation dataset. Generally, P-3B spirals range from $\sim$ 300 $\mathrm{m}$ to $\sim$ 3.63 $\mathrm{km}$ in height above the ground level (a.g.l.). As shown in Fig. S4, both the 36 $\mathrm{km}$ and nested 4 $\mathrm{km}$ WRF simulations simulate temperature well with $R^{\mathrm{2}}$ $=$ 0.98. Both WRF simulations show good agreement with P-3B measurements in $\mathit{U}$ wind (36 $\mathrm{km}$: $R^{\mathrm{2}}$ $=$ 0.77; 4 $\mathrm{km}$: $R^{\mathrm{2}}$ $=$ 0.76), $\mathit{V}$ wind (36 $\mathrm{km}$: $R^{\mathrm{2}}$ $=$ 0.79; 4 $\mathrm{km}$: $R^{\mathrm{2}}$ $=$ 0.78), wind speed (36 $\mathrm{km}$: $R^{\mathrm{2}}$ $=$ 0.67; 4 $\mathrm{km}$: $R^{\mathrm{2}}$ $=$ 0.67), and wind direction (Figs. S4 and S5). We further compare the temporal evolutions of vertical profiles for temperature, potential temperature, RH, $\mathit{U}$ wind, and $\mathit{V}$ wind below 3 $\mathrm{km}$ from the P-3B observations with those from the 36 $\mathrm{km}$ and nested 4 $\mathrm{km}$ WRF simulations in Fig. S6. Both WRF simulations capture the temporospatial variations in P-3B-observed vertical profiles well except that RH below 1.5 $\mathrm{km}$ is significantly underestimated between 09:00 and 17:00 LT in both WRF simulations. The evaluations above suggest that WRF-simulated wind fields are good and comparable at 4 and 36 $\mathrm{km}$ resolutions, but potential dry biases exist in both WRF simulations.

The NCEP Stage IV precipitation dataset provides hourly precipitation across the contiguous United States (CONUS) with a resolution of $\sim$ 4 $\mathrm{km}$ based on the merging of rain gauge data and radar observations (Lin and Mitchell, 2005; Nelson et al., 2016). The Stage IV dataset is useful for evaluating model simulations, satellite precipitation estimates, and radar precipitation estimates (Davis et al., 2006; Gourley et al., 2011; Kalinga and Gan, 2010; Lopez, 2011; Yuan et al., 2008). We obtain the Stage IV precipitation data for July 2011 from the NCAR/UCAR Research Data Archive (https://rda.ucar.edu/datasets/ds507.5/, last access: 28 December 2019). As shown in Figs. S7 and S8, both the 36 $\mathrm{km}$ and nested 4 $\mathrm{km}$ WRF simulations generally predict much less precipitation (in precipitation amount and duration) compared to the Stage IV data in July 2011 around the DISCOVER-AQ campaign region, especially for the nested 4 $\mathrm{km}$ WRF simulation, consistent with the aforementioned underestimated RH and dry bias in WRF simulations. The precipitation biases in the WRF model will affect REAM simulations of trace gases, leading to high biases of soluble species due to underestimated wet scavenging. Clouds interfere with satellite observations. Therefore, the precipitation bias does not affect model evaluations with satellite measurements of ${\mathrm{NO}}_{\mathrm{2}}$. Aircraft measurements were also taken on non-precipitating days.

Figure 5

Diurnal cycles of surface (a, c) ${\mathrm{NO}}_{\mathrm{2}}$ and (b, d) ${\mathrm{O}}_{\mathrm{3}}$ concentrations on (a, b) weekdays and (c, d) weekends during the DISCOVER-AQ campaign in the DISCOVER-AQ region (the 36 $\mathrm{km}$ grid cells over the 11 inland Pandora sites shown in Fig. 1). Black lines denote the mean observations from all the 11 ${\mathrm{NO}}_{\mathrm{2}}$ surface monitoring sites and 19 ${\mathrm{O}}_{\mathrm{3}}$ surface sites during the campaign (Fig. 1), as mentioned in Sect. 2.5. “REAM-raw” (blue lines) denotes the coincident 36 $\mathrm{km}$ REAM simulation results with WRF-YSU-simulated $k_{\text{zz}}$ data, and “REAM-kzz” (red lines) is the coincident 36 $\mathrm{km}$ REAM simulation results with updated $k_{\text{zz}}$ data. See the main text for details. Vertical bars denote corresponding standard deviations.

[Figure omitted. See PDF]

3.2

Effect of boundary layer vertical mixing on the diurnal variations in surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations

3.2.1

36 $\mathrm{km}$ model simulation in comparison to the surface observations

Figure 5a and b show the observed and 36 $\mathrm{km}$ REAM-simulated diurnal cycles of surface ${\mathrm{NO}}_{\mathrm{2}}$ and ${\mathrm{O}}_{\mathrm{3}}$ concentrations on weekdays in July 2011 in the DISCOVER-AQ region. REAM with WRF-YSU-simulated vertical diffusion coefficient ($k_{\text{zz}}$) values significantly overestimates ${\mathrm{NO}}_{\mathrm{2}}$ concentrations and underestimates ${\mathrm{O}}_{\mathrm{3}}$ concentrations at night, although it captures the patterns of the diurnal cycles of surface ${\mathrm{NO}}_{\mathrm{2}}$ and ${\mathrm{O}}_{\mathrm{3}}$: an ${\mathrm{O}}_{\mathrm{3}}$ peak and an ${\mathrm{NO}}_{\mathrm{2}}$ minimum around noontime. Here, YSU denotes the Yonsei University planetary boundary layer (PBL) scheme (Shin and Hong, 2011) used by our WRF simulations (Table S2). At night, the reaction of ${\mathrm{O}}_{\mathrm{3}}$ $+$ $\mathrm{NO}$ $\to$ ${\mathrm{O}}_{\mathrm{2}}$ $+$ ${\mathrm{NO}}_{\mathrm{2}}$ produces ${\mathrm{NO}}_{\mathrm{2}}$ but removes ${\mathrm{O}}_{\mathrm{3}}$. Since most ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions are in the form of NO, the model biases of low ${\mathrm{O}}_{\mathrm{3}}$ and high ${\mathrm{NO}}_{\mathrm{2}}$ occur at the same time. Since there are no significant chemical sources of ${\mathrm{O}}_{\mathrm{3}}$ at night, mixing of ${\mathrm{O}}_{\mathrm{3}}$-rich air above the surface is the main source of ${\mathrm{O}}_{\mathrm{3}}$ supply near the surface. Therefore, the nighttime model biases with WRF-YSU-simulated $k_{\text{zz}}$ data in Fig. 5 indicate that vertical mixing may be underestimated at night.

Figure 6

ELF-observed and model-simulated diurnal variations in MLH at the UMBC site during the Discover-AQ campaign. “ELF MLH” denotes ELF-derived MLHs by using the covariance wavelet transform method. “WRF-YSU MLH” denotes the 36 $\mathrm{km}$ WRF-YSU $k_{\text{zz}}$-determined MLHs, and “Updated MLH” denotes updated $k_{\text{zz}}$-determined MLHs. See the main text for details. Vertical bars denote standard deviations. For the ELF MLHs, there are 13 506 1 $\min$ measurements in total during the campaign, and we bin them into hourly data. The green line corresponding to the right $y$ axis shows the diurnal variations in the number of hourly ELF data points.

[Figure omitted. See PDF]

During the DISCOVER-AQ campaign, WRF-simulated vertical wind velocities are very low at night and have little impact on vertical mixing (Fig. S9a). The nighttime vertical mixing is mainly attributed to turbulent mixing. However, Hu et al. (2012) found that the YSU scheme underestimated nighttime PBL vertical turbulent mixing in WRF, which is consistent with Fig. 6, showing that WRF-YSU $k_{\text{zz}}$-determined mixed-layer heights (MLHs) are significantly lower than lidar observations in the late afternoon and at night at the UMBC site during the DISCOVER-AQ campaign (Knepp et al., 2017). Here, the $k_{\text{zz}}$-determined MLH refers to the mixing height derived by comparing $k_{\text{zz}}$ to its background values (Hong et al., 2006) but not the PBLH outputs from WRF. UMBC is an urban site (Table S1), surrounded by a mixture of constructed materials and vegetation. The UMBC lidar MLH data were derived from the Elastic Lidar Facility (ELF) attenuated backscatter signals by using the covariance wavelet transform (CWT) method and had been validated against radiosonde measurements ($N$ (number of data points) $=$ 24; $R^{\mathrm{2}}$ $=$ 0.89; bias (ELF – radiosonde) $=$ 0.03 $\pm$ 0.23 $\mathrm{km}$), radar wind profiler observations ($N$ $=$ 659; $R^{\mathrm{2}}$ $=$ 0.78; bias $=$ $-$0.21 $\pm$ 0.36 $\mathrm{km}$), and Sigma Space mini-micropulse lidar data ($N$ $=$ 8122; $R^{\mathrm{2}}$ $=$ 0.85; bias $=$ 0.02 $\pm$ 0.22 $\mathrm{km}$) from the Howard University Beltsville Research Campus (HUBRC) in Beltsville, Maryland (38.058${}^{\circ }$ N, 76.888${}^{\circ }$ W) in the daytime during the DISCOVER-AQ campaign (Compton et al., 2013). It is noteworthy that although CWT is not designed to detect the nocturnal boundary layer (NBL), it does consider the residue layer (RL) and distinguish it from MLH in the early morning after sunrise, which is similar to nighttime conditions. Therefore, CWT can detect nighttime MLHs, although with large uncertainties due to the hard-coded assumption of RL $=$ 1 $\mathrm{km}$ in the algorithm and insufficient vertical resolution of the technique. In addition, the sunrise and sunset time in July 2011 is about 05:00 and 19:30 LT (https://gml.noaa.gov/grad/solcalc/sunrise.html, last access: 27 May 2021), respectively. Figure 6 shows that WRF-YSU $k_{\text{zz}}$-determined MLHs are significantly lower than ELF observations after sunrise between 05:00 and 08:00 LT and before sunset between 18:00 and 20:00 LT. Even if we do not consider MLHs at night (19:30–05:00 LT), we can still conclude that WRF-YSU underestimates vertical mixing in the early morning after sunrise and the late afternoon before sunset, enabling a reasonable assumption that WRF-YSU also underestimates nighttime vertical mixing. Moreover, the nighttime MLHs in Fig. 6 are comparable to those measured by the Vaisala CL51 ceilometer at the Chemistry And Physics of the Atmospheric Boundary Layer Experiment (CAPABLE) site in Hampton, Virginia (Knepp et al., 2017). Finally, we want to emphasize that different definitions of NBL can result in significantly different NBL heights (Breuer et al., 2014). In this study, we follow Knepp et al. (2017) to use MLHs derived from aerosol backscatter signals as the measure of vertical pollutant mixing within the boundary layer, which is simulated by $k_{\text{zz}}$ in REAM.

To improve nighttime PBL vertical turbulent mixing in REAM, we increase $k_{\text{zz}}$ below 500 $\mathrm{m}$ between 18:00 and 05:00 LT to 5 $\mathrm{m}{\mathrm{s}}^{-\mathrm{2}}$ if the WRF-YSU-computed $k_{\text{zz}}$ $<$ 5 $\mathrm{m}{\mathrm{s}}^{-\mathrm{2}}$, which significantly increases the $k_{\text{zz}}$-determined MLHs at night (Fig. 6), leading to the decreases in simulated surface ${\mathrm{NO}}_{\mathrm{2}}$ and the increases in surface ${\mathrm{O}}_{\mathrm{3}}$ concentrations at night (Fig. 5). The assigned value of 5 $\mathrm{m}{\mathrm{s}}^{-\mathrm{2}}$ is arbitrary. Changing this value to 2 or 10 $\mathrm{m}{\mathrm{s}}^{-\mathrm{2}}$ can also alleviate the biases of model-simulated nighttime surface ${\mathrm{NO}}_{\mathrm{2}}$ and ${\mathrm{O}}_{\mathrm{3}}$ concentrations (Fig. S10). Considering the potential uncertainties in nighttime ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions, an alternative solution to correct the model nighttime simulation biases is to reduce ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions, which can decrease the consumption of ${\mathrm{O}}_{\mathrm{3}}$ through the process of ${\mathrm{NO}}_{\mathit{\text{x}}}$ titration mentioned above (${\mathrm{O}}_{\mathrm{3}}$ $+$ $\mathrm{NO}$ $\to$ ${\mathrm{O}}_{\mathrm{2}}$ $+$ ${\mathrm{NO}}_{\mathrm{2}}$). Our sensitivity tests (not shown) indicate that it is necessary to reduce ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions by 50 %–67 % to eliminate the model nighttime simulation biases, but we cannot find good reasons to justify this level of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission reduction only at night.

The updated REAM simulation of surface ${\mathrm{NO}}_{\mathrm{2}}$ diurnal pattern in Fig. 5a is in good agreement with previous studies (Anderson et al., 2014; David and Nair, 2011; Gaur et al., 2014; Reddy et al., 2012; Zhao et al., 2019). Daytime surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations are much lower compared to nighttime, and ${\mathrm{NO}}_{\mathrm{2}}$ concentrations reach a minimum around noontime. As shown in Fig. S11, under the influence of vertical turbulent mixing, the surface layer ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission diurnal pattern is similar to the surface ${\mathrm{NO}}_{\mathrm{2}}$ diurnal cycle in Fig. 5a, emphasizing the importance of turbulent mixing on modulating surface ${\mathrm{NO}}_{\mathrm{2}}$ diurnal variations. The highest boundary layer (Fig. 6) due to solar radiation leads to the lowest surface layer ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions (Fig. S11), and, therefore, the smallest surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations occur around noontime (Fig. 5a). Transport, which is mainly attributed to advection and turbulent mixing, is another critical factor affecting surface ${\mathrm{NO}}_{\mathrm{2}}$ diurnal variations (Fig. S11). The magnitudes of transport fluxes (Fig. S11) are proportional to horizontal and vertical gradients of ${\mathrm{NO}}_{\mathit{\text{x}}}$ concentrations and are therefore generally positively correlated to surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations. However, some exceptions exist, reflecting different strengths of advection ($\mathit{U}$, $\mathit{V}$, and $\mathit{W}$) and turbulent mixing ($k_{\text{zz}}$) at different times. For example, in the early morning, ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations peak between 05:00 and 06:00 LT (Fig. 5a), while transport fluxes peak between 07:00 and 08:00 LT (Fig. S11). The delay of the peak is mainly due to lower turbulent mixing between 05:00 and 06:00 LT than other daytime hours in the model (Fig. 6). Chemistry also contributes to surface ${\mathrm{NO}}_{\mathrm{2}}$ diurnal variations mainly through photochemical sinks in the daytime and ${\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}$ hydrolysis at nighttime. Chemistry fluxes in Fig. S11 are not only correlated to the strength of photochemical reactions and ${\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}$ hydrolysis (chemistry fluxes per unit ${\mathrm{NO}}_{\mathit{\text{x}}}$) but are also proportional to ${\mathrm{NO}}_{\mathit{\text{x}}}$ surface concentrations. Therefore, chemistry fluxes in Fig. S11 cannot directly reflect the impact of solar radiation on photochemical reactions. It can, however, still be identified by comparing afternoon chemistry contributions: from 13:00 to 15:00 LT, surface layer ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions and ${\mathrm{NO}}_{\mathrm{2}}$ concentrations are increasing (Figs. S11 and 5a); however, chemistry losses are decreasing as a result of the reduction in photochemical sinks with weakening solar radiation. The contributions of vertical mixing and photochemical sinks to ${\mathrm{NO}}_{\mathrm{2}}$ concentrations can be further corroborated by daytime variations in ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles and TVCDs discussed in Sects. 3.3 and 3.4.

Figure 5c shows the diurnal variation on weekends is also simulated well in the improved 36 $\mathrm{km}$ model. The diurnal variation in surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations (REAM: 1.5–10.2 $\mathrm{ppb}$; observations: 2.1–9.8 $\mathrm{ppb}$) is lower than on weekdays (REAM: 2.4–12.2 $\mathrm{ppb}$; observations: 3.3–14.5 $\mathrm{ppb}$), reflecting lower magnitude and variation in ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions on weekends (Fig. 3). Figure 5d also shows an improved simulation of surface ${\mathrm{O}}_{\mathrm{3}}$ concentrations at nighttime due to the improved MLH simulation (Fig. 6).

Figure 7

Diurnal cycles of observed and simulated average surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations over Padonia, Oldtown, Essex, Edgewood, Beltsville, and Aldino (Table S1) on (a) weekdays and (b) weekends. Black lines denote mean observations from the six sites. Red lines denote coincident 36 $\mathrm{km}$ REAM simulation results, and blue lines are for coincident 4 $\mathrm{km}$ REAM simulation results. Error bars denote standard deviations.

[Figure omitted. See PDF]

3.2.2

4 $\mathrm{km}$ model simulation in comparison to the surface observations

The results of 4 $\mathrm{km}$ REAM simulations with original WRF-YSU $k_{\text{zz}}$ (not shown) are very similar to Fig. 5 since WRF-simulated nocturnal vertical mixing is insensitive to the model horizontal resolution. Applying the modified nocturnal mixing in the previous section also greatly reduced the nighttime ${\mathrm{NO}}_{\mathrm{2}}$ overestimate and ${\mathrm{O}}_{\mathrm{3}}$ underestimate in the 4 $\mathrm{km}$ REAM simulations. All the following analyses are based on REAM simulations with improved nocturnal mixing. Figure 7 shows that mean surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations simulated in the 4 $\mathrm{km}$ model are higher than the 36 $\mathrm{km}$ results over Padonia, Oldtown, Essex, Edgewood, Beltsville, and Aldino (Table S1), leading to generally higher biases compared to the observations in the daytime. A major cause is that the observation sites are located in regions of high ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions (Fig. 2). At a higher resolution of 4 $\mathrm{km}$, the high emissions around the surface sites are apparent compared to rural regions. At the coarser 36 $\mathrm{km}$ resolution, spatial averaging greatly reduces the emissions around the surface sites. On average, ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions ($\mathrm{molecules}{\mathrm{km}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$) around the six surface ${\mathrm{NO}}_{\mathrm{2}}$ observation sites are 67 % higher in the 4 $\mathrm{km}$ than the 36 $\mathrm{km}$ REAM simulations (Table S1). The resolution dependence of model results will be further discussed in the model evaluations using the other in situ and remote sensing measurements.

Figure 8

Temporal evolutions of ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles below 3 $\mathrm{km}$ on (a, c, e) weekdays and (b, d, f) weekends from the (a, b) P-3B aircraft and (c, d) 36 $\mathrm{km}$ and (e, f) 4 $\mathrm{km}$ REAM during the DISCOVER-AQ campaign. Horizontal bars denote the corresponding standard deviations. In (a, b), dots denote aircraft measurements, while lines below 1 $\mathrm{km}$ are based on quadratic polynomial fitting, as described in Sect. 2.6. The fitting values are mostly in reasonable agreement with the aircraft and surface measurements in the boundary layer. On weekends, no aircraft observations were made between 06:00 and 08:00 LT, and therefore no corresponding model profiles are shown.

[Figure omitted. See PDF]

Figure 9

Contributions of emission, chemistry, transport, and dry deposition to ${\mathrm{NO}}_{\mathit{\text{x}}}$ TVCD diurnal variations over the 11 inland Pandora sites (Table S1 and Fig. 1) on weekdays in July 2011 for the (a) 36 $\mathrm{km}$ and (b) 4 $\mathrm{km}$ REAM simulations. “Chem” refers to net ${\mathrm{NO}}_{\mathit{\text{x}}}$ chemistry production; “Emis” refers to ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions; “Drydep” denotes ${\mathrm{NO}}_{\mathit{\text{x}}}$ dry depositions; “Transport” includes advection, turbulent mixing, lightning ${\mathrm{NO}}_{\mathit{\text{x}}}$ production, and wet deposition. “Total (${\mathrm{NO}}_{\mathit{\text{x}}}$)” is the hourly change in ${\mathrm{NO}}_{\mathit{\text{x}}}$ TVCDs (($\mathrm{TVCD}$) $=$ ${\mathrm{TVCD}}_{\mathit{\text{t}}+\mathrm{1}}$ $-$ ${\mathrm{TVCD}}_{\mathit{\text{t}}}$). “Total (${\mathrm{NO}}_{\mathrm{2}}$)” is the hourly change in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs, and “Total (NO)” is the hourly change in NO TVCDs.

[Figure omitted. See PDF]

3.3

Diurnal variations in ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles

Figure 8a and c show the temporal variations in P-3B-observed and 36 $\mathrm{km}$ REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles in the daytime on weekdays during the DISCOVER-AQ campaign. The 36 $\mathrm{km}$ REAM reproduces the observed characteristics of ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles well in the daytime ($R^{\mathrm{2}}$ $=$ 0.89), which are strongly affected by vertical mixing and photochemistry (Y. Zhang et al., 2016). When vertical mixing is weak in the early morning (06:00–08:00 LT), ${\mathrm{NO}}_{\mathrm{2}}$, released mainly from surface ${\mathrm{NO}}_{\mathit{\text{x}}}$ sources, is concentrated in the surface layer, and the vertical gradient is large. As vertical mixing becomes stronger after 08:00 LT, ${\mathrm{NO}}_{\mathrm{2}}$ concentrations below 500 $\mathrm{m}$ decrease significantly, while those over 500 $\mathrm{m}$ increase from 06:00–08:00 to 12:00–14:00 LT. It is noteworthy that MLHs and ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions are comparable between 12:00–14:00 and 15:00–17:00 LT (Figs. 3 and 6); however, ${\mathrm{NO}}_{\mathrm{2}}$ concentrations between 15:00 and 17:00 LT are significantly higher than between 12:00 and 14:00 LT in the whole boundary layer, reflecting the impact of the decreased photochemical loss of ${\mathrm{NO}}_{\mathit{\text{x}}}$ in the late afternoon. In fact, photochemical losses affect all the daytime ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles, which can be easily identified by ${\mathrm{NO}}_{\mathrm{2}}$ TVCD process diagnostics discussed in Sect. 3.4 (Fig. 9).

Figure 8b and d also show the observed and 36 $\mathrm{km}$ REAM-simulated vertical profiles on weekends. Similar to Figs. 5 and 7, observed and simulated concentrations of ${\mathrm{NO}}_{\mathrm{2}}$ are lower on weekends than on weekdays. Some of the variations from weekend profiles are due to a lower number of observations (47 spirals) on weekends. The overall agreement between the observed vertical profiles and 36 $\mathrm{km}$ model results is good on weekends ($R^{\mathrm{2}}$ $=$ 0.87). Between 15:00 and 17:00 LT, the model simulates a larger gradient than what the combination of aircraft and surface measurements indicates. It may be related to the somewhat underestimated MLHs in the late afternoon in the model (Fig. 6).

On weekdays, most simulated vertical profiles at the 4 $\mathrm{km}$ resolution (Fig. 8e) are similar to 36 $\mathrm{km}$ results in part because the average ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions over the six P-3B spiral sites are about the same, 4 % lower in the 4 $\mathrm{km}$ than the 36 $\mathrm{km}$ REAM simulations (Table S1). A clear exception is the 4 $\mathrm{km}$ REAM-simulated vertical profile Between 15:00 and 17:00 LT when the model greatly overestimates boundary layer ${\mathrm{NO}}_{\mathit{\text{x}}}$ mixing and concentrations. The main reason is that WRF-simulated vertical velocities ($\mathit{W}$) in the late afternoon are much larger in the 4 $\mathrm{km}$ simulation than the 36 $\mathrm{km}$ simulation (Fig. S9), which can explain the simulated fully mixed boundary layer between 15:00 and 17:00 LT. Since it is not designed to run at the 4 $\mathrm{km}$ resolution, and it is commonly assumed that convection can be resolved explicitly at high resolutions, the Kain–Fritsch (new Eta) convection scheme is not used in the nested 4 $\mathrm{km}$ WRF simulation (Table S2); it may be related to the large vertical velocities in the late afternoon, when thermal instability is the strongest. Appropriate convection parameterization is likely still necessary for 4 $\mathrm{km}$ simulations (Zheng et al., 2016), which may also help alleviate the underestimation of precipitation in the nested 4 $\mathrm{km}$ WRF simulation as discussed in Sect. 3.1.

The same rapid boundary layer mixing due to vertical transport is present in the 4 $\mathrm{km}$ REAM-simulated weekend vertical profile (Fig. 8f), although the mixing height is lower. Fewer spirals (47) and distinct transport effect due to different ${\mathrm{NO}}_{\mathrm{2}}$ horizontal gradients between the 4 and 36 $\mathrm{km}$ REAM simulations (discussed in detail in Sect. 3.6) may cause the overestimation of weekend profiles in the 4 $\mathrm{km}$ REAM simulation.

3.4

Daytime variation in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs

We compare satellite, P-3B aircraft, and model-simulated TVCDs with Pandora measurements, which provide continuous daytime observations. The locations of Pandora sites are shown in Table S1 and Fig. 1. Among the Pandora sites, four sites are located significantly above the ground level: UMCP ($\sim$ 20 $\mathrm{m}$), UMBC ($\sim$ 30 $\mathrm{m}$), SERC ($\sim$ 40 $\mathrm{m}$), and GSFC ($\sim$ 30 $\mathrm{m}$). The other sites are 1.5 $\mathrm{m}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ To properly compare Pandora to other measurements and model simulations, we calculate the missing TVCDs between the Pandora site heights and ground surface by multiplying the Pandora TVCDs with model-simulated TVCD fractions of the corresponding columns. The resulting correction is 2 %–21 % ($\frac{\mathrm{1}}{\mathrm{1}-\text{missing TVCD percentage}}$) for the four sites significantly above the ground surface, but the effect on the averaged daytime TVCD variation at all Pandora sites is small (Fig. S12). In the following analysis, we use the updated Pandora TVCD data.

Figure 10

Daily variations in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs on (a) weekdays and (b) weekends during the DISCOVER-AQ campaign. “REAM-36km” refers to the 36 $\mathrm{km}$ REAM simulation results over the 11 inland Pandora sites. “REAM-4 $\mathrm{km}$” refers to the 4 $\mathrm{km}$ REAM simulation results over the 11 inland Pandora sites. “Pandora” refers to updated Pandora TVCD data. “Flight” denotes P-3B-aircraft-derived ${\mathrm{NO}}_{\mathrm{2}}$ VCDs below 3.63 $\mathrm{km}$. “NASA-OMI” denotes the OMI ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs retrieved by NASA over the Pandora sites; “KNMI-OMI” denotes the OMI ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs from KNMI; “KNMI-GOME2” is the GOME-2A ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs from KNMI. “OMI-retrieval” and “GOME2-retrieval” denote OMI and GOME-2A TVCDs retrieved by using the KNMI DOMINO algorithm with corresponding 36 $\mathrm{km}$ REAM vertical profiles, respectively. The vertical bars denote corresponding standard deviations for all data except the 36 $\mathrm{km}$ REAM simulation results, the standard deviations of which are shown with pink shading. We list ${\mathrm{NO}}_{\mathrm{2}}$ TVCD values at 09:30 and 13:30 LT in Table S3.

[Figure omitted. See PDF]

The weekday diurnal variations in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs from satellites, Pandora, 4 and 36 $\mathrm{km}$ REAM, and the P-3B aircraft are shown in Fig. 10a. We calculate aircraft-derived TVCDs by using Eq. (1): 1$${\mathrm{TVCD}}_{\mathrm{aircraft}}\left(t\right)={\displaystyle \frac{\sum c_{\text{aircraft}}\left(t\right)\cdot {\mathit{\rho }}_{\text{REAM}}\left(t\right)\cdot V_{\text{REAM}}\left(t\right)}{A_{\text{REAM}}}},$$ where $t$ is time, $c_{\text{aircraft}}$ ($v/v$) denotes aircraft ${\mathrm{NO}}_{\mathrm{2}}$ concentrations (mixing ratios) at each level at time $t$, ${\mathit{\rho }}_{\text{REAM}}$ ($\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{3}}$) is the density of air from 36 $\mathrm{km}$ REAM at the corresponding level, $V_{\text{REAM}}$ (${\mathrm{cm}}^{\mathrm{3}}$) is the volume of the corresponding 36 $\mathrm{km}$ REAM grid cell, and $A_{\text{REAM}}$ (${\mathrm{cm}}^{\mathrm{2}}$) is the surface area (36 $\mathrm{km}$ $\times$ 36 $\mathrm{km}$). In the calculation, we only use ${\mathrm{NO}}_{\mathrm{2}}$ concentrations below 3.63 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ because few aircraft measurements were available above this height in the campaign. Missing tropospheric ${\mathrm{NO}}_{\mathrm{2}}$ above 3.63 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ in the aircraft TVCD calculation has little impact on our analyses as the 36 $\mathrm{km}$ REAM model simulation shows that 85 % $\pm$ 7 % of tropospheric ${\mathrm{NO}}_{\mathrm{2}}$ is located below 3.63 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ between 06:00 and 17:00 LT in the DISCOVER-AQ region, which is roughly consistent with the GMI model results with 85 %–90 % tropospheric ${\mathrm{NO}}_{\mathrm{2}}$ concentrated below 5 $\mathrm{km}$ (Lamsal et al., 2014). It should be noted that only six P-3B spirals are available during the campaign, less than the samplings of 11 inland Pandora sites.

The 4 $\mathrm{km}$ REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs are mostly higher than the 36 $\mathrm{km}$ results and the observations in daytime on weekdays (Fig. 10a). However, since the standard deviations of the data are much larger than the model difference, the 4 and 36 $\mathrm{km}$ model results generally show similar characteristics relative to the observations. REAM simulation results are in reasonable agreement with Pandora, P-3B aircraft, and satellite daytime ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs, except that NASA-derived OMI (OMNO2) TVCDs are somewhat lower than other datasets, which may be partly due to biased a priori vertical profiles from the GMI model in the NASA retrieval in the campaign (Lamsal et al., 2014, 2021). TVCDs derived by using the DOMINO algorithm and 36 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles are in agreement with those from KNMI, which indicates that the TM4 model from KNMI provides reasonable estimates of a priori ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles on weekdays in the DISCOVER-AQ region in summer.

We find evident decreases in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs from GOME-2A to OMI in Fig. 10a, which is consistent with Pandora, REAM results, and previous studies that showed decreasing ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs from SCIAMACHY to OMI due to photochemical losses in summer (Boersma et al., 2008, 2009). P-3B aircraft TVCDs also show this decrease feature but have large variations due in part to the limited aircraft sampling data.

Pandora ${\mathrm{NO}}_{\mathrm{2}}$ TVCD data have different characteristics from REAM-simulated and P-3B-aircraft-measured TVCDs between 05:00 and 07:00 and between 14:00 and 18:00 LT (Fig. 10a). Between 05:00 and 07:00 LT, Pandora data show a significant increase in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs, but REAM and aircraft TVCDs generally decrease, except for 4 $\mathrm{km}$ REAM TVCDs, with a slight increase from 06:00 to 07:00 LT on weekdays. Between 14:00 and 18:00 LT, Pandora TVCDs have little variation, but REAM and aircraft TVCDs increase significantly. The relatively flat Pandora TVCDs in the late afternoon compared to REAM and P-3B aircraft measurements are consistent with Lamsal et al. (2017), which found that Pandora VCDs were 26 %–30 % lower than UC-12 ACAM measurements from 16:00 to 18:00 LT during the DISCOVER-AQ campaign. We show the simulated effects of emission, chemistry, transport, and dry deposition on ${\mathrm{NO}}_{\mathit{\text{x}}}$ TVCDs in Fig. 9. The simulated early-morning slight decrease in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs is mainly due to the chemical transformation between ${\mathrm{NO}}_{\mathrm{2}}$ and NO favoring the accumulation of NO under low-${\mathrm{O}}_{\mathrm{3}}$ and low-${\mathrm{HO}}_{\mathrm{2}}$ and low-${\mathrm{RO}}_{\mathrm{2}}$ conditions; thus NO TVCDs increase significantly, but ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs continue decreasing slowly during the period. The increase in the late afternoon is primarily due to the decrease in photochemistry-related sinks. The reasons for the discrepancies of ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs between Pandora and REAM results during the above two periods are unclear. Large SZAs in the early morning and the late afternoon (Fig. S1) lead to the higher uncertainties in Pandora measurements (Herman et al., 2009), although we have excluded Pandora measurements with SZA $>$ 80${}^{\circ }$. In addition, Pandora is a sun-tracking instrument with a small effective FOV and is sensitive to local conditions within a narrow spatial range, which may differ significantly from the average properties of 36 and 4 $\mathrm{km}$ grid cells depending upon the time of the day (Fig. S13) (Herman et al., 2009, 2018, 2019; Judd et al., 2018, 2019, 2020; Lamsal et al., 2017; Reed et al., 2015). As we mentioned above, $\sim$ 85 % of tropospheric ${\mathrm{NO}}_{\mathrm{2}}$ is located below 3.63 $\mathrm{km}$ in the DISCOVER-AQ 2011 region based on the 36 $\mathrm{km}$ REAM simulation results. The Pandora FOV of 1.6${}^{\circ }$ is approximately equivalent to a nadir horizontal extension of only 0.1 $\mathrm{km}$ ($\mathrm{2}\times \mathrm{3.63}\mathrm{km}\times \tan \frac{\mathrm{1.6}}{\mathrm{2}}=\mathrm{0.1}$ $\mathrm{km}$) at 3.63 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ and 30 $\mathrm{m}$ at 1.0 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ Therefore, Pandora measures different air columns of ${\mathrm{NO}}_{\mathrm{2}}$ at different times of the day, especially in the morning and afternoon, when SZA is large, as shown in Fig. S13. Considering the potential spatial heterogeneity of boundary layer ${\mathrm{NO}}_{\mathrm{2}}$, it is possible that the morning (east), noontime (nadir), and afternoon (west) ${\mathrm{NO}}_{\mathrm{2}}$ VCDs are significantly different from each other. Unlike Pandora, satellites and aircraft are far from the ground surface and cover large areas; therefore, the impact of SZA on their ${\mathrm{NO}}_{\mathrm{2}}$ VCD measurements is insignificant compared to Pandora measurements. Another possible reason is that Pandora instruments had few observations in the early morning, and the resulting average may not be representative (Fig. S2).

Figure 11

Weekday hourly variations in ${\mathrm{NO}}_{\mathrm{2}}$ VCDs at different height (a.g.l.) bins ($<$ 3.63 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$, $<$ 300 $\mathrm{m}\mathrm{a}.\mathrm{g}.\mathrm{l}.$, and 300 $\mathrm{m}$ $\sim$ 3.63 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$) based on P-3B-aircraft-derived datasets and the 36 and 4 $\mathrm{km}$ REAM results. “Flight” denotes P-3B-aircraft-derived ${\mathrm{NO}}_{\mathrm{2}}$ VCDs, “REAM-36km” denotes coincident 36 $\mathrm{km}$ REAM-simulated VCDs, and “REAM-4 $\mathrm{km}$” denotes coincident 4 $\mathrm{km}$ REAM-simulated VCDs.

[Figure omitted. See PDF]

To further understand the daytime variation in ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs, we examine P-3B-aircraft-data-derived and REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ VCD variations for different height bins (Fig. 11). ${\mathrm{NO}}_{\mathrm{2}}$ VCDs below 3.63 $\mathrm{km}\mathrm{a}.\mathrm{g}.\mathrm{l}.$ display a U-shaped pattern from 05:00 to 17:00 LT. In the morning, as vertical mixing becomes stronger after sunrise, high-${\mathrm{NO}}_{\mathit{\text{x}}}$ air in the lower layer is mixed with low-${\mathrm{NO}}_{\mathit{\text{x}}}$ air in the upper layer. The increase in ${\mathrm{NO}}_{\mathit{\text{x}}}$ vertical mixing above 300 $\mathrm{m}$ is sufficient to counter the increase in photochemical loss in the morning. Conversely, the ${\mathrm{NO}}_{\mathrm{2}}$ VCDs below 300 $\mathrm{m}$ decrease remarkably from sunrise (about 06:00 LT) to around noontime due to both vertical mixing and the increase in photochemical strength. From 13:00 to 16:00 LT, ${\mathrm{NO}}_{\mathrm{2}}$ VCDs increase slowly, reflecting a relative balance among emissions, transport, chemistry, and dry depositions. The sharp jump of the VCDs from 16:00 to 17:00 LT is mainly due to dramatically reduced chemical loss. And 4 $\mathrm{km}$ REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ VCDs at 0.30–3.63 $\mathrm{km}$ between 16:00 and 17:00 LT are much higher than 36 $\mathrm{km}$ results partly because of the rapid vertical mixing in the 4 $\mathrm{km}$ REAM simulation (Figs. 8 and S9).

Similar to ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations and vertical profiles in Figs. 7 and 8, the ${\mathrm{NO}}_{\mathrm{2}}$ TVCD variation is also smaller on weekends than on weekdays, but the day–night pattern is similar (Fig. 10). Although the 4 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs are generally higher than the 36 $\mathrm{km}$ results and observations in the daytime, considering their large standard deviations, ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs from both simulations are comparable to satellite products, Pandora, and P-3B aircraft observations most of the time on weekends. The exception is that Pandora TVCDs have different variation patterns in the early morning and late afternoon from REAM simulations, similar to those found on weekdays.

3.5

Model comparisons with ${\mathrm{NO}}_{\mathit{\text{y}}}$ measurements

${\mathrm{NO}}_{\mathit{\text{y}}}$ is longer-lived than ${\mathrm{NO}}_{\mathit{\text{x}}}$, and ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations are not affected by chemistry as much as ${\mathrm{NO}}_{\mathit{\text{x}}}$. We obtain two types of ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations from the P-3B aircraft in the DISCOVER-AQ campaign: one is ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations directly measured by the NCAR four-channel instrument, corresponding to the sum of NO, ${\mathrm{NO}}_{\mathrm{2}}$, $\sum \text{PNs}$, $\sum \text{ANs}$, ${\mathrm{HNO}}_{\mathrm{3}}$, ${\mathrm{N}}_{\mathrm{2}}{\mathrm{O}}_{\mathrm{5}}$, ${\mathrm{HNO}}_{\mathrm{4}}$, HONO, and the other reactive nitrogenic species in REAM (all the other species are described in Table 1), and the other one, which we name “derived-${\mathrm{NO}}_{\mathit{\text{y}}}$”, is the sum of NO from the NCAR four-channel instrument and ${\mathrm{NO}}_{\mathrm{2}}$ (${\mathrm{NO}}_{\mathrm{2}}\mathrm{\_}\mathrm{LIF}$), $\sum \text{PNs}$, $\sum \text{ANs}$, and ${\mathrm{HNO}}_{\mathrm{3}}$ measured by the TD-LIF technique, corresponding to NO, ${\mathrm{NO}}_{\mathrm{2}}$, $\sum \text{PNs}$, $\sum \text{ANs}$, and ${\mathrm{HNO}}_{\mathrm{3}}$ in REAM (Table 1). On average, P-3B derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations (2.88 $\pm$ 2.24 $\mathrm{ppb}$) are 17 % higher than coincident P-3B ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations (2.46 $\pm$ 2.06 $\mathrm{ppb}$), with $R^{\mathrm{2}}$ $=$ 0.75, generally reflecting consistency between these two types of measurements. As shown in Table 1, on weekdays, the 36 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations are 45 % larger than P-3B, with $R^{\mathrm{2}}$ $=$ 0.33, and the 36 $\mathrm{km}$ REAM derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations are 8 % larger than P-3B, with $R^{\mathrm{2}}$ $=$ 0.41. The 4 $\mathrm{km}$ REAM shows similar results, suggesting that REAM simulations generally reproduce the observed ${\mathrm{NO}}_{\mathit{\text{y}}}$ and derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations within the uncertainties, although the average values from REAM are somewhat larger than the observations due in part to the underestimate of precipitation in the WRF model simulations resulting in underestimated wet scavenging of ${\mathrm{HNO}}_{\mathrm{3}}$ in REAM. The concentrations of weekday NO, ${\mathrm{NO}}_{\mathrm{2}}$, and $\sum \text{PNs}$ from REAM simulations are also comparable to the observations. However, weekday $\sum \text{ANs}$ concentrations are 68 % lower in the 36 $\mathrm{km}$ REAM than observations, suggesting that the chemistry mechanism in REAM may need further improvement to better represent isoprene nitrates. It is noteworthy that, since $\sum \text{ANs}$ only account for a small fraction ($\sim$ 11 %) of observed derived-${\mathrm{NO}}_{\mathit{\text{y}}}$, the absolute difference between REAM-simulated and P-3B-observed $\sum \text{ANs}$ concentrations is still small compared to ${\mathrm{HNO}}_{\mathrm{3}}$. Weekday ${\mathrm{HNO}}_{\mathrm{3}}$ concentrations are significantly higher in REAM simulations (36 $\mathrm{km}$: 57 %, 0.65 $\mathrm{ppb}$; 4 $\mathrm{km}$: 74 %, 0.86 $\mathrm{ppb}$) than P-3B observations, which is the main reason for the somewhat larger ${\mathrm{NO}}_{\mathit{\text{y}}}$ and derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations in REAM compared to P-3B observations. The higher ${\mathrm{HNO}}_{\mathrm{3}}$ concentrations in REAM may be related to the underestimation of precipitation in the corresponding WRF simulations, as discussed in Sect. 3.1 (Figs. S7 and S8), leading to the underestimated wet scavenging of ${\mathrm{HNO}}_{\mathrm{3}}$, especially for the 4 $\mathrm{km}$ REAM simulation.

We also examine the weekday diurnal variations in derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ vertical profiles from P-3B and REAM simulations in Fig. S14. Generally, both 36 and 4 $\mathrm{km}$ REAM simulations capture the variation characteristics of observed vertical profiles, which are similar to those for ${\mathrm{NO}}_{\mathrm{2}}$ in Fig. 8. REAM derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations are comparable to P-3B observations at most vertical levels on weekdays. Some larger derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations in the model results can be partially explained by larger ${\mathrm{HNO}}_{\mathrm{3}}$ concentrations in REAM, such as those below 1 $\mathrm{km}$ between 09:00 and 11:00 LT for the 36 $\mathrm{km}$ REAM and those below 2.0 $\mathrm{km}$ between 12:00 and 17:00 LT for the 4 $\mathrm{km}$ REAM (Fig. S15).

Figure 12

Diurnal cycles of observed and simulated average surface ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations at Padonia, Edgewood, Beltsville, and Aldino on (a) weekdays and (b) weekends. Vertical bars denote the corresponding standard deviations. It is noteworthy that the mean ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions over Padonia, Edgewood, Beltsville, and Aldino are 99 % higher in the 4 than the 36 $\mathrm{km}$ REAM simulations (Table S1).

[Figure omitted. See PDF]

Figure 12 shows the comparison of the diurnal cycles of surface ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations observed at Padonia, Edgewood, Beltsville, and Aldino during the DISCOVER-AQ campaign with those from the REAM simulations. Generally, the REAM simulations reproduce the observed surface ${\mathrm{NO}}_{\mathit{\text{y}}}$ diurnal cycles except for the spikes around 17:00–20:00 LT due to still-underestimated MLHs (Fig. 6). The 4 $\mathrm{km}$ simulation results have a higher bias than 36 $\mathrm{km}$ results relative to the observations in the daytime, similar to the comparisons of ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations and TVCDs in Figs. 7 and 10 due to higher emissions around the observation sites in 4 than 36 $\mathrm{km}$ simulations (Table S1 and Fig. 2).

Figure 13

Comparisons of ${\mathrm{NO}}_{\mathrm{2}}$ (a, c) TVCDs and (b, d) surface concentrations over the 11 inland Pandora sites between the 4 and 36 $\mathrm{km}$ REAM simulations on (a, b) weekdays and (c, d) weekends for July 2011. “REAM-36km” (black lines) denotes the 36 $\mathrm{km}$ REAM simulation results; “REAM-4 $\mathrm{km}$” (red lines) denotes the 4 $\mathrm{km}$ REAM simulation results; “4 $\mathrm{km}$-regrid” (blue lines) refers to the 36 $\mathrm{km}$ values by regridding the 4 $\mathrm{km}$ REAM simulation results into 36 $\mathrm{km}$ REAM grid cells. The vertical bars denote corresponding standard deviations for all data except the 36 $\mathrm{km}$ REAM simulation results, the standard deviations of which are shown with gray shading.

[Figure omitted. See PDF]

Figure 14

Distributions of the scaled mean (a) 4 $\mathrm{km}$ REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ VCDs below the UC-12 aircraft and (b) coincident ACAM measurements on weekdays in July 2011. (c) The distribution of the scaled NEI2011 ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions on weekdays. (d) The scatterplot of the scaled 4 $\mathrm{km}$ REAM and ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs from (a, b). Panel (e) shows the relative differences between (a) and (b) ($\text{REAM}/\text{ACAM}-\mathrm{1}$). (f) The distribution of the number of data points used to calculate grid cell mean ${\mathrm{NO}}_{\mathrm{2}}$ VCDs in (a, b). Here, we scale all values (VCDs and ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions) based on their corresponding domain averages. The domain averages of ACAM and coincident 4 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs are 4.7 $\pm$ 2.0 and 4.6 $\pm$ 3.2 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}$, respectively.

[Figure omitted. See PDF]

3.6

Resolution dependence of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission distribution

We showed previously that the 4 $\mathrm{km}$ REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ and ${\mathrm{NO}}_{\mathit{\text{y}}}$ surface concentrations and ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs are higher than observations in the daytime in comparison to the corresponding 36 $\mathrm{km}$ REAM results (Figs. 7, 10, and 12). An examination of monthly mean ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations and TVCDs for July 2011 also shows that 4 $\mathrm{km}$ simulation results are significantly higher than the 36 $\mathrm{km}$ results over the 11 inland Pandora sites in the daytime (Fig. 13). The process-level diagnostics in Fig. 9 indicate that the mean contribution of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions to ${\mathrm{NO}}_{\mathit{\text{x}}}\mathrm{\Delta }\text{TVCDs}$ in the 4 $\mathrm{km}$ simulation is 1.32 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}$ larger than that in the 36 $\mathrm{km}$ simulation between 09:00 and 16:00 LT, while the absolute mean contributions of chemistry and transport (they are negative in Fig. 9, so we use absolute values here) in the 4 $\mathrm{km}$ simulation are 0.26 $\times$ 10${}^{\mathrm{15}}$ and 0.87 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}$ larger than the 36 $\mathrm{km}$ simulation, respectively. The contributions of dry deposition to ${\mathrm{NO}}_{\mathit{\text{x}}}\mathrm{\Delta }\text{TVCDs}$ are negligible compared to other factors in both simulations (Fig. 9). Therefore, the 34 % higher ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions over the 11 inland Pandora sites (Table S1 and Fig. 3) are the main reason for the larger daytime ${\mathrm{NO}}_{\mathrm{2}}$ surface concentrations and TVCDs in the 4 $\mathrm{km}$ than the 36 $\mathrm{km}$ REAM simulations (Fig. 13). The significantly different contribution changes between ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions (1.32 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}$, or about one-third) and chemistry (0.26 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}$, or about 8 %) reflect potential chemical nonlinearity (Li et al., 2019; Silvern et al., 2019; Valin et al., 2011) and transport effect. Different transport contributions between the 4 $\mathrm{km}$ and the 36 $\mathrm{km}$ REAM are mainly caused by their different ${\mathrm{NO}}_{\mathit{\text{x}}}$ horizontal gradients (Figs. 2, 14, and 15), while the impact of wind fields is small since we do not find significant differences in horizontal wind components between the two simulations except for some lower wind speeds below 1000 $\mathrm{m}$ for the 36 $\mathrm{km}$ WRF simulation compared to the nested 4 $\mathrm{km}$ WRF simulation (Fig. S16). Our sensitivity tests with the WRF single-moment three-class (WSM3) simple ice scheme (not shown) can improve the wind speed comparison below 1000 $\mathrm{m}$ between the 36 $\mathrm{km}$ and nested 4 $\mathrm{km}$ WRF simulations but still produce similar ${\mathrm{NO}}_{\mathit{\text{x}}}$ simulation results as WSM6 shown here. Therefore, the somewhat lower wind speeds below 1000 $\mathrm{m}$ in the 36 $\mathrm{km}$ WRF simulation are not the reason for the difference between the 4 and 36 $\mathrm{km}$ REAM simulations. The impact of transport on the two REAM simulations can be further verified by the comparison of ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs over the six P-3B spiral sites between the two simulations (Fig. S17). Mean ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions over the six P-3B spiral sites are close (relative difference $<$ 4 %) between the two simulations (Table S1 and Fig. S17). From 09:00 to 12:00 LT, the contributions of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions to ${\mathrm{NO}}_{\mathit{\text{x}}}\mathrm{\Delta }\text{TVCDs}$ are 2.50 $\times$ 10${}^{\mathrm{15}}$ and 2.49 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}$ for the 36 and 4 $\mathrm{km}$ REAM simulations, respectively, and the contributions of chemistry are also close between the two simulations (36 $\mathrm{km}$: $-$2.62 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}$; 4 $\mathrm{km}$: $-$2.69 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}$). However, the contributions of transport are $-$0.39 $\times$ 10${}^{\mathrm{15}}$ and 0.03 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}{\mathrm{h}}^{-\mathrm{1}}$ for the 36 and 4 $\mathrm{km}$ REAM simulations, respectively, leading to larger ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs in the 4 $\mathrm{km}$ REAM simulation than the 36 $\mathrm{km}$ REAM from 09:00–12:00 LT (Fig. S17c). Since horizontal wind fields over the six P-3B spiral sites are comparable between two simulations (Figs. S4, S5, S6, and S16), and larger ${\mathrm{NO}}_{\mathit{\text{x}}}$ horizontal gradients are found near the P-3B spiral sites for the 4 $\mathrm{km}$ REAM (Fig. 2), we attribute the different transport contributions between the two simulations to a much larger ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission gradient around the measurement locations in 4 than 36 $\mathrm{km}$ emission distributions.

We regrid the 4 $\mathrm{km}$ REAM results into the grid cells of the 36 $\mathrm{km}$ REAM, which can significantly reduce the impact of different ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission distributions and associated transport on the two simulations. Compared to the original 4 $\mathrm{km}$ REAM results, the regridded surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations and TVCDs over the 11 inland Pandora sites are much closer to the 36 $\mathrm{km}$ REAM results (Fig. 13). After regridding the 4 $\mathrm{km}$ REAM results into 36 $\mathrm{km}$ REAM grid cells, we also find more comparable ${\mathrm{NO}}_{\mathit{\text{y}}}$ surface concentrations between the regridded 4 $\mathrm{km}$ results and the 36 $\mathrm{km}$ REAM results (Fig. S18). The remaining discrepancies between the regridded results and the 36 $\mathrm{km}$ REAM results may be due to chemical nonlinearity and other meteorological effects, such as larger vertical wind in the 4 $\mathrm{km}$ REAM (Fig. S9) and their different $k_{\text{zz}}$ values in the PBL. Although other factors, such as chemical nonlinearity and vertical diffusion, may affect the 36 and 4 $\mathrm{km}$ REAM simulations differently, the difference between 4 and 36 $\mathrm{km}$ simulations of reactive nitrogen is largely due to that of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions.

The 4 and 36 $\mathrm{km}$ simulation difference depends on the location of the observations. In some regions, the ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission difference between 4 and 36 $\mathrm{km}$ simulations is small. The comparison of ${\mathrm{NO}}_{\mathit{\text{y}}}$ measurements from P-3B spirals with coincident REAM results in Table 1 suggests that the 4 and 36 $\mathrm{km}$ REAM simulations produce similar ${\mathrm{NO}}_{\mathit{\text{y}}}$ (relative difference $\sim$ 4 %) and derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ (relative difference $\sim$ 6 %) concentrations on weekdays, and both simulation results are comparable to the observations. The ${\mathrm{NO}}_{\mathit{\text{y}}}$ similarity over the P-3B spiral sites between the 36 and 4 $\mathrm{km}$ REAM simulations is consistent with the comparable ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions over (relative difference $<$ 4 %) the six P-3B spiral sites between the two simulations (Table S1). The differences between the 4 $\mathrm{km}$ model simulation results and P3-B observations are larger on weekends than on weekdays (Table 1) due to the limited weekend sampling since model-simulated monthly mean values show similar differences between the 4 and 36 $\mathrm{km}$ REAM simulations on weekends as on weekdays (not shown).

3.7

Evaluation of 36 and 4 $\mathrm{km}$ ${\mathrm{NO}}_{\mathit{\text{x}}}$ distribution with OMI, GOME-2A, and ACAM measurements

The evaluation of model simulations of surface, aircraft, and satellite observations tends to point out a higher bias in 4 than 36 $\mathrm{km}$ model simulations. We note that this comparison is based on the averages of multiple sites. ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions at individual sites are not always higher in the 4 than 36 $\mathrm{km}$ REAM, such as SERC, Fairhill, and Essex, with much higher 36 $\mathrm{km}$ ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions than 4 $\mathrm{km}$ ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions (Table S1). We conduct individual-site comparisons of surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations, surface ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations, ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles, derived-${\mathrm{NO}}_{\mathit{\text{y}}}$ vertical profiles, and ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs of the 36 $\mathrm{km}$ REAM and the 4 $\mathrm{km}$ REAM results relative to the corresponding observations in Figs. S19–S23. The 36 $\mathrm{km}$ simulation results can be larger, smaller, or comparable to the 4 $\mathrm{km}$ simulation results, and both simulations can produce higher, lower, or similar results as the observations for different variables at different sites. The varying model biases depending on the observation site reflect the different spatial distributions of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions between the 36 and 4 $\mathrm{km}$ REAM simulations (Fig. 2) and suggest potential distribution biases of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions in both simulations.

Here we examine the 4 $\mathrm{km}$ model-simulated ${\mathrm{NO}}_{\mathrm{2}}$ VCDs with high-resolution ACAM measurements onboard the UC-12 aircraft in Figs. 14 and S24, respectively. The spatial distributions of ACAM and 4 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs are generally consistent with $R^{\mathrm{2}}$ $=$ 0.35 on weekdays and $R^{\mathrm{2}}$ $=$ 0.50 on weekends. The domain averages of ACAM and 4 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs are 4.7 $\pm$ 2.0 and 4.6 $\pm$ 3.2 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}$ on weekdays and 3.0 $\pm$ 1.7 and 3.3 $\pm$ 2.7 $\times$ 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}$ on weekends, respectively. The spatial distributions of ACAM and 4 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs are highly correlated with the spatial distribution of 4 $\mathrm{km}$ NEI2011 ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions. All three distributions capture two strong peaks around Baltimore and Washington, DC, urban regions and another weak peak in the northeast corner of the domain (Wilmington, Delaware) (Figs. 14 and S24). However, Figs. 14 and S24 clearly show that ${\mathrm{NO}}_{\mathrm{2}}$ VCDs from the 4 $\mathrm{km}$ REAM simulation are more concentrated in Baltimore and Washington, DC, urban regions than ACAM, which are also reflected by the higher ${\mathrm{NO}}_{\mathrm{2}}$ VCD standard deviations of the 4 $\mathrm{km}$ REAM results than ACAM. Several Pandora sites are in the highest ${\mathrm{NO}}_{\mathrm{2}}$ VCD regions, where the 4 $\mathrm{km}$ REAM generally produces larger ${\mathrm{NO}}_{\mathrm{2}}$ VCDs than ACAM, which explains why the ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs over the 11 Pandora sites from the 4 $\mathrm{km}$ REAM simulation are higher than the observations (Fig. 10) and the 36 $\mathrm{km}$ REAM results (Fig. 13) around noontime. Horizontal transport cannot explain the ${\mathrm{NO}}_{\mathrm{2}}$ VCD distribution biases in the 4 $\mathrm{km}$ REAM simulation due to the following reasons. Firstly, horizontal wind fields are simulated as well by the nested 4 $\mathrm{km}$ WRF simulation as the 36 $\mathrm{km}$ WRF compared to P-3B measurements, as discussed in Sect. 3.1. Secondly, the prevailing northwest wind in the daytime (Fig. S5) should move ${\mathrm{NO}}_{\mathit{\text{x}}}$ eastward, but we find no significant eastward shift in ${\mathrm{NO}}_{\mathrm{2}}$ VCDs compared to ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions in both ACAM and 4 $\mathrm{km}$ REAM distributions (Fig. 14). Therefore, we attribute the distribution inconsistency between ACAM and the 4 $\mathrm{km}$ REAM to the distribution biases of NEI2011 ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions at the 4 $\mathrm{km}$ resolution since the average below-aircraft ${\mathrm{NO}}_{\mathrm{2}}$ VCDs between ACAM and the 4 $\mathrm{km}$ REAM are about the same.

It is noteworthy that the number of data points used to calculate grid cell mean ${\mathrm{NO}}_{\mathrm{2}}$ VCDs varies significantly across the domain, as shown in Figs. 14f and S24f. To mitigate potential sampling errors, we only consider the grid cells with $\ge$ 10 data points on weekdays in Fig. S25. Whether we scale ${\mathrm{NO}}_{\mathrm{2}}$ VCDs using the corresponding domain averages (Fig. S25) or not (not shown), the 4 $\mathrm{km}$ REAM generally shows more concentrated ${\mathrm{NO}}_{\mathrm{2}}$ VCDs in Baltimore and Washington, DC, urban regions but more dispersed ${\mathrm{NO}}_{\mathrm{2}}$ VCDs in rural areas than ACAM, consistent with our discussion above. In addition, about 91 % of ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCD data are measured from 08:00–16:00 LT, and only using ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs between 08:00 and 16:00 LT for the above comparison does not affect our results shown here. Moreover, to minimize the effect of overestimated afternoon vertical mixing (Fig. 8) on the 4 $\mathrm{km}$ REAM simulation results, we also examine the comparison between ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs from 09:00–14:00 LT with coincident 4 $\mathrm{km}$ REAM results, which produces similar results as shown here. Finally, considering the ${\mathrm{NO}}_{\mathit{\text{x}}}$ lifetime difference between morning and noontime, we also analyze the ${\mathrm{NO}}_{\mathrm{2}}$ VCD data between 11:00 and 14:00 LT, and similar results are found.

We also evaluate the ${\mathrm{NO}}_{\mathrm{2}}$ VCD distributions from the 4 $\mathrm{km}$ REAM simulation on weekdays and weekends with ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs below the U-12 aircraft obtained from https://www-air.larc.nasa.gov/cgi-bin/ArcView/discover-aq.dc-2011?UC12=1#LIU.XIONG/ (last access: 31 December 2019) in Figs. S26 and S27. Although the domain mean ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs in Figs. S26 and S27 are higher than coincident 4 $\mathrm{km}$ REAM results due to the different retrieval method from Lamsal et al. (2017), such as different above-aircraft ${\mathrm{NO}}_{\mathrm{2}}$ VCDs and different a priori ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles, we can still find clear distribution inconsistencies between the 4 $\mathrm{km}$ REAM and ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs. The 4 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs are more concentrated in the Baltimore and Washington, DC, urban regions than this set of ACAM data, which is consistent with the conclusions derived from the ACAM dataset retrieved by Lamsal et al. (2017).

Figure 15

Distributions of weekday ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs around the DISCOVER-AQ 2011 region for 13:30 LT in July 2011: (a) the 36 $\mathrm{km}$ REAM simulation results, (b) the NASA OMI product (OMNO2), (c) the KNMI OMI product, (d) the retrieved OMI ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs by using the KNMI DOMINO algorithm with corresponding 36 $\mathrm{km}$ REAM vertical profiles, (e) the distribution of the ${\mathrm{NO}}_{\mathrm{2}}$ TVCD differences (c minus a) between KNMI OMI and 36 $\mathrm{km}$ REAM, and (f) the difference (d minus a) between retrieved OMI ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs and the 36 $\mathrm{km}$ REAM results. The ${\mathrm{NO}}_{\mathrm{2}}$ TVCD unit is 10${}^{\mathrm{15}}$ $\mathrm{molecules}{\mathrm{cm}}^{-\mathrm{2}}$.

[Figure omitted. See PDF]

The potential distribution bias of the NEI2011 ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions at 36 $\mathrm{km}$ resolution is analyzed by comparing the 36 $\mathrm{km}$ REAM-simulated ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs with those retrieved by OMI and GOME-2A, as shown in Figs. 15 (OMI, 13:00 LT) and S28 (GOME-2A, 09:30 LT). Both KNMI and our retrievals based on the 36 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles show that OMI and GOME-2A ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs have lower spatial variations than the corresponding 36 $\mathrm{km}$ REAM simulation results. OMI and GOME-2A retrievals have lower ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs around the Baltimore and Washington, DC, urban regions and higher values in relatively rural regions than the 36 $\mathrm{km}$ REAM. The distribution bias of the 36 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs is also identified on weekends through their comparison with OMI and GOME-2A retrievals (not shown). The good agreement between simulated and observed wind suggests that the model horizontal transport error cannot explain such an urban–rural contrast between satellite observations and 36 $\mathrm{km}$ REAM simulation results. However, two caveats deserve attention. Firstly, the 36 $\mathrm{km}$ REAM cannot resolve urban areas as detailed as the 4 $\mathrm{km}$ REAM (Fig. 14), and urban and rural regions may coexist in one 36 $\mathrm{km}$ grid cell. Secondly, the OMI and GOME-2A pixels can be much larger than 36 $\mathrm{km}$ REAM grid cells, possibly leading to more spatially homogenous distributions of satellite ${\mathrm{NO}}_{\mathrm{2}}$ TVCD data.

3.8

Implications for ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions

The analysis of Sect. 3.7 indicates that the NEI2011 ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission distributions at 36 and 4 $\mathrm{km}$ resolutions are likely biased for the Baltimore–Washington region. The distribution bias of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission inventories is corroborated by the comparison of the ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission inventory derived from the CONsolidated Community Emissions Processor Tool, Motor Vehicle (CONCEPT MV) v2.1 with that estimated by the Sparse Matrix Operator Kernel Emissions (SMOKE) v3.0 model with the Motor Vehicle Emissions Simulator (MOVES) v2010a (DenBleyker et al., 2012). CONCEPT, with finer vehicle activity information as input, produced wider-spread but less-concentrated running exhaust ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions compared to MOVES in the Denver urban area for July 2008 (DenBleyker et al., 2012). In addition, Canty et al. (2015) found that CMAQ 4.7.1, with on-road emissions from MOVES and off-road emissions from the National Mobile Inventory Model (NMIM), overestimated ${\mathrm{NO}}_{\mathrm{2}}$ TVCD over urban regions and underestimated ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs over rural areas in the northeastern US for July and August 2011 compared to the OMNO2 product. The urban–rural contrast was also found in Texas during the 2013 DISCOVER-AQ campaign in the studies of Souri et al. (2016, 2018), implying distribution uncertainties in ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions, although these studies and Canty et al. (2015) focused more on polluted regions with overestimated ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions in their conclusions. The emission distribution bias may also explain why Anderson et al. (2014) have different results from our simulated concentrations in Table 1. In their study, they compared in situ observations with a nested CMAQ simulation with a resolution of 1.33 $\mathrm{km}$. It is difficult to build up a reliable emission inventory for the whole US at very high resolutions with currently available datasets due to the significant inhomogeneity of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions (Marr et al., 2013), but we can still expect significant improvements in the temporal-spatial distributions of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions in the near future as GPS-based information starts to be used in the National Emissions Inventory (NEI) estimates (DenBleyker et al., 2017).

Here, we emphasize that our study is not necessarily contradictory to recent studies concerning the overestimation of NEI ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions (Anderson et al., 2014; Canty et al., 2015; McDonald et al., 2018; Souri et al., 2016, 2018; Travis et al., 2016). Different types of observations in different periods and locations are analyzed for various purposes. This study focuses more on the spatial distribution of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions in NEI2011, while previous studies are concerned more about the ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission magnitudes in highly polluted sites, although the spatial distribution issue was also mentioned in some of the studies. If we limit our analyses to those observations in Figs. 7, 10, and 12 and the 4 $\mathrm{km}$ REAM, we would also conclude an overestimation of NEI ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions. Considering the significant heterogeneity of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions, the spatial distribution of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions is a critical factor in evaluating ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions and improving emission estimation and air quality models, which deserves more attention in future studies, especially when chemical and transport models are moving to higher and higher resolutions.

4 Conclusions

We investigate the diurnal cycles of surface ${\mathrm{NO}}_{\mathrm{2}}$ concentrations, ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles, and ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs using REAM model simulations on the basis of the observations from air quality monitoring sites, aircraft, Pandora, OMI, and GOME-2A during the DISCOVER-AQ 2011 campaign. We find that WRF-simulated nighttime $k_{\text{zz}}$-determined MLHs are significantly lower than ELF lidar measurements. Increasing nighttime mixing from 18:00–05:00 LT in the REAM simulations, we significantly improve REAM simulations of nighttime surface ${\mathrm{NO}}_{\mathrm{2}}$ and ${\mathrm{O}}_{\mathrm{3}}$ concentrations.

The REAM simulation reproduces the observed regional mean diurnal cycles of surface ${\mathrm{NO}}_{\mathrm{2}}$ and ${\mathrm{NO}}_{\mathit{\text{y}}}$ concentrations, ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles, and ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs well on weekdays. Observed ${\mathrm{NO}}_{\mathrm{2}}$ concentrations in the boundary layer and TVCDs on weekends are significantly lower than on weekdays. By specifying a weekend-to-weekday ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission ratio of $\mathrm{2}/\mathrm{3}$ and applying a less variable ${\mathrm{NO}}_{\mathit{\text{x}}}$ emission diurnal profile on weekends than weekdays, REAM can simulate the weekend observations well. Two issues are also noted. First, Pandora TVCDs show different variations from aircraft-derived and REAM-simulated TVCDs in the early morning and late afternoon, which may be due to the uncertainties in Pandora measurements at large SZAs and the small effective FOV of Pandora. Second, the weekday OMI ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs derived by NASA are somewhat lower than the KNMI OMI product, P-3B-aircraft-derived TVCDs, Pandora, and REAM results; the difference may be caused by the a priori vertical profiles used in the NASA retrieval.

While a higher-resolution simulation is assumed to be superior at a priori, the large observation dataset during DISCOVER-AQ 2011 offers the opportunity of a detailed comparison of 4 and 36 $\mathrm{km}$ model simulations. Through the comparison, we find two areas that have not been widely recognized. The first is not using convection parameterization in high-resolution WRF simulations since convection can be resolved explicitly, and most convection parameterizations are not designed for high-resolution simulations. We find that 4 $\mathrm{km}$ WRF tends to overestimate boundary layer mixing and vertical transport in the late afternoon, leading to a high model bias in simulated ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles compared to P-3B aircraft observations. The reasons for this late-afternoon bias in 4 $\mathrm{km}$ WRF simulations and model modifications to mitigate this bias need further studies.

A second issue is related to the spatial distribution of ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions in NEI2011. In general, the 4 $\mathrm{km}$ simulation results tend to have a high bias relative to the 36 $\mathrm{km}$ results on the regional mean observations. However, for individual sites, relative to the 36 $\mathrm{km}$ model simulations, the 4 $\mathrm{km}$ model results can show larger, smaller, or similar biases compared to the observations depending upon observation location. Based on process diagnostics and analyses, we find that the bias discrepancies between the 36 and 4 $\mathrm{km}$ REAM simulations are mainly attributed to their different ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions and their spatial gradients at different sites. The comparison of 4 $\mathrm{km}$ ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCD measurements from the UC-12 aircraft with coincident 4 $\mathrm{km}$ REAM results shows that 4 $\mathrm{km}$ REAM ${\mathrm{NO}}_{\mathrm{2}}$ VCDs are more concentrated in urban regions than the ACAM observations. OMI and GOME-2A data also show less spatially varying ${\mathrm{NO}}_{\mathrm{2}}$ TVCD distributions with lower ${\mathrm{NO}}_{\mathrm{2}}$ TVCDs around the Baltimore–Washington urban regions and higher TVCDs in surrounding rural areas than corresponding 36 $\mathrm{km}$ REAM simulation results. Further model analysis indicates that the 36 and 4 $\mathrm{km}$ VCD discrepancies are due primarily to the distribution bias of NEI2011 ${\mathrm{NO}}_{\mathit{\text{x}}}$ emissions at 36 and 4 $\mathrm{km}$ resolutions. Our results highlight the research need to improve the methodologies and datasets to improve the spatial distributions in emission estimates.

Data availability

The DISCOVER-AQ 2011 campaign datasets are archived on 10.5067/AIRCRAFT/DISCOVER-AQ/AEROSOL-TRACEGAS (NASA/LARC/SD/ASDC, 2014). EPA air quality monitoring datasets are from https://aqs.epa.gov/aqsweb/airdata/download_files.html (last access: 23 June 2015, EPA, 2021). The NASA OMI ${\mathrm{NO}}_{\mathrm{2}}$ product is from 10.5067/Aura/OMI/DATA2017 (Krotkov et al., 2019). The KNMI OMI ${\mathrm{NO}}_{\mathrm{2}}$ product is from https://www.temis.nl/airpollution/no2.php (last access: 14 January 2015, KNMI/ESA, 2004). We obtain the KNMI GOME-2A ${\mathrm{NO}}_{\mathrm{2}}$ VCD archives from http://www.temis.nl/airpollution/no2col/no2colgome2_v2.php (last access: 22 January 2015, KNMI/BIRA-IASB, 2011). The GMI MERRA-2 simulation results are from https://portal.nccs.nasa.gov/datashare/dirac/gmidata2/users/mrdamon/Hindcast-Family/HindcastMR2/2011/stations/ (last access: 14 May 2019, NASA, 2017). We obtain the UC-12 ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCD product by Xiong Liu from https://www-air.larc.nasa.gov/cgi-bin/ArcView/discover-aq.dc-2011?UC12=1#LIU.XIONG/ (last access: 31 December 2019, Liu, 2019). The Stage IV precipitation data are downloaded from 10.5065/D69Z93M3 (CODIAC et al., 2000). The NCEP CFSv2 6-hourly product is available at 10.5065/D61C1TXF (Saha et al., 2011). REAM simulation results for this study and the UC-12 ACAM ${\mathrm{NO}}_{\mathrm{2}}$ VCD product by Lamsal et al. (2017, 10.1002/2016JD025483) are available upon request.

The supplement related to this article is available online at: https://doi.org/10.5194/acp-21-11133-2021-supplement.

Author contributions

JL and YW designed the study. JL, RZ, and CS updated the REAM model. JL conducted model simulations. KFB developed the DOMINO algorithm, CS applied the algorithm to REAM vertical profiles, and JL updated the retrieval algorithm and did the retrieval by using REAM ${\mathrm{NO}}_{\mathrm{2}}$ vertical profiles. AW, JH, EAC, RWL, JJS, RD, AMT, TNK, LNL, SJJ, MGK, XL, and CRN made various measurements in the DISCOVER-AQ 2011 campaign. JL conducted the analyses with discussions with YW, RZ, CS, AW, JH, KFB, EAC, RWL, JJS, RD, AMT, TNK, LNL, SJJ, MGK, XL, and CRN. JL and YW led the writing of the manuscript with input from all other coauthors. All coauthors reviewed the manuscript.

Competing interests

The authors declare that they have no conflict of interest.

Disclaimer

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Acknowledgements

This work was supported by the NASA ACMAP program. We thank Chun Zhao for providing us the PNNL NEI2011 emission inventory. We thank Yuzhong Zhang and Jenny Fisher for providing the updated GEOS-Chem chemistry mechanism files and thank Yuzhong Zhang, Yongjia Song, Hang Qu, Ye Cheng, Aoxing Zhang, Yufei Zou, and Ziming Ke for discussion with Jianfeng Li. We thank Susan Strahan for providing the GMI outputs download link.

Financial support

This research has been supported by the National Aeronautics and Space Administration (ACMAP program).

Review statement

This paper was edited by Yugo Kanaya and reviewed by two anonymous referees.