The standard global emission inventory in MUSICAv0 is CAMS-GLOB-ANT, which provides monthly averages for 36 emitted compounds within 17 sectors at a spatial resolution of 0.1° × 0.1° (latitude × longitude) (Soulie et al., 2024). Here, we use CAMS-GLOB-ANT version 5.1 (Eskes et al., 2021), which combines the Emissions Database for Global Atmospheric Research (EDGAR) v5 (Crippa et al., 2019) up to 2015 with the Community Emissions Data System (CEDS) (McDuffie et al., 2020) to extend emissions from 2016 to 2021. To evaluate the sensitivity of simulated air pollutant concentrations to the temporal resolution of anthropogenic emissions, we replace the global CAMS-GLOB-ANT baseline inventory with the U.S. Environmental Protection Agency (EPA) National Emissions Inventory (NEI) over the CONUS, which provides more detailed (higher spatial resolution; hourly-varying) regional emissions. We then conduct a series of simulations that differ only in the temporal resolution of NEI emissions (monthly, daily, or hourly) to isolate the model response to the timing of emissions while holding the total emissions for July constant.

The U.S. NEI includes emissions of criteria pollutants, precursors, and hazardous air pollutants (U.S. Environmental Protection Agency, 2024). The NEI is updated every three years and constructed through the Emissions Inventory System (EIS), which collects and integrates data primarily provided by State, Local, and Tribal air agencies. We first process the 2017 NEI (U.S. Environmental Protection Agency, 2022), the most recent pre-COVID inventory available at the time of this study, to generate hourly emissions on a ∼ 0.1° × 0.1° grid over the CONUS (Sect. S3) using sector-specific diurnal profiles to capture within-day as well as day-to-day variations. The 2017 hourly NEI data are then shifted by one calendar day so that the weekday-weekend cycle matches the 2018 calendar and re-gridded (mass-conserving) to the unstructured ne0CONUSne30x8 horizontal resolution using NCAR-developed tools (National Center for Atmospheric Research, 2022a, b). All subsequent references to NEI in this study refer to this adjusted product unless otherwise noted.

Switching from CAMS-GLOB-ANT v5.1 to monthly mean NEI emissions yields widespread decreases in most emitted species across the CONUS, particularly for NO and CO (Fig. S4c). Changes in anthropogenic VOC emissions are also evident (Fig. S4c), but their magnitude remains small compared with biogenic VOC emissions (compare Fig. S5 with Fig. S4a–b; see Sect. S4 for the full list of emitted VOC species). We verified that hourly NEI emissions are read by MUSICAv0 in UTC time with the prescribed diurnal and weekday-weekend variability preserved in local time at the grid-cell level (Sect. S3; Figs. S6 and S7).

To assess the impact of using an alternate anthropogenic emissions inventory and different temporal resolutions of emissions, we conduct four one-month simulations (Table 1): (1) monthly NEI data replace CAMS-GLOB-ANT v5.1 emissions over the CONUS, with CAMS monthly means retained elsewhere (NEI_monthly); (2) as in NEI_monthly but hourly NEI emissions are applied for all species over the CONUS (NEI_hourly); (3) as in NEI_monthly, but only NO is emitted hourly (NEI_hourly_NO); (4) as in NEI_monthly, but NO emissions vary daily and thus include weekday-weekend differences (NEI_daily_NO). Here, anthropogenic NO_x emissions are provided as NO, the predominant emitted form of NO_x. Hourly NEI emissions (NEI_hourly and NEI_hourly_NO) include day-specific diurnal variability (and therefore distinguish between weekdays and weekends), whereas NEI_ daily_ NO removes within-day (diurnal) variability (Table 1). To summarize: the NEI_monthly vs. BASE comparison isolates the effect of changing emission inventories; NEI_hourly vs. NEI_monthly assesses the influence of adding sub-monthly (daily and diurnal) variability for all anthropogenic emissions, NEI_daily_NO vs. NEI_monthly isolates the effect of weekday-weekend differences in NO emissions, and NEI_hourly_NO vs. NEI_ daily_NO isolates the effect of the diurnal changes in NO emission.

We conduct two additional one-month simulations using the NEI_monthly case as the reference to analyze the O₃ production regime under idealized perturbations to anthropogenic NO_x and VOC emissions. In one simulation, we reduce anthropogenic NO emissions by 30 % (NEI_monthly_m30anthroNO) and in another we reduce anthropogenic VOC emissions by 30 % (NEI_monthly_m30anthroVOC) (Table 1). These additional sensitivity simulations provide context for those incorporating the more nuanced changes in the temporal resolution of anthropogenic emissions (Sect. S4).

We use measurements collected from State and Local Air Monitoring Stations (SLAMS) that are reported to the U.S. EPA Air Quality System (AQS) for trace gases (O₃, NO₂, CO, and SO₂) and PM_2.5 concentrations (Table S4), downloaded from the AQS AirData portal (https://aqs.epa.gov/aqsweb/airdata/download_files.html, last access: 1 August 2023). To compare MUSICAv0 surface simulations with SLAMS surface measurements, we identify the closest model grid cell based on the latitude and longitude of the SLAMS monitors. We then align hourly concentrations in local time from SLAMS with MUSICAv0 for each species across all CONUS monitors. A limitation of this evaluation is that most SLAMS sites are in urban areas and are influenced by localized effects that cannot be fully resolved by the model at 14 km resolution, which may affect the representativeness of site-level comparisons. The locations of SLAMS used in this study for each species, along with their distribution across CONUS regions and the number of monitors within selected urban grid cells, are shown in Fig. S8.

We select six monitoring stations as examples to examine diurnal patterns at individual sites (Table S5), chosen based on their proximity to major city centers across the CONUS and the availability of continuous NO₂ and O₃ measurements throughout July 2018. Nearby stations with similar diurnal behavior were excluded to avoid redundancy. We do not average across sites as we prefer to preserve distinct local features given differences in monitor availability across cities (Fig. S8).

We also use retrievals from TROPOMI, a nadir-viewing shortwave spectrometer aboard the Sentinel 5 Precursor (S5P) satellite launched in 2017 and operational in 2018. We compare tropospheric vertical column densities (VCD_Trop) of NO₂ (5.5 km × 3.5 km) and HCHO (5.5 km × 3.5 km), along with total vertical column densities (VCD_Total) of CO (5.5 km × 7 km) retrieved from TROPOMI (RPRO Version 02.04.00; last access: 1 December 2023). We select pixels with quality assurance greater than 0.75 (Table S4), which largely excludes cloudy and partly cloudy pixels (van Geffen et al., 2020; Lange et al., 2023). We re-grid the column densities and corresponding averaging kernels (AKs) to a horizontal resolution of 0.15° × 0.15°, slightly coarser than that of the model simulations over the CONUS (approximately 0.125°). TROPOMI uses a priori profiles derived from the TM5-MP global chemistry transport model (Myriokefalitakis et al., 2020), the massively parallel (MP) version of the Tracer Model version 5 (TM5), to simulate the vertical distribution of NO₂, HCHO, and CO, which are provided as supplementary data with the Level-2 products. The TM5-MP model provides data at a 1° × 1° horizontal resolution for the troposphere and upper troposphere-lower stratosphere on 34 hybrid sigma-pressure levels from the surface to approximately 0.1 hPa for retrieving the VCD_Trop of NO₂ and HCHO (Williams et al., 2017). The CO total column density retrievals are based on 50 hybrid sigma-pressure levels from TM5-MP.

To compare modeled column densities with TROPOMI retrievals, we mass conservatively re-grid the hourly MUSICAv0 HCHO, NO₂, CO, and meteorological variables to a 0.15° × 0.15° finite volume grid. We calculate the average between 01:00 and 02:00 p.m. LT (to approximate 01:30 p.m. values) for each region and apply the TROPOMI AKs, linearly interpolated vertically to the MUSICAv0 vertical resolution, to ensure a consistent application of AKs across all MUSICAv0 sensitivity simulations when calculating modeled VCD_Trop of NO₂ and HCHO and VCD_Total of CO. Our interpolation of the TROPOMI AKs to the MUSICAv0 vertical grid generally preserves their vertical structure without introducing unphysical smoothing (Fig. S9). We use the tropopause height diagnosed from the model simulations. Applying AKs to the modeled column enables a more consistent comparison between model simulations and satellite retrievals by accounting for the vertical sensitivity of the satellite instrument. We compare daily 01:30 p.m. VCD_Trop or VCD_Total values across individual grid cells, matching each valid retrieval to the nearest model grid cell. Reported averages include only grid cells with valid retrievals after quality filtering.

We focus on region-specific comparisons with observations across the CONUS (Fig. 2; Tables S2 and S3), as substantial spatial variation suggests that national-level summaries may obscure important regional differences. Across the six CONUS regions, spatial correlations between observed and modeled surface concentrations in the BASE simulation (with the global CAMS-GLOB-ANT emissions) are stronger for NO₂ and O₃ (rs typically > 0.5) than for CO (rs= 0.10–0.36) (Fig. 2a; Table S2). Modeled surface concentrations of NO₂ are biased high by 22 %–40 % (2–5 ppb) in all regions except the Mountain region, which shows an average low bias of 18 % (1 ppb). Modeled July mean surface O₃ concentrations are overestimated by 11 %–29 % (6-13 ppb). Surface CO is underestimated by 17 %–87 % (27–140 ppb) compared to SLAMS across all regions. Modeled July mean VCDs (Fig. 2b; Table S2) correlate spatially with TROPOMI NO₂ VCD_Trop (rs: 0.65–0.85) and HCHO VCD_Trop (rs= 0.60–0.92), though the correlation is weak for CO VCD_Total (rs= 0.11–0.53). Modeled NO₂ VCD_Trop is underestimated by 34 %–48 % (∼ 3 × 10¹⁴ molec. cm⁻²), while HCHO VCD_Trop is overestimated by 15 %–24 % (1–4 × 10¹⁵ molec. cm⁻²) across the six regions. CO VCD_Total is underestimated by 2 %–11 % (3–18 × 10¹⁶ molec. cm⁻²) in the West Coast, Midwest, and Northeast, but overestimated by 2 %–8 % (3–14 × 10¹⁶ molec. cm⁻²) in the Mountain, Southwest, and Southeast regions.

Compared to the BASE case, NEI_monthly (Table 1) improves spatial correlations (rs) and reduces model biases (MBE and RMSE) for surface NO₂, CO, and O₃ concentrations, particularly in the Northeast (Fig. 2a; Table S2). Differences in simulated surface NO₂ and CO between NEI_monthly and BASE (Fig. 3a) reflect emissions changes in NO and CO due to the shift from CAMS-GLOB-ANT v5.1 to NEI (Fig. S4). Lower NEI NO emissions (Fig. S4c) reduce regional mean NO₂ concentrations by 1–6 ppb, bringing high biases down to within 1 ppb but exacerbating the low bias in the Mountain region to -2 ppb (Fig. 2a). CO concentrations increase relative to the BASE simulation (Fig. 3a), improving the surface CO underestimation by 13 %–55 % (20–65 ppb), although low biases of 1 %–33 % (2–71 ppb) remain (Fig. 2a). For secondary pollutants like O₃, changes in concentrations do not directly mirror emissions perturbations. Surface O₃ concentrations decrease in NEI_monthly relative to BASE, reducing modeled surface O₃ biases relative to SLAMS by 2 %–11 % (approximately 1–6 ppb) on average across the six regions (Figs. 2a and 3a). However, model biases of 3 %–21 % (1–8 ppb) remain across all regions, particularly on the West Coast (NEI_monthly in Fig. 2a).

There are minimal to no changes in rs for HCHO VCD_Trop, while NO₂ VCD_Trop shows weaker correlations and worsening model biases (Fig. 2b; Table S2). Changes in CO VCD_Total exhibit large regional variability (Fig. 3b). Switching to NEI consistently decreases both surface and column NO₂ but not CO (Fig. 3). Compared to BASE, NEI_monthly worsens the model underestimates of NO₂ VCD_Trop by 16 %–21 % (∼ 1 × 10¹⁴ molec. cm⁻²) but decreases overestimates of regional mean HCHO VCD_Trop by approximately 2 % (0.2–4 × 10¹⁴ molec. cm⁻²) across all regions (Fig. 2b). CO VCD_Total changes little (Fig. 3b), with the sign of the model bias unchanged from BASE (Fig. 2b).

Biases in trace gas columns do not always match those at the surface. For instance, surface NO₂ is biased high in some regions especially in BASE, while NO₂ VCD_Trop are consistently biased low relative to TROPOMI (Fig. 2). These different biases in the surface versus column can arise from several factors that are independent of surface emission magnitude, including: (i) vertical sensitivity and representativeness, given that satellite columns reflect vertically weighted pixel means whereas surface observations are point measurements; and (ii) retrieval uncertainties, including assumptions in the air mass factor calculation and the stratosphere-troposphere separation applied to derive tropospheric NO₂ columns (van Geffen et al., 2020). In addition, prior global chemistry models have been shown to underestimate background NO₂ VCD_Trop relative to satellite products, partly due to uncertainties in free-tropospheric NO_x chemistry (e.g., NO/NO2 ratios and partitioning among reactive nitrogen species) (van Geffen et al., 2022; Shah et al., 2023; Silvern et al., 2018, 2019).

The surface NO₂ and O₃ concentration changes between NEI_hourly and NEI_monthly are, in some regions, similar in magnitude to those between NEI_monthly_m30anthroNO (a 30 % reduction in anthropogenic NO emissions) and NEI_monthly, but the spatial patterns differ markedly (Fig. 3a), reflecting the temporal redistribution of emissions. Although the overall agreement with observations changes only slightly (rs, MBE, and RMSE; Fig. 2 and Tables S2 and S3), the direction and magnitude of concentration changes vary across regions, especially between urban and rural areas and between the western and eastern CONUS (Figs. 3 and 4).

The largest July regional-mean difference in surface NO₂ concentrations (NEI_hourly - NEI_monthly) occurs on the West Coast (+16%, or +0.3 ppb), while NO₂ decreases most in the Northeast (-7 %, -0.1 ppb) (Fig. 3a). As a result, the model shifts from underestimation to overestimation on the West Coast, while the underestimation in the Northeast is exacerbated versus the surface NO₂ SLAMS observations (Fig. 2a). Changes in surface CO are small and spatially varied, with regional mean increases of 2 %–7 % (4–10 ppb) in the West Coast, Mountain, Midwest, and Southwest, and decreases of about 3 % (∼ 5 ppb) in the Northeast and Southeast (Fig. 3a). Changes in NO₂ VCD_Trop and CO VCD_Total largely mirror the surface patterns (Fig. 3b).

High model biases persist for surface O₃ concentrations (5 %–20 %; 3–9 ppb) after incorporating hourly variations in emissions (NEI_hourly in Fig. 2a). Monthly mean surface O₃ concentrations decrease by up to ∼ 2 % (< 1 ppb) in the Midwest, Northeast, and Southeast, while increases of up to ∼ 3 % (1–4 ppb) in other regions slightly worsen model overestimation. Previous studies also find model overestimates of near-surface O₃, with the most significant biases in the Southeast U.S., possibly reflecting uncertainties in isoprene–NO_x–O₃ chemistry, inaccuracies in dry deposition and forest canopy parameterization, and overestimates in NO_x emissions (Baublitz et al., 2020; Fiore et al., 2009; Makar et al., 2017; Martin et al., 2014; Schwantes et al., 2020; Travis et al., 2016). Although surface HCHO observations are unavailable for evaluation, modeled changes remain under 3 % and follow similar spatial patterns to those of surface O₃ (Fig. 3a). Monthly mean HCHO VCD_Trop decreases in the Midwest, Northeast, and Southeast but increases in other regions, mirroring the regional pattern of surface HCHO changes (Fig. 3), though a general high model bias of 14 %–23 % (1–3 × 10¹⁵ molec. cm⁻²) persists (Fig. 2b; Table S2).

Our model simulations (Table 1) broadly capture day-to-day variations in surface NO₂ and O₃ at the six urban sites (Fig. S10), but they tend to overestimate both the daily range and the weekday-weekend differences and misrepresent the timing of peak concentrations in most cities (Fig. S11). The exaggerated daily range reflects opposite model biases at different times of day: the model overestimates nighttime NO₂ (when concentrations are highest) and fails to capture the nighttime minimum in O₃, while at midday it underestimates NO₂ and overestimates O₃ (Fig. S11). As expected, simulations with day-to-day variations in emissions better capture weekday-weekend concentration differences compared to NEI_monthly, though the diurnal shape still does not fully capture observations (see the right column of Fig. S11). The model also misrepresents the timing of daily peaks. For example, the simulated NO₂ maximum occurs at 06:00 a.m. in Los Angeles, about two hours earlier than observed (Fig. S11).

It is challenging for models to accurately represent the complex diurnal processes that shape near-surface concentrations, including boundary-layer evolution (Adams et al., 2023). Our MUSICAv0 simulations use the standard CAM6 planetary boundary layer parameterization, the Cloud Layers Unified By Binormals (CLUBB) (Bogenschutz et al., 2018; Danabasoglu et al., 2020). While prior evaluation within the CESM2/CAM6 framework shows that the scheme generally captures the broad diurnal and seasonal structure of boundary-layer behavior, overly strong nocturnal mixing leads to errors in the timing of the morning transition although the specific biases vary across regions and seasons (Holtslag et al., 2013; Schwantes et al., 2022; Stjern et al., 2023). These biases will be reflected in the simulated diurnal amplitude and timing of near-surface concentration peaks and their relationship to the underlying temporal profile of emissions. Errors may also arise from the NEI diurnal profiles applied in this study, which are sector-specific and derived from activity-based temporal allocation methods whose accuracy at local scales is uncertain. In addition, comparisons between model grid-cell means and individual SLAMS sites inevitably involve representativeness differences that may contribute to mismatches.

To understand the broader impacts of incorporating hourly NO emissions, we begin by examining regional contrasts in July daytime (09:00 a.m.–05:00 p.m. local time) responses across the CONUS, as shown in Fig. 5. Panel a maps the differences in surface NO, NO₂, O₃, and HCHO concentrations (NEI_hourly_NO minus NEI_ daily_NO), while panel b summarizes the meridional means in 5° longitudinal bins. These plots highlight a west-to-east gradient in pollutant responses driven solely by adding the diel cycle to NO emissions. To help interpret this spatial variability, panel c presents the corresponding NEI_daily_NO daytime emissions and key meteorological variables, including surface temperature, planetary boundary layer height (PBLH; m a.g.l.), relative humidity, and NO₂ lifetime against dry deposition.

Polluted urban regions with higher NO emissions (Fig. 1) show more pronounced NO_x concentration responses (Figs. 3 and 4). In the western CONUS, July daytime (09:00 a.m.–05:00 p.m.) monthly mean surface NO and NO₂ concentrations increase by up to 8 and 6 ppb, respectively, in urban areas in NEI_hourly_NO simulation relative to NEI_daily_NO (Fig. 5a–b). In contrast, monthly mean daytime surface NO and NO₂ decrease over the eastern CONUS by up to ∼ 1 ppb, respectively, with the largest changes also concentrated in urban areas. However, changes in surface O₃ do not always coincide with the largest NO_x concentration changes. In the eastern CONUS, O₃ decreases by up to ∼ 2 ppb (Fig. 5a–b), with the largest reductions over the polluted urban centers, consistent with prior findings (e.g., Jo et al., 2023). In contrast, monthly mean O₃ increases of up to 8 ppb occur over the western CONUS along the edges of urban areas (Fig. 5a). This spatial pattern points to enhanced NO_x sensitivity at the periphery of NO_x-saturated western urban cores; similar patterns occur in the NEI_ monthly_ m30anthroNO sensitivity simulation relative to the BASE case, where the 30 % reduction in anthropogenic NO emissions yields the largest O₃ changes in surrounding rural areas (Figs. 3 and 4). In the West, the mean of the NEI_hourly_NO minus NEI_ daily_ NO differences exceeds the median (Fig. 5b), driven by large NO_x increases in a few urban grid cells. In the East, the larger mean decreases relative to the median similarly reflect strong but localized NO_x decreases. By contrast, O₃ differences show closer agreement between mean and median across both regions, suggesting more spatially uniform impacts (Fig. 5c).

These spatial patterns are shaped by a complex interplay of background anthropogenic and biogenic emissions, local photochemical conditions, meteorology (especially boundary layer dynamics and transport), and deposition. In July, the East Coast has higher background NO emissions, approximately 70 % higher over the Northeast region than over the West Coast (Fig. 1). In the western U.S., anthropogenic emissions tend to occur in more localized hotspots in contrast to the continuous, corridor-like pattern in the eastern U.S., consistent with population density and urban clustering (Fig. S4a). Biogenic emissions, dominated by isoprene and monoterpenes, are highest and widespread across the eastern and southern CONUS (Fig. S5). Switching to hourly emissions generally increases daytime NO (Fig. 6), yet the resulting impact on surface concentrations across the CONUS (Fig. 5a) differs notably from the response to a uniform 30 % reduction in anthropogenic NO emissions (Fig. 3). This finding indicates that the timing of anthropogenic emissions perturbations with respect to other processes like meteorology (Fig. 5c), biogenic emissions, and deposition shape the contrasting changes between the western and eastern CONUS. For example, shallower daytime PBLH such as over the eastern CONUS can limit vertical mixing while enhancing both photochemical reactions and surface deposition (Wu et al., 2024). Together with higher humidity, these processes may contribute to a shorter NO₂ lifetime in the eastern CONUS during July 2018. We explored the possibility of a persistent synoptic-scale east-west dipole in air stagnation, defined according to the meteorological criteria of Wang and Angell (1999), but did not find any evidence for such a feature.

To further investigate the regional contrasts introduced in Sect. 4.1, we examine two representative urban centers, Los Angeles (CA) and New York City (NY), to show how local-scale meteorology and photochemistry interact with hourly NO emission patterns to produce opposite-signed responses in NO_x and O₃ concentrations. Figure 6 compares NEI_ hourly_NO and NEI_daily_NO simulations, showing (a) time series of NO emissions and surface NO concentration differences, (b) July weekday-averaged diel differences in NO, NO₂, O₃, and HCHO, and (c) weekday-averaged diel cycles of NO emissions, surface NO, and PBLH. Weekend patterns are similar but generally smaller in magnitude (see Fig. S13). Figure 7 presents July daytime O₃ production rates (P(O₃), diagnosed in MUSICAv0 as chemical O₃ production from reactions of NO with the hydroperoxy radical and organic peroxy radicals) binned by surface NO_x concentrations for the NEI_monthly, NEI_hourly_NO, and NEI_ monthly_ m30anthroNO simulations. Box plots summarize the distribution of P(O₃) within uneven NO_x concentration bins (e.g., 0–1, 1–2, …, 5–10, 10–15, …, > 50 ppb); color-coded points highlight values during morning, noon, and afternoon periods. Figures S7, S13 and S14 include our findings for all major cities shown in Fig. 1.

Comparing NEI_hourly_NO to NEI_ daily_NO, surface NO_x concentrations decrease in New York City but increase in Los Angeles, especially during the morning rush hour when the boundary layer is growing (Fig. 6b–c). Monthly mean surface O₃, however, decreases in both Los Angeles and New York City, though the magnitudes differ (Fig. 4a). Surface O₃ decreases most in Los Angeles on weekdays between 09:00 a.m. and 03:00 p.m., outweighing a smaller late-afternoon increase (03:00–06:00 p.m.) (Fig. 6b). In New York City, the net monthly mean decrease is minimal because modest reductions from noon to evening (12:00–06:00 p.m.) only slightly exceed the morning increase (08:00 a.m.–12:00 p.m.) (Fig. 6b).

To assess the sensitivity of P(O₃) to NO_x and VOC, we compare P(O₃) in NEI_monthly_m30anthroNO and NEI_monthly_m30anthroVOC with NEI_monthly (Table 1). All three simulations indicate that New York City and Los Angeles are generally NO_x-saturated, with surface O₃ increasing as NO emissions decrease (Fig. 4a) and little change in response to the 30 % reduction in anthropogenic VOC emissions (< 0.5 % across all regions; not shown). Previous studies have shown that O₃ production tends to become more NO_x-sensitive around noon and in the afternoon, when photochemical conditions are most favorable for O₃ formation (Tao et al., 2022). In contrast, the model indicates a P(O₃) peak during the late morning in both New York City and Los Angeles, particularly in NEI_monthly_m30anthroNO (Fig. 7c, f). Compared to NEI_monthly (Fig. 7a, d), NEI_monthly_m30anthroNO (Fig. 7c, f) shifts the NO_x concentration bins associated with high P(O₃) toward lower NO_x in both Los Angeles and New York City. For example, in Los Angeles, the highest-NO_x conditions (> 30 ppb) shift into lower NO_x bins (< 30 ppb) (Fig. 7a, c).

Incorporating hourly NO emissions (NEI_ hourly_NO) reduces July mean daytime P(O₃) relative to NEI_monthly by ∼ 2 ppb h⁻¹ in Los Angeles and ∼ 0.3 ppb h⁻¹ in New York City. Using hourly NO emissions (NEI_hourly_NO) increases the density of values in the highest NO_x bins in Los Angeles while maintaining similar P(O₃) values in those bins (Fig. 7b), extending the high-NO_x tail of the P(O₃)-NO_x distribution into more strongly NO_x-saturated conditions (compare Fig. 7a vs. Fig. 7b). This enhanced high-NO_x tail reflects larger morning NO_x concentrations, when the chemical regime is most NO_x-saturated and O₃ production is suppressed. Although conditions become more NO_x-sensitive later in the afternoon (Fig. 7b), the larger O₃ decrease during late morning to early afternoon (∼ 09:00–14:00 LT; Fig. 6b) dominates the net monthly mean O₃ concentration response in Los Angeles (Fig. 4a). The more temporally concentrated (peaky) emission profile and shallower daytime PBL heights in Los Angeles compared to New York City (Fig. 6c) likely amplifies NO_x accumulation, reinforcing a NO_x-saturated regime. We find that Los Angeles, Denver, and Houston show broadly similar responses to hourly versus monthly mean NO emissions, while New York City, Chicago, and Atlanta exhibit distinct behavior (Figs. S13 and S14).

Our results in Sect. 4.1 and 4.2 show that resolving hourly NO emissions changes simulated NO₂ VCD_Trop even when the total monthly emissions are the same. Given that NO₂ VCD_Trop observations from satellites are commonly used to derive top-down constraints on NO_x emissions, differences in the timing of emissions in models used for such inversions may alter the inferred emission estimates. NO_x emissions are sometimes inferred using proportional scaling or mass-balance approaches, in which a priori emissions are adjusted based on the ratio of observed to modeled NO₂ columns, for example with chemical transport models that implicitly represent the impacts of transport and chemistry along with emissions on atmospheric concentrations (Lamsal et al., 2011, 2014; Martin et al., 2003). More formal inverse modeling and data assimilation frameworks exist but similarly depend on satellite overpass-time NO₂ columns and model representations of diurnal variability (Goldberg et al., 2019; Miyazaki and Eskes, 2013).

To estimate potential biases in the inferred NO emission magnitude arising from model representation of diurnal emission timing, we perform a simple, one-month scaling-based approach for July 2018 (details in Sect. S6). In this section, Ω denotes NO₂ VCD_Trop and E anthropogenic NO emissions. While our illustrative simple framework assumes a local relationship between Ω and E, we emphasize that Ω will generally reflect emissions from multiple model grid cells given the NO_x lifetime and transport timescales. Transport, chemistry, and lifetime effects are implicitly accounted for through the forward model but are not explicitly inverted. Accordingly, (ΔE/E)est in Fig. 8 should be interpreted as an estimate of the potential bias in inferred emission magnitude rather than as a quantitative emission inversion. By comparing NEI_hourly_NO and NEI_monthly to isolate the effect of NO emission timing alone since both simulations use identical monthly-integrated emissions, we find that the inferred timing-representation bias is small on average over the CONUS (+1.8 %) but spatially variable, with regional mean values of 1 %–8 % in the western U.S. and approximately -3 % to -2 % in the Northeast and Southeast (Fig. 8; Table S6). City-scale values are larger, reaching +26 % in Los Angeles, +30 % in Denver and -7 % in New York City (Table S6).