Ground-level ozone (O3), a key component of photochemical smog, has adverse impacts on human health and ecosystems (U.S. Environmental Protection Agency, 2019). In the United States (USA), the Clean Air Act amendments of 1970 instruct the Environmental Protection Agency (EPA) to set National Ambient Air Quality Standards (NAAQS) for criteria pollutants. Since 1979, O3 has served as the indicator species for the criteria pollutant of photochemical oxidants (44 FR 8202), and since 1997, the form of the standard has been determined by the 3-year average of the annual fourth-highest daily maximum 8 h concentration (MDA8) (62 FR 38856). In 2015, the O3 NAAQS were revised to the current level of 0.070 ppm or 70 ppb (parts per billion) (80 FR 65291). As of 2018, 52 areas in the USA had been designated as nonattainment areas for the 2015 O3 NAAQS (83 FR 25776; 83 FR 35136; 83 FR 52157).

O3 is predominantly a secondary pollutant formed from photochemical reactions of nitrogen oxides (NOx) and volatile organic compounds (VOCs). Ground-level O3 concentrations are a complex nonlinear function of the chemistry of natural and anthropogenic precursor emissions, as well as meteorology, transport, and deposition (Seinfeld and Pandis, 2016). O3 formation rates depend on the concentrations and speciation of NOx and VOCs. To reduce ambient O3 concentrations, control strategies have been enacted in the USA over the last 50 years to reduce the emissions of both NOx and VOCs (Simon et al., 2015).

The effectiveness of different control strategies on O3 production rates depends on the photochemical environment under which ozone is formed. Ozone formation environments are typically categorized as either NOx-limited or NOx-saturated, with a mixed or transitional regime between the two (Sillman, 1995, 1999; Sillman et al., 1990). In the NOx-limited regime, ambient ozone concentrations will respond more strongly to changes in NOx emissions than VOC emissions. In contrast, in a NOx-saturated (or VOC-limited) regime, ozone will increase with NOx emission controls but will decrease with VOC emission controls. Understanding the photochemical regimes of different ozone nonattainment areas and how they have changed over time is important for understanding the impacts of previous control strategies and guiding future control strategies to have the maximum health benefit with the least economic burden.

Different methods have been proposed to determine ozone formation regimes and their changes over time. One common method used to evaluate ozone formation chemistry is through day-of-week (DOW) differences in the concentration of ozone and its precursors. The DOW effects leverage NOx emission differences between weekdays and weekends (Marr and Harley, 2002a, b). In the USA, on-road vehicles are a dominant source of NOx emissions (Toro et al., 2021). Diesel vehicle traffic tends to be higher on weekdays (Monday through Friday) than on weekends (Saturday and Sunday). This results in higher NOx emissions on weekdays than weekends (Marr and Harley, 2002a, b). Daily varying emission sources such as diesel vehicles are not a major source of VOC emissions. In addition, VOC emissions in some areas are dominated by biogenic emissions that do not vary by day of week. Consequently, VOC emissions are generally similar on weekends and weekdays in most areas. The result of DOW NOx patterns is that ozone concentrations tend to be higher on weekends than weekdays in NOx-saturated areas and lower on weekends than weekdays in NOx-limited areas (Koplitz et al., 2022). DOW differences in ozone were first reported in the 1970s (Bruntz et al., 1974; Cleveland et al., 1974). In 2002 the DOW ozone differences in California were explicitly tied to DOW patterns in diesel vehicle traffic (Marr and Harley, 2002a, b). Since that time, multiple studies have used DOW ozone patterns to assess ozone chemical formation regimes in individual USA cities including Los Angeles, California (Chinkin et al., 2003; Fujita et al., 2003b, a; Gao, 2007; Gao and Niemeier, 2007; Warneke et al., 2013), Fresno, California (de Foy et al., 2020), Sacramento, California (Murphy et al., 2007), Phoenix, Arizona (Atkinson-Palombo et al., 2006), Atlanta, Georgia (Blanchard and Tanenbaum, 2006), Baltimore, Maryland (Roberts et al., 2022), and New York City, New York (Singh and Kavouras, 2022). A smaller number of studies have assessed ozone DOW patterns across multiple USA urban areas (Blanchard et al., 2008; Jaffe et al., 2022; Koo et al., 2012; Koplitz et al., 2022; Pun et al., 2003). Additionally, ozone DOW patterns have been used as a method for assessing chemical formation regimes outside of the USA in Shanghai, China (Zhang et al., 2023); the Lesser Antilles archipelago (Plocoste et al., 2018); Rio de Janeiro, Brazil (Martins et al., 2015); Santiago, Chile (Rubio et al., 2011); Andalusia, Spain (Adame et al., 2014); the Iberian Peninsula (Jiménez et al., 2005); Athens, Greece (Paschalidou and Kassomenos, 2004); and in multiple other European cities (Pires, 2012). One complication with interpreting DOW O3 patterns is that O3 concentrations in urban areas are generally impacted by a mix of transport and local formation. O3 transport can occur over a variety of timescales. In some locations, there could be a regional O3 DOW effect that might be evident as a slightly lagged timescale, depending on typical transport times from major upwind urban source areas.

Previous work has shown a substantial decrease in NOx emissions in the USA over the past 20 years as a result of national, state, and local regulations (Krotkov et al., 2016; Lamsal et al., 2015; Russell et al., 2012; Toro et al., 2021). Concurrent with the USA NOx decreases, multiple studies have found that ozone chemical formation regimes have also changed in the USA (Jin et al., 2020, 2017; Koplitz et al., 2022). In this paper, we focus on 51 areas in the USA which were designated in 2018 as nonattainment areas (https://www.epa.gov/green-book/green-book-8-hour-ozone-2015-area-information, last access: September 2022) under the 2015 O3 NAAQS (some of these areas have since been redesignated to attainment, based on clean monitoring data). We look at changes in DOW patterns in the USA over 18 years from 2002 to 2019, using both measured and modeled data to provide insights into how ozone formation chemistry has changed in the USA as a result of emission reductions and to assess how well modeling is able to capture the observed changes. This 18-year dataset, which is part of EPA's Air QUAlity TimE Series Project (EQUATES), is unique in its application of consistent emissions and modeling methodologies across the entire analysis period, providing an opportunity to assess multiyear trends.

For this assessment, we use MDA8 ozone monitoring data obtained from EPA's Air Quality System (AQS) (https://www.epa.gov/aqs, last access: September 2022) and MDA8 ozone modeling data from simulations of the Community Multiscale Air Quality model version 5.3.2 (CMAQv5.3.2). The CMAQ model data are part of EQUATES, which provides an 18-year set of modeled meteorology, emissions, air quality, and pollutant deposition spanning the years 2002 through 2019, using consistent modeling methods across years. The CMAQv5.3.2 model configuration, including input data, boundary conditions, and science options are available from U.S. EPA (U.S. EPA, 2021). The emission inventories developed for the EQUATES CMAQ modeling are described in Foley et al. (2023).

We extract CMAQ modeling data only for days and grid cells with monitoring data, such that both datasets are paired in time and location. Both datasets are subset to ozone monitors located within 51 of the 52 areas that were designated in 2018 as nonattainment for the 2015 O3 NAAQS (a list of areas is available in Tables S1 and S2 in the Supplement) (83 FR 25776; 83 FR 35136; 83 FR 52157). Because this analysis focuses on May–September data, we do not include data from the Uinta Basin nonattainment area for which violations of the NAAQS predominantly occur in winter months. Data are analyzed for the 18-year period of the EQUATES modeling dataset.

We start by analyzing changes in MDA8 ozone between weekends and weekdays pooled across all monitoring locations for each nonattainment area for 5-year rolling periods (i.e., 14 different periods covering the 18-year time series). We pool data into 5-year periods for several reasons. First, it dampens impacts of interannual meteorology that can contribute to large year-to-year changes in ozone for a given location. Previous work has shown that differential meteorological patterns on weekends versus weekdays impact ozone DOW patterns in a single year and that pooling data across multiple years can reduce this effect (Pierce et al., 2010). Second, it provides a larger sample size for calculating ozone differences between weekends and weekdays. The use of 5-year periods does, however, limit the ability of this analysis to parse out changes in weekend–weekday differences that have occurred due to emission changes in the most recent individual years analyzed. For example, any changes occurring only in 2018 and/or 2019 would be dampened in the 2015–2019 pooled data.

For the purpose of quantifying differences in weekend versus weekday O3 concentrations, we use Sundays to represent weekends (WEs) and Tuesdays, Wednesdays and Thursdays to represent weekdays (WDs). We do not include ozone on Monday and Saturday to minimize any carryover impacts on concentrations from the previous day, and we exclude Friday, as it may exhibit somewhat different emission patterns than the other weekdays.

We use two metrics to quantify differences in MDA8 ozone between weekends and weekdays. First, we quantify mean differences in MDA8 ozone across the entire distribution of days in each season (winter is December, January, and February; spring is March, April, and May; summer is June, July, and August; fall is September, October, and November; the ozone season is May–September), using Eq. (1), where O3,WE represents MDA8 O3 on Sundays, and O3,WD represents MDA8 O3 on Tuesdays, Wednesdays, and Thursdays. 1ΔO3,DOW‾=O3,WE‾-O3,WD‾ In this study, we mainly focus on differences during the May–September ozone season. Welch's t test (Welch, 1947) is used to denote whether the mean WE–WD difference is statistically different from zero (p < 0.05). Within each nonattainment area, the t test calculation was used to compare the means of every weekday and every weekend day in a 5-year window, treating each day as an independent observation. All available ozone monitoring data and model output from all monitoring locations within each nonattainment area are included in the calculation, providing a measure of average behavior across each area. We also examine 24 h average modeled formaldehyde and NOx concentrations at each of the ozone monitor locations to verify whether the model shows expected patterns of higher NOx on weekdays than on weekends and trends in these ozone precursors. Formaldehyde is used as an indicator of first-generation VOC reaction products for this purpose. We note that monitoring data for VOCs and NOx are much sparser in terms of sampling frequency and spatial density than ozone measurements, so we rely on the model alone to verify underlying day-of-week patterns in precursor compounds.

Second, similar to Jaffe et al. (2022), we look at the percent of days with MDA8 ozone values above the NAAQS level of 70 ppb. We calculate the percent of total weekends and weekdays in May–September for which MDA8 ozone concentrations exceeded 70 ppb, as shown in Eq. (2). 2ΔO3,DOW,%>70=O3,WE,%>70-O3,WD,%>70 For this calculation, a day is characterized as exceeding the NAAQS in an area if measured and/or modeled MDA8 ozone is above 70 ppb at the location of any ozone monitor within the area. In this way, we are tracking days where some portion of the area has observed or modeled MDA8 ozone above 70 ppb, but the analysis does not distinguish whether the high ozone concentrations are localized over a small portion of the area or widespread across multiple monitoring locations. This analysis also does not consider whether days with modeled MDA8 ozone above 70 ppb occur simultaneously with observed MDA8 ozone above 70 ppb. We use Fisher's exact test (Fisher, 1935; Mehta and Patel, 1983) to determine whether the proportion of days above 70 ppb differs between weekends and weekdays.

Next, we use the Theil–Sen estimator (Sen, 1968; Theil, 1992) to determine the multiyear trends in ΔO3,DOW‾ and ΔO3,DOW,%>70 for each area. This nonparametric approach was chosen due to the small sample size (n = 14 5-year windows) and the fact that the Theil–Sen estimator does not require any assumptions on the distribution of the residuals. The Mann–Kendall test (Kendall, 1975; Mann, 1945) is used to determine the statistical significance of the derived trends in WE–WD MDA8 O3 differences. For each derived trend, we also document the 95 % confidence interval. Because we use a 5-year rolling window for each area, the individual data points in the trends analysis are correlated. While this should not systematically bias the calculated slopes, it will lead to lower P values and narrower 95 % confidence intervals than would be calculated if the data points were uncorrelated. However, the P value is still informative to characterize which areas have the strongest trends. Therefore, while we do report P values we do not rely on a strict threshold for determining statistical significance.

Finally, investigation of relationships between WE–WD MDA8 O3 and meteorological parameters used the meteorological dataset developed by and described in Wells et al. (2021). Meteorological parameters were similarly compared across weekends and weekdays, matching times and locations of the ozone analysis and using the same statistical methods for comparison.

While this assessment has provided insight into the ozone formation regimes across high-ozone locations in the USA, some key questions remain about the important drivers for year-to-year changes in DOW MDA8 ozone patterns and which of those drivers are well captured by the EQUATES dataset. First, while NOx and VOC emissions have been steadily decreasing across most areas of the USA, exceptions to that pattern include increasing wildfire emissions, especially in the western USA, and increasing emissions from oil and gas activities near USA nonattainment areas in Texas, Colorado, New Mexico, and Utah. Future work could focus on areas impacted by these two emission sources to assess both the impact of these increasing emissions on ozone formation regimes and the ability of the EQUATES dataset to capture those impacts. Second, this assessment predominantly focused on MDA8 ozone values across the May–September ozone season; however, past work has identified some seasonally varying ozone biases within the CMAQ model (Appel et al., 2021). Specifically, EQUATES has a tendency to underpredict ozone during the spring and overpredict ozone later in the summer (Figs. S40 and S41). Given that ozone formation tends to be more NOx-saturated in the springtime than in the summer (Jin et al., 2020, 2017), a more in-depth assessment would be needed to fully characterize the extent that differences in observed and modeled WE–WD MDA8 ozone differences are impacted by this seasonally varying model performance. Third, we assessed DOW MDA8 ozone patterns across multiple complex urban areas that encompassed spatially heterogeneous emission sources and meteorology. For some of these areas (e.g., Los Angeles, CA, and Denver, CO) the sign of the Theil–Sen slopes in WE–WD MDA8 ozone appeared consistent across monitoring locations, while in others (e.g., New York City, NY) different monitoring locations across the area appeared to show different types of trends. Further local-scale investigation into each of these areas would be necessary to fully characterize the nuances of DOW and year-to-year variations in emission and meteorology that obscure the MDA8 ozone DOW trends in some areas but not others when aggregating across monitor locations in those areas. Finally, an intriguing trend in MDA8 ozone DOW patterns was identified in multiple rural and agricultural areas of California. Recent literature has suggested that soil NO emissions, which are unlikely to have a DOW emission pattern, are an important NOx emission source in agricultural locations of California (Almaraz et al., 2018; Zhu et al., 2023). Could the MDA8 ozone DOW trends observed in these areas be reflective of the increasing relative importance of NOx sources other than mobile sources in those locations? More assessment is needed to definitively determine whether the trend in a decreasing weekday effect is a reliable indicator of areas that are becoming more dominated by local NOx sources that do not vary by DOW, more dominated by transported ozone, or some other factor. It is important to note that transported ozone may come from nearby regional sources or from longer range sources provided the transport times are sufficient to mask any DOW patterns that would be evident in the source region.

In this analysis, we found that trends in ozone formation chemistry may not always be clearly shown by trends in DOW patterns, which are impacted by a complex set of local factors including meteorology, the mix of local emission sources, and monitor locations in relationship to land–water interfaces. Lack of trends appear more often, using observed data than modeled data (Figs. 4 and 5), meaning that, while the model accurately captures Theil–Sen slopes for ΔO3,DOW‾ and ΔO3,DOW,%>70 (Fig. 11), lower P values are less common when using observational data. This suggests that there may be some stochastic processes making observed year-to-year WE–WD MDA8 ozone differences noisy, which are not fully captured by the model. Even with these limitations, this analysis has shown that DOW patterns in ambient NOx concentrations persist in USA urban areas but have become less prominent in some areas, while others have transitioned from positive WE–WD MDA8 ozone differences to negative WE–WD MDA8 ozone differences over the 18-year period analyzed. These DOW NOx differences have resulted in distinctive DOW MDA8 ozone patterns in many of the nonattainment areas assessed. The EQUATES modeling simulations appear to show larger and more positive WE–WD MDA8 ozone differences than observational data, suggesting that ozone formation in this modeling dataset is less NOx-limited than in the observations. Despite this discrepancy, the EQUATES dataset captures year-to-year changes in WE–WD MDA8 ozone patterns, as demonstrated by high correlation of the Theil–Sen slopes for WE–WD MDA8 ozone differences. The agreement between the modeled and observation datasets is more apparent when assessing summertime mean MDA8 ozone than when analyzing extreme values using the percentage of exceedance days metric. Assessing frequencies or magnitudes of extreme values is challenging when using a dataset with a limited number of weekend and weekday days, due to the stochastic and infrequent nature of high-ozone events in many areas.

While there are multiple types of measurements and modeling assessments that can be applied to characterize local ozone formation regimes, many of these require specialized measurements or datasets that are not readily available in all areas. In contrast, assessing DOW MDA8 ozone patterns requires only routine daily ozone measurements that are widely available across urban areas in the USA and in other countries. Consequently, this type of assessment is a useful tool and may be applied in many areas using routine measurements. In locations with long-term measurements, DOW patterns offer a method to look at trends in ozone formation chemistry over time. While DOW patterns in MDA8 ozone are especially useful, given the wide availability of data required for this type of assessment, we anticipate that in the near future, additional datasets for assessing ozone chemical formation regimes will become more widely available. Specifically, O3, NO2, and HCHO data from the recently launched Tropospheric Emissions: Monitoring of POllution (TEMPO) satellite may provide the ability to better understand the relationships between WE–WD MDA8 ozone patterns and precursor concentrations.