Tropospheric ozone is a potent pollutant damaging human and ecological health. The United States Environmental Protection Agency (EPA) targets the maximum daily 8 h average (MDA8) ozone levels for regulation, with a current standard at 70 parts per billion (ppb). The Global Burden of Disease assessment (Brauer et al., 2016; Lim et al., 2012; Forouzanfar et al., 2015, 2016), however, estimates the threshold level (below which exposure to ozone is not harmful) to be between 33.3 and 41.9 ppb. Further, additional epidemiological evidence has shown that there is not a real threshold, and ozone has adverse health effects at all concentrations (Bell et al., 2006; Peng et al., 2013; Yang et al., 2012).

Chemically, surface ozone is produced in the daytime and destroyed mainly by nitrogen oxides (NOx) at night, with a transition between production and destruction in the dawn and dusk hours. Understanding ozone changes and drivers at different times of day might provide additional information to assist further ozone mitigation policymaking beyond the MDA8 regulation. Previous observational and modeling studies have revealed important impacts of varying climate conditions and anthropogenic precursor emissions on the near-surface daytime, MDA8, or daily mean ozone over the United States (US) (Jacob and Winner, 2009; Fiore et al., 2015). A particular finding for these studies is that the US emission reductions have decreased the summertime daytime peak ozone over much of the United States (Cooper et al., 2012; Lefohn et al., 2010; Simon et al., 2015; Strode et al., 2015).

In addition, most studies on US ozone trends and variability tend to focus on ozone changes in a particular season and/or over a particular region – for example, summertime ozone (Rieder et al., 2015; Lu et al., 2016; Reddy and Pfister, 2016) over the eastern US (Zhang and Wang, 2016; Rieder et al., 2015; Shen et al., 2015) or springtime ozone (Cooper et al., 2010) over the western US (M. Y. Lin et al., 2015). Lin et al. (2017) analyzed the various driving factors of the US MDA8 ozone trends by season and region, with no discussion on ozone at other hours.

Bloomer et al. (2010) analyzed the 1989–2007 changes in the diurnal cycle of ozone observed from five stations over the eastern US. They showed ozone reductions at most times of day in the warm seasons due to emission reductions, in contrast to the increases in winter. Jhun et al. (2015) used a statistical model of ozone, nitrogen oxides, and several meteorological parameters to analyze the ozone trends over 1994–2010 measured at over 100 sites across the US. They linked the observed reductions in nitrogen oxides to reductions in warm season peak ozone and to enhancements in cold season peak ozone and warm season nighttime ozone. Overall, the historical changes of ozone at night and other non-peak hours and their underlying climatic or emission causes have been much less studied compared to those for peak ozone.

Here, with the usage of hourly data observed at about 1000 sites from the Air Quality System (AQS) network, we contrasted the interannual variations of the daytime versus the nighttime US ozone over 1990–2014 and also estimated the trends for ozone at the different hours of the day. We further quantified the individual effects of interannual climate variability and anthropogenic emissions on the ozone change by using three climate indices and simulations of the chemical transport model GEOS-Chem. We focused on the large-scale features of annual mean US averaged surface ozone and the impacts of broad changes in emissions and climate. We also contrasted the ozone changes and drivers between the eastern and western US and between different seasons, with complementary discussions for individual locations and land use types (urban, suburban, and rural).

The paper is structured as follows. Section 2 introduces the observation data, climate indices, and GEOS-Chem simulations. Section 3 analyzes the linear trends for annual mean hourly, daytime mean, nighttime mean, and daily mean ozone. Section 4 compares the observed ozone trends and interannual variability with three climate indices relevant to the US air quality. Section 5 uses four GEOS-Chem simulations, with perturbed emissions and meteorological inputs, to quantify the individual effects of climate variability and anthropogenic emissions. Section 6 concludes the study.

The black line in Fig. 3 shows the 1990–2014 US average ozone trends at the individual hours of the day (local standard time). At nighttime hours, the US annual mean ozone grows relatively constantly at a statistically significant rate of about 0.2 ppb yr-1. The growth rate declines in the morning hours and increases during the late afternoon hours. The minimum growth rate is located at around 14:00, when the ozone level peaks, and is slightly negative (but insignificant) with a value of -0.01±0.16 ppb yr-1. The general characteristics of ozone trends are consistent with the results of Jhun et al. (2015). The ozone trends for individual hours tend to weaken the diurnal cycle of ozone – in particular, the diurnal range of ozone (i.e., maximum – minimum) is reduced by 15 % from 26.4 ppb in 1990–1994 to 22.4 ppb in 2010–2014. The contrasting ozone trends between the daytime and the nighttime are indicative of distinctive causes. Hereafter we will focus on trends and (detrended) variability in the daytime mean, nighttime mean, and daily mean ozone, unless stated otherwise.

Table 2 shows the 1990–2014 trends for the US annual average daytime, nighttime and daily mean ozone, and Fig. 4a shows their time series. The US annual average daily mean ozone grows at a rate of 0.16 ppb yr-1 (P value < 0.01 according to an F test). The growth mainly reflects enhanced nighttime mean ozone (at 0.21 ppb yr-1, P value < 0.01). The daytime mean ozone grows at a much lower rate of 0.09 ppb yr-1 (P value < 0.05). The implied total growth from 1990 to 2014 is about 4.1, 2.3, and 5.3 ppb for daily mean, daytime, and nighttime ozone, respectively. Similar to the enhanced US annual average ozone, increasing trends of ∼ 1 ppb yr-1 in the annual mean ozone are observed at mountainous sites (e.g., Tanimoto, 2009) and regional background stations (e.g., Wang et al., 2009) in Asia. In contrast, European annual mean ozone levels have on average been decreasing during the last 20 years (e.g., Sicard et al, 2013). Furthermore, annual mean surface ozone at a background station in eastern China has declined (Xu et al., 2008).

Table 2 also differentiates the ozone trends for individual seasons over the eastern and western United States (separated by 100∘ W). Overall, the growth rates are higher over the west than over the east, and the regional difference reaches about 0.15 ppb yr-1 in summer (June, July, and August) for daytime, nighttime, and daily mean ozone. Seasonally, the most significant growth occurs in spring, with growth rates at 0.17–0.26 ppb yr-1 for the US average daytime, nighttime, and daily mean ozone. For the nighttime ozone, the range of growth rates across the seasons is smaller over the west (0.17–0.30 ppb yr-1) than over the east (0.05–0.24 ppb yr-1). For the daytime ozone over the east, the increases in spring, fall, and winter (0.07–0.15 ppb yr-1) are contrasted by a reduction in summer (-0.11 ppb yr-1), resulting in a weakened trend for the annual mean ozone (0.06 ppb yr-1). The decreases in summertime ozone are associated with reductions in the US anthropogenic emissions of precursor gases NOx, carbon monoxide (CO), and NMVOC revealed from the National Emissions Inventory (Fig. 4b), reflecting the success in controlling peak-level ozone (Cooper et al., 2012; Lefohn et al., 2010; Simon et al., 2015). Similarly, a substantial decrease in precursor concentrations over the last decades in Europe, in line with the long-term emission declines (Colette et al., 2011; Wilson et al., 2012), has resulted in a reduction in ozone episodes (Guerreiro et al., 2014). In contrast, the warm season afternoon ozone over eastern China has been growing at rates of 1–3 ppb yr-1 over the past 20 years (Sun et al., 2016, and references therein) as a result of rapidly growing precursor emissions.

Figure 5 further shows the spatial distribution of ozone trends across the 124 grid cells for daily mean, nighttime mean, and daytime mean ozone in individual seasons. The nighttime ozone has grown or remained constant in most seasons and grid cells, although ozone reductions are also visible in the summertime eastern US and at sparse places in other seasons. For the daytime ozone, the summertime declines over the eastern United States and southern California are contrasted by the summertime increases at other places and the springtime increases over most regions. The springtime daytime increases over the western US are consistent with the MDA8 ozone growth rates (0.2–0.5 ppb yr-1) reported by Lin et al. (2017). Averaged across all seasons, the nighttime ozone has grown in most grid cells, while the daytime ozone has grown over the west with mixed trends over the east (Fig. 5b, c). The spatial distribution of the trends in the annual daily mean ozone (Fig. 5a) closely resembles the trends in the nighttime ozone. The local patterns in ozone trends reflect the effects of small-scale emission, chemical, and/or meteorological features (Cooper et al., 2012; Jhun et al., 2015; Simon et al., 2015; Strode et al., 2015).

Table 3 presents the trends at the 5th, 50th, and 95th percentiles for the daytime mean, nighttime mean and daily mean ozone, separately for urban, suburban, and rural sites. For each site, we calculated daily mean, daytime mean, and nighttime mean ozone mixing ratios in each day and then computed the 5th, 50th, and 95th ozone percentiles in each year. We then averaged the annual data over each site type (rural, suburban, or urban) for a subsequent trend analysis. Overall, the 50th percentile trends are positive for daytime, nighttime, and daily mean ozone at all site types, consistent with the trends in the US annual mean ozone (Table 2).

Table 3 shows that for the nighttime mean ozone, trends at the 5th, 50th, and 95th percentiles are relatively consistent: the 5th percentile ozone grows by 0.19–0.21 ppb yr-1 across the three site types, the 50th percentile ozone grows by 0.19–0.22 ppb yr-1, and the 95th percentile ozone increases at 0.22–0.27 ppb yr-1.

Table 3 further shows that for the daytime mean ozone, trends at the 5th and 95th percentiles differ greatly, and there are only small differences across the rural, suburban and urban sites. The 5th percentile daytime mean ozone grows rapidly by 0.28–0.29 ppb yr-1 across the three site types, while the 95th percentile ozone declines by 0.13–0.21 ppb yr-1. At the rural sites, the growth rate is about 0.29 ppb yr-1 (P value < 0.01) at the 5th percentile and -0.13 ppb yr-1 (P value < 0.01) at the 95th percentile. For the 50th percentile daytime mean ozone, trends differ greatly between rural, suburban, and urban sites: the growth rate varies from 0.02 ppb yr-1 (statistically insignificant, P value = 0.95) for the rural sites to 0.19 ppb yr-1 (P value < 0.01) for the urban sites.

Overall, the distinctive trends at the three percentiles reflect a decrease in the peak ozone contrasted by an increase in the low ozone (Table 3), a decrease in summer compensated by an increase in other seasons (Table 2), and stronger tendency of ozone growth over the west than over the east (Table 2 and Fig. 5).

Our daytime results in Table 3 are consistent with Cooper et al. (2012), who examined the afternoon ozone trends over 1990–2010 at 53 rural sites. Our 5th and 95th percentile daytime ozone trends at rural, suburban, and urban sites are also broadly consistent with Simon et al. who analyzed the trends over 1998–2013 in the MDA8 ozone (Simon et al., 2015). The summertime decreases, largest at the 95th percentile, and wintertime increases in the 50th to 5th percentiles over the eastern US are also showed in the MDA8 ozone results of Lin et al. (2017) for 70 rural sites. Note that these previous studies (Cooper et al., 2012; Simon et al., 2015; Lin et al., 2017) were focused on the high-ozone afternoon hours and excluded the low-ozone early morning hours that have experienced notable ozone growth (black line in Fig. 3).

We relate the interannual variability of ozone to AMONI, AMO, and ONI. We detrended the time series of ozone and climate indices with linear fit for a subsequent statistical analysis; detrending the ozone time series removed the effects of continuously rising Asian emissions and declining US emissions (Zhang et al., 2008; Lin et al., 2008a; Huang et al., 2008; Cooper et al., 2010; Simon et al., 2015; Jhun et al., 2015; Verstraeten et al., 2015; Lin et al., 2017).

Figure 6a–c show the time series of detrended annual AMONI, AMO, and ONI indices between 1990 and 2014, in comparison with the detrended ozone data. The three indices are interannually consistent with the daytime and daily mean ozone anomalies. The AMONI anomaly correlates strongly to the daytime (R = 0.71, P value < 0.01 by a one-sided t test) and daily mean (R = 0.62, P value < 0.01) ozone anomalies, stronger than AMO and ONI. The negative AMONI anomalies in the early 1990s, 2009, and 2014, contributed by negative AMO and positive ONI (El Niño-like), correspond to negative anomalies in the daytime and daily mean ozone. By comparison, the positive AMONI anomalies around the late 1990s and early 2000s are associated with positive ozone anomalies.

Figure 7a and b further show the correlation between detrended AMONI and detrended daily and daytime mean ozone in individual grid cells. The AMONI–ozone correlation is positive and statistically significant over most of the eastern US, and it reaches 0.82 over parts of the southeast. A positive AMONI anomaly is associated with increased temperature over the east (Fig. 8c), which enhances biogenic emissions, changes chemical reaction rates, and changes atmospheric circulation, which overall lead to increased ozone formation and buildup (Jacob and Winner, 2009; Shen et al., 2015; Fu et al., 2015; Xu et al., 2017). AMONI also correlate positively to ozone over the high-altitude west. This is because a negative ONI anomaly (La Niña-like) means a decrease in lower-tropospheric transport of ozone-poor air from the Eastern Pacific (Xu et al., 2017) and a more meandering subtropical jet and strengthened ozone transport from the stratosphere that compensates for weakened transport from Asia (M. Y. Lin et al., 2015). AMONI correlates negatively to ozone over southern California, likely reflecting reduced temperature there associated with a positive AMONI (negative ONI) anomaly (Fig. 8b, c).

Figure 6a–c show that the detrended annual nighttime ozone anomaly corresponds weakly to the three climate indices, with statistically insignificant correlations at 0.15–0.34 (P value = 0.39–0.78). Figure 7 also shows that AMONI is statistically significantly correlated to the detrended nighttime ozone only in a few grid cells. We also found statistically insignificant detrended correlations between the nighttime ozone and other climate indices such as the Pacific Decadal Oscillation (R = 0.06), the Arctic Oscillation (R = -0.05), and the North Atlantic Oscillation (R = -0.29). A possible reason for weaker correlations between climate indices and the nighttime ozone anomaly (compared to the daytime and daily anomalies) is the distinct chemistry at night (i.e., titration by nitrogen) compared to the daytime photochemistry. As the vertical mixing weakens at night, the chemical process becomes more localized and may be less sensitive to large-scale climate variability.

Table 4 further shows the seasonal and regional differences in the correlation between detrended AMONI and detrended ozone. Over the eastern US, the correlations for daytime ozone reach 0.72 in summer and 0.74 in fall. The correlations for nighttime ozone are also relatively large (0.55–0.60) due to influences by “residual” ozone transitioned from the daytime (see Sect. 5.2 for diurnal cycle of model–observation correlations). The correlations are very weak (0.06–0.18) in winter and spring for both daytime and nighttime ozone. Over the western US, the correlations do not exceed 0.50 in all seasons for daytime, nighttime, and daily mean ozone, smaller than those over the eastern US. This is likely due to compensating effects on different transport and chemical processes. For example, a positive AMONI are associated with enhanced ozone production (due to enhanced temperature and reduced precipitation in the context of a positive AMO), weakened trans-pacific ozone transport from Asia (due to a negative ONI with a La Niña pattern), and strengthened STE. Nonetheless, the correlations for the western US are higher in winter and spring (0.37–0.50) than in summer and fall (0.19–0.27). Overall, Table 4 and Fig. 7 suggest that the AMONI-associated large-scale climate variability (AMO and ENSO) affects the warm season eastern US ozone (likely via chemical processes) and cold season western US ozone (likely via dynamic transport processes).

We further used GEOS-Chem simulations to investigate the distinctive effects of interannual climate variability and anthropogenic emissions on the US daytime and nighttime ozone.

We investigated the simulated US ozone changes and driving factors from 2004 through 2012, in which years assimilated meteorological data are available to drive model simulations. Figure 9 shows that the measured annual ozone trends over 2004–2012 (Fig. 9d–f) are stronger than, but spatially consistent with, the trends over 1990–2014. For the US annual mean, the growth rates over 2004–2012 are 0.43 ppb yr-1 for the nighttime ozone and 0.14 ppb yr-1 for the daytime ozone, about twice the growth rates over 1990–2014. The diurnal cycle of US annual ozone trends over 2004–2012 is similar to the cycle for the 1990–2014 trends (red solid versus black solid line in Fig. 3), although with a stronger diurnal range (maximum – minimum).

We note here that the stronger trend in annual ozone over 2004–2012 is partly due to the choice of beginning and end years (Bacer et al., 2016). For example, the growth rates over 2005–2011 are 0.31 ppb yr-1 for the nighttime mean, 0.13 ppb yr-1 for the daytime mean, and 0.23 ppb yr-1 for the daily mean ozone (Table 2), consistent but smaller than the trends over 2004–2012. As an extreme case, the growth rates between 2002 (with a local ozone maximum) and 2014 (with a local minimum) are only 0.13 ppb yr-1 (P value < 0.05) for nighttime ozone, 0.05 ppb yr-1 for daytime ozone, and 0.08 ppb yr-1 for daily mean ozone. For seasonal ozone, the trend differences between 2004–2012 and 1990–2014 are generally similar to the differences for annual ozone (Table 2). A possible reason for the stronger ozone trend in the recent decade is that anthropogenic emissions of NOx decline much more rapidly (Fig. 4b) over 2004–2012 (at a rate of 4.1 % yr-1 relative to 2004) than over 1990–2014 (2.1 % yr-1 relative to 1990). Also, a heat wave swept much of the US in 2012, partly contributing to the high value in that year.

This work shows that reductions in the US anthropogenic emissions have effectively lowered the summertime daytime ozone from 1990 to 2014, consistent with previous studies on afternoon or MDA8 ozone. On an annual mean basis, the daytime ozone have continued to increase. Furthermore, the great sensitivity of the annual average daytime ozone to interannual climate variability increases the difficulty in projecting future ozone air quality (Jacob and Winner, 2009; Rieder et al., 2015). The daily mean and particularly the nighttime ozone have experienced substantial growth due to weakened titration by NOx. This likely implies potential growth in health risk from long-term exposure of enhanced low- and medium-level ozone (Jerrett et al., 2009; Bell et al., 2006; Peng et al., 2013). As the extent of outdoor activities differs notably at different times of day, the overall effect of ozone trends at individual hours on public health warrants further research. Nonetheless, pollution mitigation strategies might consider to address ozone changes at different times of the day and not only during peak hours.