Assessing the manageable portion of ground-level ozone in the contiguous United States

Abstract

Regional air quality models are widely being used to understand the spatial extent and magnitude of the ozone non-attainment problem and to design emission control strategies needed to comply with the relevant ozone standard through direct emission perturbations. In this study, we examine the manageable portion of ground-level ozone using two simulations of the Community Multiscale Air Quality (CMAQ) model for the year 2010 and a probabilistic analysis approach involving 29 years (1990–2018) of historical ozone observations. The modeling results reveal that the reduction in the peak ozone levels from total elimination of anthropogenic emissions within the model domain is around 13–21 ppb for the 90^th −100^th percentile range of the daily maximum 8-hr ozone concentrations across the contiguous United States (CONUS). Large reductions in the 4^th highest 8-hr ozone are seen in the regions of West (interquartile range (IQR) of 17–33%), South (IQR 22–34%), Central (IQR 19–31%), Southeast (IQR 25–34%) and Northeast (IQR 24–37%). However, sites in the western portion of the domain generally show smaller reductions even when all anthropogenic emissions are removed, possibly due to the strong influence of global background ozone, including sources such as intercontinental ozone transport, stratospheric ozone intrusions, wildfires, and biogenic precursor emissions. Probabilistic estimates of the exceedances for several hypothetical thresholds of the 4^th highest 8-hr ozone indicate that, in some areas, exceedances of such hypothetical thresholds may occur even with no anthropogenic emissions due to the ever-present atmospheric stochasticity and the current global tropospheric ozone burden.

1. Introduction

Ground-level ozone is a widespread secondary air pollutant that has been linked to harmful impacts on human and ecosystem health, and a variety of materials. It is formed in the presence of sunlight involving chemical reactions of Nitrogen Oxides (NOx, including NO and NO₂) and Volatile Organic Compounds (VOC). According to the 2014 National Emissions Inventory (NEI) report, biogenic VOC emissions account for 70% of the total VOC burden in CONUS and the remainder originate from mobile and stationary sources; NOx is mostly anthropogenically emitted (> 93%) by mobile and stationary sources such as electricity generation, industrial processes, and boilers (U.S. EPA, 2017). Moreover, processes such as stratospheric intrusions and intercontinental transport are also identified as important ozone sources (National Research Council (NRC), 2009).

Regional-scale air quality models, such as the U.S. Environmental Protection Agency (EPA)’s Community Multiscale Air Quality (CMAQ) model (Byun and Schere, 2006), are widely used to design emission control strategies needed to comply with the ozone standard. Estimated design value for the future year is derived by applying modeled change to the observed base design value (i.e., weighted average of 5 years of observed 4^th highest values) based on simulations with current and perturbed future emissions (while the future meteorological conditions are kept the same as those that occurred in the base year), and then compared with the National Ambient Air Quality Standards (NAAQS) (U.S. EPA, 2018). An extreme counterfactual emissions reduction scenario where all anthropogenic emissions within CONUS are set to zero, i.e., the “zero-out” approach, is the most common modeling approach to quantify the U.S. background ozone concentration (Jaffe et al., 2018).

In this study, we explore the manageable portion of the ozone problem as it pertains to processes that can be managed via regulatory policies (i.e., controls that can be placed on the anthropogenic emission sources). We employ two sets of CMAQ (version 5.0.2) simulations using meteorology from the Weather Research and Forecasting (WRF) model; the base case model simulation reflecting the “current” conditions for the year 2010 (hereafter referred to as BASE) and the “zero-out” scenario where all anthropogenic emissions are turned off in the entire modeling domain (hereafter referred to as EM_ZERO). Although there are numerous studies that analyzed model simulations for similar emission scenarios, this is the first study that focuses on the impact of such emission scenarios on the long-term forcing and develops a probabilistic framework for assessing the ozone exceedances. The objective of this investigation is to assess the impacts of the “zero-out” emission scenario on ozone levels within the model domain for different times of the year and at various percentiles of the ozone distribution, and quantify the contribution of anthropogenic emissions loading to the ozone extreme values (4^th highest) using the probabilistic method described by Luo et al. (2019). Further, we examine the probability of exceeding several hypothetical thresholds for the 4^th highest 8-hr ozone using 29 years of historical ozone time series data combined with model results for the base case and the case with no anthropogenic emissions in model simulations. The description of the data and method of analysis is presented in Section 2; results are discussed in Section 3, and conclusions are presented in Section 4.

2. Data and Methods

2.1. Observations and Model simulations

Observations of the daily maximum 8-hr ozone concentration time series (DM8HR) were obtained from the U.S. EPA’s Air Quality System (AQS) (last accessed April 2020 at https://aqs.epa.gov/aqsweb/airdata/download_files.html). A total of 176 stations with annual data coverage above 80% for at least 25 (25+) years during the 1990–2018 period (to provide a variety of synoptic-scale weather conditions) were analyzed. They are grouped into seven geographical regions following the U.S. Climate Regions designation (https://www.ncdc.noaa.gov/monitoring-references/maps/us-climate-regions.php): Southwest (SW), South (S), West (W), Southeast (SE), Central (C), Northeast (NE) and West North Central (WNC). The number of sites within each region can be found in the legend of Fig. 1a. The influence of emissions on ground-level ozone is also examined at rural, suburban and urban sites (Fig. 1b). The location setting (i.e., site classification) is determined based on the percentage of impervious surface in the 1-km grid cell of 2011 National Land Cover Database (https://www.mrlc.gov/data/legends/national-land-cover-database-2011-nlcd2011-legend) where the AQS monitoring site falls within: [0, 20) for rural (No development and Developed, Open Space), [20, 50) for suburban (Developed, Low Intensity), and [50, 100] for urban (Developed, Medium and High Intensity) sites. To address the vast number of sites eliminated in the SE, C, and NE regions because of the required ozone monitoring season in some states being less than an entire year, we also performed a separate assessment in which we considered a larger set of sites that were constructed by applying the same data completeness requirement for the March to October period only. Analysis based on those 353 sites for March to October (Fig. S1) is referred to as the “Mar-Oct case” and presented in the supplemental material (Fig. S5, S7–S9, S11, S12).

Figure 1.

AQS sites grouped by (a) geographical regions: Southwest (SW), South (S), West (W), Southeast (SE), Central (C), Northeast (NE) and West North Central (WNC) and (b) location settings: rural, suburban and urban. Numbers in the legends indicate the number of sites in the corresponding group.

Time series of two sets of WRF/CMAQ (version 5.0.2) model simulations for the year 2010 were employed to study the maximum influence of anthropogenic emissions on the ground-level ozone concentrations over CONUS. These simulations, covering the CONUS, the southern part of Canada, and the northern part of Mexico, were performed with a horizontal grid spacing of 12km×12km in the offline mode (i.e., uncoupled chemical transport and meteorological models). With the exception of emission inputs for EM_ZERO noted below, the set-up of these simulations is identical to the simulations described in Hogrefe et al. (2018) that were performed as part of the third phase of the Air Quality Model Evaluation International Initiative (AQMEII3) (Rao et al., 2011a; Galmarini et al., 2017). The first simulation, BASE, uses the meteorology, emissions, and lateral boundary conditions representative of the 2010 conditions, while the second simulation, EM_ZERO, reflects the condition where all anthropogenic emissions were eliminated within the WRF/CMAQ modeling domain. The EM_ZERO scenario analyzed in this study differs from the EM_ZERO scenario described in Hogrefe et al. (2018) in that the previous study eliminated not only anthropogenic but also wildfire emissions, which are largely not controllable via a regulatory policy, whereas, in this study, wildfire emissions are identical between BASE and EM_ZERO. Figure S2a indicates that DM8HR from the BASE simulation is highly correlated with those observed in 2010 with a median correlation coefficient of 0.79 across all sites. The correlation is even higher (median correlation coefficient of 0.89) for the long-term (i.e., baseline) component (see Fig. S2b; the spectral decomposition is discussed in Section 2.2.). The elevated correlation for the long-term component agrees with previous studies that models can best simulate the long-term variations embedded in daily ozone time series data (Hogrefe et al., 2001; Astitha et al., 2017). Since our approach utilizes model-predicted BL changes (see Section 2.2), correlation is the most relevant metrics for model evaluation. Further details on the configuration of the WRF/CMAQ system and evaluation of the meteorological fields and air quality variables for the BASE simulation can be found in Solazzo et al. (2017a, b) and Hogrefe et al. (2018).

2.2. Ozone decomposition and reconstruction

It is well recognized that various atmospheric processes operating on different timescales are embedded in ambient ozone time series data (see Rao et al., 1997 and Fig. 2 in Dennis et al., 2010). Different filtering techniques such as the Empirical Mode Decomposition (Huang et al., 1998), Elliptic filter (Poularika, 1998), Kolmogorov-Zurbenko (KZ) filter (Rao and Zurbenko, 1994), Adaptive Filtering Technique (Zurbenko, et al., 1996), and Wavelet (Lau and Weng, 1995) can be used to achieve scale separation in time series of meteorological and air quality variables (Hogrefe et al., 2003). Rao et al. (2020) applied a modified version of the Empirical Mode Decomposition (EMD), known as the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (Colominas et al., 2014) and the KZ filter to the observed ozone time series in 2010 and found that both methods yielded similar spectral decomposition results for the DM8HR time series data (see Fig. 2 in Rao et al., 2020). Therefore, only the results from the KZ filter are presented in this paper to assess the impacts of emissions forcing and the stochastic nature of the atmosphere on the observed 8-hr ozone time series data.

Figure 2.

DM8HR ozone decomposition and reconstruction at an example AQS site (ID 040131004) in Phoenix, AZ for (a) 2010 observed scenario (red), and (b) projected scenario with anthropogenic emissions removed (blue). In both cases, the long-term baseline component (BL) is plotted as thick colored lines and reconstructed ozone time series (O₃ Rec) are represented by thin gray lines, with associated 4^th highest (4^th Rec) for each reconstruction marked by colored crosses. In (a), the actual observed 8-hr ozone time series (O₃) and the 4^th highest 8-hr ozone concentration (4^th) is marked as the black line and the horizontal thin red line. The range (minimum to maximum) of the reconstructed 4^th highest concentrations (4^th) and the number of days exceeding the NAAQS ozone threshold of 70 ppb (ex) is shown in brackets. In (a), the number in front of each range is derived from the original 2010 observations. The dashed reference lines at 70 ppb represent the level of the 8-hr ozone threshold in current NAAQS.

Ozone can be viewed as the baseline of pollution that is created by emission sources and modulated by the prevailing meteorology (Rao et al., 2011b). To remove the short-term synoptic-scale forcing (time scale of 2–21 days) from ozone time series, the low-pass KZ filter KZ(5,5), which represents 5 consecutive repetitions of a 5-day moving average, has been employed to spectrally decompose 25+ years of observed and the above mentioned two sets of modeled DM8HR time series(Rao and Zurbenko, 1994; Eskridge et al., 1997; Rao et al., 1997; Hogrefe et al., 2000 and 2003; Hogrefe and Rao, 2001; Rao et al., 2011b; Porter et al., 2015; Solazzo and Galmarini, 2016; Astitha et al., 2017; Luo et al., 2019; Rao et al., 2020). The filtered output then becomes the long-term baseline component (BL):

$$BL\left(t\right)=KZ\left(5,5\right)$$ (1)

where t = 1, 2, …, 365 (or 366 for a leap year) denotes the days from Jan. 1 to Dec. 31. The magnitude of the BL component is largely governed by anthropogenic emissions, background ozone, and other slow-changing atmospheric processes. Note that this definition of BL, which applies to time series of both observations and model simulations, differs from alternate definitions used in the literature that are based solely on observations, e.g. the distribution of O₃ observations at a rural or remote sites that have not been influenced by recent, local emissions (HTAP, 2010).

The zero-mean synoptic component (SY), which mainly contains the synoptic-scale weather-induced variations embedded in ozone time series, is derived by subtracting the BL from the original or raw ozone time series data:

$$SY\left(t\right)=O_3\left(t\right)-BL\left(t\right)$$ (2)

Where O₃ is the original ozone time series.

The 50% cutoff frequency, meaning the energy at this frequency is separated half-half, of the KZ(5, 5) is 0.0411 cycles per day, which gives a cutoff period of 1/0.0411 ≈ 24 days between the BL and SY components (Rao et al., 1997). Note that the KZ filter works well in the presence of missing data, which is one of the main advantages of the KZ filter as compared to other filtering methods (Hogrefe et al., 2003). If more than half of the days (3, 4, or 5 days) are missing in a window such as the boundary days in early January and late December, the average value would be marked as missing. For example, Fig. 2a presents the observed DM8HR ozone time series (black line; O₃ in Eq. 2) and the decomposed BL component (thick red line; BL in Eq. 1) at a randomly chosen site in Phoenix, AZ. Although there are ~10 days missing at the beginning and the end of each year because of the edge effects of the moving average filter (e.g., for the first 5-day moving average, Jan. 1–2 and Dec. 30–31 would be marked as missing value), it does not pose a problem for the analysis on higher ozone levels as ozone levels during those periods is very low (See Section 3.1 for more details).

To build the potential distribution of ozone concentrations for a specific year, we utilize its baseline and different synoptic components embedded in ozone time series data during the 1990–2018 period to account for the stochastic nature of the atmosphere (see Rao et al., 2020) following the method proposed by Luo et al. (2019). Because the strength of the synoptic forcing (defined as the standard deviation of the time series data in the SY component, SYstd) is highly correlated with the magnitude of the baseline level (mean of the BL component, BLmean), SY components are adjusted to the target BL based on the linear relationship between observed BLmean and SYstd (Luo et al., 2019). The linear relationship can be constructed by estimating parameters p and q using 25+ pairs of observed BLmean and SYstd:

$$\mathit{\text{SYstd}}=f\left(\mathit{\text{BLmean}}\right)=p*\mathit{\text{BLmean}}+q$$ (3)

The target BL can be either a set of baseline concentration time series decomposed from 2010 observations (BL_OBS) or a projected baseline (BL_PROJ) based on the observations (BL_OBS) and the change simulated in the baseline components of modeled BASE and EM_ZERO. Since baseline projection by multiplying the BL_OBS with the ratio of the simulated baseline change could result in very high magnitude of fluctuation in the projected baseline, especially during times when observed low levels of baseline are not very accurately simulated by the model, the projected baseline is calculated by:

$$B{L}_{\mathit{\text{PROJ}}}\left(t\right)=B{L}_{OBS}\left(t\right)+[B{L}_{EM\_\mathit{\text{ZERO}}}\left(t\right)-B{L}_{\mathit{\text{BASE}}}\left(t\right)]$$ (4)

Then, the synoptic forcing can be adjusted based on the target BL using the linear relationship fitted following Eq. 3:

$$S{Y}_{y, adj}\left(\text{t}\right)=S{Y}_y\left(\text{t}\right)*\frac{f\left(\mathit{\text{BLmea}}{n}_{\mathit{\text{target}}}\right)}{f\left(\mathit{\text{BLmea}}{n}_y\right)}$$ (5)

where y is a given year within 1990–2018, indicating the year which the synoptic forcing is decomposed from. The reconstruction is as follows:

$$O_3^{y,\mathit{\text{target}}}\left(\text{t}\right)=B{L}_{\mathit{\text{target}}}\left(\text{t}\right)+{\text{SY}}_{y,adj}\left(\text{t}\right)$$ (6)

For instance, $O_3^{1993,OBS}$ represents the ozone time series reconstructed with observed BL in 2010 (BL_OBS) under the 1993 meteorological conditions SY_1993,adj. A high BL level coupled with strong SY forcing is required to observe high ozone exceedances (Astitha et al., 2017); however, if the magnitude of the BL level is low, even higher SY forcing cannot lead to ozone exceedances. Hence, BL is to be viewed as the deterministic part in the ozone time series data (Rao et al., 2020). In other words, the greater the amount of pollution created by emission sources (i.e., high BL) the greater will be the influence of prevailing meteorology (i.e., long-range pollutant transport due to high SY).

The entire year of 2010 is investigated in this study as opposed to the typical ozone season of May-September that was used in Luo et al. (2019) since large perturbation in emission forcing could possibly result in high ozone concentrations during times other than the typical ozone season (Clifton et al., 2014; Strode et al., 2015; Simon et al., 2016; Seltzer et al., 2020). However, as pointed out earlier, because of the required ozone monitoring season in some states (esp. within SE, C, and NE) being less than an entire year, estimates based on a larger set of sites meeting the same data completeness requirement for Mar-Oct are also included in our discussion (relevant graphs are presented in the supplemental documentation).

Below, we illustrate the details of our approach on the ozone reconstruction with the example monitoring site in Phoenix, AZ. We first constructed the linear relationship between its BLmean and SYstd using Eq. 3 based on 29 years of decomposed 8-hr ozone time series at this site. Then, one ozone distribution for the base year 2010 with 1990 synoptic conditions is constructed with BL_OBS as target BL (BL_target) in Eq. 6 and its average level as BLmean_target in the adjustment of 1990 synoptic forcing (Eq. 5). By repeating the reconstruction with 1991–2018 synoptic conditions, a set of 29 ozone distributions for base year 2010 could be produced. We refer to the set of ozone reconstruction and distribution based on the observed baseline in 2010 and 25+ years of observed synoptic components as “OBS”. Similarly, the set of ozone reconstruction and distribution based on the projected BL (Eq. 4), along with the same set of 25+ years of observed synoptic components is referred to as “PROJ”.

3. Results and Discussion

3.1. Modeled influence of anthropogenic emissions on ozone

To study the influence of anthropogenic emissions at different times and various percentiles of the ozone distribution, the long-term BL components of OBS and PROJ are being compared. The reasons for choosing the BL components rather than the raw DM8HR ozone time series are: 1) emissions-related contributions are mostly contained in the BL component as shown in previous studies (Astitha et al., 2017; Luo et al., 2019; Rao et al., 2020); 2) BL is the deterministic part in ambient ozone data; and 3) filtering out the high-frequency zero-mean process (i.e., SY component) denoises the ozone signals throughout the year (Fig. S3).

The contribution of anthropogenic emissions defined as the portion of BL ozone reduced from OBS to PROJ (i.e., the BASE to the EM_ZERO scenario) has a seasonal feature in most regions. It is clear that as anthropogenic emissions are removed, the period with higher levels of ozone, indicated by its baseline level, shifts to springtime in all seven geographical regions (Fig. 3) due to the active cross-pacific pollutant transport and stratospheric ozone intrusions during springtime (Lin et al., 2012; Clifton et al., 2014; Strode et al., 2015; Simon et al., 2016; Mathur et al., 2017; Hogrefe et al., 2018; Seltzer et al., 2020). The dominant peak in spring and secondary peak in fall seen in PROJ in W (Fig. 3a) is consistent with that observed at Trinidad Head, CA (Fig. S4), a location considered to reflect the background ozone level for the Western U.S. Note that the regional median BL in Fig. 3 does not represent any single site and, hence, its magnitude should not be compared directly with the denoised monthly mean time series at Trinidad Head, CA. Because of their high elevations (≥1 km), sites in the SW region tend to experience higher ozone concentrations and are more influenced by the stratospheric-tropospheric ozone exchanges, yielding a longer period of time with much higher ozone levels spanning spring and early summer with a regional median BL level of PROJ around 50 ppb (Fig. 3b). The reduction of the BL components, signifying the influence of anthropogenic emissions, remains around 20 ppb for a large portion of the typical ozone season of May to September except for WNC, where the anthropogenic emissions contribute around 7 ppb (Fig. 3 and Fig. S5). Because of the limited spatial coverage of monitoring sites, the reduction in C, SE, and NE during June-September is underestimated; the reduction in these three regions could be ~25 ppb based on a larger number of sites in the Mar-Oct study (Fig. S5 d–f).

Figure 3.

Daily reduction of BL components from OBS to PROJ: (a) West (W), (b) Southwest (SW), (c) South (S), (d) Central (C), (e) Southeast (SE), (f) Northeast (NE), and (g) West North Central (WNC). The BL reduction is first calculated by BL_OBS - BL_PROJ at each site; its daily spatial median (Med, black line) and interquartile range (IQR, 25^th and 75^th percentiles; gray shaded area) are shown against the left y-axis. The daily spatial median BL components of OBS and PROJ across all sites in the region is plotted against the right y-axis.

The wintertime decreases in median BL ozone due to the removal of anthropogenic emissions are smaller than the decreases during summer because of less active photochemistry. In some cases, removing the anthropogenic emissions could even increase wintertime BL (W, S, NE, C in Fig. 3) and, thus, the actual ozone level when SY components are added (Fig. S3). Fig. S6 shows that the increased winter ozone BL level is more pronounced in the urban and suburban regions. This is likely due to the reduced titration of O₃ when anthropogenic NOx emissions are removed. The increased wintertime ozone agrees with the increasing trend seen in Simon et al. (2015) during a period of decreasing NOx and VOC emissions (1998–2013) based on observations; they also concluded that the increasing ozone with emission reduction occurred more often in the urbanized areas during winter when the ozone is usually at lower levels.

To further examine the contribution of anthropogenic emissions at each monitored location, we calculate the BL reduction at various percentiles (5^th, 25^th, 50^th, 75^th, and 95^th) at each site (Fig. 4). The BL reduction is based on rank-ordered percentiles of both OBS and PROJ and might represent changes occurred in different seasons because of the overall seasonal shift in BL as a result of the removal of anthropogenic emissions (Fig. 3). The reductions in BL are relatively consistent with a median value of ~9 ppb at all percentiles and locations, with higher reductions at the 95^th percentiles and lower reductions or even slight increases at the 5^th percentile. At the 75^th and 95^th percentiles, the regions with higher BL levels experience the most reduction, especially for Southeastern California where the baseline can be reduced by up to 29 ppb. At the 5^th percentile, the impact of the anthropogenic emission removal has a divergent feature mostly because of the urban/rural difference (the vast majority of the increases in BL are at urban sites). Similar features are also seen in the relative reduction of the BL level at all percentiles. BL reduction based on Mar-Oct sites is also included in Fig. S7. As some lower level of BL in OBS and higher level of BL in PROJ becomes unavailable, the reduction and reduction rate of BL at all percentiles are all elevated.

Figure 4.

Observed BL, BL reduction (OBS - PROJ), and BL reduction rate [(1 - PROJ / OBS) *100 %] at (a-c) 5^th, (d-f) 25^th, (g-i) 50^th, (j-l) 75^th, and (m-o) 95^th percentiles of the entire year 2010. The three numbers in the lower-left corner of each subplot represent the corresponding minimum, median, and maximum across all sites. The negative reductions (increase of BL) from the 50^th percentiles and up is seen only at two urban sites located in the San Francisco Peninsula.

Figure 5 summarizes the continent-wide reduction at various percentiles stemming from zeroing out of all anthropogenic emissions. The median BL and ozone reduction (in ppb, blue in Fig. 5) steadily increased from 7 ppb to 10 ppb from 1^st to 80^th percentiles and then rise to 11 ppb for BL and 13 ppb for ozone at the 90^th percentile; and 14 ppb for BL and 21 ppb for ozone at the 100^th percentile. When divided by the local OBS level, the reduction rate (in %, green in Fig. 5) forms a “U” shape from lower to higher percentiles for both BL and DM8HR. Their reduction rates are almost identical at 22% and 20%) from the 20^th to 80^th percentiles for baseline and DM8HR, respectively. The DM8HR has a more pronounced reduction rate at the lower and higher percentiles because of the fluctuations caused by varying synoptic components. The spatial spread of the reduction rate is especially larger at lower percentiles for both BL and DM8HR. This signifies that with an extreme scenario of zeroing out of anthropogenic emissions, we see a decrease ranging from 20% to 26% for the BL level and from 19% to 40% for the DM8HR ozone levels. Summarization for Mar-Oct can be found in Fig. S8.

Figure 5.

Continental spatial median concentration and reduction of (a) BL and (b) ozone concentrations. Brown line in (a): median observed BL concentration across all sites at 1–100^th percentiles. The reduction (OBS – PROJ) and reduction rate [(1 - PROJ / OBS) *100 %] are first calculated at each site. Then the median across all sites (Med) is shown as blue (in ppb) and green (in %) lines with shaded area representing the corresponding interquartile ranges (IQR). The same applies in (b) except that the DM8HR ozone at a specific site is the median of 25+ reconstructed ozone time series based on the observed BL (OBS) or projected BL (PROJ).

3.2. Modeled manageable portion of ozone exceedances

The 4^th highest ozone concentration denotes the upper tail of the annual ozone distribution and is used to derive the ozone design value used in NAAQS. The extreme values are highly variable and susceptible to the stochastic nature of the atmosphere, synoptic weather conditions, and episodic natural events such as stratospheric intrusion and wildfires (Rao et al., 2020). Moreover, it is usually more difficult to accurately simulate the 4^th highest than the baseline component (Astitha et al., 2017; Porter et al., 2015; Rao et al., 2020). In this section, we utilize 25+ years of historical ozone observations (1990–2018) along with results from the modeled BASE and EM_ZERO scenarios to examine the impact of anthropogenic emissions on the 4^th highest ozone concentration towards defining the modeled manageable portion of the ozone exceedances and estimating the probabilities of exceedance for several hypothetical levels of thresholds for the 4^th highest ozone. The former is represented by the reduction rate from OBS to PROJ, i.e., 2010 base case to zero-out emissions scenario, while the latter is reflected in the probability of the 4^th highest ozone exceeding various hypothetical ozone thresholds based on the PROJ scenario (Pex). To be more specific, Pex is calculated as the ratio of the number of reconstructed 4^th highest values larger than the given ozone threshold to the total number of PROJ reconstructions (25+) at each site, multiplied by 100%.

For the 2010 base year, the median values for the 4^th highest ozone concentration based on ozone reconstructions with 25+ years of SY components ranges from 47 to 102 ppb (Fig. 6a); most of the sites with median 4^th highest ozone >70 ppb are in the S, NE, and inland W regions. By eliminating all anthropogenic emissions, all median 4^th highest ozone values fall below the current ozone threshold of 70 ppb as expected, and levels for the 4^th highest >50 ppb are mostly found in W, SW, S and NE regions (Fig. 6b). The higher 4^th highest ozone values in the projected scenario found at sites with elevations ≥1 km in SW reflect the high contribution of background ozone from the global ozone burden and stratospheric ozone intrusion (e.g., Jaffe et al., 2018). The modeled manageable portion of the 4^th highest ozone (Fig. 6d; percent reduction) shares almost identical geographical features with the reduction in the 4^th highest between OBS and PROJ (Fig. 6c). The two isolated negative reductions are seen at sites located in the San Francisco Peninsula, with an increase of 3 and 14 ppb for the median 4^th highest ozone value corresponding to an increase of 6% and 29%, respectively. Unfortunately, this analysis could not be extended to the Mar-Oct sites (Fig. S9) since high ozone levels could possibly be cut out within that time span (see high levels of BL in PROJ outside the Mar-Oct window in Fig. 3). This is especially true for the PROJ scenario and regions lacking spatial coverage (C, SE, NE) as seen from the bias maps of the median 4^th highest ozone concentration (Fig. S10).

Figure 6.

Median of 4^th highest ozone distribution at each site based on 25+ ozone reconstructions with 1990–2018 SY components: (a) with BL decomposed from observations in 2010, (b) with projected baseline PROJ; and their (c) reduction (OBS - PROJ) and (d) reduction rate [(1 - PROJ / OBS) *100 %] (i.e. the manageable portion) in by eliminating anthropogenic emissions. The three numbers in the lower-left corner of each subplot represent the corresponding minimum, median, and maximum across all sites. The two negative reduction sites in c) and d) are the same ones seen in the 50^th percentiles and up maps in Figure 4, located in the San Francisco Peninsula.

Figure 7 summarizes the manageable portion of modeled ozone for different regions. In general, the SW region sees lower levels of manageable portion in the 4^th highest ozone with a spatial median of 16%. W, S, C, SE, and NE sites have around 24–30% manageable portion of the 4^th highest ozone with interquartile ranges of 17–33%, 22–34%, 19–31%, 25–34% and 24–37% respectively. The analysis on the subsets of sites where 4^th highest ozone is >70 ppb shows that the manageable portion of the 4^th highest ozone at these locations tends to be higher than that at the entire set of sites. This also implies that the S and NE regions where higher 4^th highest ozone values are widespread (38 out of 46 sites and 16 out of 19 sites with the 4^th highest > 70ppb respectively) in the year 2010 (Fig. 7), are very likely to improve their air quality by implementing additional emission reductions. Likewise, the W sites with the 4^th highest ozone >70 ppb (Fig. 6 and 7) generally feature a large manageable portion of the 4^th highest ozone (28–38%). However, it could be difficult for some of the SW sites with the 4^th highest ozone >70 ppb to see much improvement by controlling emissions since more than 77% (100% minus the manageable portion in SW, which is at most 23% after excluding the outlier site) of the ozone problem is estimated to be not attributable to anthropogenic emissions within the modeling domain. Similar features can be seen in Fig. S11 for Mar-Oct with overestimated manageable portion of ozone because of the same reasons discussed in the previous paragraph as well as the addition of more sites in all regions except for WNC.

Figure 7.

Regional summary of manageable portion of the 4^th highest ozone concentration. Spatial boxplots of medians of the 25+ 4^th highest ozone distribution reconstructed with BL of (a) OBS and (b) PROJ; (c) theoretical manageable portion of ozone [(1-PROJ/OBS) *100 %]. The thinner boxplot in each region represents sites with the 4^th highest in 2010 exceeding the current ozone threshold of 70 ppb. The number of sites in each boxplot is shown in (b).

The probabilities of the 4^th highest ozone exceeding various lower ozone thresholds (Pex) in Fig. 8 reveal potential challenges in future revisions of the ozone NAAQS. Given the ever-present atmospheric stochasticity and current estimated global ozone burden in our model simulations, some of the SW sites may not be able to attain any stricter ozone thresholds even with drastic reductions in the anthropogenic emission loading within the model domain. For example, if the ozone threshold were to be set at 60 ppb, the probabilities of exceeding that threshold at some W, SW, and NE sites would be more than 60% (Fig. 8c). For a hypothetical ozone threshold of 55 ppb, the results of this analysis suggest that it would be difficult, if not impossible, to meet that threshold at a majority of the sites. The Pex for Mar-Oct in Fig. S12, underestimated compared to Pex in Fig. 8, provides a similar picture with a higher spatial coverage.

Figure 8.

The lowest achievable ozone thresholds represented by the probability of 4^th highest ozone based on PROJ scenario exceeding the current and hypothetical ozone thresholds (Pex) of: (a) 70 ppb, (b) 65 ppb, (c) 60 ppb and (d) 55 ppb. Value of 0 is shown as a small open circle. The three numbers in the lower-left corner of each subplot represent the corresponding minimum, median, and maximum across all sites.

4. Conclusions

In this study, we examined the maximum impacts of anthropogenic emissions on the ozone long-term (i.e., BL) component, the entire ozone concentration distribution, and the 4^th highest ozone concentration with the aid of WRF/CMAQ simulations for the 2010 base year and a projection with zero-out anthropogenic emissions. The modeled manageable portion of ozone exceedances is also estimated based on the probabilistic method described in Luo et al. (2019) under different historical synoptic forcings that were observed during1990 to 2018 to reveal the theoretical upper limit for the manageable portion of ozone exceedances. The results indicate that the typical ozone season (May to September) would witness an ozone level reduction in the range of 20–25 ppb in most regions if all anthropogenic emissions were eliminated. While higher levels of ozone are indicated by higher baseline levels, occurrences of the 4^th highest ozone tend to appear more frequently in the springtime due to the shift in the chemical regime of the ozone cycle when anthropogenic emissions are non-existent. This study provides insights into the maximum ozone improvement that is achievable given the current global tropospheric ozone burden and stochastic nature of the atmosphere as represented in the modeling system and historical observations used in this study. In general, levels of ozone at the annual 90^th −100^th percentile could be reduced by around 13–21 ppb across CONUS. Larger manageable portions of the 4^th highest ozone are seen in the W (IQR: 17–33%), S (IQR: 22–34%), C (IQR: 19–31%), SE (IQR: 25–34%) and NE (IQR: 24–37%) regions. However, some W and SW sites exhibit much less influence from the domain-wide anthropogenic emissions loading because of the influence of intercontinental ozone transport and stratospheric intrusions, indicating that it may be difficult for them to maintain the 4^th highest ozone at <60 ppb given the current global ozone burden and the degree of stochasticity present in ozone extreme values. Hopefully, in the future, more sites would expand their monitoring season in order to capture the potential seasonal shift in the high level of ozone.

Implication Statement.

Because air pollution is intricately linked to adverse health effects, National Ambient Air Quality Standards (NAAQS) have been established for criteria pollutants to safeguard human health and the environment. Areas not in compliance with the relevant standards are required to develop plans and policies to reduce their air pollution levels. Regional-scale air quality models are currently being used routinely to inform policies to identify the emissions reduction required to meet and maintain the NAAQS throughout the country. This paper examines the feasibility of the 4^th highest ozone, which is used to derive the ozone design value for NAAQS, complying with various current and hypothetical 8-hr ozone thresholds over CONUS based on the information embedded in 29 years of historical ozone observations and two modeling scenarios with and without anthropogenic emissions loading.