Air quality modeling for the NA domain and calendar year 2010 used CAMx version 6.2 (Ramboll Environ, 2015) to simulate formation and transport of O3. Gas-phase chemistry was included using the Carbon Bond (CB05) mechanism (Yarwood et al., 2005), but heterogeneous-phase chemistry was not included for efficiency and because the focus is on O3. The CAMx modeling domain covers the continental US with 459 by 299 grid cells of 12 km by 12 km resolution and 26 vertical layers (Table S1 in the Supplement). The vertical height of the first layer is 20 m. Model inputs were prepared from data provided to all AQMEII participants supplemented by other data sources (described later). The 2010 annual simulation was initialized on 22 December 2009 to limit the influence of initial concentrations.

This study uses reactive tracers with a chemical mechanism compiler (RTCMC) in CAMx to track contributions from BCs. The RTCMC module simulates explicit tracers in parallel to the host model and represents sources (emissions and BCs), transport processes (advection and diffusion), deposition, and user-defined chemistry (Yarwood et al., 2014). RTCMC chemistry can use species concentrations from the host model, e.g., OH radical, in the reactions of tracers. If no chemistry is defined for an O3 tracer, then it becomes chemically inert and deposition is the only sink. Currently, the RTCMC models dry deposition of tracers but not wet deposition, which will result in conservative estimates of BC contributions.

CAMx can also track BC O3 contributions through the ozone source apportionment technology (OSAT; Ramboll Environ, 2015) module, but we use RTRAC because it offers flexibility to model reactive and inert tracers, and to track BC O3 from specific groups of vertical layers. Vertical attribution is valuable in identifying height ranges of transported O3 that are most influential to ground-level O3. Baker et al. (2015) compared RTCMC and OSAT over North America and found that the two approaches estimated very similar BC O3 contributions in warmer months. Baker et al. also evaluated computational efficiency of RTCMC tracers. We use the same RTCMC scheme for reactive O3 tracers as Baker et al. (2015) (Table S2), which includes O3 destruction by odd hydrogen (i.e., HO2 and OH) and alkenes but not O3 destruction by NO because this would entail tracking conversion of BC O3 to/from several NOy species, including NO2, PAN and HNO3. OSAT accounts for O3 destruction by NO and Baker et al. show that omission of O3 destruction by NO in the RTCMC scheme causes some positive bias in BC O3 estimates in winter months (Baker et al., 2015).

The BC O3 contributions are analyzed in terms of MDA8 by season because the MDA8 is relevant to the US National Ambient Air Quality Standard (NAAQS) for O3, which is set at a level of 70 ppb. Seasonal averages are used to evaluate how BC contributions depend upon transport patterns and photochemistry.

Two sensitivity simulations defined by AQMEII were conducted to quantify O3 response to anthropogenic emission reductions. The GLO scenario reduced all anthropogenic emissions by 20 % globally and the EAS scenario reduced anthropogenic emissions in the East Asia by 20 %. In the GLO scenario, the CAMx NA anthropogenic emissions were reduced by 20 % for the entire modeling domain with no changes to fire and biogenic emissions. Under the EAS scenario the CAMx NA emissions are the same as in the base case. ECMWF provided BCs specific for each scenario. The ECMWF BCs inadvertently omitted H2O2 for the EAS scenario and so we followed the AQMEII suggestion to use H2O2 from the base case for both the EAS and GLO scenarios. Other modeling inputs are unchanged from the base case. The annual CAMx sensitivity simulations were conducted with the inert and active boundary O3 tracers described above. Several metrics are used when comparing sensitivity scenarios to the base case including the fourth highest maximum daily 8 h average (H4MDA8), the average of the 30 highest MDA8 O3 days (Top 30), and seasonal average MDA8. We show the results for the entire modeling domain and additionally discuss our findings for 22 selected major cities in a wide variety of climatic and geographic environments (Fig. S1 in the Supplement). Cities are represented by the monitoring site with highest H4MDA8 in the metropolitan statistical area.

We evaluate O3 model performance for the base case relative to benchmarks that are accepted in NA. Predictions of MDA8 O3 are evaluated against observations from the Clean Air Status and Trends Network (CASTNET; rural) and the Air Quality System (AQS; urban and rural). Observations are considered valid when at least 75 % of data are available. A 40 ppb O3 cutoff is applied to focus on the upper end of O3 frequency distributions. Statistical metrics used in this evaluation include two bias metrics (normalized mean bias, NMB; fractional bias, FB) and two error metrics (normalized mean error, NME; fractional error, FE) (Table 2). Table 3 shows that model biases (NMB, FB) are less than ±12 % and errors (NME, FE) up to 15 % for all seasons and both networks, well within the bias/error goals of less than ±15 % and 35 % recommended by EPA (1991). The model tends to overestimate O3 in spring and summer, while it underestimates in fall and winter. Model biases and errors are slightly improved, by less than 2 %, at CASTNET sites. The AQS network includes many urban sites that are heavily impacted by local emissions not well resolved by our modeling grid.

The seasonal spatial distributions of NMB, NME, and Pearson's correlation coefficient (r) at individual CASTNET and AQS sites are shown in Figs. S2–S5. The springtime performance is weakest (|r| < 0.5) in the western US, where high terrain prevails. Stratospheric O3 intrusions are influential in the high terrain of the western US during spring (Langford et al., 2009, 2015; Lin et al., 2012b; Emery et al., 2012) but models do not always represent this process accurately. In summer, the performance is weakest along the Gulf Coast and California coast. Model performance against AQS observations is presented separately for selected 22 major cities. CAMx performs well at these cities, with NMB of -12 to 12 % and NME of 12 to 21 % for MDA8 with zero threshold (Table S3), satisfying the bias/error goals. The performance statistics improve when applying a 40 ppb threshold (Table S4). We additionally provide quantile–quantile (Q–Q) plots (Fig. S8) which compare independently sorted (time-unpaired; space-paired) observed and modeled O3 for each city. The Q–Q plots suggest good model distributions (e.g., data pairs near the 1 : 1 line) for Los Angeles, Sacramento, Phoenix, Denver, Dallas, Houston, Pittsburgh, and Philadelphia.

We evaluated vertical profiles of O3 at Trinidad Head (41.0541, -124.151∘) on the northern coast of California (Figs. S6 and S7). This location is indicative of O3 entering the US with prevailing winds from the Pacific Ocean and useful for evaluating O3 BCs over the Pacific. CAMx reproduces the strong gradient in O3 starting at the tropopause (between ∼ 10 and ∼ 14 km) but sometimes underestimates O3 near the model top (∼ 16 km). In the free troposphere (∼ 1 to ∼ 10 km), CAMx performance is reasonable with a mix of over- and underpredictions. In the marine boundary layer (MBL; < 500 m), CAMx tends to overpredict surface O3 because the model lacks a consistently observed gradient toward lower O3 (typically lower than 50 ppb) at the surface. Potential causes are too little O3 deposition to the ocean, a lack of O3 destruction by halogen chemistry in the MBL, too strong vertical mixing in the MBL, or a combination of factors. The influence of low bias in the Pacific MBL will be confined to west coast states because this air is blocked effectively by western mountain ranges (see Sect. 3.5). However, similar biases in the MBL over the Gulf of Mexico and Atlantic Ocean could influence O3 across the southern and eastern US (see Sect. 3.2). Further analysis is required to address this uncertainty.

We found a consistent high bias for summer O3 at sites near the Gulf Coast and investigated potential causes using our tracer simulation results. Global modeling studies have reported high-O3 bias simulated in air arriving along the Texas coast (Fiore et al., 2002, 2014; McDonald-Buller et al., 2011; Zhang et al., 2011). The high-O3 bias over 10 ppb has been reported by a regional modeling study (Sarwar et al., 2015; Smith et al., 2015). Adding reactions of halogens (especially iodine; Smith et al., 2015) to the O3 chemistry in CAMx mitigated (e.g., ∼ 3 ppb reduction of MDA8 O3 at coastal sites) but did not eliminate this bias. This study provides an opportunity to examine whether bias in O3 BCs could contribute to O3 bias along the Gulf Coast.

Time series of daily MDA8 O3 together with O3 BC contributions are shown in Fig. 1 for Galveston and Sabine Pass in Texas. CAMx tends to overpredict O3 during summer months when onshore winds are prevalent (TCEQ, 2016) due to the enhancement of the Bermuda High bringing warm air from the Gulf of Mexico (Zhu and Liang, 2013). During this period CAMx has a large overprediction bias exceeding 15 ppb at the two sites when observations are low (∼ 20 ppb). Reactive O3 tracers exceed observed O3 by an average of 2–3 ppb, which explains only a portion of the total model bias, so other factors must be contributing (not examined here). Inert O3 tracers are on average higher than the reactive tracers by 10–12 ppb, demonstrating that inert tracers can overestimate BC contributions.

BC contributions based on the active tracers to seasonal average MDA8 O3 over the US are higher in spring than summer (Fig. 2). Spring contributions are over 40 ppb across the western US, which is the region most influenced by pollution transported from Asia (Jaffe et al., 2003; Zhang et al., 2011; Lin et al., 2012a; Emery et al., 2012). In particular, high-O3 events over the high terrain in the western US have been linked to intercontinental transport and stratospheric intrusions (Lin et al., 2012a, b). The highest modeled contributions occur in spring, which is consistent with observations (Parish et al., 2012; Cooper et al., 2012) and previous modeling studies (Emery et al., 2012; Fiore et al., 2014). BC contributions in the eastern US are generally below 40 ppb in spring and below 30 ppb in other seasons.

BC contributions on high-O3 days are compared to the summer average BC contributions in Fig. 3 for 22 major cities. The metrics for high O3 are the H4MDA8, which is relevant to the NAAQS but is a single day, and the Top 30, which includes a variety of conditions that can lead to high O3. They are compared to the summer average MDA8 O3 metric which includes high- and low-O3 days. In 15 of the 22 major cities (Los Angeles, Sacramento, Dallas, Kansas City, St. Louis, Chicago, Atlanta, Cincinnati, Columbus, Detroit, Pittsburgh, Baltimore, Philadelphia, New York, and Boston), the BC contribution to the Top 30 and summer average days differs by less than 15 %, indicating that higher O3 on the Top 30 days is mainly attributable to larger O3 production within the modeling domain. BC contributions to the H4MDA8 are smaller than to the Top 30 in 16 of the 22 cities, which is consistent with greater destruction of BC O3 by local photochemistry on the highest O3 days in these cities. Los Angeles and Sacramento fall in this category with BC contributions to H4MDA8 below 20 ppb. For the cities east of the Rocky Mountains the H4MDA8 contributions range from 19 ppb (St. Louis) to 34 ppb (Detroit) and the differences among the three metrics are generally within 5 ppb with no metric consistently being highest or lowest. In contrast, for cities in the Intermountain West (Boise, Phoenix, Salt Lake, and Denver) the BC contribution is consistently lower for summer average than the two high-O3 day metrics (i.e., differences are more than 8 ppb). For Denver, our model estimates 57 ppb of BC contribution to the Top 30 and 72 ppb to the H4MDA8. However, the modeled 72 ppb BC contribution to the Denver H4MDA8 is certainly overstated, because the observed MDA8 on this day was only 50 ppb, and we place more emphasis on metrics like the Top 30 that consider multiple days. BC contributions tend to be higher in the western than the eastern US because of higher terrain and deeper planetary boundary layer (PBL) that can efficiently transport mid-tropospheric O3 to ground level, and longer O3 lifetimes in the PBL (Fiore et al., 2002).

We further investigated how total O3 changes as the modeled BC contributions increase (in 10 ppb increments) as shown in Fig. 4 for several cities. The relationships vary between cities and the model captures this variation with the western cities (Denver and Phoenix) showing different patterns than eastern cities (Philadelphia and Atlanta). For Denver and Phoenix in the Intermountain West, total O3 increases with BC contribution and approaches the 1 : 1 line at higher BC contribution, revealing small groups of days when MDA8 O3 exceeded 60 ppb (11 days for Denver and 7 days for Phoenix) and the modeling indicates that BCs accounted for almost all of this O3. In other words, BC contributions alone distinguish high-O3 days from low days. These groups of high BC-contributed days are important because local emission reductions, or even US-wide emission reductions, would be ineffective at reducing O3. However, there are other days in Denver when total MDA8 O3 exceeded 60 ppb with modeled BC contribution below 30 ppb (see first and second bars in Fig. 4) on which reducing local or US emissions would lower O3. Air quality managers need methods to identify dates when emission reductions would be ineffective so that those dates can be excluded from emission strategy development. Nonetheless, these results should be interpreted with consideration given to model performance. As shown in the Q–Q plot for Denver (Fig. S8), the model can capture the O3 distribution quite well, although it underestimates MDA8 O3 over 65 ppb and overestimates MDA8 O3 over 80 ppb. For this reason, we encourage making use of multi-day metrics (such as Top 30) rather than a single-day metric (e.g., H4MDA8).

Contributions to seasonal average MDA8 from reactive and inert tracers are shown in Fig. 5, where positive differences indicate larger contributions from inert tracers in all seasons. In summer, when photochemistry is active, the differences are more than 10 ppb. At the Galveston and Sabine Pass sites, the differences frequently exceed 20 ppb in summer days (Fig. 1). During spring and fall, contributions from inert tracers are 5 ppb higher than contributions from active tracers in southern regions and less than 5 ppb elsewhere. In winter, when photochemistry is less active, the estimated contributions from the inert and active tracers are similar. Such seasonal variation is consistent across all three groups of vertical layers (Figs. S9–S11). These results emphasize the critical role of O3 chemistry and highlight the estimation bias inherent to the inert tracer approach.

The largest BC contributions to seasonal average MDA8 O3 over the US are from the MT with small contributions from the UTLS (Fig. 6). The contributions from the MT are highest in spring, followed by summer, fall, and winter. The attribution of the MT to spring maxima is over 40 ppb across the western US, but less than 25 ppb in the eastern US. The contributions from the LT are highest (more than 30 ppb) along the western boundary but decrease sharply at the coastline as a result of dilution as the boundary layer moves onshore along with higher O3 deposition velocities over land. Dissipating contributions from the model boundaries with distance are seen at all lateral sides because O3 deposits to the earth's surface and is destroyed by chemical reactions in the atmosphere. The penetration of LT BC O3 inland O3 peaks in winter, when chemistry and deposition are least active. The UTLS BC tracers contribute only a few ppb, mostly over the highest western terrain, and up to 6 ppb in summer when vertical convection is most active.

Reducing emissions in East Asia by 20 % (EAS scenario) decreases average O3 across the US in all seasons (Fig. 7, first-column panels). As expected, decreases are highest in the west because the western US is closest to Asia and has high terrain. The O3 decreases are small: below 1 ppb in spring and below 0.5 ppb in other seasons. Decreases in O3 BC reactive tracers (Fig. 7, second-column panels) are almost identical to modeled O3 decreases but slightly smaller (e.g., generally within 0.1 ppb) because the reactive tracers omit some chemical interactions.

We examine more closely the relationship between changes in O3 and reactive tracers in the EAS scenario. In each surface grid cell, we regress hourly O3 changes against reactive tracer concentration changes (summed over boundary height ranges) to compute slope and r as demonstrated in the two scatter plots for Denver in spring and summer. Slope and r values of 1 indicate that the O3 changes are explained entirely by the changes in O3 BC reactive tracers. The delta total O3 and delta tracer O3 relationship is near-linear at Denver with a slope of 0.87 and r of 0.9. The 16 panels in Fig. 7 show the regression parameters for each grid surface cell and match the scatter plots for Denver. The slope and r (Fig. 7, third- and fourth-column panels) values have similar spatial patterns in all seasons. In winter and fall the slope values are near 1 with r of 0.8 to 1 across the US, suggesting strong influence of O3 transport from Asia during these seasons. In spring, strong correlation (r = 0.8 to 1) is seen in the western US, but areas in the eastern US have a slope lower than 0.2 and r lower than 0.4, indicating that the O3 BC tracers can explain only a fraction of the total O3 change. The lowest correlation (r < 0.2) is in the summertime over the southeastern US in a region where the EAS scenario produces almost no change in surface O3, indicating that transport from Asia becomes unimportant. High correlation in the western US in all seasons emphasizes the influence of O3 transport from Asia in this region.

Reducing global emissions by 20 % (GLO scenario), including US emissions, decreases summertime average O3 by up to 4 ppb (Fig. 8). The largest reductions occur over the eastern US, where US emissions cause domestic O3 production. O3 reductions in fall and winter are small, generally lower than 1–2 ppb. Many NOx-rich areas (e.g., urban cores) show O3 increases as a result of NOx emission reductions (i.e., NOx disbenefit) in all seasons. The spatial pattern and magnitude of the changes in BC O3 tracers differs from the changes in surface O3 (Fig. 8, first- and second-column panels) except near the boundaries and in winter. The correlation between changes in surface O3 and BC tracers is low (|r| < 0.4) in all seasons except winter (Fig. 8, fourth-column panels). Overall, in the GLO scenario US surface O3 is more sensitive to domestic emission reductions than changes in BCs.

We use summer average MDA O3 to show how the two emission scenarios change BC contributions for the 22 major cities (Fig. 9). The EAS scenario reduces BC contribution in all cities with reductions range from 0.06 ppb (Houston) to 0.3 ppb (Boise). Reductions are larger in the western US and more northern latitude in the eastern US (e.g., larger reduction in Columbus than Atlanta). These reductions result mostly from smaller MT and LT BC contributions because the EAS scenario scaled back the contribution of each height range about equally. The EAS scenario changed the UTLS BC contributions by less than 0.01 ppb. The GLO scenario produced larger reductions than the EAS scenario and they range from 0.4 ppb (Boston) to 1.2 ppb (Los Angeles). These reductions are mainly driven by the MT BCs in all of the cities except Dallas and Houston. Higher influence from the LT in the GLO scenario than the EAS scenario for most cities is consistent with the GLO scenario reducing emissions just outside the CAMx domain, whereas in the EAS scenario the emission reductions occur only in East Asia.

The AQMEII activity permits comparison of BC O3 contributions in different regional models. US EPA applied the CMAQ model over the NA domain with the same model input data as we used with CAMx except for biogenic emissions. The WRF-CMAQ system was configured using WRFv3.4 and CMAQv5.0.2 (Appel et al., 2013; see also Foley et al., 2010, and Byun and Schere, 2006). Options in CMAQ include wet deposition as described in Byun and Schere (2006) and dry deposition as described in Pleim and Ran (2011). Additional details on the CMAQ configuration used in these simulations can be found in Solazzo et al. (2017). Figure 10 compares BC O3 contributions to seasonal average MDA8 O3 estimated by CAMx and CMAQ using inert BC O3 tracers (CMAQ was not run with reactive tracers). Differences between the CMAQ and CAMx inert tracer impacts are smaller than the differences between inert and reactive tracers in CAMx, which exceed 10 ppb (Fig. 5), but they are notable, in a range of 4–8 ppb in summer and 2–6 ppb in spring with CAMx being higher. Factors contributing to these differences may include fewer vertical layers in CAMx (26, compared to 35 in CMAQ), allowing greater transport of UTLS O3 to ground level (Emery et al., 2012), omission of wet scavenging for the CAMx inert tracers, treatment of deep convective transport in CMAQ, or differences in model treatments of O3 dry deposition. Liu et al. (2017) performed multi-model process comparisons with four AQMEII models and draw similar conclusions regarding factors that can contribute to differences in tracer impacts.