We select 17 ozonesonde sites, eight in North America and nine in Europe, all of which have a record of at least three profiles every month between 1990 and 2016 (Fig. 1 and Tables 1 and 2). The data are obtained from the World Ozone and Ultraviolet Radiation Data Centre (WOUDC; http://www.woudc.org, last access: 1 February 2020). Observations over most stations are obtained using electrochemical concentration cells (ECCs), which rely on the oxidation reaction of ozone with potassium iodide in solution (Komhyr et al., 1995). At Hohenpeißenberg, Germany, measurements are obtained using the Brewer–Mast instrument. The sonde ozone measurements have a vertical resolution of ∼150 m, with an accuracy of ±5 % in the troposphere (WMO, 2014).

We use a replay simulation (http://acd-ext.gsfc.nasa.gov/Projects/GEOSCCM/MERRA2GMI, last access: 1 February 2020) of the GEOSCCM with the Global Modeling Initiative (GMI) chemical mechanism (Duncan et al., 2007; Strahan et al., 2007) for gas chemistry, which includes a complete treatment of stratospheric and tropospheric chemistry, and the Goddard Chemistry Aerosol Radiation and Transport (GOCART) module (Chin et al., 2002; Colarco et al., 2010) for aerosols. The replay simulation follows the replay methodology as described in Orbe et al. (2017) and uses the RAs3 setting, which best represents overall transport. The model inputs the 3-hourly time-averaged output of MERRA-2 meteorology (U, V, T, pressure) and recomputes the analysis increments, which are used as a forcing to the meteorology at every time step over the 3 h replay interval. The replay simulation is run at a MERRA-2 native resolution of ∼50 km in the horizontal dimension and 72 vertical levels. This replay simulation is referred to as the “MERRA2-GMI” simulation.

The MERRA2-GMI simulation was run from 1980 to 2018. The emissions in this run include anthropogenic, biofuel, biomass burning and biogenic emissions. The values for fossil fuel and biofuel emissions are taken from the MACCity inventory (2011) until 2010 and then derived by following the Representative Concentration Pathway (RCP) 8.5 scenario after 2010. The MACCity anthropogenic emissions are derived by interpolating the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) emissions (Lamarque et al., 2010) on a yearly basis between the base years 1990, 2000, 2005 and 2010. For the years 2005 and 2010, the interpolation follows the RCP 8.5 emission scenario. Biomass burning emissions are taken from the Global Fire Emissions Database (GFED) version 4s (Giglio et al., 2013) after 1997. Prior to 1997, biomass burning emissions are based on a GFED4s climatology with year-to-year variability imposed using regional scale factors derived from the Total Ozone Mapping Spectrometer (TOMS) aerosol index (Duncan et al., 2003). The simulation used the Model of Emissions of Gases and Aerosols from Nature (MEGAN) (Guenther et al., 2006) to simulate biogenic emissions, including isoprene, within the model. The lightning parameterization in the model (Allen et al., 2010) is constrained by the MERRA-2 detrended cumulative mass flux, with seasonal constraints from the Lightning Imaging Sensor (LIS)/Optical Transient Detector (OTD) v2.3 climatology (Cecil et al., 2014). Methane is specified using latitude and time-dependent surface observations from the NOAA Earth System Research Laboratory (ESRL) Global Monitoring Division (GMD) network (Dlugokencky et al., 2011). Our initial model evaluations suggest that the MERRA2-GMI ozone simulations are in good agreement with the means and variability of the total and tropospheric column ozone from satellite observations (Figs. S1 and S2 in the Supplement).

A StratO3 tracer is included in the model to diagnose the stratospheric ozone influence on the troposphere. StratO3 is set equal to simulated ozone in the stratosphere and is removed in the troposphere based on interannually varying monthly mean loss rates and surface deposition fluxes archived from a standard full chemistry simulation. StratO3 tracer is defined relative to a dynamically varying tropopause, which is derived from an artificial tracer, e90, introduced by Prather et al. (2011). The e90 tracer is set to a uniform mixing ratio (100 ppb) at the surface with a 90 d e-folding lifetime everywhere in the atmosphere. This lifetime is long enough for the tracer to be well mixed throughout the troposphere but short compared to the transport timescales in the stratosphere, resulting in sharp e90 tracer gradients across the tropopause. In our simulations, the e90 tropopause value is set to 90 ppb. Prather et al. (2011) shows that the e90 tropopause not only matches the other traditional definitions of the tropopause but also has the advantage of being able to capture complex features such as tropopause folds. Therefore, the e90 tropopause is optimal in effectively separating stratospheric from tropospheric air from a chemical composition perspective. The e90 tracer has been used in many studies of STE as an accurate tropopause definition and an ideal transport tracer in the upper troposphere and lower stratosphere (UTLS) (e.g., Hsu and Prather, 2014; Liu et al., 2016, 2017; Pan et al., 2016; Randel et al., 2016).

The MERRA2-GMI simulation has hourly output for ozone and 3-hourly output for StratO3 at each model level. When comparing to the ozonesonde measurements, the model outputs are sampled at the nearest grid point and launch time for each sonde.

Previous studies have shown that the relative contribution of stratospheric ozone to tropospheric ozone is greatest in the free troposphere during winter (e.g., Holton et al., 1995; Stohl et al., 2000; Skerlak et al., 2014, 2015) and at the surface during spring (e.g., Lin et al., 2012, 2015). In summer, the relative contribution of stratospheric ozone is low due to the increased chemical ozone production in the troposphere. Here, we focus on the winter (DJF) and spring (MAM) seasons to examine the interannual variations in the strength and vertical extent of stratospheric ozone impact on the tropospheric ozone.

Figure 2 compares the anomalies of modeled and ozonesonde measured ozone at 200, 400 and 700 hPa in the winter and spring seasons from 1990 to 2016 averaged over sites from North America and Europe. Anomaly at each site is calculated by removing the respective seasonal mean climatology from 1995 to 2016 and then averaged over all sites for each region (see the Supplement). The shaded area represents the 95 % confidence interval (CI) of the calculated mean from daily observations over all the selected stations. To quantify the magnitude of IAVs, we adopt the standard deviation (SD) of these ozone anomalies. We perform the standard statistical F test to assess the regional and seasonal differences in the ozone IAVs. The calculated standard deviations and F test statistics are shown in Tables S1 and S2.

At 200 hPa, the model reproduces well the observed IAV in both seasons over both regions (r≥0.91, Fig. 2a–d). There are no significant differences in the magnitude of ozone IAV between North America and Europe (Table S1). But over both regions, the ozone IAVs show significant seasonal differences with greater magnitude in spring than in winter (Table S2). Negative ozone anomalies occur in the early 1990s and at the end of the record from 2014 to 2016, while positive anomalies are obtained for most years between 1998 and 2013. The negative ozone anomalies during the period of 1992–1996 are consistent with the chemical and dynamical perturbations following the 15 June 1991 eruption of Mt. Pinatubo (Hadjinicolaou et al., 1997; Stenchikov et al., 2002; Rozanov et al., 2002). The negative ozone anomaly in 2015–2016 is associated with stratospheric circulation changes caused by the unusually warm ENSO event coinciding with a disrupted quasi-biennial oscillation (QBO) during that period (Tweedy et al., 2017; Diallo et al., 2018).

At 400 hPa, the model reproduces the IAV of the observations, with negative anomalies in the early 1990s and mostly positive anomalies thereafter. The observed negative ozone anomaly after the Mt. Pinatubo eruption has a maximum amplitude of 7 ppb (-13 % relative anomaly) in the winter of 1992–1994. The model underestimates the observed peak depletion in the winter of 1992, with the simulations falling outside the 95 % CI of the observations from 1992 to 1994 (Fig. 2e). In spring, the model reproduces well the timing of observed ozone depletion but again underestimates its amplitude (Fig. 2f). At 700 hPa, the observations from the North American sites show a similar negative ozone anomaly in 1992–1994 to that obtained at 200 and 400 hPa, with prevailing positive anomalies thereafter. The model results for the sign of the interannual variations are in relatively good agreement with observations but again underestimate the magnitudes of the negative anomalies in the early 1990s after the Mt. Pinatubo eruption.

Over European sites, the observed ozone IAV, excluding year 1990–1991, exhibits a similar pattern to the one at 200 hPa after 1991, although the minima after the Mt. Pinatubo eruption are not as pronounced as over North America (Fig. 2g–h). The maximum positive anomaly, observed in 1990–1991, is not reproduced by the model. The model–observation correlation coefficients increase significantly if we omit these 2 years (from 0.18 to 0.58 in the winter and from 0.43 to 0.58 in the spring). At 700 hPa, the model reproduces the magnitude of observed ozone depletion after the Mt. Pinatubo eruption. Unlike at 200 hPa, the differences in the magnitude of ozone anomalies between North America and Europe are significant at 400 and 700 hPa, with smaller ozone anomalies over European sites than over North American sites in both seasons (Table S1). The magnitude of ozone anomalies does not show significant seasonal difference between DJF and MAM, except over North America at 400 hPa (Table S2).

We use explained variance (square of correlation coefficient: r2) to determine the fraction of the ozone variance in the troposphere that can be attributed to the variance in stratospheric ozone. Table 3 shows the explained variances in the winter and spring ozone anomalies between 200 and 400 hPa, 200 and 700 hPa for the observations and simulations averaged over the North American and European stations. Both the model and observations suggest that about 27 % of North American ozone variations at 400 hPa are accounted for by changes at 200 hPa in the winter. The 200–400 hPa ozone relationship is higher in the spring (r2=0.4 in the observation and 0.46 in the simulation). Over Europe, the 200–400 hPa ozone relationship in the observations is relatively low (r2=0.1 in DJF and 0.02 in MAM), due to the phase shift of these two-time series during the first 2 years, where observed ozone anomalies are negative at 200 hPa but reach a maximum at 400 hPa. The explained variance increases to 0.45 after removing these 2 years in the winter but not that much in the spring (r2=0.05). High correlations of the ozone anomalies between 200 hPa and 400 hPa are seen in the model in both seasons. The highest relationship between 200 and 700 hPa is found over the North American sites in the spring with r2=0.21 and 0.17, respectively, in the observation and simulation, which is consistent with the previous findings of the deep STE hotspots along western US in the spring season (Lin et al., 2012; Skerlak et al., 2014; Langford et al., 2015).

The correlations between the stratosphere and troposphere IAV in both observations and model simulations suggest a potential impact of stratospheric ozone on tropospheric ozone variations. Previous studies have found high correlations between ozone in the lower stratosphere with that in the middle and lower troposphere, with the largest effects in late winter and spring. Correlation does not necessarily mean causality, and to date, model investigations of this correlation (Terao et al., 2008; Hess and Zbinden, 2013) have not used a model with both stratospheric and tropospheric chemistry and up-to-date stratospheric circulation. The MERRA2-GMI simulation has both of these attributes, detailed dynamic diagnostics and StratO3 as described in Sect. 2.2. In the next section, we use the StratO3 tracer to examine the contribution of stratospheric ozone to the IAV of tropospheric ozone, as a function of altitude, season and location. We will also use diagnostics from the model to explore the influence of transport on the stratospheric ozone contribution to the tropospheric ozone and its IAV.

Figure 3 shows the same comparison between the observed (black lines) and simulated ozone (red lines) anomalies as in Fig. 2 but adding the anomalies of simulated StratO3 (green lines). As expected, the StratO3 anomalies at 200 hPa are almost identical to the simulated ozone anomalies, since most measurements are in the stratosphere at this level.

The variability in the amount of tropospheric ozone that was transported from the stratosphere as inferred by StratO3 is due to both the variability in the lower stratospheric ozone reservoir and the variability in the net downward mass flux (Albers et al., 2018). These two variabilities may either cancel or reinforce each other, depending on their relative phases. At 400 hPa, over the North American stations, the minimum and maximum of StratO3 tracer is highly correlated with the minimum and maximum of simulated ozone. The IAV of StratO3 explains more than 60 % of simulated ozone variations (Fig. 3e, f; r=0.77 in DJF and 0.81 in MAM), suggesting that the changes in stratospheric ozone input strongly impact the simulated ozone IAV in the upper troposphere. The correlation between StratO3 and observed ozone is slightly lower (0.44) than that with simulated ozone in DJF over North America. The decreased correlation is mainly due to the model–observation discrepancy between 1992–1994. The sondes at 400 hPa show a similar ozone depletion through 1992–1994 as seen at 200 hPa after the Mt. Pinatubo eruption, while the model shows an ozone increase after 1992 through 1994, which is driven by changes in the stratospheric ozone contribution to the modeled ozone (Fig. 3e). This suggests that the impact of the negative anomalies of stratospheric ozone (200 hPa) may be counterbalanced by an increase in downward mass flux from the stratosphere, thus leading to the model underestimation of the negative anomaly in observations at 400 hPa. In MAM, the StratO3-measured O3 correlation is high (0.74) over North America. Over European sites, a similar correlation is observed between simulated ozone and StratO3 at 400 hPa in the winter (r=0.78), with a slightly smaller value in the spring (r=0.61). The correlation decreases when comparing StratO3 to the observed ozone, mainly because of the model–observation discrepancy during the first 2 years. Omitting the first 2 years gives a fair correlation between StratO3 and observed ozone (0.66 in DJF and 0.34 in MAM). The fair-to-good correlations between StratO3 and observed ozone indicate a significant impact of stratospheric ozone on the tropospheric ozone variations at 400 hPa over both North America and Europe..

Figure 3i–l compare the simulated StratO3 anomalies to the observed and simulated ozone anomalies at 700 hPa over North American and European ozonesonde sites during winter and spring. Over North America, the observed ozone anomalies stay low in the early 1990s and increase thereafter in both seasons, which is underestimated in the model. In the winter, StratO3 anomalies decrease slightly in contrast to increases in both observed and simulated ozone anomalies. The winter StratO3–O3 correlation is ∼0. In spring, both observed and simulated ozone exhibit similar IAV, which agree with the phase of the StratO3 anomalies after the Pinatubo period (1991–1995). The StratO3–O3 correlation increases from 0.07 to 0.33 in winter and from 0.36 to 0.58 in spring from 1996 to 2016.

Over North America, our model results are in good agreement with the observed IAV at all levels except right after the Mt. Pinatubo eruption. The model only reproduces about half of the observed tropospheric depletion over North America. As discussed above, this could be due to an excessive mass flux from the stratosphere in the MERRA-2 analysis during this period. Model results are in better agreement with the magnitude of observed ozone depletion after the Mt. Pinatubo eruption in the middle and lower troposphere over Europe. There is no significant relationship between StratO3 and simulated ozone at 700 hPa. This is expected since the impact of stratospheric ozone decreases, and the impact of ozone production from its precursors becomes more important at lower altitudes. In summary, our model analysis identifies differences in the strength and vertical extent of stratospheric ozone impact on the tropospheric ozone IAV between North America and Europe. Over North America, the StratO3 IAV has a significant impact on the tropospheric ozone IAV from the upper to lower troposphere and explains 60 % and 66 % of the simulated ozone IAV at 400 hPa and 11 % and 34 % at 700 hPa in winter and spring, respectively, after 1995. Over Europe, the influence is limited to the middle to upper troposphere and becomes much less at 700 hPa.

The differences in the stratospheric ozone impact between North America and Europe are likely due to variations in the net downward flux associated with planetary-scale waves. Previous studies have suggested that the IAV of the STE mass flux is likely correlated to changes in the tropopause height (e.g., Gettelman et al., 2011). Figure 4a–d show the comparison of the observed ozone mixing ratio anomalies at 400 hPa and the tropopause pressures derived from the observed ozone profiles following the criteria in vertical gradient and ozone mixing ratio given by Browell et al. (1996). The tropopause pressure was estimated to be at the pressure where a linear regression line passing through the lower stratospheric ozone profile (150–400 ppb, lower than 100 hPa) intersects with the 100 ppb ozone level. Figure 4e–h compare the simulated ozone and StratO3 anomalies at 400 hPa with the tropopause pressures derived from simulated ozone profiles following the same criteria as for the observations. As expected, the IAV of ozone and StratO3 positively correlates with that of the derived tropopause pressure (anticorrelates with the tropopause height) in both model and observation. In general, during years with a lower tropopause, stratospheric ozone influence at 400 hPa increases and results in a positive ozone anomaly. During years with a higher tropopause, decreased stratospheric ozone influence leads to a negative ozone anomaly at 400 hPa.

The above high correlations between the IAV of tropopause pressure and StratO3 raise the question of what dynamical conditions control the higher/lower tropopause pressures, STE mass fluxes and the subsequent impact of stratospheric ozone on tropospheric ozone. These questions are particularly important if these dynamical conditions change in the future as a result of climate change. In the next section, we rely on the model's 3-D dynamical diagnostics, including air mass flux and horizontal wind patterns, to examine both the vertical and horizontal transport influence of the stratospheric ozone contribution on the tropospheric ozone and its IAV. We also examine the longitudinal difference in the model's dynamics to explain the identified differences in the stratospheric ozone impact between North America and Europe.

Planetary-scale Rossby waves, superimposed on the mean westerly zonal flow, are the dominant dynamical variability over northern midlatitudes in winter and spring. Troughs occur where the flow moves equatorward. Tongues of stratospheric polar air extend equatorward associated with frequent STE processes. Ridges occur where the flow moves poleward, bringing in warm tropospheric air. The Northern Hemisphere is typically encircled by several of these waves, with troughs (ridges) likely occurring over eastern (western) continental edges (e.g.,Thorncroft et al., 1993). Homeyer and Bowman (2013) have shown that a Rossby wave in the upper troposphere can affect the flow in the lower levels and plays an important role in the meridional transport of both tropical and subtropical air masses. Ozone transport associated with these wave disturbances are responsible for a large fraction of ozone temporal and spatial variability in winter and spring (e.g., Kinnersley and Tung, 1998; McCormack et al., 1998; Lozitsky et al., 2011; Zhang et al., 2015).

In this section, we examine the linkage of the vertical and horizontal transport to the stratospheric ozone contribution in the troposphere using the model's 3-D air mass flux and horizontal winds. Our analysis focuses on the year 1993, when there is a major discrepancy with the observations at 400 hPa as shown in Fig. 3.

Figure 5 illustrates the relationship of the StratO3/O3 ratio at 400 hPa to the horizontal winds at 400 hPa and the vertical air mass flux near the seasonal mean tropopause pressure in the year 1993. The seasonal mean tropopause pressure in the model averaged from 30 to 80∘ N is around 250 hPa in winter and around 300 hPa in spring. Because of the different tropopause heights, different pressure levels are shown in the figures. The vertical air mass flux is calculated by multiplying omega (dp / dt; units: pa s-1) with the density of air. The sign of calculated air mass flux is reversed so that positive values represent upward fluxes; negative values represent downward fluxes. Figure 5a and b show the simulated StratO3/O3 ratio in the winter and spring of 1993; prevailing wind patterns at 400 hPa are superimposed on this ratio. The jet, the location of maximum winds, is indicated by thick red lines. Figure 5c and d show the anomalies of simulated StratO3/O3. Figure 5e and h show the vertical air mass flux around the tropopause pressure (blue color represents the downward motion, and red color represents the upward motion near the tropopause) and their anomalies (blue color represents an increase in downward flux or a decrease in upward flux; red color represents a decrease in downward flux or an increase in upward flux around the tropopause).

In the winter of 1993, the jet at 400 hPa exhibited a typical wave pattern, with a trough over eastern North America and ridges over western North America and western Europe (Fig. 5a). Strong northwesterly winds prevailed north of 50∘ N over western North America. They converged with the westerlies around 45∘ N in eastern North America. The winds then changed direction to southeasterly and flowed into the North Atlantic and Europe, bringing warmer tropospheric low ozone air into western Europe (Fig. 5a). The maps of the air mass flux and its anomalies (Fig. 5e and g) suggest that North America between 50 and 70∘ N was dominated by more vigorous downward mass fluxes of stratospheric air. Meanwhile, the northwesterly winds brought ozone-rich air from high latitudes, resulting in a positive anomaly of stratospheric ozone influence at 400 hPa (Fig. 5c). Although the lower stratospheric ozone level decreased significantly in the winter of 1993 due to the Pinatubo eruption (Fig. 2), the enhanced downward mass fluxes across the tropopause in the model counteracted the depletion and led to a positive ozone anomaly at 400 hPa over North America between 50 and 70∘ N (Fig. S3). Over the high latitudes (>70∘ N), where there is less dynamic perturbation (including both vertical and horizontal transport), the stratospheric ozone contribution at 400 hPa was largely driven by the depletion of the ozone concentration in the lower stratosphere and showed a strong negative anomaly in 1993 (Fig. 5c). Most of the European region was covered by the increased downward air mass flux near the tropopause in the winter of 1993. However, a negative anomaly of the StratO3 contribution at 400 hPa was seen over western Europe. It is likely that the combined negative effects of the ozone depletion in the lower stratosphere and the downwind of the warmer tropospheric low ozone air from the subtropical North Atlantic Ocean exceeded the positive effect of the increased downward air mass flux over this region.

In the spring of 1993, regions with maximum StratO3/O3 at 400 hPa were located further north compared to those in the winter and centered around the northern part of the Labrador Sea. Southwesterly wind prevailed south of 65∘ N over western North America, bringing in low-StratO3 oceanic air from the subtropics. The winds changed direction to northwesterly around 120∘ W and 65∘ N and flowed westerly around the Hudson Bay until reaching the west coast of Europe. The winds then bifurcated into two branches: one passed by the northern side of Europe and the other flowed around the southern side of Europe. In North America south of 50∘ N, there were three cells with increased upward air mass fluxes near the tropopause ranging from 110 to 50∘ W (Fig. 5f, h). Regions downwind of these cells exhibited a negative anomaly of stratospheric ozone contribution at 400 hPa (western North America, eastern North America from the Great Lakes and west North Atlantic, Fig. 5d). Over Hudson Bay and the central US, the increased downward mass flux near the tropopause contributed to the slightly positive anomaly of the stratospheric ozone contribution at 400 hPa. Over the north and west of Europe, the combined negative effects from the increased upward mass flux across the tropopause (Fig. 5f, h) as well as downwind of the westerly flows with low stratospheric ozone (Fig. 5b) led to strong negative anomalies in the stratospheric ozone contribution at 400 hPa.

Figure 6 shows a similar analysis as Fig. 5, except for 1998, when stratospheric ozone levels have recovered from the Mt. Pinatubo eruption and reached a regional maximum (Fig. 2). In the winter of 1998, a poleward shift of the jet occurred over most of North America. The jet location as well as the regions with the maximum StratO3/O3 ratio moved to the north by about 7∘ compared to the winter of 1993 (Fig. 6a). With the poleward shift of the jet, the StratO3/O3 anomaly exhibited a negative (positive) maximum along the equatorward (poleward) side of the jet. Southwesterly winds prevailed over regions equatorward of the jet at 400 hPa, bringing in tropical oceanic low ozone air to North America. Around the tropopause, there were increased upward air mass fluxes along the west coast of North America and decreased downward air mass fluxes over western and central North America. Therefore, although there was an increase in the stratospheric ozone concentration in 1998, the wind patterns in the troposphere and the vertical mass flux around the tropopause associated with the northward shift of the jet system caused a less favorable transport of stratospheric ozone into troposphere. This resulted in a negative anomaly of StratO3/O3 at 400 hPa over most of North America. In the spring of 1998, a similar poleward shift of the jet occurred over North America as that in the winter, with a negative StratO3/O3 anomaly over most of North America north of 45∘ N (Fig. 6b, d).

Our analysis suggests that significant interannual variations exist in both the regional wind patterns associated with the westerlies waves and the strength of downward air mass fluxes across the tropopause. The IAV of stratospheric ozone influence in the troposphere reflects a combined effect of the IAV in the 3-D dynamics as well as that in the lower stratospheric ozone concentration, which may either oppose or reinforce each other.

Our analysis based on data and model sampled at sonde sites identified differences in the strength and vertical extent of stratospheric ozone impact on tropospheric ozone IAV between North America and Europe. Over North American sites, a significant impact of the StratO3 IAV on tropospheric ozone extends to the lower troposphere. Over European sites, the influence is limited to the middle to upper troposphere. In this section, we examine whether the longitudinal differences seen over the ozonesonde sites are a large-scale phenomenon, by extending our analysis to a broader spatial domain. Figures 7 and 8 show the latitudinal average (30 to 80∘ N) of tropopause pressure, geopotential height at 400 hPa and the StratO3/O3 ratio at 400 hPa at each longitude from 180∘ W to 180∘ E from 1990 to 2016 in winter and spring. The geopotential heights and tropopause pressure are good diagnostics of large-scale circulation patterns. All of them show strong longitudinal difference between North America (120–60∘ W) and Europe (10∘ W–26∘ E), with lower geopotential height, higher tropopause pressure (lower tropopause height) and greater stratospheric ozone contribution over North America than over Europe. The longitudinal gradients between North America and Europe are slightly weaker in spring than in winter. The spatial map of StratO3/O3 climatology at 400 hPa averaged from 1990 to 2016 suggests that the longitudinal difference is persistent over most of the mid-high latitudes (Fig. S4) and is closely related to the wavelike pattern in jets. The climatology of jet meanders to the south over central and eastern North America and brings in cold polar air with more stratospheric subsidence. The jet moves to the north over Europe and brings in warm air with less stratospheric ozone influence. Skerlak et al. (2014) identified the deep STE hotspots along western North America using the ERA-Interim reanalysis data set from the European Centre for Medium-Range Weather Forecasts (ECMWF) from 1979 to 2011. Therefore, over North America, the stratospheric subsidence inside the polar vortex as well as deep stratospheric intrusion events results in a deeper and greater stratospheric ozone influence on the tropospheric ozone than over Europe.

A modulating factor in the IAV is the Artic Oscillation (AO) – the primary mode of dynamical IAV in the troposphere during winter. Several studies have examined the mechanism for downward transport from the stratosphere to the troposphere and attributed changes in the strength of lower-stratospheric polar vortex to AO anomalies at the surface, with a positive AO phase linked to a more isolated and stronger polar vortex (Ambaum and Hoskins, 2002; Perlwitz and Harnik, 2003) and lower tropopause heights. Lamarque and Hess (2004) found that the AO explains up to 50 % of the IAV in tropospheric ozone over North America in January–March but did not find any significant correlation in European sonde data, with similar results from the Model for OZone And Related chemical Tracers (MOZART) model. They argued that the correlation may be caused by the influence of the AO on its modulation of STE as well as transport of ozone and its precursors. Kivi et al. (2007) found that changes in the AO explained most of the tropospheric ozone trends in January–April, based on analysis of Arctic ozonesonde data. Figure 9 shows the longitudinal variations in simulated ozone and AO correlation profiles averaged over 30 and 80∘ N from 1000 to 200 hPa. Over North America (120 to 60∘ W), the correlation between simulated ozone and the AO index is negative and stays low above 400 hPa. The anticorrelation increases with increased pressure and reaches its maximum near the surface around 90∘ W. The anticorrelation averaged over Europe (10∘ W–26∘ E) stays low above 400 hPa, increases slightly from 400 to 700 hPa, then decreases sharply near the surface. This is similar to the correlations obtained from the ozonesonde profiles (Fig. S5). The similarity of correlation patterns over sonde sites and their surrounding broader regions indicates that the AO-related stratospheric subsidence is a large-scale phenomenon.