Surface ozone-meteorology relationships: Spatial variations and the role of the jet stream

We investigate the relationships among summertime ozone (O3), temperature, and humidity on daily timescales across the Northern Hemisphere using observations and model simulations. Temperature and humidity are signiﬁcantly positively correlated with O3 across continental regions in the mid-latitudes (˜ 35 - 60@N). Over the oceans, the relationships are consistently negative. For continental regions outside the mid-latitudes, the O3-meteorology correlations are mixed in strength and sign but generally weak. Over some high latitude, low latitude, and marine regions, temperature and humidity are signiﬁcantly anticorrelated with O3. Daily variations in transport patterns linked to the position and meridional movement of the jet stream drive the relationships among O3, temperature, and humidity. Within the latitudinal range of the jet, there is an increase (decrease) in O3, temperature, and humidity over land with poleward (equatorward) movement of the jet, while over the oceans poleward movement of the jet results in decreases of these ﬁelds and vice versa. Beyond the latitudes where the jet traverses, the meridional movement of the jet stream has variable or negligible eﬀects on surface-level O3, temperature, and humidity. The O3-meteorology relationships are largely the product of the jet-induced changes in the surface-level meridional ﬂow acting on the background meridional O3 gradient. Our results underscore the importance of considering the role of the jet stream and surface-level ﬂow for the O3-meteorology relationships, especially in light of expected changes to these features under climate change.


Introduction
Ambient surface-level ozone (O 3 ) plays a prominent role in atmospheric chemistry (Fiore et al., 2015;Pusede et al., 2015), while posing significant threats to human health (Landrigan et al., 2018) and ecosystem productivity (Tai & Martin, 2017).Long-term trends in observed O 3 in the Northern Hemisphere mid-latitudes reveal sustained, yearround increases in baseline O 3 concentrations (Parrish et al., 2012), underpinning the need for a better understanding of the drivers of O 3 variability.Meteorology strongly a↵ects O 3 concentrations and chemistry through both variations in prevailing weather conditions on daily, seasonal, or interannual timescales as well as long-terms trends associated with climate change (e.g., Jacob & Winner, 2009;Fiore et al., 2015;Otero et al., 2016;Lefohn et al., 2018).The meteorological, or transport-related, phenomena that a↵ect O 3 are not cause-and-e↵ect relationships in the same sense as emissions or chemical kinetics and energetics (i.e., temperature-dependent reaction or emissions rates).Rather, the link between O 3 and meteorology reflects a joint association (e.g., high temperatures are often associated with slow-moving anticyclones).
Previous studies have focused on characterizing the relationship between O 3 and temperature or humidity in historical data.Generally these studies found a positive O 3temperature relationship (e.g., Rasmussen et al., 2012Rasmussen et al., , 2013;;Pusede et al., 2015) and a variable O 3 -humidity relationship with substantial latitudinal variability (e.g., Camalier et al., 2007;Tawfik & Steiner, 2013;Kavassalis & Murphy, 2017).However, the majority of past studies on the O 3 -meteorology relationships focused on populated, industrialized portions of the Northern Hemisphere mid-latitudes, potentially overlooking important variations of these relationships elsewhere.These studies have been conducted for di↵erent and often non-overlapping time periods during which changes of O 3 precursors could a↵ect chemical background conditions (Kim et al., 2006;Derwent et al., 2010;Cooper et al., 2012;Simon et al., 2015;Lin et al., 2017).Finally, past studies have used di↵erent methodologies (e.g., O 3 -relationships derived from hourly, daily, or seasonal data; see Brown-Steiner et al. (2015) for additional information).All these factors complicate direct comparisons from study to study; thus, it is di cult to piece together a comprehensive sense of how the O 3 -meteorology relationships vary across the globe and what processes drive these relationships.Recent work by Kerr et al. (2019) and Porter and Heald (2019) suggests that greater than 50% of the covariance of O 3 and temperature in the United States (U.S.) and Europe on daily timescales stems from meteorological phenomena, not chemistry or emissions.It is an open question whether this also holds for the O 3 -humidity relationship.
There have been several meteorological mechanisms proposed to link O 3 with temperature and humidity.However, little consensus exists as to which mechanism is the most important and the regions or timescales over which it operates.Baroclinic cyclones can disperse built-up concentrations of pollution by entraining polluted air from the planetary boundary layer (PBL) into the free troposphere (e.g., Mickley, 2004;Leibensperger et al., 2008;Knowland et al., 2015Knowland et al., , 2017)).Quasi-stationary anticyclones such as the Bermuda High can influence regional climate and O 3 (e.g., Zhu & Liang, 2013).Properties of the PBL, such as its height, or temperature inversions and mixing within the PBL, have also been suggested as transport-related mechanisms that a↵ect surface-level O 3 (e.g., Dawson et al., 2007;He et al., 2013;Reddy & Pfister, 2016;Barrett et al., 2019).Winds near the earth's surface or aloft can ventilate pollution away from its source region (e.g., Camalier et al., 2007;Hegarty et al., 2007;Tai et al., 2010;Sun et al., 2017).Interactions among the atmosphere, land surface, and biosphere have been proposed to explain the O 3 -humidity relationship in North America (Tawfik & Steiner, 2013;Kavassalis & Murphy, 2017).The jet stream is a pronounced feature of the general circulation of atmosphere in both the Northern and Southern Hemisphere mid-latitudes and is characterized by a region of strong eastward wind aloft.Its existence arises from momentum and heat fluxes forced by transient eddies, and the jet extends throughout the depth of the troposphere (Woollings et al., 2010).The variability of surface-level summertime O 3 as well as its relationship with temperature have been linked to the latitude of the jet stream over eastern North America (Barnes & Fiore, 2013;Shen et al., 2015).Similar connections between the jet position, persistence of the jet in a given position, and wintertime particulate matter with a diameter < 2.5 µm (PM 2.5 ) have also been demonstrated in Europe (Ordóñez et al., 2019).
The aim of this paper is to document the relationships of surface-level temperature and specific humidity (henceforth "humidity") with O 3 in the Northern Hemisphere during boreal summer and explore the processes responsible for spatial variations of these relationships.Through our model simulations, we demonstrate that variations in transportrelated processes drive the covariance of O 3 with temperature and humidity on daily timescales.
We build o↵ of the previous regionally-focused work of Barnes andFiore (2013), Shen et al. (2015), and Ordóñez et al. (2019) to show the connections between the position of the jet stream and surface-level temperature, humidity, and O 3 variability hold across the Northern Hemisphere.Finally, we develop and test hypotheses that tie the jet stream to the surface-level relationships among O 3 , temperature, and humidity.

Model Simulations
The majority of our analysis of the O 3 -meteorology relationships is performed using simulations of NASA's Global Modeling Initiative chemical transport model (GMI CTM; Duncan et al., 2007;Strahan et al., 2007Strahan et al., , 2013)).The GMI CTM is driven by meteorological fields from the Modern-Era Retrospective analysis for Research and Applications, version 2 (MERRA-2;Gelaro et al., 2017).GMI CTM simulations used in this study have 1 latitude x 1.25 longitude horizontal resolution (⇠ 100 km) with 72 vertical levels, extending from the surface to 0.01 hPa.
The chemical mechanism of the CTM includes tropospheric and stratospheric chemistry with approximately 120 species and over 400 reactions.In addition to the spectrum of chemical processes dependent upon the model meteorology, several aspects of O 3 production and destruction also depend on the meteorology: biogenic emissions (temperature, photosynthetically active radiation), soil emissions of NO x (temperature, precipitation), lightning emissions of NO x (convective mass flux), wet deposition (wind, clouds, precipitation), and dry deposition (wind, clouds, temperature, pressure).Additional information about the natural and anthropogenic emission inventories and model parameterizations (e.g., biogenic emissions, lightning NO x , etc.) is provided in Kerr et al. (2019) and Strode et al. (2015).
The GMI CTM is a proven model to understand surface-level O 3 variability and its drivers (e.g., Duncan et al., 2008;Strode et al., 2015;Kerr et al., 2019).Kerr et al. (2019) evaluated the CTM with observations from an in-situ network in the U.S. and showed that the model skillfully simulated the observed daily variability of O 3 during the summer despite a high model bias in the eastern U.S. and low model bias in the western U.S; these biases are common among CTMs (e.g., Brown-Steiner et al., 2015;Guo et al., 2018;Phalitnonkiat et al., 2018).
In this study we focus on the O 3 -meteorology relationships in the Northern Hemisphere for a three-year period (2008 2010) during boreal summer (1 June 31 August).
We use O 3 from the model's surface level, which has a nominal thickness of ⇠ 130 m.
CTM output from the early afternoon (mean 1300 1400 local time), coinciding with the overpass time of the Afternoon Constellation ("A-Train") of Earth observing satellites, was archived as gridded fields, whereas hourly output was archived only at select sites.We consequently use modeled O 3 from this early afternoon period, noting that this time of day typically represents a time in which the PBL is well-mixed (e.g., Cooper et al., 2012) and daily O 3 concentrations reach their maximum (e.g., Schnell et al., 2014).
Considering O 3 during this early afternoon period versus longer averaging periods leads to similar results (Kerr et al., 2019).
Two simulations are analyzed in this study.The first is a control simulation with daily (or sub-daily) variations in meteorology, chemistry, and natural emissions.Anthropogenic emissions in this simulation vary from month to month.Unless otherwise indicated, all subsequent figures and analysis use this control simulation.In a second simulation referred to as "transport-only," we isolate the role of transport.Meteorological fields related to transport such as pressure, wind, convection, PBL height, and precipitation (as it a↵ects the vertical transport of O 3 via wet deposition) all vary on daily and sub-daily timescales in this transport-only simulation.The daily variations of other meteorological fields that a↵ect chemistry and emissions (e.g., temperature, clouds and albedorelated variables, surface roughness, specific humidity, soil moisture, and ground wetness) are removed by using a single, monthly mean diurnal curve for each of these fields at each grid cell.Therefore, any process that relies on these variables (e.g., photolytic and kinetic reaction rates, biogenic emissions, dry deposition) is identical for a given time of the diurnal cycle for all days in a particular month.Other non-biogenic emissions are fixed to a single monthly mean value with no diurnal variations.We note that although there are no day-to-day variations in emissions-and chemistry-related processes within a given month in the transport-only simulation, there is still seasonal and interannual variability.This transport-only simulation is similar to the "Transport" simulation discussed in Kerr et al. (2019) with the exception that specific humidity is also averaged to a monthly mean diurnal cycle.

Observations
We use in-situ observations of O 3 across North America, Europe, and China to examine the observed variations of the O 3 -meteorology relationships and assess the accuracy of the GMI CTM.We choose these regions because their in-situ networks, described  (Hjellbrekke & Solberg, 2019).
For China we use observations from the Chinese Ministry of Ecology and Environment (MEE) for summers 2016 2017 (Li et al., 2019).Observations are primarily from urban centers, and if a particular Chinese city has > 1 monitor, a city-wide average was computed following Z.Zhao and Wang (2017), resulting in data from 360 Chinese cities.
The choice of this 2016 2017 time period is because this Chinese observational network did not come online until the mid-2010s.Accordingly, when we assess the performance of the GMI CTM and discuss the observed O 3 -meteorology relationships in China, we use model simulations (Section 2.1) and reanalysis data (Section 2.3) for 2016 2017 rather than the 2008 2010 period used elsewhere in this study.

Meteorological Reanalysis
In addition to providing meteorological input to drive the GMI CTM, MERRA-2 is also used to determine the relationships between O 3 and meteorology.Several of the observational networks detailed in Section 2.2 lack co-located meteorological observations, and Varotsos et al. (2013) commented that lack of co-located O 3 and temperature (or other meteorological) observations necessitates the use of gridded products to examine the relationships between O 3 and meteorology.
MERRA-2 meteorological fields are not available at the satellite overpass times sampled by the GMI CTM simulations (Section 2.1).We calculate daily averages from the following MERRA-2 fields: hourly surface-level (10-m) zonal (U 10 ) and meridional (V 10 ) wind, three-hourly 2-m specific humidity (q), three-hourly 500 hPa zonal wind (U 500 ), and hourly PBL height (P BLH).Daily 2-m maximum temperature (T ) is computed as the maximum of hourly values.Our use of daily maximum temperature follows Zhang and Wang (2016) and Meehl et al. (2018).
There are uncertainties associated with an assimilated product like MERRA-2, but Bosilovich et al. (2015) presented evidence that MERRA-2 provides a very good quality reanalysis data set.As the MERRA-2 data have higher horizontal resolution than the GMI CTM (0.5 latitude ⇥ 0.625 longitude for MERRA-2 versus 1 latitude ⇥1.25 longitude for the CTM), we degrade the MERRA-2 data to the resolution of the CTM using xESMF, a universal regridding tool for geospatial data (Zhuang, 2018).

Statistical analysis
We use the Pearson product-moment correlation coe cient and the slope of the ordinary least squares (OLS) regression (denoted r(x, y) and dy/dx for variables x and y, respectively) to (1) quantify the O 3 -meteorology relationships on daily timescales and (2) evaluate the ability of the GMI CTM to accurately simulate observed O 3 from the in-situ networks detailed in Section 2.2.The correlation coe cient is a parametric test that measures the degree of linear correlation between x and y, and the OLS regression describes the linear relationship between x (explanatory variable) and y (dependent variable).
The serial dependence (persistence) in our meteorological and chemical data reduces the e↵ective sample size by an amount not known a priori and inhibits the use of traditional hypothesis testing methods such as t-tests to evaluate significance (Zwiers & von Storch, 1995;Wilks, 1997;Mudelsee, 2003).Therefore, we use moving block bootstrapping to quantify the significance of the correlation coe cient.While traditional bootstrapping resamples individual, independent values of the time series, moving block bootstrapping resamples continuous subsets of the time series with blocklength L and does not destroy the ordering responsible for the persistence (Wilks, 2011).At each grid cell we synthetically construct a null distribution of 10000 bootstrapped realizations of the correlation coe cient (Mudelsee, 2014) and use L = 10 days.As a rule of thumb, blocklengths should generally exceed the decorrelation time.More rigorous methods for optimizing L exist, but we find that L = 10 is adequate for our application and our results are not sensitive to the exact value of L. To evaluate the significance, we estimate the 95% confidence interval using the percentile method of the bootstrapped values (i.e., the 95% confidence interval of our 10000 realizations is given by the 250th and 9750th sorted values).If this confidence interval does not contain zero, we declare the correlation coe cient significant.

Jet stream position
We define the latitude of the jet ( jet ) as the latitude of maximum zonal winds at 500 hPa (U 500 ) on each day.This approach to determine jet follows Barnes and Fiore (2013) but di↵ers in two ways: (1) Barnes and Fiore (2013) determined jet using U 500 averaged over the eastern North America zonal sector.We determine jet locally (at each longitudinal grid cell) and between 20 70 N; (2) After finding the maximum U 500 for each longitude, we employ a simple moving average that is essentially a convolution of daily jet of a general rectangular pulse with width ⇠ 10 .This approach removes large changes (abrupt latitudinal shifts) in jet with longitude.Using smoothed versus unsmoothed data or di↵erent pulse widths yields similar overall findings in this study.

Cyclone detection and tracking
To assess the impact of extratropical cyclones on surface-level O 3 , we use the MAP Climatology of Mid-latitude Storminess (MCMS) database to locate cyclones (Bauer & Genio, 2006;Bauer et al., 2016).Within MCMS, cyclones are detected as minima in the ERA-Interim sea level pressure (SLP) dataset (Dee et al., 2011) and are subject to additional filters to screen for spurious detections.Once detected, MCMS tracks cyclones with criteria that require gradual changes in SLP, no sudden changes in direction, and cyclones travel distances less than 720 km over single six-hourly time steps.Additional details can be found in Bauer and Genio (2006) and Bauer et al. (2016).
We begin with an analysis of the distribution and variability of modeled surfacelevel O 3 during summer (Figure 1a).Concentrations of O 3 are highest (⇠ 30 60 ppbv) in a broad mid-latitude band over continental regions extending from 20 50 N.The GMI CTM suggests that O 3 is not zonally-symmetric within this mid-latitude band and that the highest mean concentrations (> 50 ppbv) are in the Middle East and central and eastern Asia.Outside of the mid-latitudes, the CTM simulates lower O 3 concentrations (< 30 ppbv), and the lowest concentrations in the hemisphere (< 15 ppbv) are found in the remote tropical marine atmosphere.We characterize the daily variability of O 3 by the standard deviation, and two levels (8 and 10 ppbv) are highlighted with the thin dashed and thick contours in Figure 1a.The hemispheric distribution of mean summertime surface O 3 and its variability in Figure 1a is consistent with simulations from other models in a recent model intercomparison (Turnock et al., 2020).
To illustrate the possible influence of anthropogenic emissions on the spatial variability of mean O 3 concentrations, we show mean annual anthropogenic NO x emission data from the Emissions Database for Global Atmospheric Research (EDGAR; Crippa et al., 2018) at their native resolution (0.1 latitude x 0.1 longitude) in Figure 1b.EDGAR is used in the GMI CTM, but is overwritten by regional inventories, if available.To first order, regions with the highest O 3 concentrations and largest O 3 variability generally coincide with industrialized regions that have high precursor emissions (Figure 1).
We evaluate whether the modeled O 3 distribution shown in Figure 1a is realistic using the correlation coe cient, calculated for CTM grid cells containing in-situ monitors (Section 2.2).The temporal correlation between modeled and observed O 3 > 0.5 in the vast majority of grid cells (Figure 2).The strength of the correlation is slightly weaker in central China than other parts of China or Europe and North America (compare Figures 2c and 2a-b), but there are no other readily-detectable spatial patterns regarding the strength of the correlation.
The primary goal of our study is to document the O 3 -meteorology relationships in terms of the strength of the temporal correlation of O 3 with temperature and humidity.Thus, the model's ability to accurately reproduce this covariance (Figure 4) is the relevant litmus test for model performance.Recent studies by Strode et al. (2015) and Kerr et al. (2019) have shown that the GMI CTM can reproduce the meteorological-and emissions-driven variability of summertime O 3 as well as the O 3 -temperature relationship over the U.S. On account of these studies and our analysis in Figure 2, the GMI CTM is a suitable tool to address our research questions.The agreement between the observed and modeled O 3 -meteorology correlations will be explored in the following section (Section 4), and this analysis will also support our use of the GMI CTM to simulate the covariance of O 3 with temperature or humidity.

O 3 -meteorology relationships
In this section we describe the relationships among O 3 , temperature, and humidity on daily timescales in the Northern Hemisphere during summer.We primarily use the GMI CTM but also compare the modeled relationships to observed values.As discussed in the Introduction (Section 1), other studies have focused mainly on subsets of the Northern Hemisphere mid-latitudes, but our examination of the relationships across the entire hemisphere allows us to have a more holistic sense of the synoptic-scale variations of these relationships.
In the mid-latitudes (⇠ 30 60 N), statistically-significant positive values of r(T, O 3 ) are simulated by the CTM throughout North America and Eurasia (Figure 3a), but over virtually all the oceans r(T, O 3 ) is negative.Poleward of the mid-latitudes, the strength of r(T, O 3 ) decreases nearly monotonically over land, reaching either weak values or sig-nificantly negative correlations (Figure 3a).The O 3 -temperature relationship is varied equatorward of the mid-latitudes; but, in the zonal mean, r(T, O 3 ) decreases to negative values south of 30 N. Previous work by Rasmussen et al. (2012) and Brown-Steiner et al. (2015) in the U.S. and Han et al. (2020) and Lu, Zhang, Chen, et al. (2019) in China showed a similar latitudinal gradient of r(T, O 3 ).Despite the general tendency of a positiveto-negative relationship between O 3 and temperature with decreasing latitude, there are regions at low latitudes with significant positive correlations between O 3 and temperature (Central America, Sahel, the south coast of the Arabian Peninsula, Indo-Gangetic Plain; Figure 3a).
The sign of r(q, O 3 ) generally transitions from significantly positive in the continental mid-latitudes to significantly negative over continental regions at higher and lower latitudes and over the oceans (Figure 3b).Unlike r(T, O 3 ), the sign of r(q, O 3 ) outside of the mid-latitudes is more spatially uniform.The only exceptions to the widespread negative correlations occur over small parts of the Mediterranean Sea and Caribbean and Indian Oceans (Figure 3b).These results are supported by modeling and observational studies in the U.S. and China, which indicate r(q, O 3 ) > 0 in the northern U.S. and China and r(q, O 3 ) < 0 in southern U.S. and China (e.g., Tawfik & Steiner, 2013;Kavassalis & Murphy, 2017;Li et al., 2019).
Many other studies report dO 3 /dT (Rasmussen et al., 2012;S. Zhao et al., 2013;Brown-Steiner et al., 2015;Kerr et al., 2019;Porter & Heald, 2019), and we also present dO 3 /dT and dO 3 /dq in Figure S1a-b for comparisons with these other studies.The spatial variations of the slopes shown in Figure S1a-b are qualitatively similar to r(T, O 3 ) and r(q, O 3 ) shown in Figure 3, as is expected by construction.We also note that the large-scale patterns in Figure 3 are preserved whether r(T, O 3 ) and r(q, O 3 ) or dO 3 /dT and dO 3 /dq are calculated with daily data aggregated over summers 2008 2010 or with daily data from individual summers.
To test whether the modeled O 3 -meteorology relationships are realistic, we calculate r(T, O 3 ) and r(q, O 3 ) from the in-situ networks described in Section 2.2.The strength of the zonally-averaged values of observed and modeled r(T, O 3 ) and r(q, O 3 ) generally reaches a maximum around 50 N across four distinct regions (Figure 4).In Europe and the eastern U.S., the CTM slightly overestimates the strength of r(T, O 3 ) and r(q, O 3 ) by ⇠ 0.1 0.3, similar to other studies (e.g., Brown-Steiner et al., 2015;Kerr et al., 2019).
Since we used temperature from MERRA-2 to calculate the observed r(T, O 3 ) and r(q, O 3 ) (some of the observational networks lack co-located meteorological measurements), differences in the O 3 -meteorology relationships are driven by di↵erences in simulated versus observed O 3 rather by temperature.Observations are sparse outside of the mid-latitudes.
A small number of AQS monitors in Alaska and NAPS monitors in northern Canada supports the transition of r(T, O 3 ) and r(q, O 3 ) from positive to negative at high latitudes that is suggested by the model (Figure 4).
In summary, the observation-and model-based analysis of the relationships among surface-level O 3 , temperature, and humidity reveals substantial variability across the Northern Hemisphere during summer.The terrestrial mid-latitudes (⇠ 30 60 N) stand out as the largest, most spatially-coherent region with significant positive relationships of O 3 with temperature and humidity (Figures 3-4).The O 3 -meteorology relationships are negative over nearly all marine regions, while they are mixed in sign and often not significant at high and low latitude continental regions (Figures 3-4).

Factors causing the O 3 -meteorology relationships
The O 3 -meteorology relationships in Figure 3 are far from uniform, and their spatial structure begs the question: what factors drive these relationships?In Section 1, we discussed several direct and indirect drivers that have been linked to O 3 variability, such as emissions, chemistry, and transport.Recent work has shown that transport-related processes are key contributors to the O 3 -temperature relationship in the U.S. and Europe (Kerr et al., 2019;Porter & Heald, 2019), and we expand on these previous findings and examine the covariance of O 3 with temperature and humidity over the Northern Hemisphere.We do this using the transport-only GMI CTM simulation in which the daily variability of chemistry and emissions are fixed (Section 2.1).
The transport-only simulation achieves similar mean O 3 concentrations as the control simulation (compare Figures 1a and S2a There are a few regions such as the eastern U.S. and southeast Asia where the daily variability of chemistry and emissions appears important for the O 3 -meteorology relationships (Figures 5, S1e-f).In these regions the strength of the correlation and the magnitude of the slopes decreases up to ⇠ 50% in the transport-only simulation and the correlation coe cient switches from significant to not significant.We note that these regions have high levels of anthropogenic emissions (e.g., NO x ; Figure 1b) and biogenic emissions (e.g., isoprene; Guenther et al., 2012).Further work is warranted to understand how emissions (and the chemical processes linking emissions to O 3 production) contribute to the O 3 -meteorology relationships in these regions.
Although daily variations in chemistry-and emissions-related processes do not drive the O 3 -meteorology relationships across the Northern Hemisphere, the importance of chem-istry and emissions in setting the background state should not be ignored.To illustrate this, we return to Figure 3 and draw attention to the stark land-ocean contrasts in the O 3 -meteorology relationships with marine regions generally characterized by negative correlations.These marine regions are largely low NO x environments (Figure 1a).In this type of chemical regime, an increase in humidity or temperature is expected either to not impact or decrease O 3 (e.g., Johnson et al., 1999;Coates et al., 2016).The transportonly simulation still includes this low NO x marine environment relative to other regions, just without day-to-day variations.
These results answer our original question whether daily variations in transport, chemistry, or emissions are primarily responsible for the O 3 -meteorology relationships, but they also raise the question of which aspect(s) of transport links temperature and humidity to O 3 .In the next section we investigate the role of the jet stream on surfacelevel temperature, humidity, and O 3 , and we also develop and test hypotheses to link synoptic-scale flow aloft to meteorology and composition at the surface.

5.1
The role of the jet stream Barnes and Fiore (2013) determined that the largest O 3 variability and peak strength of r(T, O 3 ) are located near jet in the eastern U.S.These results were further explored by Shen et al. (2015) who found that O 3 responded to seasonal variations in the position of the jet stream and that a poleward shift of the jet increased O 3 concentrations south of the jet.In this section we expand upon this previous work and document the response of surface-level O 3 , temperature, and humidity to daily changes in jet across the Northern Hemisphere.
The time-averaged latitude of the jet stream ( jet ) is shown by the scatter points in Figure 1, and jet averaged over the entire hemisphere is 50.1 N. The variability of the jet, cast in terms of the standard deviation, averaged over the Northern Hemisphere is 10.5 , but its variability is not constant throughout the hemisphere (vertical bars in Figure 1).Rather, we note the largest variability over continental regions, particularly Eurasia (⇠ 20 ), and smaller variability over maritime regions, coinciding with the Atlantic and Pacific storm tracks.The position of the jet is only one metric to describe the jet stream, and other jet-related measures exist (e.g., strength of the jet, waviness).Our focus on jet rather than other metrics is based on Ordóñez et al. (2019) who found that jet exerts a stronger influence than the strength of the jet on surface-level pollution extremes.
The maximum variability of O 3 (Figure 1a) and the strength of the O 3 -meteorology correlations (Figures 3-5) peak at or slightly south of jet , and jet also separates regions with elevated O 3 concentrations to its south from regions with low (< 30 ppbv) concentrations to its north (Figure 1a).These results are consistent with Barnes and Fiore (2013); however, it is worth pointing out a couple of exceptions: (1) In Asia, O 3 variability peaks over a broader latitudinal range, extending from jet to ⇠ 20 N (Figure 1).( 2) There are regions with significant positive values of r(T, O 3 ) such as the Sahel and India that do not coincide with jet (Figure 3a).Our current work also reveals the weak-to-negative correlation between O 3 and humidity or temperature for marine environments and some high and low latitudes.
To further examine the role of the jet stream on the O 3 -meteorology relationships, we segregate summer days into two subsets: days when the jet stream is in poleward (PW) and equatorward (EW) position.Days classified as PW (EW) are days in which jet exceeds (is less than) the 70th (30th) percentile of all daily jet at each longitudinal grid cell.We construct composites of O 3 , temperature, and humidity by identifying the average value of these fields on days with a PW or EW jet stream and thereafter calculate the di↵erence of these PW and EW composites.
The di↵erence in the PW and EW composites (PW -EW) of O 3 , temperature, and humidity are positive in the mid-latitudes over land (Figure 6), which indicates that these fields increase when the jet is in a more northerly position.The positive values are generally significant (hatching in Figure 6), coincide with the latitudinal band over which the jet stream migrates, and persist 10 15 north and south of jet over land.Outside the continental mid-latitudes, the association between the position of the jet and O 3 , temperature, or humidity is weak and not statistically significant (Figure 6).
In contrast, there is a di↵erence in the response of O 3 to the jet stream versus temperature and humidity over the mid-latitude ocean basins.In the case of O 3 , a poleward movement of the jet decreases O 3 over the oceans but increases it over land, while temperature increases over both land and oceans (Figure 6a).This sharp land-ocean contrast, akin to the land-ocean contrasts in the O 3 -meteorology correlations (Figure 3), could reflect land-ocean asymmetries in O 3 and its precursors and will be further explored in Section 5.3.On the other hand, temperature and humidity increase over the oceans as the jet shifts poleward, akin to the behavior of these variables over land (Figure 6b-c).
The impact of the jet stream on O 3 , temperature, and humidity outside of the mid-latitudes is largely not significant (Figure 6).
For completeness, maps of the correlation of jet distance with the variables in Figure 6 are shown in Figure S3.We note that the strength of the correlation between jet and O 3 and meteorology is weaker than r(T, O 3 ) and r(q, O 3 ), and the spatial extent of areas with significant correlations is smaller (compare Figures 3 and S3).
While the response of O 3 and meteorological fields to the meridional movement of the jet stream is consistent in its sign in the mid-latitudes over land, there are some regions outside of the continental mid-latitudes where jet movement leads to increases of one variable and decreases of another.China is an example of this.As the jet migrates poleward, O 3 significantly increases, as it does throughout the mid-latitudes; however, temperature remains more or less constant, and humidity slightly decreases (Figures 6,     S3).This discrepancy and others evident in Figures 6 and S3, particularly those at lower latitudes and over the oceans, are beyond the scope of this study, but future studies should further examine and address regions where O 3 , temperature, and humidity are decoupled from the jet in this manner.
Having uncovered the dominant role of transport and the connections with the jet, we next explore transport-related processes that might be responsible for the relationships among surface-level O 3 , the jet stream, and meteorology.As cyclones are commonlyinvoked to explain O 3 variability, we begin by showing the impact of the jet stream on cyclone frequency and, in turn, the e↵ect of cyclones on O 3 .We then explore and discuss how the jet stream a↵ects the surface-level meridional flow and commensurate changes in O 3 , temperature, and humidity.

Cyclones
Mid-latitude baroclinic cyclones follow a storm track dictated by the jet stream, and changes in jet a↵ect the location of this storm track (e.g., Shen et al., 2015).To assess the dependence of cyclone frequency on jet , we show the spatial distribution of the climatological frequency of cyclones detected by MCMS (Section 2.4.3) in Figure 7a.
The highest frequency of mid-latitude cyclone detections largely follows jet and is o↵set north of the jet by ⇠ 10 over North America.In other regions such as eastern Asia the peak cyclone frequency occurs in a broader latitudinal band, extending north and south of jet by ⇠ 15 (Figure 7a).
We identify the subset of days with a poleward-shifted or equatorward-shifted jet using the 70th and 30th percentiles of the daily latitudes of the jet stream, as previously described, to determine the dependence of cyclones on jet .We thereafter determine the frequency of cyclones on these subsets of days and show the di↵erence (Figure 7b).The meridional movement of the jet a↵ects cyclones in two di↵erent ways.First, the total number of cyclones on days when the jet is in a poleward position is 15% less than on days when the jet is equatorward.Second, the storm track shifts alongside the jet, and cyclones are more highly concentrated about jet when the jet is equatorward compared with when it is poleward (Figure 7b).
The decrease and latitudinal shift in cyclone frequency with meridional movements of the jet stream could be the transport-related mechanism responsible for the above O 3meteorology relationships.The cold fronts associated with mid-latitude cyclones have been suggested as a mechanism for the ventilation of the eastern U.S. (Mickley, 2004), and Knowland et al. (2015) and Jaeglé et al. (2017) demonstrated how cyclones redistribute O 3 , its precursors, and other pollutants vertically and horizontally in the atmosphere.We assess the impact of cyclones on surface-level O 3 by further filtering the cyclones from the MCMS dataset (Section 2.4.3),requiring that a particular cyclone (1) occurs over land and ( 2) is detected for 2 six-hourly time steps to allow us to calculate the direction of propagation.We then rotate cyclones following Knowland et al. (2015) and Knowland et al. (2017) such that they propagate to the right of Figure 8 to account for the impact of di↵erent ascending and descending airstreams within the cyclones.Applying these filters to cyclones in summers 2008 2010 yields ⇠ 730 cyclones with an average lifetime of ⇠ 54 hours.The mean direction of cyclone propagation is east-southeast (⇠ 120 , where 0 is north).Though we have only considered cyclones occurring over land in this analysis, compositing all land-and ocean-based cyclones produces O 3 anomalies of similar magnitude.
The largest negative O 3 anomaly occurs in the "cold sector" of the cyclone, and the largest positive anomaly occurs in the "warm sector," However, these positive and negative anomalies cancel each other when averaged over the footprint of the cyclones, leading to a net ⇠ 0 ppbv change in O 3 (Figure 8).Comparing our results with conceptual models and case studies of baroclinic cyclones (e.g., Cooper et al., 2004;Polvani & Esler, 2007) hints that the positive anomalies in Figure 8 occur near the warm conveyor belt (WCB), where there is likely polluted air entrained from the PBL and lower troposphere.On the other hand, the largest negative anomalies are found in the vicinity of the dry intrusion (DI) and could be influenced by cleaner air entrained from the upper troposphere or lower stratosphere.The roles of the WCB in ventilating pollution from the PBL and the cleaner air brought to the PBL by the DI could cancel each other out and be one reason for the small increases and decreases in surface-level O 3 .
If cyclones were the mechanism that linked jet to surface-level O 3 , we might expect that the cyclones-driven impact on O 3 would be > 6 ppbv in the mid-latitudes, similar to the impact that jet has on O 3 (Figure 6a).However, our analysis in Figure 8 indicates that, on average, cyclones have a much weaker e↵ect on surface-level O 3 , despite the connections between cyclones and the jet stream (Figure 7b).There is, though, substantial variability among individual cyclones (the standard deviation of the O 3 anomaly is a factor of ⇠ 6 greater than the largest anomaly; Figure 8).As such, some cyclones might be e↵ective at reducing surface-level O 3 , but this is far from the case for all cyclones.
Other studies support the small role of cyclones on surface-level O 3 .Knowland et al. (2015) showed that the surface-level O 3 anomaly associated with springtime cyclones in the North Atlantic and Pacific is small (i.e., 5 < O 3 < 5 ppbv); however, they found a larger impact when examining the mid-to upper-level O 3 anomalies.Moreover, Leibensperger et al. (2008) found a negative correlation between the number of O 3 pollution events and the number of mid-latitude cyclones passing through the southern climatological storm track (⇠ 40 50 N) over eastern North America on interannual timescales, but Turner et al. (2012) demonstrated that the cyclone-O 3 correlation is weak, and cy-clone frequency explains less than 10% of the variability of O 3 pollution events in the region.
In summary, while the storm track dictating the preferred location of baroclinic cyclones shifts with the jet (Figure 7b), cyclones are likely not the key mechanism controlling O 3 variability in the Northern Hemisphere mid-latitudes as they only explain a small fraction of the changes of O 3 associated with daily migrations of the jet (Figure 8).

Meridional transport
The ventilation and dilution of the PBL, the surface-level zonal flow (U 10 ), or the total wind (U 10 ) could link the position of the jet stream to surface-level O 3 .However, an analysis of P BLH, U 10 , and U 10 rules out these variables as drivers of the O 3 -jet relationship (Text S1-S2, Figures S4-5).To summarize: jet is not significantly correlated with variations in P BLH and U 10 throughout the majority of the Northern Hemisphere.
Similar to our analysis of cyclones in Figures 7-8, U 10 is significantly correlated with jet throughout parts of the mid-latitudes but not correlated with O 3 independently of the jet.
However, the surface-level meridional flow (V 10 ) is significantly correlated with the position of the jet in the mid-latitudes (Figure 9a).When the jet is in a poleward position, V 10 increases by more than 2 m/s throughout the mid-latitudes with the largest increases centered over the oceans (Figure 9a).In the mid-latitudes, time-averaged V 10 is varied in sign but generally weak ( 0.5 < V 10 < 0.5), so the large values of V 10 accompanying a poleward jet represent a large increase in the southerly flow.
In addition to its connections with jet , V 10 is significantly positively correlated with O 3 in the continental mid-latitudes (Figure 9b).Here, the strength of r(V 10 , O 3 ) rivals that of r(T, O 3 ) and r(q, O 3 ) (compare Figures 9b and 3), suggesting that the surfacelevel meridional flow is also a key covariate of O 3 variability on daily timescales.In parts of the mid-latitudes such as eastern North America or Asia, a unit increase in V 10 is associated with an increase of O 3 that is roughly one-third of its total daily variability (Fig- ure S6).Equatorward of the mid-latitudes (particularly for ⇠ 10 30 N), V 10 is significantly negatively correlated with O 3 , while r(V 10 , O 3 ) is not significant poleward of the mid-latitudes.Thus far we have shown significant positive relationships among jet , V 10 , O 3 , and the meteorological variables in the mid-latitudes (Figures 6a, 9).When the jet is poleward, surface-level meridional flow becomes strongly southerly, and there is significant poleward advection of O 3 , temperature, and humidity.
We posit that the relationships of O 3 with jet and the meteorological variables are largely the product of surface-level meridional flow acting on the latitudinal background gradients.Ozone generally peaks south of jet (Figure 1a), so there are negative gradients in the vicinity of the jet (Figure 9b).These negative gradients are wellaligned with the regions where there is increased southerly flow at the surface when the jet is poleward (Figure 9a).This configuration serves to advect higher concentrations of O 3 into the mid-latitudes when the jet is poleward (Figures 9b, S6).Although not shown here, the latitudinal gradients of temperature and humidity are broadly similar to dO 3 /d inasmuch as they are positive south of the mid-latitudes.When surface-level southerly flow increases, these gradients favor increases of temperature and humidity in the midlatitudes, as is evident in Figure 6b-c.
The importance of the background gradient can also partially explain the negative O 3 -jet relationships over the oceans.Latitudes where dO 3 /d > 0 often extend farther poleward over the oceans than over land.For example, over the Pacific storm track dO 3 /d > 0, while dO 3 /d < 0 between ⇠ 20 and 40 N in the Pacific (Figure 9b).Under these conditions, increased southerly flow associated with the poleward movement of the jet would decrease O 3 (i.e., a negative O 3 -jet relationship).Other factors also may be im-portant in marine environments.For example, strong surface-level zonal winds in the vicinity of the Atlantic and Pacific storm tracks may lead to zonal gradients that are as important as the meridional background gradients investigated in this section.
The importance of both meridional flow and the latitudinal background gradient has been the subject of recent studies for O 3 and other trace gases.Keppel-Aleks et al. (2012) showed that the daily variability of total column carbon dioxide was dominated by non-local e↵ects and primarily reflects the synoptic scale latitudinal carbon dioxide gradient.Changes in the mean meridional circulation (specifically the extratropical stratosphericto-tropospheric transport associated with the Southern Hemisphere Hadley Cell) have been suggested to explain recent trends in Southern Hemisphere tropospheric O 3 (Lu, Zhang, Zhao, et al., 2019).On smaller spatial scales, transport-related features favoring southerly flow (e.g., the nocturnal low-level jet in the U.S.) are important for explaining O 3 in the PBL (Taubman et al., 2004).Our future work will further elucidate the main physical features that link the jet stream, surface-level meridional flow, and background tracer gradients.
In the mid-latitudes, the meridional vacillation of the jet stream impacts the surfacelevel meridional flow (Figure 9a).The meridional flow, in turn, plays a profound role in surface-level O 3 variability (Figure 9b).Temperature, humidity, and O 3 are generally higher south of jet , and the meridional flow acts on their background gradients and leads to the coupling between the jet stream and O 3 and the meteorological variables shown in Figure 6.

Conclusions
The primary intent of this study was to document the relationships among surfacelevel O 3 , temperature, and humidity and explore the cause(s) of these relationships.Both observations and the GMI CTM support substantial spatial variations in r(T, O 3 ) and r(q, O 3 ).In continental regions of the mid-latitudes (⇠ 30 60 N), the O 3 -meteorology relationships are significantly positive (Figures 3-4).The O 3 -meteorology relationships are significantly negative over the oceans (Figure 3).For other continental regions outside the mid-latitudes, r(T, O 3 ) and r(q, O 3 ) are generally weak and often not statistically significant, but we have shown regions at low latitudes (e.g., Central America, Sahel) that are exceptions to this rule-of-thumb (Figure 3).
Our transport-only GMI CTM simulation indicates that the O 3 -temperature and O 3 -humidity relationships are largely driven by transport-related phenomena on daily timescales (Figure 5).We stress that these findings do not trivialize the importance of chemistry and emissions.Chemistry-and emissions-related processes are essential for setting the background state for the production of a secondary pollutant such as O 3 ; however, daily variations in these processes are not the dominant drivers of O 3 variability or its covariance with temperature and humidity.Our results showcasing the dominant role of transport are in line with previous work by Kerr et al. (2019) and Porter and Heald (2019), which showed that a majority of the O 3 -temperature relationship in the U.S. and Europe derive from meteorological phenomena.
The variability of surface-level O 3 , temperature, and humidity are linked to the meridional movement of the jet stream in the Northern Hemisphere mid-latitudes.This result extends previous work focusing on the eastern U.S. (e.g., Barnes & Fiore, 2013;Shen et al., 2015) to the entire Northern Hemisphere.Over land in the mid-latitudes, a poleward (equatorward) shift of the jet is associated with increased (decreased) surface-level O 3 , temperature, and humidity (Figures 6, S3).Over the oceans, temperature and humidity respond to this meridional vacillation of the jet in the same fashion as over land, but the poleward (equatorward) movement of the jet decreases (increases) O 3 .
We ultimately found that the jet influences these surface-level fields by means of changes in the surface-level meridional flow.On days when the jet is in a poleward position, the pronounced southerly flow in the mid-latitudes together with the latitudinal gradients of O 3 , temperature, and humidity generally lead to increases of O 3 , temperature, and humidity in the mid-latitudes (Figures 6, 9).We have shown clear land-ocean di↵erences in the relationships among O 3 , temperature or specific humidity, and the jet stream (Figures 3,6,S3).We partially attribute these the land-ocean contrasts to differences in the latitudinal gradient of O 3 over land versus over the ocean (Figure 9b).
Establishing the spatial variations of the O 3 -meteorology relationships is a prerequisite to understand which regions could experience an "O 3 -climate penalty" (Wu et al., 2008) under future climatic changes.As the O 3 -meteorology relationships in the presentday climate are far from uniform in both magnitude and sign, it is unlikely that future changes in the climate will a↵ect O 3 uniformly.Furthermore, as the relationships among O 3 , temperature, and humidity are driven by an indirect association with transport, caution should be used when applying any measures of the current sensitivity of O 3 to meteorological variables (e.g., dO 3 /dT or dO 3 /dq from Figure S1) to future climatic changes.
Overall, our results demonstrate the importance of the position of the jet stream and surface-level meridional flow on O 3 variability in the Northern Hemisphere, both of which will be a↵ected by the future climate (e.g., Barnes & Polvani, 2013;Shaw & Voigt, 2015;Grise et al., 2019).A robust poleward displacement of the jet stream is expected in the twenty-first century, while changes to other properties of the jet (i.e., variations in speed; north-south movement) will exhibit spatial heterogeneity (Barnes & Polvani, 2013).The e↵ect of these changes on surface-level O 3 needs to be explored.September 6, 2020, 1:47am

Acknowledgments
). Percentage di↵erences in mean O 3 between simulations are generally less than ±5%, suggesting that the non-linearities underpinning O 3 chemistry do not drastically change mean O 3 concentrations when day-to-day variations in chemistry-and emissions-related processes are removed.Regions in Figure 1a with high NO x (and presumably other precursor) emissions such as the eastern U.S., Europe, and China experience the largest decrease in mean O 3 concentrations as the daily variability of chemistry-and emissions-related processes are removed (Figure S2).The O 3 -meteorology relationships calculated with O 3 from the transport-only simulation are remarkably similar to the same quantities from the control simulation (e.g., compare Figures S1a-b and S1c-d), emphasizing the dominance of transport on these relationships.As the transport-only simulation used monthly mean values or monthly averaged diurnal cycles for processes related to chemistry and emissions, it is possible that some of the daily correlations over the three summers in our measuring period could be due to month-to-month or interannual variations in temperature or humidity coupled to chemistry or emissions.However, Porter and Heald (2019) found a similar dominance of transport in their simulations where summertime averaged values were used (rather than monthly averages).Furthermore, when we repeat the correlation analysis (i.e., Figure 3) using daily data from individual months (rather than the combined nine months) we find good agreement between the correlations from control and transport-only simulations with both showing the key features (e.g., positive correlations over mid-latitude continental regions, negative values over the oceans).Taken all together, these results indicate transport is the dominant process driving the O 3 -meteorology relationships across Northern Hemisphere mid-latitudes.Over most of the oceans and a majority of the continental regions in the Northern Hemisphere, the strength of the O 3 -meteorology relationships slightly increases in the transport-only simulation (negative values in Figures 5, S1e-f).The hatching in Figures 5 and S1e-f indicates that the significance of the O 3 -meteorology relationships is largely retained when only daily variations in transport-related processes are considered.
below, measure and archive O 3 hourly.Since the model outputs O 3 averaged over 1300-1400 hours (local time), comparing this output with hourly O 3 observations averaged over the same time of the day represents the most direct comparison.The lack of in-situ networks with observations at a high temporal frequency in many other parts of the world hinders our ability to examine model performance over other regions.