Arctic Tropospheric Ozone Trends

Observed trends in tropospheric ozone, an important air pollutant and short‐lived climate forcer (SLCF), are estimated using available surface and ozonesonde profile data for 1993–2019, using a coherent methodology, and compared to modeled trends (1995–2015) from the Arctic Monitoring Assessment Program SLCF 2021 assessment. Increases in observed surface ozone at Arctic coastal sites, notably during winter, and concurrent decreasing trends in surface carbon monoxide, are generally captured by multi‐model median trends. Wintertime increases are also estimated in the free troposphere at most Arctic sites, with decreases during spring months. Winter trends tend to be overestimated by the multi‐model medians. Springtime surface ozone increases in northern coastal Alaska are not simulated while negative springtime trends in northern Scandinavia are not always reproduced. Possible reasons for observed changes and model performance are discussed including decreasing precursor emissions, changing ozone dry deposition, and variability in large‐scale meteorology.

• Coherent ozone trend analysis methodology applied to multi-decade, pan-Arctic surface and ozonesonde datasets and multi-model medians • Increasing winter Arctic tropospheric ozone overestimated by models in the free troposphere, and spring surface changes not captured • Spring (summer) decreases (increases) in observed ozone throughout the troposphere, not always simulated by models

Supporting Information:
Supporting Information may be found in the online version of this article.
production of anthropogenic or natural emissions of O 3 precursors, including nitrogen oxides (NO x ), carbon monoxide (CO) and methane (CH 4 ), in the Arctic, or following air mass transport from mid-latitudes, as well as transport of O 3 from the stratosphere (Law et al., 2014;Schmale et al., 2018).Sinks include photochemical destruction, including reactions involving halogens leading to so-called ozone depletion events (ODEs) (Barrie, et al., 1988;Simpson et al., 2007), and surface dry deposition (Clifton et al., 2020).Growth in anthropogenic emissions since pre-industrial times has led to increases in tropospheric O 3 throughout the Northern Hemisphere (NH) (Cooper et al., 2020;Gaudel et al., 2018;Tarasick et al., 2019;Turnock et al., 2020) contributing to observed global and Arctic warming over the past century (e.g., Griffiths et al., 2021;Szopa et al., 2021).Since the mid-1990s, a mix of relatively weak positive and negative trends (+1 to −1 parts per billion by volume (ppbv) per decade) have been reported in the NH at the surface and in the free troposphere (FT), with largest increases over south and eastern Asia, associated with increasing anthropogenic emissions (Cooper et al., 2020;H. Wang et al., 2022).
To date, only a few studies have focused on assessing tropospheric O 3 trends in the Arctic.While positive O 3 trends were diagnosed at several surface sites, results do not always have high certainty, and both positive and negative trends were reported at some Canadian sites (Cooper et al., 2020;Sharma et al., 2019;Tarasick et al., 2016).In the Arctic FT, studies found significant positive trends (B.Christiansen et al., 2017;H. Wang et al., 2022), no trends (Tarasick et al., 2016), or mixed trends in different seasons (Bahramvash Shams et al., 2019).Differences in the periods analyzed, sign or magnitude of trends, based on different methodologies, data averaging, etc. emphasizes the need to further examine trends using the same methodology.Coherent estimation of observed trends, and evaluation of modeled trends, is needed to better understand O 3 changes and impacts on Arctic climate that are sensitive to the altitude where O 3 perturbations occur (Rap et al., 2015).This study assesses annual/decadal and monthly trends, together with possible evolution in seasonal cycles, of Arctic tropospheric O 3 over the last 20-30 years.Observed changes are also compared to results from atmospheric chemistry-climate models run as part of the recent Arctic Monitoring and Assessment Programme (AMAP) SLCF assessment (AMAP, 2021;Whaley et al., 2022;von Salzen et al., 2022), taking into account reported model deficiencies (Whaley et al., 2023).Results are discussed in light of possible changes in sources and sinks of Arctic tropospheric O 3 .

Measurements
The location of surface and ozonesonde sites used in this study are displayed in Figure 1, together with the Arctic Circle at 66.6°N, used to define the Arctic.

Trend Analysis
Observed monthly and annual/decadal trends in surface O 3 concentrations at different sites are determined using a non-parametric Mann-Kendall method based on the 90th and 95th confidence limits (CLs) and Sen's slope methodology (Sen, 1968;Theil, 1950) 1 and Tables S2-S5 in Supporting Information S1.

Modeled Trends
Modeled trends at the surface and different altitudes are calculated for 1995-2015 using results from four global chemistry-climate models (CMAM, GISS-E2.1,MRI-ESM2, UKESM1) and two chemistry-transport models (DEHM, EMEP MSC-W) run using the same ECLIPSEv6b anthropogenic emissions, and nudged with meteorological reanalyses as part of AMAP ( 2021).Details can be found in Whaley et al. (2022), Text S3 and Table S1 in Supporting Information S1.Simulated monthly mean O 3 volume mixing ratios from the model grid box containing the measurement location are used to compute multi-model medians (MMMs).For ozonesonde comparisons, modeled vertical profiles are interpolated onto the same vertical bins as the measurements before trends are computed.

Observed Ozone Trends
Annual and decadal trends are calculated for 1993-2019, or for the longest period with sufficient data, for all the sites (see Figure 1, Table S2 in Supporting Information S1).
First considering high Arctic sites at coastal locations that exhibit a winter maximum with low spring concentrations attributed to ODEs, as discussed in Whaley et al. (2023).Alert has positive O 3 annual trends (p = 0.044), as does Villum (p = 0.034) for the shorter time period 1999-2019, while annual trends at Utqiaġvik are not apparent (see Figure 1   Continental northern Scandinavian sites exhibit a different behavior with Tustervatn (p = 0.003), and Pallas, with lower certainty (p = 0.067), showing negative annual trends but no clear annual (or monthly) trends at Esrange (p > 0.151) over any of the periods considered.The shape of the seasonal cycle for the earlier versus the later period is similar at these sites, which also have a spring maximum like Zeppelin.O 3 appears to be decreasing throughout the year when comparing earlier and later periods although negative trends are only evident at Pallas (March, December), and at Tustervatn in spring and early summer for 1999-2019 trends (Figure S4 in Supporting Information S1).Summit is more representative of the FT and samples air masses transported from North America and Asia, or of stratospheric origin (Dibb, 2007;Schmeisser et al., 2018).No clear annual trend, calculated over the shorter 2001-2019 record, is seen, but negative monthly trends are estimated for January, March-May and September (p ≤ 0.060).S4 of Supporting Information S1.Observed trends are more frequently diagnosed over 1993-2019 (Figure 2) than over the shorter period ending in 2015 (Figure 3).While the MMMs simulate O 3 seasonal cycles reasonably well, low O 3 concentrations are missed in spring, and wintertime O 3 is underestimated (Whaley et al., 2023).The MMMs simulate positive trends at Zeppelin (Jan., p = 0.048) and negative trends at Esrange (May, p = 0.017), respectively, but not observed positive trends at Utqiaġvik (April, p = 0.035).Trends are simulated, but not observed, at Alert (January, December, p = 0.058, 0.014), Zeppelin (April, p = 0.032), Villum (Sept., p = 0.035), and Tustervatn (March, p = 0.057).

Observed Vertical Trends
This analysis focuses on O 3 changes in the lower and mid-troposphere.Figure 4 shows observed relative trends at six Arctic ozonesonde sites from 925 to 400 hPa for 1993-2019 (see p-values in Table S5 of Supporting Information S1).Absolute trends above and below 400 hPa, and relative trends from 925 to 100 hPa, providing information on changes in the upper troposphere (UT) and lower stratosphere (LS), are also calculated (Figures S7a and S7b in Supporting Information S1).Overall, while there are few high confidence trends, there seems to be a "dipole effect" with positive trends in winter and summer, and negative trends in spring and autumn.Positive winter (notably Jan.) trends are found up to 400 hPa at most sites (except Resolute and Sodankyla), and at Scoresbysund in early spring.Positive wintertime trends are more evident in the earlier period in the UTLS (Figure S8 in Supporting Information S1).Eureka, Resolute, and Sodankyla have periods with negative trends especially during spring and early summer in the lower troposphere (LT).Resolute decreases extend up to 500 hPa in March-April.Relative trends vary from −1.5% to +0.5-1.0%per year (Figure 4 and Figure S7b in Supporting Information S1) while stronger negative trends are diagnosed in later years (1999-2019) compared to 1993-2013 at all sites (Figure S8 in Supporting Information S1).

Comparison of Observed and Modeled Vertical Trends
Figure 5 shows observed ozonesonde and MMM trends for 1995-2015 up to 400 hPa (see Figure S9 in Supporting Information S1 for results up to 100 hPa).Only results from five models are used, since EMEP MSC-W only provided surface O 3 .The MMMs appear to capture the observed "dipole effect" seen in the observed trends.Models also capture observed increases in the winter but trends are overestimated at most sites, especially at Ny Ålesund.Negative winter trends at Resolute are not simulated.This may be linked to positive modeled winter trends above 500 hPa at all sites (see also Figure S9 in Supporting Information S1).Summertime positive MMM trends are larger than observed trends at some sites, for example, Resolute and Ny Ålesund.

Discussion and Conclusions
Increasing annual surface O 3 trends at Arctic coastal sites, and at Zeppelin, are in qualitative agreement with Cooper et al. (2020), but in contrast to negative or non-significant surface trends at Canadian sites (Tarasick et al., 2016).A notable finding is that positive trends occur mainly in the winter months.While such increases were reported previously at Utqiaġvik (A.Christiansen et al., 2022;Cooper et al., 2020) and Alert (Sharma et al., 2019), we confirm this tendency over the wider Arctic.Emission reductions of NO x in Europe and North America, and more recently over eastern Asia, have led to increasing wintertime O 3 at mid-latitudes due to less nitrogen oxide titration of O 3 (Bowman et al., 2022;Jhun et al., 2015;T. Wang et al., 2022).This can explain observed increases in wintertime surface Arctic O 3 , influenced primarily by transport of air masses from Europe (Hirdman et al., 2010).Evidence for declining O 3 precursor trends is supported by decreases in observed CO in the Arctic during autumn and winter (Figure S10 in Supporting Information S1).At the same time, CH 4 continues to increase globally contributing to rising O 3 in the NH (Zeng et al., 2022) (see also Text S4 in Supporting Information S1 on Arctic O 3 precursor trends).Another intriguing finding is springtime surface O 3 increases at Utqiaġvik (especially over 1999-2019, Figure S4 in Supporting Information S1), but no discernible trends at Alert and Villum.Changes in O 3 concentrations at this time of year may be driven by changes in ODE frequency linked to climate change or weather patterns (Oltmans et al., 2012).ODEs lead to zero or very low springtime O 3 due to bromine released from frost flowers or blowing snow (on sea-ice) (Simpson et al., 2007;Yang et al., 2008Yang et al., , 2010) ) or iodine compounds with a possible oceanic source (Benavent et al., 2022).Increasing prevalence of first year sea-ice leading to increasing sea-spray aero sols from blowing snow (Confer et al., 2023) may explain increases in springtime tropospheric bromine oxide, observed from satellites, along the north coast of Greenland and central Arctic Ocean (Bougoudis et al., 2020).Indeed, the frequency of low springtime O 3 concentrations has been increasing at Canadian high Arctic sites (see Figure S11 in Supporting Information S1) but no clear springtime monthly trends are determined at Alert or Villum in our analysis.Springtime increases at Utqiaġvik could be due to stronger transport from mid-latitudes to this site during periods with a more northerly extension of the Pacific storm track, hampering conditions for ODEs (Koo et al., 2012).They could also be due to an increasing influence from local emissions, such as shipping or Alaskan petroleum extraction, when photochemistry becomes active in spring (Gunsch et al., 2017).
Decreases in springtime/early summer O 3 in northern Scandinavia, especially over the later 1999-2019 period, are consistent with negative trends reported at Tustervatn since 1995 (Cooper et al., 2020), and in northern Sweden during summer (Andersson et al., 2017).These decreases are associated with lower maximum O 3 concentrations linked to reductions in European precursor emissions leading to less photochemical O 3 production (Cooper et al., 2020) although no clear trends in observed springtime CO are found at Zeppelin (Figure S10 in Supporting Information S1).Springtime negative trends at Summit may also be due to emission reductions over North America.Our results do not suggest a shift in the O 3 seasonal cycle toward higher concentrations in the spring (i.e., moving back toward pre-industrial O 3 seasonality) as reported at NH mid-latitudes (Bowman et al., 2022).
Another explanation for decreasing springtime O 3 at the surface could be that reductions in snow cover due to climate warming (Mudryk et al., 2020) are leading to more O 3 dry deposition to Scandinavian forests.
The observed and modeled surface trend comparison covers 1995-2015, thereby missing the later time period when stronger observed O 3 trends are found, especially positive trends in winter.MMMs capture wintertime O 3 increases at Zeppelin, but overestimate at Alert although they simulate decreasing surface winter CO at Alert and Utqiaġvik (Figure S10 in Supporting Information S1).This suggests that while anthropogenic emission changes may be captured, other model deficiencies may be contributing, such as modeling shallow boundary layers, O 3 deposition or NO x lifetimes, as noted by Whaley et al. (2023).The MMMs miss springtime increases at Utqiaġvik.This could be due to incorrect simulation of transport patterns (Oltmans et al., 2012) or missing surface halogen chemistry leading to incorrect modeled seasonality (Whaley et al., 2023).Negative springtime (May) Scandinavian trends are not always reproduced, possibly reflecting issues in the emission trends or modeled dry deposition.
Positive FT O 3 trends in winter at most Arctic sites are found, in common with several coastal Arctic surface sites, and in-line with increases reported at NH mid-latitudes (Gaudel et al., 2018), and at Canadian ozonesonde sites, except Resolute (H.Wang et al., 2022).Patterns in observed trends are quite well captured by the MMMs over 1995-2015, notably positive trends in winter and summer, although they tend to be overestimated.Positive summer trends may be linked to increased photochemical production from increased lightning and boreal fires due to climate warming (Veraverbeke et al., 2017).Observed negative trends in spring, extending from near the surface into the FT, are generally reproduced, and are likely to be due to decreasing NO x emissions leading to lower FT O 3 where photochemical production is NO x -limited.Negative LT trends could also be due to increasing ODEs extending over 100 kms and up to 1.5 km (Yang et al., 2020;Zilker et al., 2023).Overestimation of winter trends contrasts to previous studies where models underestimated NH trends (A.Christiansen et al., 2022;H. Wang et al., 2022).This may be due to differences in model transport or O 3 precursor emission trends, including NO x reductions (see also Text S4 in Supporting Information S1).AMAP models overestimate mid-latitude FT O 3 (Whaley et al., 2023), possibly suggesting a larger sensitivity to precursor emission changes.
Observed trends in the UT (LS) appear to have switched from positive to negative since 1993 in winter/spring, which may explain stronger positive FT trends in the earlier part of the record .More frequent positive phases of the Arctic Oscillation in recent years may be contributing with a weaker Brewer-Dobson circulation leading to less transport of stratospheric O 3 into the Arctic UTLS, a higher tropopause height, and thus lower O 3 concentrations in this region (Zhang et al., 2017) Overall, this study identifies trends with high-medium certainty in observed Arctic tropospheric O 3 that are generally quite well captured by MMM results, Further investigation into the causes of observed trends, and model performance, are needed taking into account uncertainties in the observations and models (Fiore et al., 2022;Young et al., 2018), including known model issues (Whaley et al., 2023).

Figure 1 .
Figure 1.Left: Location of surface (bold) and ozonesonde (italic) sites and showing the Arctic Circle (66.55°N).Right: O 3 trends at surface sites in ppbv per decade and p-values.Trends (>90% confidence limit, p ≤ 0.1) are in bold.Geographical coordinates for all sites are provided in Whaley et al. (2023).See text for details.

Figure 2 .
Figure 2. Observed surface O 3 trends and seasonal cycles.Left: seasonal cycles of monthly median O 3 (ppbv) at (a) Alert, (b) Utqiaġvik, (c) Villum, (d) Zeppelin, and (e) Pallas for 1993-2000 (red lines) versus 2012-2019 (blue lines).Shaded areas show upper and lower quartiles of hourly values.Right: monthly trends for 1993-2019.Boxes represent the slope of the trend in ppbv per year with red boxes having 95th% confidence limit (CL), blue boxes 90th% CL, and black boxes are trends with low certainty.Error bars show 95th% CLs.Results are shown for shorter periods depending on data availability.

Figure 3
Figure 3 compares observed monthly and MMM trends for 1995-2015, or the closest possible time interval in case of years with missing observations.Results for other sites are shown in Figure S6 of Supporting Information S1 and p-values in TableS4of Supporting Information S1.Observed trends are more frequently diagnosed over 1993-2019 (Figure2) than over the shorter period ending in 2015 (Figure3).While the MMMs simulate O 3 seasonal cycles reasonably well, low O 3 concentrations are missed in spring, and wintertime O 3 is underestimated(Whaley et al., 2023).The MMMs simulate positive trends at Zeppelin (Jan., p = 0.048) and negative trends at Esrange (May, p = 0.017), respectively, but not observed positive trends at Utqiaġvik (April, p = 0.035).Trends are simulated, but not observed, at Alert (January, December, p = 0.058, 0.014), Zeppelin (April, p = 0.032), Villum (Sept., p = 0.035), and Tustervatn (March, p = 0.057).

Figure 3 .
Figure 3.Comparison of observed (left) and multi-model median (right) surface O 3 trends and seasonal cycles at (a) Alert, (b) Utqiaġvik, (c) Villum, (d) Zeppelin, and (e) Esrange.Upper panels: seasonal cycles for 1995-2004 (red lines) versus 2005-2015 (blue lines).Shaded areas show upper and lower quartiles of monthly values (observations only).Lower panels: monthly median trends in ppbv per year for 1995-2015, or shorter periods depending on data availability.Box coloring and error bars same as Figure 2.
. However, Liu et al. (2020) did not detect any trend in the stratospheric O 3 flux into the Arctic UT.On the other hand, H. Wang et al. (2022) attributed FT increases in NH mid-high latitude O 3 to increases in aircraft NO x emissions.