Ozone concentrations are still increasing in many locations in the Northern Hemisphere mostly due to an increase in Asian precursor emissions . Since tropospheric ozone is an effective greenhouse gas with a relatively long lifetime, its main impact on climate and air quality is within mid-latitude regions. Several studies have shown that ozone also makes an important contribution to the Arctic surface temperature increases due to direct local warming in the Arctic as well as heat transport following warming due to ozone at mid-latitudes . The Arctic ozone budget still requires better quantification and is complicated by the interplay between the long-range transport, including the downward transport of stratospheric ozone , removal of boundary layer O3 due to halogen chemistry, especially in springtime , and photochemical production due to local sources, such as boreal forest fires , NOx enhancement from snowpack emissions , local summertime production from peroxyacetyl nitrate (PAN) decomposition , or ship emissions . The O3 distribution over North America for the spring period at high latitude has been discussed using the Tropospheric Ozone Production about the Spring Equinox (TOPSE) and Arctic Research of the Composition of the Troposphere from Aircraft and Satellites (ARCTAS) data set in several publications showing (i) frequent occurrence of ozone depletion events (ODE) in the planetary boundary layer (PBL), (ii) a net O3 photochemical production rate equal to zero throughout most of the troposphere, and (iii) a latitudinal increase of tropospheric O3 concentrations due to transport from mid-latitudes and stratosphere–troposphere exchange (STE).

For the summer period ozone photochemical production is expected according to the numerous studies conducted at mid-latitudes , but little attention has been given to the high latitude distribution during this season. During the ARCTAS-B and POLARCAT (Polar Study using Aircraft, Remote Sensing, Surface Measurements, and Models of Climate Chemistry, Aerosols, and Transport) campaigns , many ozone measurements were carried out over Canada and Greenland from 15 June to 15 July 2008 via aircraft and regular ozone soundings. This allows a detailed analysis of the ozone regional distribution at high latitudes between 55 and 90∘ N and a discussion about the relevant ozone sources driving the summer ozone values. So far, knowledge of the relative influence of the main ozone summer sources has been mainly derived from modeling studies, e.g., summer simulations of global model simulation of the ozone source attribution , or regional modeling of biomass burning case studies for North American fires or Asian fires .

In order to interpret the measurements, we use a hemispheric simulation performed using the regional chemical transport model, WRF-Chem . WRF-Chem simultaneously produces a meteorological forecast, including the dynamics of the upper troposphere and lower stratosphere (UTLS) region, and online chemistry to predict ozone (and other trace gas and aerosol) concentrations. Here, we use the WRF-Chem model to investigate dynamics that determine ozone concentrations as a function of latitude, including STE processes, which can bring high-ozone air from the stratosphere into the upper troposphere in the Arctic. In addition, we use the model to compare directly predicted and measured ozone in summer 2008 and use CO as a tracer to separate air influenced by surface emissions and subsequent ozone formation via photochemistry from air influenced by stratosphere–troposphere mixing processes.

The objectives of this paper are thus twofold: (i) to establish the summer tropospheric ozone latitudinal variability over Greenland and Canada based on the large number of ozone measurements available in the free troposphere during June-July 2008, and (ii) to explore the role of photochemistry and STE in the observed high latitude ozone distribution by extracting the potential vorticity (PV) and CO distribution from a 2 month WRF-Chem regional model simulation. The WRF-Chem simulation also provides the modeled ozone distribution to verify the coherence between the modeled PV and CO with the observed ozone distributions. The ozone data set and the WRF-Chem simulation are described in Sects.  and , respectively. The model vs. measurement ozone comparison is discussed in Sect. , while the latitudinal distributions of ozone, CO, and PV are presented in Sect. .

We use three different ozone measuring instruments to build the data set considered in this work: airborne lidar, in situ aircraft ozone analyzer, and electrochemical concentration cell (ECC) ozonesonde. Two airborne ozone differential absorption lidars (DIAL) are considered: the NASA-DC-8 instrument and the ALTO lidar on the French ATR-42. In situ ozone analyzers have been installed onboard three different aircraft: the NASA DC-8, the DLR Falcon-20, and the French ATR-42. We also considered seven ground-based stations that participated in the ARC-IONS campaign over Canada (latitudes > 50∘ N in the longitude range between -70 and -160∘ W) and over Greenland (latitudes > 55∘ N in the longitude range between -60 and -20∘ W). The times and positions of the June-July 2008 measurements included in our study are given in Table  for the aircraft observations and in Table  for the ozonesondes. The aircraft flight paths over Canada and Greenland are shown in Fig.  and the measurements are always performed during daytime.

The German DLR Falcon-20 was based in Greenland and used an UV absorption instrument (Thermo Environment Instruments,Inc.; TEI49C) to measure O3 with an uncertainty of ±2 ppbv (±5 % of the signal) . The ATR-42 aircraft was also based in Greenland and the ozone measurements were made using a similar instrument (TEI49-103) calibrated against a NIST (National Institute of Standards and Technology)-referenced O3 calibrator, Model 49PS, at zero, 250, 500, and 750 ppbv . A 4 ppbv negative bias related to O3 loss in the ATR-42 air inlet (e.g., 10 % for 40 ppbv and 5 % for 80 ppbv) has been corrected in this study. On 14 July 2008, the comparison of the ATR-42 and the DLR Falcon-20 ozone data at two altitude levels over Greenland (near 67∘ N) shows an uncertainty better than 2 ppbv (Fig. ). The NASA DC-8 O3 measurements were made with a chemiluminescence technique, with the instrument calibrated by additions of O3 determined by UV optical absorption at 254 nm. Uncertainties in the DC-8 O3 data are typically ±2 ppbv (±5 % of the signal). The NASA DC-8 flew mainly over Canadian forest fire regions in summer . The comparison between the NASA DC-8 and the DLR Falcon-20 O3 data during an aircraft intercomparison flight on 9 July over northern Greenland shows very good precision, but a 4 ppbv positive difference (Fig. ). None of the Falcon-20 and NASA DC-8 data were corrected.

The ALTO lidar was mounted in a zenith viewing mode making ozone vertical profiles above the aircraft, thus limiting the number of O3 data available at altitudes below 3 km. The lidar measurement altitude range is of the order of 6 km above the aircraft flight level with a 300 m vertical resolution and a 10 km horizontal resolution (i.e., a 2 min integration time). The system is fully described in , and the instrument performances for different examples of daytime airborne measurements are discussed in (urban pollution in the boundary layer, ozone in the UTLS, and long-range transport in the free troposphere). Several comparisons with in situ measurements (ECC ozonesonde and airborne UV photometer) show no specific biases in clear air measurements. Measurements taken near clouds or thick aerosol layers are not included here since corrections of systematic errors related to aerosol interference become very large . These represent 20 % of the lidar profiles recorded during the campaign.

The NASA-DC-8 Ozone DIAL system and configuration implemented during the campaign is described by . The instrument provides simultaneous zenith and nadir profiles to cover the troposphere and lower stratosphere. The O3 measurement resolution is 300 m in the vertical and approximately 70 km (3 min) in the horizontal for daytime measurements. On all field experiments, the airborne DIAL O3 measurements are compared with in situ O3 measurements made on the DC-8 during ascents, descents, and spirals, while comparisons to ozonesondes during coincident overflights were also conducted. In the troposphere, the DIAL O3 measurements have been shown to be accurate to better than 10 % or 2 ppb, whichever is larger . More recently, the precision of the DIAL O3 measurements during the high-latitude SAGE-III Ozone Loss and Validation Experiment in 2003 (SOLVE II) was found to be better than 5 % from near the surface to about 24 km and the accuracy was found to be better than 10 % in comparison with the Ny-Ålesund lidar, ozonesondes, and in situ DC-8 measurements . DIAL and Microwave Limb Sounder (MLS) O3 measurements from the Polar Aura Validation Experiment (PAVE) in 2005 were found to agree within 7 % across the 12–24 km altitude range . Comparisons between DIAL and MLS were also examined in the upper troposphere and lower stratosphere (215–100 hPa region) from data obtained during the International Intercontinental Chemical Transport Experiment (INTEX-B) field experiment, and these results show good agreement in the lower stratosphere, with decreasing performance of the MLS measurements into the troposphere .

The ozonesonde monitoring was intensified in 2008 over North America in the framework of the ARCIONS initiative and the characteristics of the ozone measurements are fully described in . Nearly daily soundings were made during the aircraft flight period at four stations over Canada between 53 and 63∘ N, and at Summit, Greenland, while weekly soundings were made at the high latitude stations Alert and Resolute.

The latitudinal cross sections of the ozone data set selected in this study are shown in Fig.  considering two domains to produce a bidimensional latitude–altitude plot over Canada (-160 to -70∘ W) and Greenland (-70 to -20∘ W). The horizontal distributions of the data set used for producing the latitude–altitude plots are shown in Fig. . For latitudes higher than 80∘ N, the data comes from a limited number of observations: eight sondes from Alert and three DC-8 flights from 8 to 10 July. The locations of the ozonesonde stations are shown as red vertical bars in the latitudinal cross sections. In the troposphere at altitudes less than 8 km over Canada, lidar, in situ, and ozonesonde measurements contribute 67, 20, and 13 % of the ozone data set, respectively, while they correspond to 26, 69, and 5 %, respectively, over Greenland (see the number of observations using each technique in Table ). The aircraft data (lidar and in situ) therefore strongly contribute to the high measurement density even though they correspond to a limited number of flying days, i.e., 18 days from 26 June to 18 July. In this work, ozone data are averaged hourly when they are in the same cell of a grid, with 0.5×0.5∘ and 1 km vertical resolution in order to avoid an oversampling of similar air masses. Only hourly averages are considered in Table .

For both regions similar ozone vertical distributions were observed: low O3 mixing ratio ≤ 40 ppbv below 3 km and high O3 concentrations typical of the UTLS in the altitude range 8–11 km between 60 and 80∘ N (i.e., red and dark region with O3 mixing ratio > 150 ppbv). The average O3 mixing ratio in the altitude range 4–8 km is of the order of 65 ppbv for both regions but the latitudinal O3 gradients are more visible over Greenland than over Canada. Two mid-tropospheric ozone branches are seen in the latitude band 65–73 and 78–85∘ N over Greenland, the former being tilted to the south at 70∘ N and the latter to the north at 80∘ N.

The WRF-Chem ozone mixing ratios corresponding to the times and locations of the June–July observations have been used to produce latitudinal cross sections comparable to the results shown in Fig. . The modeled O3 latitudinal cross sections are shown for Canada and for Greenland in Fig. , while the corresponding relative differences between the model results and the measurements are shown in Fig. . The altitude and latitude ozone variability of the model is comparable to the observations over both regions. The vertical structure, i.e., the transition between the low ozone values below 3 km and the higher ozone concentration in the free troposphere, and the UTLS ozone variability, are well reproduced. The latitudinal gradient of the ozone concentration in the 4–8 km altitude range over Greenland is also visible in the simulation results where the two mid-tropospheric ozone branches are also visible at 70 and 80∘ N. This structure is slightly shifted in the modeled cross section explaining both positive and relative differences in the free troposphere over Greenland (Fig. ). The agreement is less good for the latitudinal gradient over Canada where the low ozone values (< 50 ppbv) seen in the model at 65∘ N in the mid-troposphere are not seen in the observations, although the signs of the latitudinal gradients seem correct.

To quantify the observation–model agreement the scatter plot of modeled vs. measured ozone is also presented in Fig.  using a PV color scale to distinguish the tropospheric and stratospheric contributions. The measured vs. modeled O3 correlation (red line in Fig. ) is of the order of 0.9 in the altitude range 0–15 km over both regions because the occurrence of stratospheric ozone intrusions are very well reproduced by WRF-Chem. The correlation is however between 0.5 and 0.7 in the troposphere only using observations with PV values less than 1 PVu (green line in Fig. ). Considering that the ozone variability in the free troposphere is not very large (< 50 ppb), the spatial and temporal variability is still well reproduced by WRF-Chem even below the tropopause region. Even though the UTLS temporal variability is well reproduced by the model (Fig. ), there is a significant underestimation of ozone by a factor 1.5 in the WRF-Chem simulation for the lowermost stratosphere (PV > 2 PVu) (Figs.  and ). In the troposphere there is also a negative bias of the model data of the order of -6 to -15 ppbv with the largest differences over Canada (see Table ). A fraction of this tropospheric underestimate by WRF-Chem is likely related to the O3 underestimate in the lowermost stratosphere, causing the stratospheric ozone source in the upper troposphere to be too small. The modeled O3 underestimate in the lower stratosphere likely originates from the O3 climatology used to initialize the model in the stratosphere. The other part of this 10-ppbv bias is due to an underestimate of the lightning NOx contribution in the WRF-Chem simulation and/or an underestimate of the vertical transport of continental emissions to the mid- and upper troposphere. have shown that the lightning NOx source and vertical transport of continental emissions both contribute to 15 % of the O3 concentration in the free troposphere at latitudes higher than 60∘ N in summer. Despite these flaws, the WRF-Chem simulations are still very valuable because the spatial O3 variability compares rather well with the observations. The PV and CO distribution available from the WRF-Chem simulations will therefore be used to examine the respective roles of photochemistry and STE in the observed ozone distribution in the following sections.

In this section, we investigate the mean latitudinal and vertical ozone gradient in the troposphere using the POLARCAT observations. To quantify these gradients, the O3 cross sections plotted in Fig.  have been divided into several regions using the model CO latitudinal cross sections. The vertical boundaries are defined according to the mean ozone vertical profile in the troposphere, i.e., the depth of the low ozone values layer in the lower troposphere and the downward extent of the tropopause region. The variability of the CO distribution is only taken from the model, since CO data is not available for many of the ozone observations (ozonesonde and lidar profiles).

The purpose of this work is to provide a complete picture of Arctic ozone using measurements available during the summer 2008 POLARCAT campaigns over Canada and Greenland, where three aircraft were deployed and seven ozonesonde stations intensified their ozone vertical profiling. This is the first case of such complete temporal and geographical coverage. We take advantage of the large number of airborne lidar profiles (representing 67 % of the O3 measurements over Canada and 26 % over Greenland). The measured ozone climatology established in this paper can also be used for future model evaluation at high latitudes. For example, in our work, while good correspondence of the measured O3 vertical and latitudinal distribution is found with model results from WRF-Chem, a negative O3 bias of -6 to -15 ppbv between the model and the observations is found in the free troposphere over 4 km, especially over Canada. This deficiency is partly related to the WRF-Chem model stratospheric ozone initialization.

The WRF-Chem model simulation is also used to discuss the relative influence of tropospheric ozone sources at high latitude in summer. Ozone average concentrations are of the order of 65 ppbv at altitudes above 4 km both over Canada and Greenland, while they are less than 50 ppbv in the lower troposphere. For Canada, the analysis of the modeled CO distribution and the weak correlation (< 30 %) of O3 and PV suggest that stratosphere–troposphere exchange (STE) is not the major contribution to tropospheric ozone at latitudes less than 70∘ N, where transport of North American biomass burning (BB) emissions took place during the 2008 summer. Conversely, significant STE is found over Greenland according to the better O3 vs. PV correlation (> 40 %) and the higher value of the 75th PV percentile. This is related to the persistence of cyclonic activity over Baffin Bay during the summer. A weak negative latitudinal summer ozone gradient of -6 to -8 ppbv is found over Canada in the mid-troposphere between 4 and 8 km because the O3 photochemical production from BB emissions mainly takes place at latitudes less than 65∘ N, while STE plays a larger role at latitudes higher than 70∘ N. A positive ozone latitudinal gradient of 12 ppbv is observed in the same altitude range over Greenland not because of an increasing latitudinal influence of STE, but because of different long-range transport from multiple mid-latitude sources (North America, Europe, and even Asia for latitudes higher than 77∘ N).

For the Arctic latitudes (> 80∘ N), free tropospheric O3 concentrations are related to a mixture of stratospheric O3 transport across the tropopause and Asian pollution, as already suggested by using a case study of aircraft observations in the Arctic. Our study, which uses more ozone measurements, leads to the same conclusions.

The CNRS ATR-42 aircraft ozone lidar observations can be downloaded from ftp://polarcat@ftp.aero.jussieu.fr/Kanger/LidarO3/ and the the metadata file is called “ReadmeO3lidardata”. The CNRS ATR-42 ozone in-situ measurements from the MOZART instrument are deposited on ftp://ftp.aero.jussieu.fr/Kanger/Data_ATR/ and the meta-data file is called “Readme_MOZART.txt”. Requested password is “kanger” to access the CNRS data. The data from the NASA DC-8 aircraft have been downloaded from the NASA archive website, using a direct Link to the DIAL data archive (http://www-air.larc.nasa.gov/cgi-bin/ArcView/arctas#HAIR.JOHN/, ) and to the ozone in-situ measurements (http://www-air.larc.nasa.gov/cgi-bin/ArcView/arctas#WEINHEIMER.ANDREW/, ). No passwords are required as this is open and public access. The POLARCAT ozone data for DLR Falcon-20 is available from the HALO/Falcon database (https://halo-db.pa.op.dlr.de/). ECC ozonesonde data have been downloaded from the ARCIONS web site ftp://es-ee.tor.ec.gc.ca/pub/ftpdt/ARC-IONS Data/summer/ICARTT/ and the WOUDC web site (http://woudc.org/data/explore.php?lang=en). Data files from the WRF model simulation are too big to be publicly available but could be obtained from LATMOS. Air mass trajectories are calculated with ECMWF meteorological analysis downloaded from ECMWF MARS data base which is not publicly available.