In the spring season, the main snowmobile route from Longyearbyen to the east coast of Spitzbergen goes along the road through the Adventdalen valley, and therefore there is daily snowmobile traffic nearby the CO2 laboratory belonging to the University Centre in Svalbard (UNIS CO2 lab, coordinates: 78.20247∘ N, 15.82887∘ E). The station is located at a distance of approximately 5 km to the south-east of the coal power plant and major crossroads in Longyearbyen and thus is representative of monitoring of air pollution from snowmobiles. The chemiluminescence NO/NO2/NOx analyser (model T200) was installed at this laboratory for the period from 23 March to 15 May 2017. The inlet of the sampling hose was secured outside from the window, while the temperature inside the laboratory was kept constant with the help of a thermostat to maintain stable conditions needed for correct functioning of the analyser. The sensor was calibrated weekly using a zero-air generator and a certified NO gas with known concentration (800 ppb), and the hourly average NOx data were scaled linearly to eliminate zero drift. The automatic weather station (UNIS AWS) is located nearby the UNIS CO2 lab, and the data from that station have been used to assess local meteorological conditions.

In addition to the meteorological parameters from the Adventdalen station, data from UV monitors installed at the UNIS roof in Longyearbyen have been used. The sensors SKU 420 UV-A (315–380 nm) and SKU 430 UV-B (280–315 nm), produced by the SKYE Instruments, were calibrated on 24 August 2016.

Continuous NOx measurements are performed by the Norwegian Institute for Air Research (NILU) in the framework of the air quality monitoring programme in Ny-Ålesund . The analyser is installed in the middle of the settlement (coordinates: 78.92470∘ N, 11.92641∘ E), 100 m to the north-west of the meteorological station operated by the Norwegian Meteorological Institute and 300 m to the south-south-east of the diesel power plant. Similarly to the measurements in Adventdalen, weekly calibrations with zero air and span gas are performed at the Ny-Ålesund station, and the hourly average NOx data are corrected for drift. The hourly O3 gas monitor data from the Zeppelin observatory, located nearby the mountain top (474 m a.s.l.), 2 km to the southwest of Ny-Ålesund (coordinates: 78.90719∘ N, 11.88606∘ E), were used for comparison with the O3 measurements in Barentsburg. The UV data obtained using the multifilter radiometer GUV 541 at the Sverdrup station in Ny-Ålesund and local meteorological observations from the Zeppelin station were provided by NILU as well. The GUV radiometer is checked and corrected against a travelling reference instrument every year.

In addition to this, daily radiosonde and weekly ozonesonde data from the French–German AWIPEV research station in Ny-Ålesund have been used. Since temperature inversion may be an important factor promoting accumulation of local pollution in the atmospheric boundary layer, the method for its detection in the radiosonde vertical profiles, described by , has been applied: the days when the temperature was increasing with height by more than 0.3 ∘C in the lowest 500 m were defined as days with temperature inversions. In order to compare the O3 sonde measurements with ground-level observations, the O3 mixing ratio in units of parts per billion per volume (ppbv) were calculated from the O3 partial pressure and atmospheric pressure measured by the radiosonde .

Daily radiosonde launches were operated at the AWIPEV station (AWI), using Vaisala RS92 radiosondes until April 2017 and Vaisala RS41 radiosondes afterwards. In this study, we apply radiosonde data for March 2017, post-processed according to the principles of the GCOS Reference Upper-Air Network (GRUAN; ). The RS41 data for April–May 2017 are processed with the manufacturer's software.

The Russian Arctic and Antarctic Research Institute (AARI) performs the measurements in Barentsburg independently in the frame of the air quality monitoring programme. The equipment installed in the settlement includes chemiluminescence NOx AC32M and UV photometric O3 O342M analysers produced by Environnement S.A. and the Vaisala HydroMet system MAWS201. The observational site is located on the narrow terrace 40 m above sea level (coordinates: 78.06070∘ N, 14.21718∘ E) and 600 m to the north-east of the coal power plant. The analysers continuously gather the data and transmit them to the laboratory facility of the Russian Scientific Centre in Barentsburg. The data with 20 min time resolution were averaged to obtain hourly data. The analysers were installed and initially calibrated by the manufacturer's accredited specialists in December 2016. After the installation, zero control was performed regularly using the NOx analyser's built-in automatic zero-air function for the correction of zero drift lines. The maintenance of the converter filter was done at the frequency recommended by the manufacturer. However, in contrast to the equipment in Ny-Ålesund and Longyearbyen, the NOx and O3 analysers in Barentsburg were not calibrated manually during the field campaign. Therefore, the data from this station may be prone to drift. This is especially important to keep in mind when studying NOx concentrations, since the NOx values are usually very low in the remote Arctic environment . On the other hand, the UV O3 monitor is more stable and does not demand as frequent calibration as chemiluminescence instruments , and thus data from this instrument are more reliable.

Previous studies showed the influence of large-scale weather phenomena on long-range-transported and local air pollution. demonstrated how positive and negative phases of the North Atlantic Oscillation control long-range transport of air pollution to the Arctic. The modelling study of introduced four weather regimes based on ERA40 data and described the aerosol budget variations associated with different regimes and feedback of aerosol distribution on the weather regime persistence. classified nine weather regimes using self-organizing maps and cluster analysis of principal components and investigated the influence of the prevailing large-scale meteorological conditions on the air quality in Berlin. In the current work, we apply Christian Grams' weather regime classification that is based on the 6-hourly ERA-Interim data. This classification was previously used to investigate the frequency of poleward moisture transport events by atmospheric rivers and southward transport of Arctic air during cold-air outbreaks and for assessment of Europe's wind power output . Thus, this approach is suitable for the study of long-range transport of air masses and local dispersion efficiency that depends on atmospheric stability and wind speed. The following three-step procedure has been implemented to assess the effect of the prevailing synoptic meteorological situation on long-range transport of pollutants and on the local meteorological conditions affecting concentrations of O3 and NOx in Svalbard.

Firstly, the whole measurement period was divided into sub-periods based on the prevailing large-scale atmospheric circulation patterns. In the climatological mean, the large-scale conditions are characterized by weak ridging of absolute geopotential height at 500 hPa over the eastern North Atlantic and westerly upper level flow over Svalbard. Such a regime is placed in the “no regime” category in the classification (their Fig. S1h). The deviations from these mean conditions are classified into seven distinct weather regimes that represent the variation of the large-scale circulation patterns over the North Atlantic and European region.

Secondly, the hourly meteorological data, O3 concentration and planetary boundary layer height (PBL) from the global ERA5 reanalysis data set with 31 km spatial resolution were used to investigate the prevailing large-scale weather conditions for the identified sub-periods. The ERA5 O3 mass mixing ratio is estimated based on the assimilated satellite observations and external 2-D photochemical model . The PBL in the ERA5 data set is defined by the height when the Richardson number for the adjacent vertical model layers exceeds a critical value of 0.25, and the air becomes stably stratified .

Thirdly, the FLEXible PARTicle (FLEXPART) V8.2 air parcel trajectory data set was utilized for the same sub-periods to study the long-range atmospheric transport to Svalbard. Previously, FLEXPART data have been used to investigate long-range transport of black carbon and sulfates and mercury and O3 in the Arctic. The current FLEXPART data set contains a 3-dimensional Lagrangian dispersion simulation with 5 million air parcels driven with the meteorological data from the ERA-Interim reanalysis with a spatial resolution of approximately 80 km and temporal resolution of 6 h . The 10 d backward trajectories starting within the lowest 500 m above the ground in the region covering Ny-Ålesund, Longyearbyen and Barentsburg (from 77.5 to 79.5∘ N and from 10 to 20∘ E) were extracted from this data set. The trajectory model data were combined with BrO total column data derived from GOME-2 (ir)radiance satellite observations to identify regions with elevated concentration of this O3-depleting substance.

Additionally, to study long-range transport of extremely O3-depleted or O3-enriched air masses, the following procedure has been implemented to detect O3 decrease and increase events occurring simultaneously in Barentsburg and at the Zeppelin station:

Since the distance between the Zeppelin observatory and Barentsburg is more than 100 km, a time lag in correlation between the data from the two stations is expected. The acceptable time lag has been defined based on the lagged linear correlation between the data sets. The maximum time lag for which the correlation coefficient is higher than or equal to the coefficient calculated for the zero-hour lag is defined as the maximum allowable time lag.

The backward air mass ensemble trajectories have been simulated using the Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) model for these joint events for 240 h back in time to identify the source regions of the air masses . This 10 d simulation time has been chosen as a compromise between the average lifetime of tropospheric O3, which may be 3 to 4 weeks , and the uncertainty of modelled air mass trajectories that increases with travelled distance . The standard ensemble simulation with 27 members was calculated in the READY system by offsetting the Global Data Assimilation System (GDAS) meteorological data with a 0.5∘ resolution by a fixed grid factor in the horizontal and vertical dimensions to take into account the uncertainty of the trajectory forecast .

Previously, the only collocated NOx and O3 measurements from Ny-Ålesund were performed at the Zeppelin station from February to May 1994. The results were published in . In that study, the combination of NOx data and concentration of particles with diameter below 10 nm, atmospheric stability and wind direction was used to identify possible local pollution events. In spring 1994, the local pollution was detected at the Zeppelin station during 6.4 % of the measurement time, and the number of these events increased with increased insolation in May. As there were no O3 data from Ny-Ålesund available, the O3, CO and differential mobility particle sizer (DMPS) data from the Zeppelin station were used to study the influence of local NOx emissions in Ny-Ålesund on the O3 concentration. CO indicates the presence of other pollutants emitted simultaneously in the process of fossil fuel burning, and although the correlation between NOx and CO concentration in the plumes depends on the engine and fuel type, age of the plume and environmental conditions , we expect elevated CO concentrations in the fresh plumes arriving at the Zeppelin station. Therefore, a local pollution effect was defined for O3 measurements at the Zeppelin station when all four conditions were fulfilled:

The wind direction measured both in Ny-Ålesund and at the Zeppelin station was northerly (above 270∘ or below 90∘) since the diesel power plant is located 300 m to the north–north-west of the NOx monitor in Ny-Ålesund and 2 km to the north–north-east of the Zeppelin station.

To assess how the NOx emissions in Barentsburg affect the local O3 concentration there, the NOx and O3 data from the Barentsburg station have been compared. Positive anomalies in O3 concentration were found for the same wind directions where increased NOx concentrations were observed, but this may be due to higher concentrations of O3 in air masses, which were transported to Svalbard from the south and south-west. Since there are multiple sources of local pollution in Barentsburg (coal power plant, ships and cars), another method has been implemented:

The hours when NOx concentrations were above average in Barentsburg have been defined.

The hourly NOx and O3 concentrations and meteorological data from all three stations are shown in Figs.  and . Nine distinct large-scale weather regimes have been identified for the campaign period (marked with numbers from I to IX). A detailed description of the weather regimes is given in Sect. 3.2 of the current paper.

Despite alteration in the large-scale circulation, little variability was found in CO data from the Zeppelin station (solid olive line in the Fig. c) and O3 values in ERA5 reanalysis (orange line in the Fig. c). The threshold of CO median value ± median absolute deviation (132.6±6.7 ppb) is shown by the dashed olive line in the Fig. c. There were no sharp peaks in the concentration of this gas, indicating that there was little influence of local pollution on the measurements at the mountain station, but the levels of this compound were stably high in the beginning of the campaign and showed a gradual decline from the end of March to the middle of May. This is a response to increased insolation throughout the fieldwork period (from 0.03 to 0.3 W m-2 and from 7.9 to 28.9 W m-2 UV-B and UV-A irradiation observed in Ny-Ålesund, respectively), as CO is rapidly oxidized by the OH radical produced in the O3 photolysis. A pronounced decline in CO concentration and sharp drop in O3 values in ERA5 reanalysis may be observed in sub-period VIII. The O3 concentration in the ERA5 reanalysis exceeds the values observed at the Zeppelin station (dark blue line in the Fig. c) most of the time except this sub-period. Thus, the O3 ERA5 reanalysis data are not representative of the regional Arctic processes or short-term variability in long-range transport but show strong sensitivity to photochemical destruction.

There is a weak statistically significant positive correlation between NO, NO2 and NOx values measured in Adventdalen and in Ny-Ålesund (the Pearson correlation coefficients are rNO = 0.13, rNO2 = 0.15 and rNOx = 0.13; p<0.0001 for all compounds). Conversely, no correlation is present with NOx data from Barentsburg. Low correlation between the NOx values at the three stations indicates the importance of local emission sources and micrometeorology (wind channelling and spatial variation in atmospheric stability) rather than synoptic meteorological conditions. The background NOx concentrations observed in Svalbard in previous studies using different measurement techniques are below 0.4 ppb, and thus, the natural variability in NOx values due to long-range transport to Svalbard would be undetected in the NOx data sets presented in the current study.

The Barentsburg O3 data contain some abrupt peaks with magnitude of up to 9 ppbv and duration of just 1 h (light blue line in Fig. c), while they are absent in the Zeppelin O3 data. Indeed, the Barentsburg station is located inside the settlement, and the O3 data there are more prone to be influenced by local NOx pollution, while the Zeppelin station is situated far from the local emission sources. In order to investigate the significance of this local impact, a 6 h moving average filter has been applied to the O3 Barentsburg data, and the results are shown with a dark blue line in Fig. (c). The smoothed and original O3 data from Barentsburg have been compared statistically: both the two-sided Wilcoxon rank sum (WRS) test and the t test show that the application of the low-pass filter on the O3 data from Barentsburg does not result in significant change in the concentration distribution. The correlation between O3 concentrations at the Zeppelin station and in Barentsburg is moderate (Pearson correlation coefficient r=0.69 both for smoothed and unsmoothed data; p<0.001). This indicates that O3 concentrations at both stations are highly influenced by the meteorological conditions on the synoptic scale, and local impacts are of minor importance.

We have applied methods described in Sect. 2.5 to define the effect of local NOx emissions in Ny-Ålesund and Barentsburg on the O3 concentrations in the settlements. As a result, 5 % of the O3 data from the Zeppelin station might have been influenced by the local pollution from Ny-Ålesund, and the statistically significant (p<0.0001) decrease in the O3 mean (31.6 vs. 36.1 ppb) and median (34.4 vs. 38.0 ppb) concentrations has been revealed for this group. However, northerly wind that may transport local pollution from Ny-Ålesund also brings air masses which have a lower O3 background value. Therefore, when one compares potentially locally polluted air masses with the background air masses coming from the north, the difference in mean and median O3 concentrations becomes statistically insignificant, 31.6 vs. 33.3 ppb and 35.6 vs. 34.4 ppb, respectively. Following the method of for the local pollution event detection, the concentration of particles with a diameter of 10 nm routinely measured by the DMPS at the Zeppelin station and a threshold of >95 percentile have been used to identify peaks in concentration of newly formed particles. Similarly to the results of , the peak events were detected at the Zeppelin station only in the second part of the 2017 measurement campaign (from 24 April to 13 May). The northerly wind direction was present only during 12 out of 45 h with peak particle concentration; however, none of these cases was characterized by increased CO concentration at the Zeppelin station. Thus, these peaks in concentration of small particles might have been related to natural rather than anthropogenic emission sources. Indeed, described the offset in new particle formation towards late spring and summer when biological emissions become important sources for this process. Therefore, both statistical comparison of the O3 concentrations in clean and potentially polluted air masses mentioned above and the absence of coinciding peaks in particle concentration and CO concentration indicate that the O3 observations at the Zeppelin station were not significantly affected by local NOx pollution during the 2017 campaign.

The difference between the original and smoothed O3 data from Barentsburg varies from -19 % to 11 % of the smoothed value, and there is a moderate negative correlation between the magnitude of NOx peak and the reduction in O3 concentration (r=-0.65, p<0.0001). Despite this sensitivity of O3 concentration to local pollution in Barentsburg, the median NOx concentrations observed there were low, and average reduction of O3 concentration in comparison to the smoothed values was only 1 %. This effect is not statistically significant, and therefore other factors, such as variation in concentrations within long-range-transported air masses, may be more important for explanation of difference between the O3 Zeppelin and Barentsburg data sets.

The comparison of the vertical O3 data from the O3 sondes (dark blue squares in Figs. c and c) from Ny-Ålesund and the ground-based measurements at the Zeppelin station and in Barentsburg reveals that the discrepancy in the data between the two stations may be explained by the fact that the stations are located at different heights and measure air masses with an uneven distribution of O3 in the lowest 500 m. If contrasting the closest point to the sounding time in the observations made in Barentsburg and in Ny-Ålesund, similar tendencies as in the O3 sonde data may be observed. For example, there is a significant difference between the data from Barentsburg and Ny-Ålesund (33.74 ppb vs. 38.88 ppb) for the measurement closest to the time of sounding on 10 May, and an increase of O3 concentration with height between 1000 hPa (50 m) and 950 hPa (500 m) is noticeable in the sounding data as well (Fig. a). One can see in the potential temperature profiles (Fig. b) that the pronounced atmospheric inversion tends to be noticeable in the O3 sonde profiles as well. For example, a simultaneous increase in O3 concentrations and potential temperature is pronounced at the level of 850 hPa (1300 m) on 26 April, 860 hPa (at 1200 m) on 5 April and 900 hPa (at 1000 m) on 12 April (Fig. a and b). The O3 concentration in the ERA5 profiles was overestimated and showed little variability in the lowermost layer, except the profile for 10 May (dashed dark grey line in the Fig. a) when the reanalysis and observational values coincided for 1000 hPa level, but the O3 concentration in ERA5 was underestimated by almost 40 % at the height of 925 hPa (Fig. a). The virtual potential temperature profiles in reanalysis resemble O3 sonde profiles closely (Fig. b). The PBL height from ERA5 reanalysis (marked with stars in the Fig. b) was above the low-level temperature inversions detected on 5 and 12 April and below the level of the most pronounced virtual temperature inversion in the O3 sonde profiles.

In addition to hourly values, average concentrations of measured compounds have been calculated for each hour of the day. The diurnal variation in NO, NO2 and O3 concentrations at the stations is shown in Fig. . Mean and median daytime (from 06:00 to 17:00 UTC) and night-time (from 18:00 to 05:00 UTC) concentrations are shown in Table  (here the daytime and night-time are defined based on the snowmobile traffic pattern in the Adventdalen valley).

The NO2/(NO+NO2) ratio is quite high in Adventdalen and in Barentsburg and exhibits diurnal variation, while it is much lower in Ny-Ålesund, and there is no statistically significant difference between its day and night values. This may be explained by the fact that the measurement station in Ny-Ålesund was located much closer to the diesel power plant, a constant source of fresh NOx emissions, where the NO2/NOx ratio is much lower irrespective of the time of the day . The lowest hourly NO2/NOx ratio of 0.29 and the highest peak of NOx were observed in Ny-Ålesund at 17:00 UTC 28 April (Fig. ). The concentration of NO and NO2 was 87.8 and 16.4 ppb, respectively, which indicates the presence of a strong emission source, for example snowmobiles, in the immediate vicinity from the station. Since this was a single NOx peak in the data, the NO2/NOx ratio was unusually low, and the meteorological conditions were untypical for pollution accumulation in Ny-Ålesund (south-easterly wind with moderate speed of 5.3 m s-1); hence this value has been excluded from further statistical analysis.

NO is a primary product of fossil fuel combustion , and a higher NO/NOx ratio is expected close to the emission source. The station in Adventdalen is located at a distance of 5 km from the coal power plant, and snowmobile traffic there is a temporarily emission source present mostly during daytime. In contrast, measurement stations in Barentsburg and Ny-Ålesund are located near the power plants, constantly releasing combustion products at a variable rate. Thus, it is noticeable in Adventdalen that the NO concentration is close to zero during the night (dark blue bar in Fig. a) in the absence of fresh traffic emissions and photochemical conversion of NO2 to NO. As the traffic intensity increases during the day, the NO concentration rises, however, and so does the NO2 concentration (red bar in Fig. a), since there is a rapid conversion of NO to NO2 by the reaction with O3.

There is a slight increase in the O3 values in Barentsburg (grey line in Fig. b) during daytime. In contrast, a slight decrease in daytime O3 concentration is observed at the Zeppelin station. The discrepancy in the diurnal O3 profiles from the two stations may be explained by the difference of the altitude of each of the two stations and response of the measurements to the boundary layer dynamics. The ERA5 data show diurnal variation in the PBL height for both locations with highest average values of 462 and 293 m, in Barentsburg and Zeppelin, respectively, for 13:00 UTC. The Barentsburg station is located at the altitude of 50 m a.s.l. and samples air within the ABL. During the daytime, the vertical mixing between the atmospheric surface layer and the air masses aloft enhances, and the boundary layer height increases. This mixing process may enrich the surface layer with O3. A similar influence of convection on replenishing the O3 in the Arctic ABL after the depletion events has been described in the work of . In contrast, the Zeppelin station is located at the altitude of 474 m a.s.l. and mostly samples air from the free troposphere with higher O3 concentration, and thus the data from this station do not exhibit similar diurnal variation as the Barentsburg station. However, the magnitude of these effects is small, and according to the t test and the WRS test, there is no statistically significant difference between the night-time and daytime O3 concentrations measured at the stations (Table ).

The average NO and NO2 concentrations measured at the stations are distributed unevenly over the wind directions. In Adventdalen, the average wind speed was 5.1±3.1 m s-1. The south-easterly wind dominated during the field campaign (Table ), and there was no significant difference between the daytime and night-time observations. The highest average daytime NO and NO2 concentrations were observed when the wind was from north-east and south-east in Adventdalen (Fig. ). To test if the size of snowmobile motorcade has an influence on the NOx concentration in Adventdalen, manual observations of the number of snowmobiles were done in 19 d. In general, the effect of a large number of snowmobiles was only noticeable in the NOx data when there was low wind speed. For example, in the evening of 1 May 2017, the wind speed was 1.9 m s-1, and the NO2 concentration increased sharply to 7.3 ppb due to 21 snowmobiles passing by the station. A group of a similar size was passing by in the evening of 2 May 2017, but the effect on NO2 values was 3 times lower as the wind speed was higher (4.0 m s-1). The maximum hourly NO2 concentration of 11.4 ppb was measured during the Easter holiday on 13 April 2017. In that day, the combination of increased recreational traffic and mild weather conditions (wind speed below 1 m s-1 and air temperature -8 ∘C) led to accumulation of concentration 13 times higher than daytime hourly average measured during the field campaign. Such low wind speed is untypical for the wind regime in Adventdalen, where normally ventilation is sufficient to effectively disperse NOx emitted by the usual amount of motorized traffic. The highest average night-time NO2 concentrations were detected when the wind was from north-west, which reveals possible influence of the coal power plant. The average night-time concentrations of NO were very low, regardless of the wind direction.

Figure a and b illustrate distribution of NO and NO2 concentrations over wind directions in Ny-Ålesund and Barentsburg, respectively. South-easterly wind with average speed of 3.6 and 3.9 m s-1 in daytime and night-time, respectively, was dominating in Ny-Ålesund. However, the highest average NOx concentrations in Ny-Ålesund were measured when the wind was coming from the north (Fig. a). This points clearly to the local diesel power plant being the main emission source. Similar results regarding the influence of the local power plant in Ny-Ålesund on NOx concentrations were presented in and . During the field campaign, the prevailing wind in Barentsburg was from south and south-east, with an average speed of 2.5 m s-1 and from south-east and east with mean speed of 2.3 m s-1 in daytime and night-time, respectively. The NOx concentrations measured there were much lower and more evenly distributed over different wind directions than in Ny-Ålesund (Fig. b). The coal power plant operates day and night and, in the light wind conditions, may contribute to accumulation of local pollution in the settlement even in the absence of south-westerly wind.

An overview of median concentrations of measured compounds and prevailing local meteorological conditions for the nine sub-periods defined based on the prevailing weather regimes is given in Table . The wind directions observed at the stations during each of the sub-periods have been sorted into 16 bins with a 22.5∘ interval. The three main wind directions for each sub-period and for the whole campaign are stated with letters in Table . A detailed analysis of the meteorological conditions prevailing during each of the weather regimes and their influence on the concentration of the compounds measured at the three station is given below.

The “no regime” conditions were present for almost 20 d or 37 % of total campaign duration. The sub-period I lasted for 14 d and was characterized by the lowest median temperatures and UV irradiance at all stations and by synoptic-scale north-easterly flow (Fig. a). However, the conditions were inhomogeneous for this period: there were two quick passing cyclones on the 26 March and 3 April (pressure drops in Figs. b and b) that led to an increase in local air temperatures and wind speeds. In the sub-period I, the temperature inversions were observed in 46 % of the radiosonde profiles from Ny-Ålesund, but the median inversion strength was below 0.95 ∘C (median for the whole campaign).

The sub-periods II–V were characterized by Atlantic ridge (AR) and Scandinavian trough (ScTr). These regimes are characterized by the varying degree of geopotential height ridging over the North Atlantic at 500 hPa. During the AR regime, an upper-level north-westerly flow prevails . The strongest wind speed was observed in Adventdalen during the AR regimes, since the synoptic-scale lower-level flow (Fig. b and d) was parallel to the Adventdalen valley (Fig. ). During the transition from AR to ScTr, the upper-level ridge weakened and shifted southwards , as did the cyclonic systems on the lower level (Fig. c and e). The AR and ScTr regimes were characterized by the PBL height being below the median value for the campaign at all sites. The temperature inversions were observed for the regimes II, III and IV with an inversion frequency of 67 %, 57 % and 13 %, respectively, and the inversion strength was above the median for the campaign.

The sub-period VI was a 3 d “no regime” transition between the AR and ScTr and two blocking regimes: Scandinavian blocking (ScBL) and Greenland blocking (GL). This sub-period was characterized by the synoptic-scale westerly wind bringing warm Atlantic air over Svalbard (Fig. f), increasing local temperature and PBL height and adding the westerly component to the wind direction at all stations.

During the sub-period VII (ScBL), the positive geopotential height anomaly was located over northern Scandinavia, and part of the upper-level flow was deflected poleward around the blocking anticyclone . The anticyclonic movement was pronounced in the lower-level flow (Fig. g), and the synoptic-scale north-westerly flow prevailed over the western part of Svalbard. The local wind speed, temperature and PBL height decreased.

The sub-period VIII (GL) was characterized by the strong positive anomaly in the geopotential height at 500 hPa over Greenland and the prevailing upper level north-westerly wind . The lower-level blocking over Greenland promoted north-easterly flow over Svalbard (Fig. h).

The sub-period IX (“no regime”) was characterized by the strong anticyclone over the Barents Sea, that led to pronounced transport of warm Atlantic air with southerly flow to Svalbard (Fig. h). No temperature inversions were detected in the radiosonde data from Ny-Ålesund for the sub-periods VII, VIII and IX.

The elevated NOx concentrations were observed in Ny-Ålesund and Adventdalen during the sub-periods I–V. This may be explained by the enhanced accumulation of locally emitted NOx in the ABL due to suppressed vertical mixing on cold days associated with the AR and ScTr regimes. In Adventdalen, NOx concentrations are not dependent on the wind direction (Fig.  and Table ), and east-south-easterly and south-easterly are the dominant wind directions for all regimes except for the period VI. The highest median values of NOx were observed during the periods with the lowest PBL height, ScTr regimes. During the periods VI and VII, the PBL height increased, but westerly and west-north-westerly wind might have brought pollution from the coal power plant and the town of Longyearbyen to the Adventdalen valley (Fig. ). In Ny-Ålesund, the highest median NOx value was observed in sub-period I due to the presence of north-north-westerly wind that brought the plume from the power plant to the measurement station (Fig. a).

In contrast to Adventdalen, the boundary layer height and cold temperatures played a secondary role for the NOx concentrations in Barentsburg, and the controlling factor was south-westerly component of the wind for the most polluted periods. The highest median NOx values were detected during the sub-period VI, when south-south-westerly wind was dominating. The major emission sources in Barentsburg are located on the seashore, and warmer marine air from west and south-west may bring local pollution to the station situated on the hill above these sources (Fig. b). However, the second highest median NOx value was observed for the sub-period II, the period with the lowest PBL height and easterly wind at this station. The wind direction was not from the coal power plant, but the wind speed was very low, and thus, the local pollution could accumulate in the ABL if a strong inversion was present aloft.

The O3 data show similar variability for the Zeppelin station and Barentsburg. The concentrations below the campaign's median were observed for the sub-periods V, VII and VIII. The FLEXPART 10 d backward trajectory probability contours show that for these sub-periods, the air masses passed over the region north of Svalbard where the concentration of BrO was elevated (Fig. e, g, h). Conversely, the sub-periods III, IV, VI and IX, with O3 concentration above the median at both stations, are characterized by the air masses arriving from the south-east, east, west and south-west (Fig. c, d, i), respectively. In the sub-period I, 24 % of the data from Barentsburg were missing. This sub-period's concentrations at the Zeppelin station were slightly higher than the campaign's median, despite the most significant O3 depletion episode occurring on 31 March–1 April (Fig. ). The trajectory contours show possible influence of the local depletion in the Svalbard region (Fig. a). In the sub-period II, 67 % of the data from the Zeppelin station were missing (Fig. ). The O3 concentration in Barentsburg was above the median for this sub-period, and the trajectory data show the air masses arriving from the south-east (Fig. b). As in previous studies of , the downward transport of O3-enriched air masses from higher altitudes played a significant role during the 2017 campaign. The percentage of trajectory points reaching elevations above 2000 m was highest for the sub-periods III, VI and IX (27 %, 33 % and 24 % of the total number trajectory points for each sub-period respectively). In contrast, during the sub-period VIII with the lowest O3 concentration at both stations, the percentage of elevated trajectory points was minimal, only 4 %. One can also see that the percentage of elevated trajectories varies for the same type of weather regime and determines importance of the downward air mass transport for the measured surface O3 concentrations in different sub-periods (e.g. ScTr regime in Fig. c and e and Table ).

Two joint extreme O3 depletion events (31 March and 6 May 2017) and three increase events (13 April, 28 April and 3 May 2017) have been detected (Figs.  and ). The HYSPLIT trajectory analysis shows that these O3 depletion events occurred when the cold air masses from the central Arctic reached Svalbard. The trajectory for the strongest depletion episode is shown in Fig. (a). The concentration of O3 in the Arctic air masses may be lower because of a lack of sunlight and O3 precursors such as NOx, hydrocarbons and CO needed for the O3 formation. Further depletion may have occurred due to photochemical reactions with bromine species over the sea ice in the period from 30 March 2017 at 10:00 to 31 March 2017 at 17:00 when the trajectories passed the region with elevated BrO concentration between 80∘ N and 85∘ N. The simulated median sun flux was quite low (67 W m-2) but probably sufficient enough to support the halogen-induced O3 destruction, which might occur even under low-light conditions . The trajectories for the increase events revealed southerly origin of the air masses, but source regions were different for all three cases. In the first case, air masses were arriving from the northern part of Russia and in the second one from North America and Iceland. However, the highest O3 concentrations at both stations were observed on 3 May 2017 when the air masses were transported from Europe (Fig. b). The air masses arrived in Svalbard from the west and did not pass over the areas with elevated BrO concentration.