Surface ozone at the Caucasian site Kislovodsk High Mountain Station and the Swiss Alpine site Jungfraujoch: data analysis and trends (1990–2006)

Long-term ozone measurements of two background mountain sites, namely the Kislovodsk High Mountain Station in Caucasus, Russia and the Jungfraujoch a.s.l.) are compared. Despite of more than 1.5 km altitude di ﬀ erence ozone concentrations are comparable at JFJ an KHMS in the beginning of measurements (1990–1993) while the annually averaged levels at JFJ are around 15 ppb higher than the ones at KHMS for the most recent years (1997–2006). Averaged for di ﬀ erent periods ozone concentrations at KHMS are comparable with the respective values observed at the elevated sites in the midlatitudes, situated in the altitude range 10 1600–2400 m a.s.l. Distribution function of the hourly concentrations has two peaks at JFJ and it is close to Gaussian distribution in the case of KHMS. Seasonality at both sites is characterized by double spring-summer maximum. Spring maximum at both stations is more pronounced for the air masses with the longest contact with upper free troposphere and stratosphere. Average concentrations increased at JFJ 15 but decreased at KHMS for the period 1990–2006. Trends are more pronounced for the 1990s ( + 0.73 ± 0.20 ppb/year at JFJ and − 0 . 91 ± 0.17 ppb/year at KHMS for the period 1991–2001) in comparison with later years ( + 0.04 ± 0.21 ppb/year at JFJ and − 0 . 37 ± 0 . 14 ppb/year at KHMS for the period 1997–2006). Trends show a distinct seasonality, which is di ﬀ erent for the di

concentrations are comparable at JFJ an KHMS in the beginning of measurements (1990)(1991)(1992)(1993) while the annually averaged levels at JFJ are around 15 ppb higher than the ones at KHMS for the most recent years (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006). Averaged for different periods ozone concentrations at KHMS are comparable with the respective values observed at the elevated sites in the midlatitudes, situated in the altitude range 10 1600-2400 m a.s.l. Distribution function of the hourly concentrations has two peaks at JFJ and it is close to Gaussian distribution in the case of KHMS. Seasonality at both sites is characterized by double spring-summer maximum. Spring maximum at both stations is more pronounced for the air masses with the longest contact with upper free troposphere and stratosphere. Average concentrations increased at JFJ 1 Introduction 5 Ozone plays a crucial role in tropospheric chemistry as it is the most important compound of photooxidant air pollution, it determines the oxidation capacity as the main precursor for OH radical and it is a significant greenhouse gas (IPCC, 2007). Surface ozone concentrations are very variable both in space and time, on both long and short scale. Trend determination of tropospheric/surface ozone is often a difficult task be-10 cause the accuracy of the ozone gas analyzers is sometimes smaller than observed long-term changes (Virgazan, 2004;Oltmans et al., 2006;Brönnimann et al., 2002;TOR-2 report, 2003;Jaffe and Ray, 2007). Nevertheless it is well established that surface ozone concentration at unpolluted sites in Europe increased by more than a factor of two between the 1950s and the early 1990s (Staehelin et al., 1994), most prob- 15 ably because of large increase in ozone precursor emissions. Since the late 1980s measures are undertaken to reduce ozone precursor emissions in Western European countries. During the 1990s anthropogenic NO x emissions deceased in Germany and Switzerland by more than 30%, the VOC decrease was even larger (EMEP, 2004). Problems of the former Soviet Union (USSR) in the beginning of the 1990s caused an 20 economic crisis which led to a dramatic decrease of ozone precursors' emission in all countries of the former SU, and especially in the industrial centers.
Ozone concentration in ambient air does not show a simple linear response to the ozone precursor (nitrogen oxides (NO x : NO+NO 2 ) and volatile organic compounds (VOCs)) emission changes. It has been shown, that peak ozone concentrations over ozone concentration was documented for European high alpine sites (Jungfraujoch, Zugspitze, Sonnblick) by Brönnimann et al. (2002) and Ordóñez et al. (2007). Thouret et al. (2006) and Zbinden et al. (2006) reported from regular aircraft measurements MOZAIC an overall increase in ozone concentration in the upper troposphere and the lower stratosphere of about 1%/y for 1994-2003 (extending over Europe, North Atlantic and Eastern US). Similar results were obtained during the GASP program (Schnadt Poberaj et al., 2007). Jaffe and Ray (2007) reported increasing ozone levels for most of the elevated locations in the Eastern rural USA regions for the 1990s.
Several reasons can cause long-term ozone changes at particular receptor site, including the response on the emission changes (both natural and anthropogenic), in winter ozone at polluted sites (due to decrease in titration) and the decrease in the high summer ozone episodes was less than expected. Therefore it was suggested that changes in ozone concentration advected to Europe could have partially compensated the expected decrease. This was corroborated by Andreani-Aksoyoglu et al. (2008) for model analysis of Swiss ozone trends. Among the reasons considered to explain 5 positive ozone trend in the Northern Hemisphere the dramatic increase of South-East Asia emissions since 1990 was considered by several authors. Estimates obtained with state of art global numerical simulations (e.g. Auvray and Bey, 2005) can only partially explain ozone winter increase at high alpine sites. Ordóñez et al. (2007) suggested that an increased transport of ozone from the stratosphere could be responsible for 10 a substantial part of the increase in the background ozone found at European mountain sites.
In this study we compare long-term ozone measurements at Kislovodsk High Mountain Station (2070 m a.s.l.) located at the Eastern border of Europe at Caucasus Mountains, and those of the mountainous site Jungfraujoch (3580 m a.s.l.) located in the 15 Swiss Alps with the aim to understand the reason of the different ozone behavior at two background stations. Trajectory analysis is used as a tool to separate the air masses of different origin and to study the contribution of the different source regions to the observed variability of the surface ozone concentration at two mountain regions. Air Pollution Monitoring Network (NABEL) using a commercially available instrument (Thermo Environmental Instruments, Model 49C, UV absorption). The instrument is regularly compared to a transfer standard (TEI 49C PS) which is traced back to a NIST standard reference photometer. The detection limit is 0.5 ppb, the measurement uncertainty is determined to be ±2% (1 sigma), neglecting the uncertainty of the absorption 5 coefficient.

Measurements and methods
Statistical characteristics of the ozone datasets and distribution functions are based on the hourly mean data. All ozone subsets connected with trajectory analysis are 5 hourly averages calculated on the basis of hourly mean selected from 2 h before to 2 h after air arrival to the respective station. This procedure does not impact obtained 10 estimates of trends and characteristics of the seasonal cycles, but improves the data statistics in the cases of the data gaps directly at the time of air masses arrival.

Backward trajectories
To attribute variability characteristics to the properties of the air masses we use 3-D trajectories calculated with a help of the trajectory tool LAGRANTO (Wernli and Davies, 15 1997). The trajectories are based on the three-dimensional wind fields of the recent 40years reanalysis data set (ERA-40) of the European Centre for Medium Range Weather Forecast; for the period after ERA-40 (i.e. after August 2002) operational ECMWF analyses is used instead. Trajectories are calculated for 10 days back in time and have 6 h temporal resolution. The altitudes of the sites are taken into account by choosing 20 the appropriate arrival level, i.e. 650 hPa for JFJ and 750 hPa for KHMS.
Trajectories are used to trace the origin of the air mass arriving to the station both in vertical and horizontal direction. For this aim the diagnostic parameters potential vorticity (PV), altitude along trajectory and PBL height from meteorological re-analysis data are used to discriminate different vertical source areas. Air parcels coordinates 25 along the trajectories are used to study the horizontal transport patterns.
Several filters were applied to select the air masses being in the contact with the free troposphere and the stratosphere (in the text referred to as FT/ST cases), namely: 911 the altitude along the trajectory at least once exceeds 400 hPa level; the altitude along the trajectory at least once exceeds 500 hPa level and PV value exceeds 1.3 PVU; the altitude along the trajectory at least once exceeds 500 hPa level and PV value exceeds 1.6 PVU;

5
the altitude along the trajectory at least once exceeds 500 hPa level and PV value exceeds 2 PVU.
Note, that altitude and PV criteria should be fulfilled simultaneously at some point of trajectory. The trajectory is considered as belonging to the mentioned subset independently of the time when the criteria were satisfied. This makes the selected 10 classes non-uniform. The use of PV value alone as an indicator of the free tropospheric/stratospheric air is not sufficient due to complex topography of the studied locations, which can create some local flows with higher PV. Two filters are applied to trace the air masses which were in recent contact with continental PBL:

15
the altitude along the trajectory is at least once lower than PBL height east of 10 W; the altitude of the trajectory is lower than PBL height at least during two days of the last 5 days before arrival to the site (without geographical limitation).
The trajectories filters are applied to the 5-hourly averaged ozone concentration at 20 each site (see above). It should be noted that all the cases (FT and PBL) are studied independently, i.e. the cases filtered out with a stronger criteria are included in the cases selected based on the weaker but similar criteria. The monthly mean ozone concentrations are calculated for the individual subsets. The filtered monthly means are used for average seasonal cycles calculation and for trends analysis both for annual means and for different seasons. Because the concentration changes are not uniform in time we consider and compare two different time periods, namely 1991-2001 with large concentration changes at both site and 1997-2006 when ozone changes were much smaller.
To analyze the impact of the horizontal advection on the observed variations of the 5 surface ozone concentration the trajectories arriving to the sites are classified. Cluster analysis is performed for the trajectories of the total length (10 days back) and for the whole period (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) so that the main transport directions are the same for the different considered periods (1991-2001 and 1997-2006). The classification is done by means of k-mean clustering of the horizontal air parcel coordinates (latitude and longitude) as the most impacting variables (the algorithm can be found for example in Cape et al., 2000, andin Tarasova et al., 2007, andreferences there in). Vertical transport occurs much slower which makes the vertical coordinate less variable and hence less efficient for trajectory's bunches discrimination. Similar number of clusters was selected for both sites (7). 15 One can argue that transport patterns may have changed from the period 1991-2001 to the period 1997-2006, so classification should have been performed for the selected periods separately. But selected procedure can take these changes into account by means of the interannually changing frequency, while classification of the separate periods makes it difficult to compare the obtained clusters.  Figure 2 shows the monthly mean concentrations with standard deviation at JFJ and KHMS, calculated based on the hourly mean concentrations. In the beginning of the dataset (1990)(1991)(1992)(1993) the measured concentrations at the two elevated sites are quite 25 close to each other. In 1991 the lowest annual concentration is observed at JFJ for the 913 whole 16-years period and in contrast the highest concentration is observed at KHMS. Since 1991 till 1996 a strong increase of the surface ozone concentration is observed at JFJ and strong decrease occurs the same time at KHMS, whereas in the later period ozone concentrations at both sites stabilized. The decrease at KHMS for the 90-s was first documented by Senik and Elansky (2001), while the most recent data were not 5 published yet. Average rates of change for the whole period 1990-2006 are shown in Table 1 together with the other statistical characteristics of the datasets based on the hourly mean concentrations. Due to anomalous ozone variations in 2005 and 2006, statistics for these particular years is presented separately in the last two columns of the Table 1. Average ozone concentration exhibits pronounced vertical gradient in 10 spite of rather high elevation of both sites. In Europe a distinct vertical gradient is mostly reported for the measurements up to 1200 m a.s.l. (Chevalier et al., 2007) while the picture is less consistent for more elevated sites (see Table 2).

Concentration statistical characteristics and distribution functions
Comparing the levels of the surface ozone concentration at KHMS with literature data we can see (Table 2), that average concentration at KHMS (2070 m a.s.l.) for the 15 period 2001-2004 is close to the observations at Arosa (1840 m a.s.l.). Le Casset (1750 m a.s.l.), another station, which is even lower than KHMS, has slightly higher average concentration but still in the limits of 1 standard deviation of the average concentration estimate (Chevalier et al., 2007). Comparing KHMS ozone levels with the data reported for USA sites by Jaffe and Ray (2007) we can find comparable concen-20 trations (see Table 2) at Craters of the Moon, ID (1815 m a.s.l.) and Yellowstone N.P., WY (2400 m a.s.l.). Hence, the average ozone concentration observed at Caucasus region is consistent with the other observations at the elevated sites of the northern mid-latitude (in the altitude range 1600-2400 m a.s.l.). At the same time among the reported sites at the levels 1600-2800 m a.s.l. quite substantial variability of the aver-25 age ozone concentration is observed, which may be connected with different pollution level at the sites and different impact of complex orography. Nevertheless approaching to 3000 m a.s.l. average ozone concentration usually exceeds 50 ppb. Among the less elevated stations the average concentration higher than 50 ppb is only reported for 914 Monte Cimone, which is strongly affected by ozone and precursors advection from the heavily polluted Po basin (Campana et al., 2005;Cristofanelli et al, 2007).
The variance of the hourly concentrations is bigger at JFJ in comparison with the Caucasian station (Table 1) both for the whole data series and for 2005-2006 subset. Minimum hourly mean concentrations are comparable for JFJ and KHMS for the 5 whole series, while in 2005-2006 ozone minimum concentration at JFJ is higher than at KHMS. This may be related with strong positive trend of the surface ozone concentration at JFJ, which is more pronounced for minimum values (Brönnimann et al., 2002).
The frequency distribution of the hourly mean values is close to Gaussian probability 10 function for KHMS (see skewness of the distribution function in Table 1, and Fig. 3). Distribution of the hourly mean concentrations is asymmetric for JFJ and can be approximated by a superposition of two Gaussian functions (characteristics of the distribution are given in Fig. 3). Despite of the proximity of the KHMS' distribution function to the Gaussian shape, there are still some signs of the secondary peak formation which 15 looks like a "shoulder" in the range of bigger concentrations (Fig. 3). The presence of two maxima in the distribution function can indicate the existence of the two typical regimes of ozone concentration. Double peak ozone distribution function is also reported for Mt. Cimone and Mt. Waliguan (Lee et al., 2007;Wang et al., 2006) with the first peak centers at 43 ppb and 45 ppb correspondingly. These values are in agree-20 ment with the main peaks of the ozone distribution at considered sites (45.5 ppb at JFJ and 43.4 ppb at KHMS).

Seasonal variations
The seasonal cycles at the KHMS and JFJ sites show higher ozone values in the warm season (for both locations two maxima are pronounced) and lower values in the 25 cold season as typically observed at rural, remote and elevated sites in northern midlatitudes (Tarasova et al., 2007), except the remarkably low values reported at KHMS in September and October in 1996 (Fig. 2a). Spring and summer maxima in the seasonal 915 cycle are more distinguishable for KHMS than for JFJ. The highest ozone concentrations in the Caucasus and the Alps occur between March and August, depending on the particular year (Fig. 4). This is also the time period when mountain venting occurs transporting air from lower altitudes by convection to high altitudes (Henne et al., 2004;. High ozone values observed in the summer 2003 at JFJ are connected 5 to the very high temperatures of mentioned summer, which seem not only have affected ozone in the European Planetary Boundary layer (e.g. Ordóñez et al., 2005) but also ozone concentrations at JFJ. Record low ozone values were found at KHMS in September/October 1996 (the low values were confirmed by two instruments operated simultaneously) and the highest concentrations were registered in summer at JFJ in 10

2006.
A rather extreme amplitude of the seasonal cycle is observed at KHMS in 1996 and2003 (27.7 ppb and 20 ppb). If not considering these outlying years the average amplitude of the seasonal cycle at Caucasian station has a tendency to decrease from 18.3 ppb in 1990-1992 to 12.8 ppb in 1997-2001 (Fig. 3). This change in seasonal 15 variation amplitude is related to a strong decrease in spring/summer values whereas the decrease in ozone concentrations in the cold season is much smaller (Senik and Elansky, 2001;Tarasova et al., 2003).
Seasonal cycle of the surface ozone concentration at JFJ is also characterized by spring-summer maximum. The amplitude of the seasonal variations (difference be- The origin of the spring and summer ozone maxima at rural and remote sites has been discussed since many years. Historical records show spring maxima in earlier years (Linvill et al., 1980;Monks, 2000;Nolle et al., 2005), although the shape of the 916 Interactive Discussion cycle is likely to be sensitive to pollution conditions (see e.g. Scheel et al., 2003). In the earlier time surface ozone maximum was typically found in May (e.g. at Arosa) and it was attributed to the mixing with stratospheric air (e.g. Götz and Volz, 1951;Staehelin et al., 1994). Surface measurements and ozone sounding at rural and semipolluted sites in north America and Europe from the 1980s (Logan, 1985) and the 5 1990s (Tarasova et al., 2007, and references therein) often show a shift of the seasonal maximum to summer, which is commonly attributed to photochemical ozone production related to anthropogenic emissions of ozone precursors.
While analyzing the variability and the formation of the seasonal cycle it is important to consider which role is played by air transport in the mentioned processes. Figure 5 10 shows the results of backward trajectory analyses attempting to discriminate between the effect of mixing with stratospheric air and recent contact with the (polluted) planetary boundary layer. Air parcels defined as "UTLS" (i.e. PV values larger than 2 PVU and altitude higher than 500 hPa at least once along the trajectory) show a general tendency to have higher ozone concentrations than the averages, while the air parcels 15 defined as having contact with PBL (at least 2 days of the last 5 before arrival at the station the air parcel was inside PBL, which altitude is retrieved from reanalysis data along trajectory) have ozone concentration a bit lower than average, except for July. The only slight enhancement in the UTLS class as compared to the mean monthly concentrations suggests strong mixing with tropospheric air prior to the arrival to mea-20 suring site. It is also interesting to note that the strongest enhancement for the "UTLS" subset at both sites is observed in May, when ozone concentration is the highest in the lower-most stratosphere in the northern extratropics. Monthly mean concentrations in the PBL subsets are a bit lower than average during the cold season which probably reflects the effect of ozone poor PBL air (e.g. caused by ozone dry deposition and 25 titration with NO).
The major transport ways to each station are presented in Fig. 6. As it can be seen for both locations Western clusters are prevailing. At the same time even Western clusters are affected by the different areas of impact, i.e. the sampling is more often 917 Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion over the Atlantic Ocean for JFJ and over different European regions for KHMS. It should be noted that the clusters representing the transport of the longest range are observed in total in less than 5% of cases for both locations. Substantial number of cases (around 18% at JFJ and 22% at KHMS) corresponds to the transport on the regional/local scale. Average seasonal cycles are calculated for the different clusters of the horizontal 5 advection (Fig. 7). Two important features can be seen in the graphs, which are similar for both sites (JFJ and KHMS). The highest spring maximum is observed in May in the cluster, originating in the East Asia (cluster 1 for KHMS and cluster 5 for JFJ). For the other clusters representing long-range transport and traveling in the upper part of the troposphere the seasonal variations are the same, i.e. with prevailing spring maximum 10 (clusters 5 and 3 for JFJ and cluster 6 for KHMS). Statistical significance of the relative contribution of the mentioned clusters to the average seasonal cycle is not very high due to low frequency (less than 3%). Second important feature, which is the same for both locations, is the excess of the summer maximum above the spring one in the cluster of the local/regional transport 15 (cluster 4 for JFJ and cluster 3 for KHMS). These clusters for both stations have the highest frequency of the contact with the polluted continental PBL and are associated with a very slow transport (stagnation condition). Summer maximum which is developed in local clusters is associated with ozone photochemical production in the polluted air, hence indicating the connection of the summer maximum with photochemical pro-20 cesses. Slight excess of the summer maximum above the spring one is observed in the other clusters, which spent long time over the continent (for example, in cluster 7 for JFJ). For the clusters impacted by the European PBL arriving at KHMS spring and summer maxima are of comparable magnitude.

Trend analysis 25
As it has been mentioned above, long-term trends of the surface ozone concentration have different tendency, which value also depends on season (Fig. 8). In the following chapter the trends of the surface ozone concentration are analyzed at Kislovodsk 918 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion (Sect. 3.3.1) and at Jungfraujoch (Sect. 3.3.2) making use of the data obtained by backward trajectory analysis. The trends are calculated for different time periods (1991-2001 and 1997-2006) and transport subsets. Vertical subsets are discussed first so that the main finding can be used in the interpretation of the trends in the clusters of the horizontal advection. Analysis of the frequencies of the cases for each particular sub-5 set did not reveal consistent long-term changes to explain the systematic ozone trends (Fig. 9b, d), while the variability of the shorter scale (inter-annual) correlates to a certain degree with the variability of the transport patters. Working on the trends explanation we also considered the impacts of the changes of tropopause height and geopotential height at 500 hPa to understand the role of the dynamical processes better. We 10 found that the trends of the mentioned parameters have different signs at KHMS and JFJ, showing the difference in regional dynamics. But to the moment we can not draw a conclusion on the role of these factors in resulting trends formation.

Trend analysis for ozone measurements at Kislovodsk High Mountain Station
Ozone trends at KHMS separated by the vertical transport classes (including planetary 15 boundary layer contact and stratospheric origin, see Table 3 and Fig. 9a) are negative and statistically significant in 1991-2001 for all selected classes. Difference between trends in FT and PBL classes is insignificant. Maximum negative trends at KHMS (Table 3) are observed in summer and minimum negative trends are observed in winter (FT cases) or in autumn (PBL cases). Such trends seasonality (most negative in 20 summer and the least negative is winter) can be explained by ozone precursors emissions reduction which occurred in 1990s in Europe due to implied regulations and in the Newly Independent States due to USSR breakdown accompanied by economical crisis. KHMS is situated further deep in the continent which provide higher probability of the arriving air masses to contact planetary boundary layer anywhere over the 25 continent and bear the signs of emissions decrease both in Europe and more local regions. The area from which the air is sampled in the subset of the strongest PBL criteria (2 days of the last 5 before arrival to the station should be spent in the PBL) 919 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion covers mostly the Northern Caucasus region (not shown here). The trends in the FT subsets at KHMS are also negative despite of the reported increase in the 1990s of the stratospheric ozone contribution in the troposphere . Among the reasons to explain at least the part of this bias to the negative values the following could be mentioned: 1) quite substantial impact of the PBL on the ozone levels at 5 KHMS (especially in summer with developed convection over the continent, hiding the effects of ozone increase at the higher levels in the troposphere); 2) poor description of the mixing processes in the trajectory model which leads to misinterpreting the FT cases; 3) geographically different FT sampling area in comparison with JFJ (the picture is not shown). Nevertheless, the selected approach should be considered as a compromise, which let us see the signs of the stratospheric air in the admixture measured at the site. One more point is that for KHMS total time of the contact with the upper troposphere/lower stratosphere (PV>2 PVU and p<500 hPa) in comparison with JFJ in 1991-2001 is much smaller (31578 total hours against 64038 h at JFJ), so the effect of the ozone concentration growth in the free troposphere over Atlantic is less important 15 for KHMS. Ozone trends in the different vertical subsets are much smaller at KHMS in 1997-2006 in comparison with the earlier period (Table 4, Fig. 9a), being in the range from −0.3 to −0.6 ppb/year. Pronounced trend differentiation is observed between the average PBL (more negative) and FT trends. But this difference can be opposite for Interactive Discussion production or even its increase (note, emissions of the former Soviet Union did not decrease further or started increasing again since 1997-1998 due to economics stabilization). Moreover the most significant changes of the seasonal trends occurred in particular during summer (from the range −1.2. . .−0.9 ppb/year in 1991-2001, to the range −0.24. . . +0.12 ppb/year in 1997-2006). Winter trends in the strictest PBL sub-  15 and for all seasons are negative and statistically significant (Table 5) with somewhat wider range than in the vertical subsets. This means that the air masses can be better segregated accordingly to the geographical origin. Annual trends are in the range from −1.09 to −0.7 ppb/year. On average the most negative trend is observed (among the most contributing clusters) in the cluster 4, covering the Southern and Central Eu-20 rope and originating over Central Atlantic. This may be connected with ozone decrease downwind of the area (Central and Southern Europe) with strong emissions regulations of ozone precursors. Similar negative trends are found in cluster 5 (−0.89 ppb/year), which originates over Central Europe, confirming the idea that negative trends at KHMS may be connected not only with local emissions decrease (see below) but also can be (refinery region). For the period from 1990 till 1995 oil production in the region decreased nearly 3 times (from around 9 million tons to 3.78 million tons as reported by the State Committee on Statistics "Goskomstat" (1999) which means the decrease of accompanying emissions. Weather this decrease was sharp or gradual is unknown due to absence of statistical information. Therefore high ozone concentrations at the 5 KHMS station were caused by ozone production in the air advected from PBL polluted by oil industry (see seasonal cycle in the corresponding cluster).
The difference between trends in different advection clusters at KHMS is larger if considering the individual seasons ( Table 5). The largest trends scatter is observed in summer (from −1.43 to −0.67 ppb/year), while winter trends are much closer to each 10 other (from −0.78 to −0.42 ppb/year). In general in most of clusters (except cluster 2) maximum negative trends are observed in summer, while the minimum negative trends are observed in winter/autumn. In summer cluster 3 is the most frequent one and ozone trends in this cluster are attributable to the decrease of ozone precursor's emissions of the local scale. Moreover maximum impact of the PBL as shown above is expected 15 in this cluster. Summer trends in the clusters 5 and cluster 4 might be impacted by Southern and Western European emissions decrease due to legislation. The trend of the surface ozone due to emissions regulation (at least of nearby emissions) should be positive in winter, while at KHMS all the trends are negative which may be attributed to the underestimated mixing in the trajectory model or with particular winter chemistry 20 at the low ozone precursors levels. Interesting to note, that the bigger area is covered by a cluster, the stronger is the negative trend for both winter and summer (in winter −0.78 ppb/year for cluster 4, −0.62 ppb/year for cluster 5 and −0.42 ppb/year for cluster 3; in summer −1.43 ppb/year, −1.29 ppb/year and −1.04 ppb/year for the corresponding clusters), while the difference between summer and winter trends is similar 25 for considered clusters. Smaller winter ozone trend in the cluster of the local advection may indicate either much higher rate of the local emission reduction in comparison with European regulations or different structure of the emission reduction (mostly industrial, oil-gas associated).

922
Unlike the discussed above clusters of the air advection to KHMS in 1991-2001, cluster 7 (originating at the West coast of US) has different properties. In this cluster the most negative trend is observed in autumn (−1.23 ppb/year) and the least negative trend is observed in winter (−0.62 ppb/year). The reasons for such a seasonal pattern of the trend are unclear. Relative contribution of the air arriving in this cluster 5 accounts on average for less than 15%. It is likely that several processes contribute simultaneously into formation of the trend in the cluster 7.
Comparing the trends of the surface ozone concentration at KHMS for the period 1997-2006 (Table 6) with the earlier period, discussed above, we can see substantial decrease of the trends absolute values. For most of the clusters (clusters 2, 3, 4, 5 10 and 7) trends remain negative and statistically significant. The least negative trend is observed in the cluster 4 (originating in Central Atlantic), up to −0.18 ppb/year. Moreover, the most substantial changes of the trends' values and their seasonality are observed in this cluster (cluster 4). Taking into consideration that ozone source areas for KHMS are situated much lower than for JFJ, the shape of the trends seasonality (slight 15 negative trend in winter and slight positive trend in summer might be the signs of the increased ozone production in the PBL over the Atlantic due to ship emission increase. In 1997-2006 spring trends became mostly statistically insignificant (with the exception of the cluster 7, for which it became just a bit less negative in comparison with the one for the period 1991-2001, i.e. changed from −0.75 ppb/year to −0.57 ppb/year). 20 Proximity to 0 of the summer trends in the main advection clusters, i.e. cluster 3 (local), cluster 5 (Central Europe) and cluster 4 (Central Atlantic) and weak but statistically significant negative trends in winter in the considered clusters probably indicated the growth of the emissions to the South and West of the KHMS location, transported then to the KHMS and impacting ozone concentration levels.

Trend analysis for ozone measurements at Jungfraujoch
General features of the long-term ozone concentration evolution at JFJ for different seasons are presented in Fig. 8c. Summary of the trends in different vertical subsets at 923 JFJ is given in Tables 7 and 8 for the periods 1991-2001 and 1997-2006, respectively. Table 7 shows that for the period 1991-2001 the trends at JFJ are positive and statistically significant for all the subsets (including seasonal). The value of the annual trends at JFJ ranges from +0.82 to +0.59 ppb/year (first column of Table 7). Trends are very close in all FT cases while they are smaller for PBL cases (especially for the 5 air which was in the contact with PBL at least 2 days of the last 5 before arriving to the station). As far as we do not set a spatial criterion to the area of the sampling in the case of the last mentioned subset, the PBL may also b sampled over Atlantic.
All the trends at JFJ during 1991-2001 (Table 7) have pronounced seasonality, characterized by the most positive trends in spring and the least positive trends in summer 10 and autumn. An exception is the subset with the longest contact with PBL, for which the strongest positive trend is observed in winter and the least positive trend is observed in summer. Ozone increase in winter is consistent with the ozone response to NO emissions reduction as expected from air pollutants abatement regulation (less titration of ozone in winter and less production in summer). However, the increase of the surface 15 ozone concentration in the warm season contradicts to emission regulation strategy, which took place during the early 1990s. One may need to take into account that the PBL contact cases happen over large geographical regions, which are not uniform in the sense of emissions. PBL contact may occur both over the polluted Northern part of Italy and Spain, which probably impacts the most ozone concentration at the station 20 and over the large parts of the North Atlantic, which is less polluted than continental Europe.
Seasonality of the trends in the FT cases (Table 7) can provide some more insight into the reasons of trends: 1) the trends are more positive in the FT/ST group indicating that at the higher levels in the troposphere the growth is more substantial. Ozone 25 growth in the upper troposphere is connected with increased transport from the stratosphere (see Ordóñez et al., 2007); 2) as far as stratospheric ozone has seasonal maximum in spring the strongest response on the increased influx should be expected for the spring months, which is consistent with our analysis (i.e. the most positive trends in FT subsets are observed in spring). The geographical area, where the stratospheric air is sampled before arrival to JFJ (not shown here) suggests that the UTLS regions over North Atlantic/USA East coast and Canada provide the strongest impact on the monthly means in the FT subsets. Therefore we can conclude that for the period 1991-2001 the trends at JFJ may mainly be caused by two factors, namely in situ emissions regu-5 lations, causing ozone decrease in summer and increase in winter in PBL overlapping with systematic increase due to the growth of the stratospheric contribution (mostly seen in spring). This conclusion is in line with the finding of Ordóñez et al. (2007) who provided evidence that ozone concentration at JFJ has increased in the 1990s in the free troposphere over Atlantic due to increased contribution of the stratospheric ozone.

10
The trends values are substantially different at JFJ for the later period, i.e. 1997-2006. On average the trends at JFJ (see Table 8, Fig. 9c) are statistically nonsignificant and close to zero (from +0.18 to −0.01 ppb/year). Change of the absolute values of the trends is accompanied by the change in the trends seasonality. The trends remain mostly positive and statistically significant is winter, unlike the other sea- 15 sons. The only FT subset where trend remains positive and statistically significant in summer corresponds to the cases with "p<400 hPa", i.e. for the air which travels quite high but not necessarily had contact with the stratosphere. The cases which fulfill this criterion may correspond to the long-range transport of precursors and ozone from Asia. Note, that for the earlier period (Table 7) positive trend at JFJ was also the high-20 est in summer for the mentioned FT subset. The least changes of the trends without changes of their seasonality between 1991-2001 and 1997-2006 are observed at JFJ in the subset with the longest contact with PBL (the lowest row in Table 8), hence ozone response to the emissions control over Europe is consistent for the two periods. The trends in PBL may be slightly weaker due to the fact that the rate of emission changes 25 has decreased in 1997-2006 in comparison with 1991-2001. Indeed, according to the emission inventories presented by the EMEP program (http://www.emep.int) the rate of the EU15 (excluding Greece) emission decrease rate as reported by Parties declined from 1970 Tg/y to 1584 Tg/y for CO, from 326.3 Tg/y to 252.9 Tg/y for NO x , from 507.6 Tg/y to 391.2 Tg/y for NMVOC during 1991NMVOC during -2001NMVOC during and 1997NMVOC during -2006 The strongest change in the magnitude of the spring trends is observed for the strictest FT subset, which may indicate that contribution from the stratosphere is not changing anymore and that the average trend is driven by emission regulations mostly. Absence of the positive trends during the warm season in 1997-2006 indicate that the 5 earlier trends are likely to be only slightly driven by the increased ozone production over Eastern Asia, because emissions in the region were and are rising very fast (see for example van der A et al., 2008). As the response to the increasing emissions ozone production should continue rising while this is not clearly seen in the data.
The number of FT or PBL cases at JFJ has not changed much (Fig. 9d)  substantial changes in the source area. Surface ozone trends at JFJ in the different advection clusters are summarized in Tables 9 and 10. As we have seen in the discussion of the vertical subsets, ozone trends at JFJ are positive and statistically significant for the period 1991-2001 in all advection clusters (Table 9). The most positive annual trends are observed in the clus-20 ters 1, 3 (both observed very seldom) and 6 (observed on average in 16% of cases), i.e. in the "longest" clusters traveling quite high. It should be noted that in the cluster 5, also originating in Asia positive trend is the least as the air is sampled in the higher latitudes. In general the closer the origin area to Europe is, the smaller are the positive trends.

25
In most clusters ozone trends seasonality (except for the clusters 5 and 7) is characterized by the strongest positive trend in spring. The most local clusters 4 (Europe) and 7 (Central Atlantic) are characterized by the lowest among the other clusters spring trend. On average, air arriving in these two clusters has the least often contact with 926 UTLS (5.2% and 4.6% of cases, respectively). On the other hand even stratospheric air in these clusters is sampled in the other geographical regions in comparison with the clusters of longer length. It is likely that more substantial change (increase) of STE occurred closer to the US coast, but not over Europe (we could see above analyzing FT and long-range transport cases at KHMS). Moreover, considering the clusters 5 which center is situated to the North of the pathways of the other clusters (cluster 5) and which has quite high number of contacts with UTLS (12.6%) we do not observe an increase in spring, moreover, the spring trend in cluster 5 is close to 0.
Central Atlantic (cluster 7) trends are rather similar to the ones observed in the local European cluster (cluster 4). In these classes the trends might be viewed as superposition of European emission regulations (more positive in winter, less positive in summer) and the general increase of the ozone levels in the Atlantic troposphere. The role of the emission regulations seems to be less important (due to trends similarity in cluster 4 and cluster 7) in comparison with general level increase over Atlantic, but likely important enough to modulate the seasonal shape of the trend. 15 Comparing the trends in the advection clusters for the later period (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) with the earlier period (1991-2001, see Tables 9 and 10) the following features can be seen. Annual trends in all clusters become close to 0 with a level of statistical significance comparable to the earlier period. The strongest changes of the trends' values occurred in spring. Proximity of the spring trends to 0 may indicate that the processes, which had 20 provided ozone growth in spring are no longer relevant in the second period, i.e. stratospheric contribution in the upper troposphere is not increasing any more. Nevertheless in winter ozone trends in most of clusters remain positive and statistically significant, being the most positive in the cluster 4 (local European) and cluster 7 (Central Atlantic). Assuming that both of these clusters are substantially impacted by European 25 air we can conclude that in winter the small increase of ozone concentration may be associated with European emissions decrease. Proximity of the summer trends to 0 may also support this conclusion. Among the substantially contributing, only in the cluster 7 (Central Atlantic) summer trend is still positive and statistically significant. This may 927 be connected with increasing ozone production over Atlantic where ship emission are rising (Eyring et al., 2007). Note that summer trend in the cluster, in which air circulates over Europe, is close to zero.

Conclusions
Ozone variability at two elevated sites situated in the same latitude belt but at differ- Distribution functions of the hourly concentrations were studied for both sites. Ozone distribution at JFJ can be fitted by bi-modal distribution function, while for KHMS the distribution of hourly mean concentrations is closer to the Gaussian function. The primary maxima of the distribution functions are close for two sites, being 45.5 ppb at JFJ 20 and 43.4 ppb at KHMS and comparable to other data, presented in Lee et al. (2007).
Both sites are characterized by the wide spring-summer seasonal maximum. The shape of the seasonal cycle varies from year to year at both locations (JFJ and KHMS) and this variability is possibly caused by variations in global atmospheric dynamics.
Analysis of the 3-D trajectories for the whole measurement period showed that for the 25 subsets more impacted by stratosphere (with PV exceeding 2 PVU along trajectory) the spring maximum is dominating, while summer maximum is more controlled by ozone 928 production in PBL (selected as the cases when trajectory spent more than 2 days of the last five before arrival at the site in the contact with PBL). Analysis of the seasonal cycle for the different horizontal advection clusters showed that spring maximum prevails at both JFJ and KHMS in the clusters associated with long-range transport and originating in the free troposphere above East Asia. In the clusters of the local/regional advection 5 summer maximum is prevailing also at both locations showing the importance of ozone photochemical production in the polluted air masses which were in the contact with European PBL. The trends of the surface ozone at JFJ and KHMS were studied for two different periods, namely 1991-2001 with strong concentration changes at both stations and 1997- To interpret the origin of the trend at each station we used a trajectory analysis. 20 The subsets were organized separating the arriving air masses in accordance with vertical or horizontal air parcels position along the trajectories. So, the cases of the air contact with PBL, with upper free troposphere and stratosphere and having different geographic origin areas were studies separately. Analyzing different subsets we come to the conclusion that the ozone trends at 25 KHMS in the 1991-2001 are connected with strong emissions reduction in the regions, surrounding station (from local to European scale). This conclusion is supported by the trends value in PBL cases and in the local and European horizontal air advection clusters. Nevertheless, the trend is unlikely to be totally due to regional photochem-929 istry and impact of the other factors is possible (ex., underestimated mixing, difference from JFJ in the areas of the stratospheric air sampling and less frequency of such events). Only the change in emissions is not able to reproduce the magnitude of the negative trends, for example the trends in the subset for the upper troposphere/lower stratosphere and for the clusters of the long-range transport traveling quite high are 5 still negative, while air spent very limited time in the contact with regions of the strong emissions reduction. There are some indirect indications that stratospheric air contribution at KHMS differs from the one at JFJ (negative spring trend in the ozone subset filtered for the cases of PV>2 PVU and p<400 hPa remains nearly invariable between 1991-2001 and 1997-2006) at KHMS.

10
Positive trends of the surface ozone at JFJ in the 1991-2001 are connected with an increased contribution from the stratosphere over Atlantic, which is confirmed by the analysis of the free tropospheric/stratospheric subsets, by the highest positive trends in the longest and the highest traveling horizontal clusters and by trends seasonality (maximum positive trends are observed in spring). The response to the regional Euro-15 pean emissions decrease is less important but it is contributing to the seasonality of the trend. During the later years the trends became close to zero. Increase of the winter concentrations in the later years at JFJ may be connected with emissions regulations as far as decrease in NO titration and decrease in ozone production are likely to occur under this scenario. 20 It should be noted that the changes in the number of cases of different criteria performability are not able to explain dramatic changes of the trends at the two locations between 1991-2001 and 1997-2006, so the changes in the advection patters can play an important role in the formation of the inter-annual variability but unlikely to impact the systematic ozone changes. The reason of the trends difference at KHMS and JFJ 25 is mainly due to the difference in the position of the stations and difference in the source areas affecting ozone variations in the Caucasus and Alps. Position of KHMS close to the Caucasus Ridge and far from the border of the continent makes this location very sensitive to the wider range of factors, controlling ozone in the continental planetary 930 boundary layer. Both substantial emissions decrease in 1990s due to USSR break down and measures in Europe to control emissions as well as different dynamical processes could be seen at the station creating a complex and not clearly understood tendency. Being higher in altitude and closer to Atlantic ocean JFJ is more sensitive to the background ozone changes in the free troposphere over the ocean, while emis-5 sion changes can also play some role in the trends formation. Rise of the shipping emissions in Atlantic may be seen more clearly in the coming years in the JFJ ozone levels.          Fig. 1. Position of the ozone measurement stations which data are used in the paper. The Caucasus region is presented in more details in the upper corner. The map is compiled from two maps of UNEP/GRID-Arendal Maps and Graphics Library (UNEP/GRID-Arendal. The Caucasus ecoregion, topographic map. UNEP/GRID-Arendal Maps and Graphics Library. 2008. Available at: http://maps.grida.no/go/graphic/the-caucasus-ecoregion-topographic-map. Accessed June 12, 2008 and UNEP/GRID-Arendal. How the comb-jelly (Mnemiopsis leidyi) is spreading through European seas (invasive species).
UNEP/GRID-Arendal Maps and Graphics Library.
2007. Available at: http://maps.grida.no/go/graphic/ how-the-comb-jelly-mnemiopsis-leidyi-is-spreading-through-european-seas-invasive-species. Accessed 12 June 2008).  Gaussian approximations are given for the distribution functions calculated on the long-term measurements. In the graphs "x" corresponds to the distribution curve center and "w" gives a fit curve width.