Extreme events in total ozone over Arosa – Part 2: Fingerprints of atmospheric dynamics and chemistry and effects on mean values and long-term changes

Abstract. In this study the frequency of days with extreme low (termed ELOs) and extreme high (termed EHOs) total ozone values and their influence on mean values and trends are analyzed for the world's longest total ozone record (Arosa, Switzerland). The results show (i) an increase in ELOs and (ii) a decrease in EHOs during the last decades and (iii) that the overall trend during the 1970s and 1980s in total ozone is strongly dominated by changes in these extreme events. After removing the extremes, the time series shows a strongly reduced trend (reduction by a factor of 2.5 for trend in annual mean). Excursions in the frequency of extreme events reveal "fingerprints" of dynamical factors such as ENSO or NAO, and chemical factors, such as cold Arctic vortex ozone losses, as well as major volcanic eruptions of the 20th century (Gunung Agung, El Chichon, Mt. Pinatubo). Furthermore, atmospheric loading of ozone depleting substances leads to a continuous modification of column ozone in the Northern Hemisphere also with respect to extreme values (partly again in connection with polar vortex contributions). Application of extreme value theory allows the identification of many more such "fingerprints" than conventional time series analysis of annual and seasonal mean values. The analysis shows in particular the strong influence of dynamics, revealing that even moderate ENSO and NAO events have a discernible effect on total ozone. Overall the approach to extremal modelling provides new information on time series properties, variability, trends and the influence of dynamics and chemistry, complementing earlier analyses focusing only on monthly (or annual) mean values.


Introduction
Downward trends in global stratospheric ozone during recent decades have been shown to be directly linked to increasing surface UV-radiation (e.g. Calbo et al., 2005). The successful implementation of the Montreal Protocol (e.g. WMO, 2007) started the discussion on ozone recovery (e.g. Austin and Wilson, 2006;Eyring et al., 2007;Harris et al., 2008;Shepherd, 2008;Hegglin and Shepherd, 2009;Waugh et al., 2009). In the past, long-term ozone trends were determined from homogenized data series by fitting with multiple linear regression models, in which suitable independent variables (so-called explanatory variables) were used to represent atmospheric variability, such as the Quasi-Biennial Oscillation (QBO), the 11-year solar cycle, and a linear trend attributed to anthropogenic ozone depletion (e.g. Staehelin et al., 2001;WMO, 2003). Previous studies have identified a number of other processes influencing total ozone at mid-latitudes, such as synoptic-scale meteorological variability (e.g. Dobson and Harrison, 1926;Steinbrecht et al., 1998;Shepherd, 2008), decadal or long-term climate variability (e.g. Hood and Zaff, 1995;Chandra et al., 1996;Hood, 1997), described e.g. by the Northern Atlantic Oscillation (NAO) (e.g. Appenzeller et al., 2000;Orsolini and Doblas-Reyes, 2003), the Arctic Published by Copernicus Publications on behalf of the European Geosciences Union.
Several studies (e.g. WMO, 2007;Harris et al., 2008) showed that the major cause of decline of total ozone at northern mid-latitudes from 1970 to the mid-1990s was the increase of stratospheric concentrations of ODS. From 1992 to 1994 the Mt. Pinatubo eruption of 1991 had a significant influence on ozone in the northern mid-latitudes. This effect has been larger than that of El Chichón in 1982 (e.g. Harris et al., 2008). Although equivalent effective stratospheric chlorine (EESC) had the largest influence on total ozone changes over the northern mid-latitudes (e.g. Mäder et al., 2007;Harris et al., 2008) dynamical changes had a strong influence too. Dynamical changes have been more important over Europe than in other regions of the world (e.g. Appenzeller et al., 2000;Staehelin et al., 2001;WMO, 2003). According to  and Maeder et al. (2007), dynamic processes could account for about 30% of the ozone decline starting in the 1970s. Koch et al. (2005) found that the increase in fast isentropic transport of tropical air to midlatitudes contributed to ozone changes over Europe between 1980 and1989. Other studies performed within the EU project CANDIDOZ (Harris et al., 2008) and summarized in the WMO/UNEP Ozone Assessment (WMO, 2007) indicate that the increase in total ozone after the mid-1990s is mainly caused by dynamic changes, while the change in ODS only contributed insignificantly to the observed ozone increase at northern mid-latitudes (e.g. Hood and Soukharev, 2005;Harris et al., 2008;Shepherd, 2008;Hegglin and Shepherd, 2009).
This study deals with the frequency distribution of extreme events in low and high total ozone (ELOs, EHOs), their influence on mean values and trends, and "fingerprints" of major atmospheric processes in total ozone. This paper is complementary to its companion paper, which deals with the basic concept of applying extreme value theory to stratospheric ozone (Rieder et al., 2010, hereafter called Part 1).

Data
In this study we analyze the world's longest total ozone record from Arosa, Switzerland, which is based on direct sun measurements with Dobson spectrophotometers and has been homogenized (for details see Staehelin et al., 1998a, b). Data on total ozone from Arosa is provided by the World Ozone and Ultraviolet Radiation Data Centre (WOUDC, http://www.woudc.org).
Several indices describing atmospheric dynamics (e.g. El Niño Southern Oscillation (ENSO), North Atlantic Oscillation (NAO)) and chemistry (volcanic eruptions, ODS in general and polar vortex ozone loss in particular) have been used in this study. Table 1 contains an overview of the datasets used (including their sources).

Methods
The Generalized Pareto Distribution (GPD) is commonly used in the analysis of rare events. It arises as the natural distribution of the excess of a variable above (below) a sufficiently high (low) threshold (e.g. Davison and Smith, 1990). The use of the GPD for modelling extremes is justified by asymptotic arguments (e.g. Pickands, 1975;Coles, 2001), as the GPD is the limiting distribution of normalized excesses beyond a threshold, as the threshold approaches the endpoint of the distribution.
The GPD (see Eq. 1) of a variable xis defined by 3 parameters: the threshold value u, the scale parameter σ and the shape parameter ξ : Here, u defines which values of the variable x can be considered as extremes, while σ takes account of the scale of the variable x (i.e., is determined by its units), and ξ determines the shape of the tail of the distribution. In the present application the task is to fit the GPD to the tails of daily total ozone data from Arosa (i.e., to very low and high values) on monthly basis. In an accompanying paper (Part 1) the application of extreme value theory, including threshold selection (based on mean residual life plots and threshold choice plots) and threshold evaluation (based on Quantile-Quantile comparisons of observed and modeled extremes and density plots) is described in more detail.  personal communication, 2009) to daily values by linear interpolation (see Part 1). Sensitivity analysis showed that the time periods for threshold selection do not significantly affect the frequency of ELOs and EHOs and that the main findings of the study are robust and do not depend on the time windows used for threshold determination (see Part 1 and its supplementary material).

Frequency of extreme events in total ozone on annual and seasonal scale
The classification of total ozone observations on days with extreme low (ELOs), extreme high (EHOs) and non-extreme days (NEOs) was described in an accompanying paper (see Part 1). The following discussion of the frequency of extreme events in total ozone on annual and seasonal basis, considering threshold values for anthropogenically and volcanically unperturbed  and perturbed  periods, is based on this classification. The observed frequency for all three groups is plotted as a time series in Fig. 1c. The earliest part of the total ozone time series of Arosa (before 1932) suffers from a calibration uncertainty and large gaps in the measurements (see Staehelin et al., 1998a, b). It appears that the first 20 years of the Arosa record reveal natural variability in ELOs and EHOs (on average ∼10%). In 1940-1942 a strong enhancement in EHOs is visible (due to a strong ENSO event) (Brönnimann et al., 2004a, b), followed by a second period of approximately 30 years of moderate natural variability. Overall the number of EHOs is slightly larger than the number of ELOs during the first 40 years of the Arosa record. EHOs are indicative for northerly advection of O 3 -rich airmasses from the high-latitude lowermost stratosphere, while ELOs are indicative for southerly winds of O 3 -poor airmasses from the direction of the subtropical jet (e.g. Koch et al., 2005). Therefore, the predominance of EHOs over Arosa, which is situated north of the Alpine main ridge, may be interpreted as a predominant influence of northerly intrusions. From the late 1970s a decrease in the number of EHOs and an increase in the frequency of ELOs are observed. This is in general agreement with the notion that the tropical band widened and the polar regime shrank over the past decades (Hudson et al., 2006;Seidel et al., 2008). During the last 5-10 years a decrease in the frequency of ELOs and a stabilization or small increase in the frequency of EHOs can be observed over Arosa, in general agreement with the notion that anthropogenic ozone depletion has come to a halt.   Table 1. total ozone and smallest frequency of days with extreme high total ozone is found in the years following the eruption of Mt. Pinatubo and in years with strong contributions of ozonepoor air after breakdown of the polar vortex. The highest frequency of EHOs and lowest frequency of ELOs is found during phases with strong El Niño Southern Oscillation.
Dynamical effects (e.g. related to ENSO or NAO) and chemical effects (e.g. contributions of strong polar vortex ozone loss) as well as combined dynamical/chemical effects (e.g. major volcanic eruptions) on column ozone over the northern mid-latitudes have been discussed in various studies (e.g. ENSO in 1940-1942(Brönnimann et al., 2004a; NAO (e.g. Appenzeller et al., 2000;Orsolini and Limpasuvan, 2001); mixing of airmasses after the breakdown of the polar vortex (e.g. Knudsen and Grooss, 2000;Fioletov and Shepherd, 2005); volcanic effects (e.g. Brasseur and Granier, 1992)). However "fingerprints" of these events have so far been identified only in particular cases, but not coherently, irrespective of the magnitude of the events. The current study enables such identification by applying extreme value theory (compare Fig. 3). Solar cycle and QBO effects (regularly included as explanatory variables in statistical regression analysis on total ozone, e.g. Mäder et al., 2007;Vyushin et al., 2007;WMO, 2007) however, are not included in our study as they show continuous, but small (1-2%), influence on total ozone. Below we show how extreme value modeling helps to attribute ozone variability to the dynamical and/or chemical mechanisms. Expected and observed dynamical and chemical effects on total ozone are discussed in detail in the scientific literature (see Sect. 3.2). From this knowledge we deduce (by visual inspection) whether "fingerprints" of such events are visible in the frequency of extremes (ELOs and/or EHOs on annual/seasonal basis). "Fingerprints" of these mechanisms are discernible in the extreme part of the Arosa time series even when no attribution is possible from mean value analysis (see Fig. 3). Below we discuss these different factors in more detail.

Atmospheric dynamics and chemistry
Figure 1 provides an overview of the temporal evolution of the various indices used to describe atmospheric dynamics and chemistry.

El Niño/Southern Oscillation (ENSO)
The El Niño/Southern Oscillation (ENSO) phenomenon is one of the most prominent modes of climate variability (Diaz and Markgraf, 2000), well known to affect climate and weather in large parts of the world. Strong ENSO events (e.g. 1940-1942 are triggered by the high contrast between high tropical and low extra-tropical Pacific sea surface temperatures (Brönnimann et al., 2004a;Brönnimann et al., 2004b), as these events are known to affect the North Pacific and surrounding areas via changes in the Hadley circulation and Rossby wave generation (e.g. Trenberth et al., 1998;Alexander et al., 2002). However, the precise extent of the influence of ENSO on European winter climate and the northern stratosphere remains a matter of debate (e.g. Friedrich and Müller, 1992;Pozo-Vazquez et al., 2001;Greatbatch et al., 2004). Other studies have shown that the ENSO winter signal over Europe consists of cold temperatures in Northern Europe, high sea level pressure from Iceland to Scandinavia, and low sea level pressure over central and western Europe (e.g. Friedrich and Müller, 1992;Merkel and Latif, 2002;Greatbatch et al., 2004). Several studies have shown that strong ENSO events are associated with a weak(er) polar vortex and more frequent stratospheric warming (van Loon and Labitzke, 1987;Labitzke and van Loon, 1999), as a stronger wave activity flux diminishes the vortex strength and strengthens the meridional circulation in the middle stratosphere, leading to enhanced transport of ozone from the tropics to the extra tropics, stronger descent over the polar region, a warmer lower stratosphere and enhanced total ozone in the Arctic in late winter (e.g. Newman et al., 2001;Randel et al., 2002). Influence of strong ENSO events on surface temperatures and column ozone over the northern mid-latitudes was shown by Brönnimann et al. (2004b).

North Atlantic Oscillation (NAO)
Interannual and decadal changes in tropospheric meteorology and lower stratospheric dynamics may be related to the strengthening of the Arctic Polar Vortex, the weakening of the Brewer-Dobson circulation, changes in the Arctic Oscillation (AO) and North Atlantic Oscillation (NAO), and to reductions in tropopause height and temperatures in the lower stratosphere (e.g. Pawson and Naujokat, 1999;Hurrell, 1995;Thompson and Wallace, 1998;Steinbrecht et al., 1998;Randel and Wu, 1999;Newman and Nash, 2000;Wallace, 2000). Total ozone in the Northern Hemisphere is influenced by modes in the variability of the atmospheric circulation, such as the Arctic Oscillation (AO) or the North Atlantic Oscillation (NAO), both affecting winter climate in Europe and the strength of the winter Arctic vortex (Appenzeller et al., 2000;Thompson and Wallace, 2000;Orsolini and Limpasuvan, 2001;Hadjinicolaou et al., 2002;Orsolini and Doblas-Reyes, 2003). Conversely, changes in ozone can also feed back to the global circulation (Volodin and Galin, 1998;Hartmann et al., 2000), although evidence that this is a major factor in the Atmos. Chem. Phys., 10, 10033-10045, 2010 www.atmos-chem-phys.net/10/10033/2010/ northern mid-latitudes is lacking (Braesicke and Pyle, 2003). The NAO is the leading mode of the interannual variability in the Euro-Atlantic sector and shows strong effects on the northern hemispheric troposphere, as it influences many meteorological parameters (e.g. temperature, precipitation). The NAO influences changes in the direction and intensity of the dominant westerly tropospheric jet stream over the Atlantic (e.g. Orsolini and Limpasuvan, 2001) and is the main driver of the inter-annual variability in storm tracks over the Atlantic during the cold season (e.g. Lau, 1988). Tropopause pressure is higher during positive NAO phases at high latitudes and lower mid-latitudes, as expected from an enhanced Icelandic low and Azores high pressure system (e.g. Appenzeller et al., 2000). During its positive phase the NAO is associated with an enhancement of ozone over Labrador/southern Greenland and a lowering of ozone over Europe (Orsolini and Doblas-Reyes, 2003), while during its negative phase (weak Icelandic low and strong Azores high) it is associated with enhancement of ozone over Europe (e.g. Brönnimann et al., 2004a;Brönnimann et al., 2004b). The NAO index (and the related AO) has changed since the 1960s (e.g. Hurrell, 1995; Thompson and Wallace, 2000;Appenzeller et al., 2000), while the Arctic polar vortex strengthened. A stronger northern polar vortex (similar to its Antarctic counterpart) is concomitant with a slower Brewer-Dobson circulation and, consequently, with lower column ozone in high latitudes (Chipperfield and Jones, 1999;Hadjinicolaou et al., 2002). Figure 1f shows the evolution of the NAO index (principal components of the leading empirically-determined orthogonal function of seasonal sea level pressure anomalies over the Atlantic sector (20 • N-80 • N, 90 • W-40 • E) from Hurrel, 2009). Several NAO fingerprints are visible during spring and winter in the extremes record (see Fig. 1 and Fig. 2a,  d). We see "fingerprints" of NAO in its negative phase (NAO Index smaller −1), leading to higher column ozone over Europe, in 1936Europe, in , 1940Europe, in -1942Europe, in , 1952Europe, in , 1955Europe, in , 1960Europe, in , 1963Europe, in , 1966Europe, in , 1969Europe, in -1970Europe, in , 1977Europe, in 1979Europe, in -1980Europe, in and 1996 as well as during its positive phase (NAO Index larger than 1), leading to lower column ozone over Europe in 1975Europe in , 1981Europe in , 1983Europe in , 1985Europe in , 1989Europe in -1990Europe in , 1992Europe in -1993Europe in , 1995Europe in , 2000Europe in , 2005Europe in , 2007Europe in and 2008. However, the influences of ENSO and contributions from strong polar ozone depletion and NAO often occur in the same years, which renders discrimination of their effects difficult. We fitted generalized linear models with quasi-Poisson error structure (e.g. McCullagh and Nelder, 1989;Hastie and Pregibon, 1992) to assess the relation between the NAO Index and the numbers of days with extreme high and low total ozone. These models show that higher NAO significantly (at 5%) increases the numbers of ELOs and decreases the numbers of days with EHOs in winter and spring, and also on an annual basis (see Table 2).

Volcanic eruptions
Gaseous compounds including SO 2 and H 2 S enter the stratosphere through volcanic eruptions (e.g. Staehelin et al., 2001). While H 2 S is very rapidly converted to SO 2 , this gas is subsequently oxidized to H 2 SO 4 , which undergoes bimolecular nucleation with water to form new aqueous H 2 SO 4 droplets or condenses on pre-existing H 2 SO 4 -H 2 O droplets. Large enhancements of aqueous sulfuric acid aerosols by volcanic eruptions strongly affect the ozone layer via two mechanisms: first, they impact the chemistry of the lower stratosphere, mainly by providing surfaces for heterogeneous processes resulting in ozone depletion; second, they alter stratospheric dynamics due to their infrared absorptivity leading to lower stratospheric heating (Brasseur and Granier, 1992;Kodera, 1994;Hadjinicolaou et al., 1997). The temporal evolution of the Sato Index (Sato et al., 1993), which describes volcanic aerosol in terms of its mean optical thickness is shown in Fig. 1g. Throughout all seasons "fingerprints" of the volcanic eruptions of El Chichón (March/April 1982) and Mount Pinatubo (June 1991) are visible in the extremes of the time series, through a strong increase in ELOs and almost complete erosion in EHOs, see Figs. 1c, 2 and 3b. After the eruption of El Chichón strong influence of ELOs on observed ozone mean values is visible mainly during winter 1982/1983 and spring 1983 (see Fig. 2a and d). Even larger influence on the frequency of extreme events and their influence on observed mean values (>20 DU) is obvious after the eruption of Mount Pinatubo in 1992 (all seasons) and winter and spring of 1993. For the third largest volcanic eruption, Gunung Agung in May 1963, the extremes record shows less clear "fingerprints". However, a strong ENSO event that occurred in the same year may disguise the influence of the volcanic eruption on column ozone.

Strong polar vortex ozone loss and ozone depleting substances
In the early 1950s the industrial chemical production of ozone-depleting substances (ODS) began. Shortly after the publications of Molina and Rowland (1974) and Rowland and Molina (1975), the deployment of CFC-11 and CFC-12 as spray propellants was prohibited in some countries, leading to a reduction in their worldwide production. However, emissions of these compounds increased again from 1982 up to 1987. The negative impact on column ozone of anthropogenically produced CFCs has been widely studied (e.g. Molina and Rowland, 1974;Cicerone et al., 1974;Rowland and Molina, 1975;WMO, 2003WMO, , 2007. In 1987 the representatives of national governments reached an agreement, the Montreal Protocol, to reduce the global emission of substances leading to stratospheric ozone destruction (e.g. Staehelin et al., 2001). It led to a decrease in global production and therefore also in release of CFCs and other ODS. Depending on their physical and chemical properties, the various halogen compounds have different potentials to deplete ozone. Mostly they are described in terms of EESC (equivalent effective stratospheric chlorine) which is the sum of chlorine containing species multiplied by the number of Cl atoms contained in the compound and weighted by their fractional stratospheric release rate. EESC also accounts for the effects of bromine species (e.g. WMO, 2003). EESC increased from the 1950s on, peaking in 1997 and then started to decline slowly. The evolution of EESC in the Earth's atmosphere is shown in Figure 1h. A drawn-out "fingerprint" of the influence of ODS is visible within the increasing frequency of ELOs in the annual means (Fig. 1c) and in all seasons except autumn (Fig. 2a, b, d) from the 1970s/80s onward.
Enhanced ozone depletion at high northern latitudes, propelled by heterogeneous chemistry on polar stratospheric clouds (PSCs), is regularly observed in late winter and early spring. Ozone depletion is caused by activation of chlorine on PSCs that then catalytically destroys ozone inside the polar vortex (e.g. Peter, 1997;Solomon, 1999). In the chemically undisturbed stratosphere the breakdown of the polar vortex is expected to cause higher ozone values in late winter and spring than under chemically disturbed conditions (ODS). This ODS-induced polar ozone loss has a significant influence on ozone trends observed at mid-latitudes due to mixing of polar air masses after the breakdown of the vortex in northern springtime (Knudsen and Grooss, 2000;Fioletov and Shepherd, 2005).
Contributions of strong polar vortex ozone loss to column ozone over Arosa can be identified from the extremes record. Values of polar ozone loss contributions to total ozone at the northern mid-latitudes (see Fig. 1j) have been taken from . From the mid-1980s onward, during late winter and spring, a strong decline in EHOs is visible for the Arosa record (see Fig. 1c), in particular in spring and winter (see Fig. 2a, d). This decline in EHOs may be related to transport of ozone-rich air from high latitudes; however, since the massive increase in ODS in the early 1980s these ozone-rich events were gradually quenched by high latitude ozone depletion, affecting the Arosa measurements, in particular after the breakdown of the polar vortex. This impact peaks in the spring seasons of 1989-1990, 1993, 1995-1997, 2000 and 2002, when (almost) no EHOs occurred at all in the Arosa record. While the frequency of EHOs dropped during these periods the frequency of ELOs peaked at the same time. Polar ozone loss might have been so large that EHOs in the chemically unperturbed period were replaced by ELOs when winter temperatures were particularly cold. Furthermore transport by southerly winds in the lowermost stratosphere (Koch et al., 2005), also led to an enhancement in ELOs. The latter could be due to dynamic changes such as a broadening of the tropical belt (Hudson et al., 2006), or due to in situ processing in mid-latitudes (Solomon, 1999), or to both.

Influence of extreme events on observed means and trends in total ozone
This section considers the impact of extreme events on observed mean values and trends in total ozone. This is an interesting question, as we could show that physical and chemical phenomena leave stronger "fingerprints" in the tails of the ozone distribution than in the averaged quantities derived from the dataset (see Fig. 3).
To analyze the influence of extreme events on observed mean values, we excluded (i) the ELOs, (ii) the EHOs, and (iii) both ELOs and EHOs from the Arosa time series. The seasonal averages (Fig. 4a-d) and the annual average (Fig. 4e) of total ozone are shown for the original observed time series (red curves), the ELOs removed from the time series (light blue curves), the EHOs removed from the time series (dark blue curves) and the time series after removing both extremes (black curves). Figure 4f shows smoothed annual data series before and after removal of extremes to highlight long-term trends. Here the data were smoothed by applying LOESS (LOcally wEighted Scatterplot Smoothing) (e.g. Cleveland, 1979;Cleveland and Devlin, 1988;Cleveland et al., 1990) with a smoothing parameter q = 2/3. First, we note that the inter-annual variability is strongly reduced when the extremes are removed, indicating that the infrequent extremes strongly influence annual values (as well as seasonal means). During the first decades the observed seasonal averages and the seasonal averages of the ELOsremoved time series follow each other rather closely throughout all seasons (Fig. 4a-d). Conversely, the EHOs-removed averages are in good agreement with the observed averages from the mid-1980s on. Furthermore, the linear downward trend (for 1970-1990) is much weaker throughout all seasons after both extremes are removed, and the annual trend is reduced by a factor of 2.5, see Fig. 4f and Table 3. The strongest trend reduction on the seasonal scale can be observed in spring (about a factor of 3.5), when polar processes add to the general mid-latitude ozone depletion. Table 3 Atmos. Chem. Phys., 10, 10033-10045, 2010 www.atmos-chem-phys.net/10/10033/2010/ provides a comparison of linear trends per decade for the original time series and the series after removal of extremes on annual and seasonal basis. From these results it can be concluded that variability and trend in total ozone observations at Arosa are dominated by extreme events in low and high total ozone. As discussed in Sect. 3.2 the influences of ENSO and NAO (in its negative phase) are captured in the frequency of days with extreme high total ozone, while the influence of NAO (in its positive phase) and influence of volcanic eruption are captured in the frequency of days with extreme low total ozone. The influence of ODS and contributions of strong polar vortex ozone loss after its breakdown through mixing of air masses and transport to northern mid-latitudes are discernible in both tails. However, contributions from polar vortex ozone loss are revealed more clearly in the decreasing frequency of EHOs. Solar cycle and QBO effects are not included in our analysis as they show continuous (not related to the tails) and only small (1-2%) influence on total ozone. For annual means, as well as for spring-summer-autumn, we find a stronger trend reduction for the time series without EHOs than for those without ELOs, while in winter the trend reduction is larger when ELOs are removed. Trend reduction is largest in spring, where especially contributions of low ozone from north (after the breakdown of the polar vortex) to Europe can be considered as major influencing variable. In contrast, in winter the time series without ELOs shows a stronger trend reduction, which partly might be related to the mode of the North Atlantic Oscillation, as reported by Appenzeller et al. (2000). Ranking of the influencing factors is difficult in an extreme value analysis context, and determination of the exact contribution of dynamical and chemical factors on ozone changes is difficult as they sometimes occur simultaneously, partly compensating or amplifying each other. As ozone depleting substances are considered to affect both types of extremes (due to enhanced chemical ozone depletion in mid-latitudes and enhanced polar vortex ozone loss) a significant effect of anthropogenical chemical influence on column ozone changes is obvious. However, "fingerprint" analysis shows that dynamical factors influence column ozone too and much more frequently and strongly than previously thought. Several studies estimated an influence of 1/3 (or even higher) of dynamical contributions to ozone changes (e.g. Hood and Soukharev, 2005;WMO, 2007;. From the frequent "fingerprints" found for ENSO and NAO within the presented analysis we conclude that the presented results are in general agreement with these studies. For the individual seasons it seems that the winter trend is more strongly influenced by dynamical changes (through changes in the polar vortex) than those of spring. Importantly, strong ENSO events also contribute significantly to high column ozone during springtime, an effect that might be disguised by the dominating influence of polar vortex contributions during the past two decades.
To quantify the importance of ELOs and EHOs for the Arosa record the influence of ELOs (Eq. 2) and EHOs (Eq. 3) on seasonal and annual averages in total ozone were calculated: Here I ELOs (I EHOs ) is the influence of extreme events in low (high) total ozone on the seasonal or annual mean value, M ELOsex (M EHOs(ex) ) is the average total ozone with ELOs (EHOs) excluded from the time series, M AROSA denotes the average total ozone observed over Arosa and t denotes to the averaging time period (seasonal or annual basis). The influence of ELOs, EHOs and both extremes on annual means of total ozone over Arosa is shown in Fig. 5e, while Fig. 5a-d shows the influence on seasonal averages. In general, patterns identified on seasonal scale, described below, persist also for the influence of ELOs and EHOs on observed annual averages.
The variability in the column ozone, the extremes and the influence of ELOs and EHOs are all generally smaller during fall and summer than during spring and winter. Throughout all seasons, a dominating influence of EHOs is visible until the late 1970s. The influence of ELOs increases from the mid-1970s on, becoming dominant during the 1980s.
ELOs have strongest influence (e.g. >−20 DU) on observed mean values in winter (see Fig. 5d) in 1932in , 1949in , 1953in -1954in -1993in and spring (see Fig. 5a) in 1938in -1939in , 1985in , 1993 In almost all of these years strong influence of NAO (positive phase) and/or contributions of polar vortex ozone loss can be identified. Furthermore, the eruption of Mt. Pinatubo (June 1991) led to an increase in the total number of ELOs and their contribution to mean values in the years after the eruption. The average influence of ELOs on observed mean values of column ozone is about −9 DU in spring, −3 DU in summer, −5 DU in autumn and −12 DU in winter.

Conclusions
We found in all seasons that a large fraction of the variability and trend in the total ozone time series at Arosa (Switzerland) can be attributed to extreme events (ELOs and EHOs). Clear "fingerprints" of atmospheric dynamics (ENSO, NAO), the major volcanic eruptions of the 20th century (Gunug Agung, El Chichón and Mt. Pinatubo) and atmospheric chemistry (contributions of ODS and in particular polar vortex ozone loss) are discernible in the frequency distributions of extreme events. Fingerprint analysis of the tails in the ozone distribution function, i.e. of the extreme events, allows attribution of the influence of the chemical and dynamical factors in a more systematic manner than a mean value analysis can achieve. Fingerprint analysis reveals that these chemical and dynamical factors affect total ozone much more frequently and strongly than thought previously. In comparison to previous studies (e.g. Appenzeller et al., 2000;Orsolini and Limpasuvan, 2001;Orsolini and Doblas-Reyes, 2003;Brönnimann et al., 2004b;Sassi et al., 2004) we showed that moderate ENSO and NAO events also have a significant influence on total ozone. During years with strong polar vortex ozone loss we could further show that the frequency distribution of total ozone changes significantly; especially interesting is the almost complete erosion in extreme high values during the past three decades. Strong influence of volcanic eruptions has previously been documented in the scientific literature; however, from our results we note to see that their impact influences both tails (strong increase in ELOs and erosion in EHOs), suggesting that the depletion is ubiquitous. It is further important to note that various chemical and dynamical factors sometimes occur simultaneously, partly compensating or amplifying each other (e.g. ENSO, NAO, and polar vortex ozone depletion), so that the individual contributions to a specific fingerprint are hard to distinguish. Further it was shown that the influence of EHOs and ELOs on seasonal and annual averages changed strongly with time. While EHOs show a strong influence on seasonal and annual means during the first half of the 20th century, the influence of ELOs on the means became larger starting in the mid-1980s. The Atmos. Chem. Phys., 10, 10033-10045, 2010 www.atmos-chem-phys.net/10/10033/2010/  Switzerland, 1927Switzerland, -2008 comparison of the means of the original observation record with those from the time series with (i) ELOs, (ii) EHOs, and (iii) both extremes removed confirmed these findings and showed further that trends are strongly reduced (on average by a factor of 2.5) when both extremes are removed. The application of extreme value theory provides deeper insight into time series properties and a more complete attribution of the influence of dynamical and chemical factors on total ozone variability than the analysis using standard metrics would allow.