The development of tracer-tracer relations in the polar stratosphere is analyzed during the period when the vortex forms and a westerly circulation develops after polar summer (the setup phase of the polar vortex). We consider high southern latitudes from March to June for winter 1997 and 2003 and high northern latitudes from September to October 2003. ILAS and ILAS-II satellite observations and model simulations are used to investigate chemical changes in O3, NO2 and HNO3 during these periods. Tracer-tracer relations and meteorological analyses consistently indicate a separation of the incipient polar vortex into two parts. The area within the edge of the inner vortex is isolated from the outer part that is still influenced by mixing with air of midlatitude origin. In the Antarctic in April, ozone concentrations vary by about 0.5 ppmv within the isolated inner vortex between 500 and 600 K potential temperature. This inhomogeneous distribution of ozone is likewise obvious in MIPAS satellite measurements. Box model simulations explain that the low ozone concentrations in April are caused by chemical ozone loss due to catalytic cycles which are mainly driven by NOx at this time of the year. The simulations also explain the observed conversion of NOx to HNO3 during the setup phase of the 2003 Antarctic vortex. During June in the Antarctic, the internal vortex transport barrier disappears and ozone mixing ratios become homogeneous throughout the entire vortex. At that time, no further ozone loss occurs because of the lack of sunlight.
 During the past decade much research has been devoted to chemical ozone loss in the Arctic winter lower stratosphere (LS). Ozone mixing ratios in the polar winter are controlled by both dynamical and chemical processes in the polar vortex. The descent of air in the polar region of both the Arctic and Antarctic vortex [e.g., Russell et al., 1993; Rosenfield et al., 1994], which is most effective at the beginning of the winter [e.g., Müller et al., 1996; Kawamoto et al., 2004; Konopka et al., 2004], increases ozone mixing ratios at a given altitude in the LS during winter and spring. According to current understanding, polar ozone loss in the LS occurs in winter and spring within the polar vortex and is mainly due to ClOx and BrOx chemistry, starting with the illumination of the polar region after the darkness of winter [e.g., Solomon, 1999]. Owing to the existence of polar stratospheric clouds (PSCs), chlorine is activated from its reservoirs. Effective catalytic cycles driven by active chlorine and bromine destroy ozone. In spring, the polar vortex weakens, becomes distorted, and finally breaks down. A rapid easterly acceleration occurs in conjunction with warming of polar air resulting in the formation of the summertime anticyclone. During polar summer, ozone is destroyed by NOx-catalyzed cycles [e.g., Farman et al, 1985a; Brühl et al., 1998; Fahey and Ravishankara, 1999; Crutzen and Brühl, 2001]. Therefore ozone mixing ratios decrease in the LS until the polar vortex starts to form again in fall.
 At present, few observations are available for investigating the dynamical and chemical processes controlling ozone and other tracers during the formation of the polar vortices. Kawa et al.  analyzed the evolution of ozone during the setup phase of the Northern Hemisphere (NH) polar vortex in fall 1999, using in situ data from the Observations of the Middle Stratosphere (OMS) balloon platform, satellite data from the Polar Ozone and Aerosol Measurement III (POAM III) instrument and a chemistry and transport model (CTM). The study by Kawa et al.  shows that in late summer minimum ozone values and their variance in high northern latitudes are the result of two processes. First, with increasing wave-driven transport in fall, ozone-rich air is transported from lower latitudes poleward whereas ozone-poor air is located close to the pole [Brühl et al., 1998; Kawa et al., 2002]. Second, net photochemical loss is accelerated by decreasing solar illumination because the shorter wavelengths necessary for ozone production are absorbed more strongly at low solar elevations in autumn whereas ozone loss from radical catalytic cycles is still effective [Perliski et al., 1989; Farman et al., 1985b]. Therefore enhanced chemical ozone destruction occurs during this time of the year. In the study by Kawa et al. , only one particular Arctic fall was investigated.
 In the present study, the formation process of the vortex is discussed for two Antarctic and one Arctic winter using analyses from the UK Meteorological Office (MetO) [Swinbank and O'Neill, 1994]. The existence of a polar vortex edge is defined by the existence of a maximum vorticity gradient (∇PV) constrained by the wind velocity (v) with regard to equivalent latitudes, as proposed by Nash et al. . Here, the value of the maximum vorticity gradient times wind velocity (∇PV · v) is employed to identify an additional transport barrier located within the vortex edge. Measurements of ILAS (Improved Limb Atmospheric Spectrometer) aboard ADEOS (Advanced Earth Observing Satellite) and ILAS-II aboard ADEOS-II show a separation of air masses located in the different regions of the vortex using the tracer-tracer correlation method, a method for separating chemical ozone loss from dynamical processes [e.g., Proffitt et al., 1993; Tilmes et al., 2003, 2004; Müller et al., 2005]. A study by Lee et al.  found two distinct regions of the Antarctic vortex between July and the end of October, based on model simulations, where the inner vortex remains isolated between late winter and spring. Here, we show that the inner vortex was only intermittently present during May and June in winters 2003 and 1997 and during the entire Antarctic winter 2003 [Tilmes et al., 2006b].
 To identify chemical changes and calculate chemical ozone loss during winter and spring using tracer-tracer correlations, an early winter reference function is derived to describe the chemically undisturbed conditions inside a sufficiently isolated early vortex [e.g., Müller et al., 2002, 2005; Tilmes et al., 2003, 2004]. A threshold value used to define a sufficiently impermeable vortex is given below. The development of the early Arctic vortex 1996–1997 from November to January was earlier analyzed by Tilmes et al.  on the basis of version 6 ILAS observations. In that study, it was shown that it is important to carefully determine the time in winter when ozone mixing ratios are homogeneous at a certain N2O level to determine a valid early winter reference function. This is the time when the vortex is isolated and compact O3/N2O correlations are established. In a very weak early vortex, mixing across the vortex edge may change tracer-tracer correlation in such a way that larger ozone mixing ratios result and chemical ozone loss may therefore be underestimated. Müller et al.  discussed the influence of mixing processes on tracer-tracer correlations in the polar region in detail and showed that mixing processes cannot change the deduced chemical ozone loss significantly in an isolated polar vortex. Here, it will be shown that tracer-tracer correlations within the vortex of the Southern Hemisphere (SH) in June 2003 describe a compact relationship at a time when the transport barrier at the edge of the vortex is being established.
 In this paper, ILAS and ILAS-II observations will be described in section 2. The setup phase of the 2003 Antarctic vortex will be investigated in section 3 on the basis of the analysis of MetO data and on ILAS-II observations. Further, a comparison with simulations of the Chemical Lagrangian Model of the Stratosphere (CLaMS) box model [McKenna et al., 2002a, 2002b] and with the Michelson Interferometer for Passive Atmospheric sounding (MIPAS) observations [e.g., Fischer and Oelhaf, 1996] is performed for the 2003 Antarctic winter. We utilize the ESA MIPAS reprocessed Level 2 data set with Instrument Processing Facility version 4.61. In section 4, the setup phase of the 1997 Antarctic vortex is described on the basis of meteorological analyses and ILAS observations and a comparison is made between the two Antarctic winters 2003 and 1997. The additional use of ILAS-II observations in September and October 2003 and ILAS version 6 observations will allow further insights into the setup phase of the Arctic vortex as described in section 4. Finally, conclusions are given in section 5.
2. Measurements by ILAS and ILAS-II
 Measurements by the two instruments ILAS (Improved Limb Atmospheric Spectrometer) aboard ADEOS and ILAS-II aboard ADEOS-II were used in this study. These two instruments measured atmospheric trace gases using the solar occultation technique at high latitudes in both hemispheres (between 54°N and 71°N and 64°S to 88°S). Observations of the altitude distribution of ozone, HNO3, N2O, CH4, and of the aerosol extinction coefficient at 780 nm data from cloud top up to 70 km are available. Vertical profiles of other atmospheric trace gases, measured by ILAS-II, such as NO2, H2O, ClONO2, and N2O5, were also retrieved during the entire measurement period, but have not been validated yet and will not be used in this study. The accuracy of the ozone is estimated to be better than ±10% [Sugita et al., 2006].
 ILAS performed observations for 8 months between October 1996 and June 1997. These measurements cover the entire Arctic winter 1996–1997 [Suzuki et al., 1995; Sasano et al., 1999]. ESA MIPAS reprocessed Level 2 data set with Instrument Processing Facility version 4.61. ILAS-II observations covered the period from March to October 2003, during an entire Antarctic winter period and the setup phase of the 2003 Arctic vortex [Nakajima et al., 2006]. To investigate the development of the Antarctic polar vortex in 1997 and in 2003, March to June observations from both data sets are used.
 ILAS-II version 1.4 and ILAS version 6.0 ozone, N2O, HNO3, and NO2 (only from ILAS (T. Yokota et al., Improved Limb Atmospheric Spectrometer-II (ILAS-II) version 1.4 algorithm for retrieval of gas and aerosol profiles in the stratosphere, manuscript in preparation, 2006)) are used in this study. Applying the tracer-tracer correlation method, N2O is used as a long-lived tracer for both the ILAS and ILAS-II analysis. For ILAS, the uncertainty for N2O data is better than 10% over 10–30 km, which is of sufficient quality to be used here [Kanzawa et al., 2003]. For ILAS-II, N2O is up to 20% too low for mixing ratios less than 100 ppbv during March to June 2003 in the SH compared to measurements of the Submillimeter Radiometer (SMR) (v1.2) aboard the Swedish minisatellite ODIN [Ejiri et al., 2006]. This is also the case during September and October 2003 in the NH. However, no significant seasonal change is found in this low bias. Thus, although N2O likely has an instrumental offset compared to SMR measurements (which depends on altitude), since the offset does not vary seasonally, it will not affect the results of tracer-tracer correlations.
3. Setup Phase of the Antarctic Vortex 2003
 Tracer-tracer correlations are used to investigate the development of ozone, HNO3 and NO2 inside the incipient polar vortex. (Here, the setup phase of the vortex, the period September/November for the Arctic [Kawa et al., 2002], and March/May for the Antarctic, is referred to as the “incipient” vortex. In earlier work, the phase of the developing vortex in March/April in the Antarctic and in November in the Arctic has been referred to as the proto-vortex [e.g., Michelsen et al., 1998].) This section presents results from March to June 2003 for the Antarctic.
 The edge of the polar vortex is determined here according to the definition of Nash et al. , whereby the edge is defined as being colocated with the maximum potential vorticity gradient (∇PV) that is closest to the polar night jet. Nash et al.  used 15.2 m/s as an empirically determined limit below which a vortex edge cannot be defined. As soon as a vortex edge can be identified in fall according to this definition, we assume that a vortex has started to form. At first, the equivalent latitude filter of the Nash et al.  algorithm was used to filter transport barriers between 45 and 80°. In this way, the detected vortex edged switched between two transport barriers that are almost equally strong (not shown). Therefore, in order to identify both the two potential transport barriers at the same time, we applied this equivalent latitude filter twice, for the interval between 40° and 70° and for the interval poleward of 70°. The threshold value of 70°S is empirically determined considering the pattern of ∇PV · v with regard to equivalent latitudes and time (as described below, see Figure 1).
 We define the edge derived poleward of 70°S as the “inner vortex edge” and the area inside as the “inner vortex.” Further, the edge derived equatorward of 70°S is defined as the “outer vortex edge” and the area inside is referred to as the “outer vortex.” If two vortex edges are present at the same time, the region within the poleward edge of the outer/inner vortex is referred to as “outer/inner vortex core.” If only one vortex edge exists the enclosed area is referred to as the “entire vortex” as shown in Figure 2. (One edge at exactly 70°S will also be referred to as the entire vortex edge.)
 The potential vorticity gradient, ∇PV (solid line), the wind velocity (dotted line) and the value ∇PV · v (dashed line) are shown in Figure 2. A wind jet is colocated with both the inner and outer maximum PV gradient suggesting the existence of a double edge structure of the vortex during this time of the year.
 In Figure 1 ∇PV · v is shown as a function of time over the setup phase of the vortex at 475, 550, and 650 K for the Antarctic winter 2003. Here and in the following, the modified PV [Lait, 1994] is used to allow the comparison of ∇PV · v at different altitudes. White crosses show the location of the vortex edge according to the definition by Nash et al. . However, the existence of a vortex edge derived with this criterium is not sufficient to assume an impermeable polar vortex edge. Steinhorst et al.  performed a CLaMS model study to estimate a threshold value of the maximum PV gradient (∇PV), which characterizes a “sufficiently impermeable” vortex edge for the Arctic winter 1999–2000. Using the modified potential vorticity [Lait, 1994], this threshold value was found to be 1.5 PVU/degree. Further, Steinhorst et al.  showed that if the maximum ∇PV is less than this threshold value, mixing might occur locally in some segments with weak PV gradients, but mixing into the vortex was not found to occur across the whole contour. Here, we assume that the polar vortex is isolated if the maximum of modified PV gradient times wind velocity exceeds a threshold value of 22.8 PVU/degree · m/s, derived using the threshold value 1.5 PVU/degree [Steinhorst et al., 2005] times 15.2 m/s [Nash et al., 1996].
 The polar vortex started developing at the beginning of March 2003 at altitudes above 475 K potential temperature. In the first half of March, there is one vortex edge at 70°S equivalent latitude (see Figure 1, top and middle, white plus signs). However, the maximum ∇PV · v is far below the threshold value for an isolated vortex. Therefore the degree of isolation of the vortex is expected to be small. During the second half of March and the first half of April, ∇PV · v shows two relative maxima that exist almost continuously poleward of 50°S at 650 K. Correspondingly, there is an inner and outer vortex edge (see Figure 1, top). At the end of March the polar vortex consists of two parts that are separated, because each of the maxima exceeds the threshold value for an isolated vortex (defined above) and might constitute a transport barrier. At 550 K, during the first half of April, the inner vortex edge exists almost continuously whereas an isolated outer vortex does not exist continuously before mid-April 2003 (see Figure 1, middle).
 The described synoptic situation is illustrated for 3 days in Figure 3. The PV distribution at two different potential temperature levels (650 K and 550 K) indicates that the vortex consists of two isolated regions. Further, the different characteristics of the two air masses, that is lower ozone mixing ratios within the inner vortex and higher ozone mixing ratios outside the inner vortex edge, are obvious at the beginning of April (as described in the following).
 Indeed, in Figure 4 back trajectory calculations show that air masses observed by ILAS-II poleward of 75°S equivalent latitude stay within the area of the inner vortex between 10 March and 2–4 April 2003, at altitudes above the 150 ppbv N2O level; ≈ at 500 K potential temperature during this time of the year (see Figure 5, top).
 Therefore the inner vortex is well isolated from the outer vortex during this period. At altitudes below 500 K, air masses arrive on average from equivalent latitudes equatorward of 75°S. This is in agreement with the fact that no vortex edge was calculated at 475 K so that the polar area is less isolated at these altitudes.
 An outer vortex edge has existed continuously since the end of March at 650 K and since mid-April at 475 and 550 K. The entire vortex can now be expected to be isolated from air outside the vortex, because the local maximum ∇PV · v clearly exceeds the threshold value (defined above), a behavior also described in numerous earlier studies [e.g., Juckes and McIntyre, 1987; Pierce and Fairlie, 1993; Manney et al., 1994]. In section 3.1, the tracer development during the setup phase of the vortex will be discussed within this meteorological context.
 In Figure 6, the averaged solar illumination in sunlit hours per day is shown at 550 K for different equivalent latitudes between 55°S and 90°S. Average values of sunlit hours for different time intervals and equivalent latitudes are summarized in Table 1. In early March, the area within equivalent latitudes poleward of 55°S is strongly illuminated with more than 12 sunlit hours per day. In the second half of March and the first half of April, solar illumination decreases depending on equivalent latitude. From the second part of April, the vortex is hardly illuminated by the sun anymore at 80°S. After May 2003, the entire area poleward of 80°S is in darkness. Interestingly, the inner vortex edge seems to disappear at the time when the polar night reaches the location of the inner vortex edge.
Table 1. Sunlit Hours Per Day at 550 K Calculated From UK Meteorological Data Within Different Equivalent Latitudes
 At the beginning of the Antarctic winter in March, MIPAS observations taken over the entire polar region of the SH show very inhomogeneous ozone concentrations in the recently established and thus weak inner vortex at 650 K potential temperature (see Figure 7a). Further, ozone-rich air is transported to the polar region by the Brewer-Dobson circulation and increases the ozone mixing ratios in the polar region. During the second half of March and April, patches of air with low ozone mixing ratios mix with those with higher ozone mixing ratios within the inner vortex (see Figures 7b–7d). During April, an inner and an outer transport barrier seems to exists, although partly intermittently as derived using the Nash et al.  criterion (see Figure 1, top). In May, although the inner vortex edge no longer exists, air masses with smaller ozone mixing ratios are still close to the pole surrounded by air with greater ozone mixing ratios. Air masses slowly mix and homogenize within the established entire vortex. During June, ozone mixing ratios in the entire vortex are almost homogeneously distributed, only a few patches of air with low ozone mixing ratios still exist, as shown in Figures 7g and 7h.
 During the development of the vortex, both changing transport conditions (described above) and changing chemical processes, because the solar radiation that reaches the area around the south pole decreases, influence the mixing ratios of the chemical constituents. A detailed description of these processes is given in the following.
3.1. O3/N2O Development During the Setup Phase of the Antarctic Vortex 2003
 O3/N2O profiles measured inside the inner vortex core and inside the core of the entire vortex between 475 and 650 K from March to June 2003 are presented in Figure 8. Additionally, ozone at different equivalent latitude regions for two specific N2O levels is shown as a time series between 19 March and 22 May to visualize the temporal development of ozone mixing ratios as a function of equivalent latitudes (see Figure 9).
 The first ILAS-II observations inside the developing Antarctic vortex were obtained from 19 and 22 March 2003 (see Figure 8). During this period, no inner polar vortex exists below 550 K. The O3/N2O relationship for the profiles measured in the core of the entire vortex can be described by an empirically derived reference function that is established for the time prior to polar night (see Figure 8a, black line). Using N2O as a long-lived tracer (mixing ratios in ppbv) and ozone (mixing ratios in ppmv) this incipient vortex reference function (valid for 30 ppbv < N2O < 300 ppbv) is:
The standard deviation of the O3/N2O relationship is σ = 0.18 ppmv, which will be defined as the range of uncertainty of the reference function. Between 1 and 14 April 2003, an inner vortex edge is observed above 475 K and additionally, an outer vortex edge exists above 550 K, as shown in Figure 1. ILAS-II profiles are available for both of these vortex regions. A separation between profiles inside the inner vortex region (Figure 8b, green profiles) and inside the outer vortex is clearly observed for N2O < 150 ppbv (Figure 8b) and above 500 K in Figure 5 (bottom). N2O profiles show the lowest mixing ratios at a given potential temperature level for profiles measured within the inner vortex core, which means the vortex air masses that descended most strongly at the highest latitudes.
 Profiles inside the inner vortex region and poleward of 80°S in early April scatter below the incipient vortex reference function estimated from March profiles (see Figure 8b) by up to 0.5 ppmv ozone mixing ratios, especially at the 100 ppmv N2O level (550 K, see Figure 5 (bottom)). Figure 9 shows low ozone mixing ratios poleward of 77.5°S equivalent latitude with a variability of more than 0.5 ppmv during the first part of April 2003. Although the inner polar vortex is isolated (as shown above), ozone in the inner vortex region is not homogeneously distributed during the development of the polar vortex, as shown in Figure 7. Further, measurements located equatorward of the inner vortex edge and 77.5°S (Figure 9) are characterized by significantly greater ozone mixing ratios (up to ≈1 ppmv) compared to values described by the incipient vortex reference function. These values also show a variability of more than 0.5 ppmv around their average values. These observations are consistent with the interpretation that the outer region of the entire vortex is influenced by mixing in of ozone-rich air from outside the vortex, because no outer vortex edge was observed at the beginning of April below 650 K (see Figure 1).
 During the first part of April, several ILAS-II ozone measurements show mixing ratios that are significantly smaller than the values describing the incipient vortex reference function. Patches of very low ozone concentrations exist at lower latitudes that are not sampled by ILAS-II in March, as shown in Figure 7b. However, during April, some of the ILAS-II observations are located within the very low ozone patches (see Figures 7c and 7d). Therefore the significantly smaller ozone mixing ratios in early April visible in both the O3/N2O scatter plots and the potential temperature profiles may be a result of different sampling of the ILAS-II instrument in late March and early April. On the other hand, a reduction of ozone in the isolated inner vortex can be caused by chemical ozone loss [Kawa et al., 2002], see section 3.3.
 In a similar manner to the deduction of chemical ozone loss in late winter and early spring [e.g., Tilmes et al., 2004], the tracer-tracer correlation method can be employed to derive chemical ozone loss in the isolated inner vortex in April 2003 using ILAS-II observations. For this purpose, we include the possible effect of a different sampling in the analysis. It has to be kept in mind that the influence of possibly occurring isentropic mixing across the inner transport barrier of the vortex might result in increasing ozone mixing ratios. This was shown for the polar vortex in late winter by Müller et al. . The amount of chemical ozone depletion that occurs at this time of the year will be further investigated using CLaMS box model simulations (see section 3.3).
 During 15–30 April, the separation between ozone concentrations at different equivalent latitudes is more pronounced (Figures 8c and 9). The profiles located inside the inner vortex core (Figure 8c, green symbols) still show the lowest ozone mixing ratios with a slight increase compared to early April at altitudes corresponding to N2O ≈ 100 ppbv (500 K, see Figure 5). The entire polar vortex is isolated from the surrounding midlatitude air, as described above, and, correspondingly, ozone mixing ratios of the entire vortex do not further increase. For the inner vortex, no further decrease of ozone mixing ratios, i.e., further chemical ozone depletion, is detected in accordance with the lack of insolation (see Figure 6).
 Between the end of April and the beginning of May, ozone-tracer relations indicate a decrease of ozone measured at equivalent latitudes between 70°S and 75°S (Figures 8c and 8d, blue diamonds, and Figure 9). Likewise, during May and the beginning of June, ozone mixing ratios decrease slightly between 60°S and 70°S for N2O values smaller than 150 ppbv (Figures 8d–8f, red diamonds). Thus the effect of ozone depletion owing to decreasing solar radiation is observed at all equivalent latitudes sampled by ILAS-II. Further, at the end of May 2003, observations made outside the vortex display completely different ozone-tracer relations characterized by greater ozone mixing ratios (not shown). Thus the entire vortex is isolated at this time of the year. By June, insolation goes to zero for all equivalent latitudes in the entire vortex and ozone mixing ratios become slowly uniform because of mixing within the vortex. Still in early June, the lowest ozone mixing ratios are measured at the equivalent latitudes furthest poleward (Figure 8f, cyan and blue profiles). However, the very low ozone mixing ratios of ≈2 ppmv and less for N2O < 150 ppbv observed during the first 2 weeks of April have vanished.
 A new compact O3/N2O relationship has developed by mid-June, when the vortex is well isolated and ozone in the entire vortex has become uniform. This new relationship, characterized by larger ozone mixing ratios compared to the incipient relationship, is used as an early winter reference function to derive chemical ozone depletion in late winter and spring, after the polar night as in the work by Tilmes et al. [2006b].
3.2. HNO3/N2O Development During the Setup Phase of the Antarctic Vortex 2003
 During the setup phase of the 2003 Antarctic vortex, any observed ozone depletion cannot be a result of chemistry due to activated halogen compounds because no supercooled ternary solution particles (STS) or polar stratospheric clouds (PSCs) were observed by the MIPAS before mid-May [Spang et al., 2005]. Below 40 km, the most important ozone loss cycle involves NOx radicals [Crutzen, 1970]:
 With decreasing illumination between March and April, a larger amount of nitrogen is contained in longer-lived odd-nitrogen reservoir gases and therefore NOx decreases. Besides N2O5, the major nitrogen reservoir is HNO3. During the polar night NO2 is very low. However, during April, enough sunlight, 4.2 sunlit hours per day at 70°S (see Table 1), is still present and NO2 and N2O5 are photolyzed rapidly so that a chemical equilibrium between NO2 and NO develops. Thus ozone can be chemically destroyed by the NOx-catalyzed cycles (equation (2)).
 As for O3/N2O, an empirical relationship of HNO3/N2O can be derived for 19–22 March 2003 for the incipient vortex (see Figure 10a, black line). In the case of HNO3, two reference functions were developed, one valid for altitudes above about 500 K, that is, between the 30 and 125 ppbv N2O level, and a second for altitudes below about 500 K, that is, between the 125 and 300 ppbv N2O level. Using empirical polynomial functions to describe the HNO3/N2O relationship, the best fit was derived by separating the relationship into two N2O intervals.
 Between 19 and 22 March (Figure 10a) and 1 and 14 April (Figure 10b) an increase in HNO3 mixing ratios is observed at all equivalent latitudes for N2O < 150 ppbv (below 500 K). The increase of HNO3 likely occurs in accordance with decreasing NOx owing to decreasing solar illumination at high latitudes, as was observed for the Antarctic winter 1997 (see section 4, below).
 Between April (Figure 10b) and the end of May (Figure 10e), HNO3 mixing ratios increase from 12 ppbv to 15 ppbv above the 80 ppbv N2O level (above 550 K). The largest increases occur poleward of 75°S equivalent latitude (Figure 10e). HNO3/N2O profiles in April separate smoothly as a function of equivalent latitude up to mid-May in Figure 10d. With decreasing sunlit hours per day at lower latitudes, HNO3 builds up during the second part of May. HNO3 mixing ratios for different equivalent latitudes indicate a rather strong scatter above the 80 ppbv N2O level.
 However, a few observations between 70°S and 75°S scatter significantly below the reference function at altitudes between 50 and 150 ppbv N2O level (between 450 and 600 K). The first STS particles were detected at the end of May by MIPAS (ENVISAT) [Spang et al., 2005; Tilmes et al., 2006b] and may have absorbed gaseous HNO3, which would result in decreasing HNO3 mixing ratios. Correspondingly, ILAS-II aerosol extinction coefficients increase at this time of the year.
3.3. Local Ozone Loss Using the CLaMS Box Model
 Box model simulations with the Chemical Lagrangian Model of the Stratosphere (CLaMS) were performed to aid the interpretation of the ILAS-II measurements and to calculate chemical ozone loss in the setup phase of the 2003 Antarctic vortex. The model is described in detail by McKenna et al. [2002b, 2002a]. Here, we use the CLaMS modules for trajectory and chemistry calculations. The reaction rates are used according to Sander et al. . The trajectories are calculated using MetO meteorological analyses. The diabatic vertical motion is calculated using the radiation scheme described by Morcrette  and Zhong and Haigh . For the period from 20 March to 31 May 2003, ten representative single air parcel trajectories are selected from the 3100 trajectories initialized at 600 and 650 K such that the considered air parcels stay close to a given latitude. The average latitudes of these trajectories are 60°S, 65°S, 70°S, 75°S, and 82°S for both levels.
 Chemical tendencies along the trajectories are calculated within CLaMS. The initial chemical composition of the air parcels is derived from the average mixing ratio of 15 ILAS-II profiles (N2O, Ozone, and HNO3) obtained on 20 and 21 March at equivalent latitudes poleward of 75°S. H2O and CH4 are taken from a climatology (monthly means) based on HALOE data [Grooß and Russell, 2005]. Total inorganic chlorine (Cly) is taken from a relation with CH4, derived from observations on board the ER-2 aircraft in spring 2000 [Grooß et al., 2002] and total inorganic nitrogen (NOy) from a correlation with N2O, derived from in situ observations in the 2002–2003 Arctic winter [Grooß et al., 2004]. N2O5 is set to 0.15 ppb and NOx is assigned to the difference between HNO3 + 2 × N2O5 and NOy. The obtained NOx mixing ratios (3.8 ppb and 5.3 ppb at 600 and 650 K, respectively) are consistent with the HALOE climatology [Grooß and Russell, 2005] (3.1 ± 1.3 ppb and 4.9 ± 1.5 ppb, respectively). Total inorganic bromine (Bry) is set to 22 ppt [Sinnhuber et al., 2002]. These simulations were performed to investigate the latitude dependence of ozone depletion during the time of vortex formation. Further, other chemical aspects, like the interaction between NOx and HNO3, were investigated with this simulation.
Figure 11 shows the simulated ozone concentration along the five trajectories initialized on the isentropic levels 650 K (Figure 11, top) and 600 K (Figure 11, bottom) corresponding to mixing ratios of 74 ± 4 ppbv and 97 ± 5 ppbv N2O, respectively. In the first weeks of the simulation, the net ozone depletion rate is largest for the most poleward trajectories. This is the case because ozone production rates and ozone loss rates depend on the solar zenith angle in different ways. For lower solar elevation at polar latitudes, no significant ozone production by oxygen photolysis can occur while ozone loss cycles are still active [Kawa et al., 2002]. In mid-April, the most poleward trajectory enters the polar night and therefore ozone depletion ceases. Trajectories further equatorward require more time to enter the polar night and, therefore, air parcels are characterized by a longer period of large solar zenith angles, compared to the most poleward trajectory. Therefore ozone depletion is smallest at the most poleward trajectory, as shown in Figure 11 (purple line). The simulated ozone depletion is larger for altitudes corresponding to the 74 ppbv N2O level (≈580 K in mid-April, 2003) than for the 97 ppbv level (≈540 K in mid-April, 2003) owing to larger NO2 values at higher altitudes.
 Contributions from different catalytic cycles lead to the simulated chemical ozone loss. For both altitudes considered, simulated ozone depletion is dominated by NOx-catalyzed cycles. The fraction of NOx-catalyzed ozone loss in the sum of all ozone loss cycles in early April is 64% to 78% for the trajectories starting at 650 K and 56% to 69% for the trajectories starting at 600 K. Also, the fraction of ozone loss due to NOx-catalyzed cycles increases from 60°S poleward.
 The ozone loss determined using CLaMS is now compared with the estimates derived from ILAS-II observations using the tracer correlation for all profiles measured inside the core of the entire vortex, as defined above (Figure 12). The ozone loss is calculated for all profiles measured by ILAS-II inside this region. The error bars correspond to the standard deviation of the daily averaged ozone loss. ILAS-II observations did not sample the very low ozone concentrations in March measured by MIPAS that are at most 0.2 ppmv smaller at the 600 K level and 0.1 ppmv at the 650 K level compared to the lowest concentrations at ILAS-II locations, as described above. Therefore, in the calculation of chemical ozone loss, we correct for the effect of the different sampling of ozone-poor areas by ILAS-II in March and April by decreasing the apparent chemical ozone loss in April by 0.2 ppmv and 0.1 ppmv at the 74 ppbv and 97 ppbv N2O level, respectively. Indeed, the purpose of this calculation is to give a rather rough estimate of chemical ozone loss for comparison with CLaMS simulations.
 The averaged ozone loss between 580–620 K (corresponding to ≈74 ppbv N2O and the 650 K level in March) for the period 2 to 7 April derived from ILAS-II measurements is 0.12 ppmv with a standard deviation of 0.17 ppmv (≈5% of the available ozone). At lower altitudes, at 530–570 K (corresponding to ≈97 ppbv N2O) ozone loss for the same time interval is 0.10 ppmv with a standard deviation of 0.17 ppmv, which is even less than 5% chemical ozone loss. The simulated ozone loss of 0.15 ppmv at the 74 ppbv N2O level and of 0.08 ppmv at the 97 ppbv level is in good agreement with the values derived from the ILAS-II measurements within the variability of the deduced ozone loss. However, owing to the inhomogeneity of ozone concentrations with the inner vortex edge, the uncertainty of the derived chemical ozone loss is large, for example on 8 and 9 April. In summary, from this analysis there is no indication of any problem with the NOx chemistry for large solar zenith angles.
 Additionally, on both levels, simulated and measured HNO3 agrees quite well both in the absolute mixing ratios and in the temporal development of the conversion from NOx to HNO3 (see Figure 13).
3.4. Tracer/N2O Development During the Setup Phase of the 1997 Antarctic Vortex
 The Antarctic polar vortex in winter 1997 started developing in March 1997, according to the definition from section 3. An inner vortex edge exists during March at very high equivalent latitudes (85°S) at altitudes between 475 and 650 K, but the maximum ∇PV · v is below 22.8 PVU/degrees*m/s, see Figure 14. Therefore the vortex is likely permeable up to the end of March 1997. During April, the inner vortex edge moves toward 75°S. In the first half of April, an outer and inner vortex exist at altitudes above 475 K. The maximum ∇PV · v exceeds the threshold value for an isolated vortex at this time. In the second half of April and May, only the outer vortex edge moves from 75°S to 65°S by the end of May at all altitudes considered. The inner vortex boundary is also found intermittently during May 1997.
 In Table 1 the averaged solar illumination in sunlit hours per day averaged over certain time intervals and equivalent latitudes is shown at 550 K for both 2003 and 1997. During April 1997, there are slightly fewer averaged sunlit hours per day than during April 2003 for equivalent latitudes between 70°S and 80°S (see Table 1). However, significantly larger values for sunlit hours per day occur in April 1997 compared to April 2003 at equivalent latitudes between 80°S and 90°S, especially in late April. The larger illumination of equivalent latitudes poleward of 80°S in April 1997 is likely a result of a shift of the polar vortex toward lower geographical latitudes in this year.
 In the period March 16 to 30 1997, the ILAS instrument only performed measurements within equivalent latitudes between 80°S and 90°S (see Figure 15a, orange squares). These O3/N2O profiles that are located within the inner vortex core poleward of 85°S equivalent latitudes show a compact relationship, whereas the other O3/N2O profiles show a strong scatter at altitudes above the 150 ppbv level of N2O (≈500 K). At this time, only an inner vortex edge exists at ≈85°S and thus there is no transport barrier between midlatitudes and the outer vortex. It has to be noted that between 120 and 150 ppbv N2O a small number of profiles scatter below this relation in March 1997 with a maximum deviation of 0.4 ppmv.
 During the first half of April, ozone mixing ratios inside the inner vortex are up to 0.7 ppmv smaller at altitudes above 480 K (i.e., the 100 ppbv N2O level) compared to values describing the reference function at the same equivalent latitudes. This may be a result of the different sampling of an inhomogeneous incipient vortex distribution and chemical ozone loss as described for winter 2003 (see section 3.3).
 The profiles measured inside the entire vortex poleward of 75°S in the first half of April are up to 2 ppmv larger compared to observations inside the partly isolated inner vortex. The ozone mixing ratios measured in the outer vortex decrease between March and April, possibly because of a combination of the increasing isolation of the entire vortex and ongoing chemical ozone loss. During the second half of April, only one vortex edge is established at an equivalent latitude of ≈75°S. Therefore air masses poleward of ≈75°S mix within the entire vortex. This is why ozone mixing ratios in the inner and outer vortex (see Figure 15c, large and small diamonds) are no longer separated. A possible further decrease in ozone, owing to the solar illumination during late April, is masked by isentropic mixing between outer and inner vortex air.
 During May 1997, the low ozone mixing ratios below 2 ppmv vanish owing to mixing within the vortex for equivalent latitudes equatorward of 85°S. No measurements were obtained at equivalent latitudes poleward of 85°S since the beginning of May 1997. At this time of the year, the separation between observations inside the inner vortex and the entire vortex is still clear. The difference between observations inside the inner vortex and the outer vortex in May 1997 is larger than in May 2003 because in 1997 the transport barrier at the inner vortex edge might be still present during May (see Figure 14). During June 1997, the ozone-tracer correlations inside the entire vortex become uniform (Figures 15f and 15g). An early winter reference function for the calculation of chemical ozone loss in spring can be derived for June 1997 to obtain chemical ozone loss for the entire 1997 Antarctic winter as done by Tilmes et al. [2006a].
 The incipient vortex reference functions for NO2/N2O and HNO3/N2O are derived from all the ILAS observations at high southern latitudes during the second half of March 1997 (Figure 16, black lines). Although NO2 has a strong diurnal cycle, it is possible to consider the tracer-tracer relations of NO2/N2O, because ILAS measurements are always taken at the same solar zenith angle, namely at sunset in the Antarctic.
 During the second half of March, the NO2/N2O and HNO3/N2O correlations are very uniform and compact (Figure 16, black lines). In contrast, the O3/N2O correlations show a rather large scatter (Figure 15a). At this time of the year, the entire area between 55°S and 90°S is highly illuminated, see Table 1, and therefore no strong meridional gradients in HNO3 and NO2 exist. Indeed, NO2 is short-lived and will adjust quickly to the photochemical equilibrium. During April, HNO3 mixing ratios increase by up to 5 ppbv and NO2 mixing ratios decrease by ≈3–4 ppbv. Note that in accordance with increasing HNO3, NOx (defined as NOx = NO2 + NO) decreases. Therefore the NO2 decrease must be less than the HNO3 increase. The scatter in NO2/N2O profiles increases (Figure 16, bottom left) because the profiles may reflect the different amount of solar illumination received, depending on their location. As long as some NOx is left, NOx-catalyzed cycles are still effective, which is the case during the second part of April 1997.
3.5. Comparison of the Tracer Development During the Setup Phase of 1997 and 2003 Antarctic Vortices
 The vortices in both Antarctic winters 1997 and 2003 started forming in March according to the Nash et al.  criterion. In 1997, an inner vortex boundary formed at very high equivalent latitudes. In 2003, a vortex edge was detected at equivalent latitudes around 75°S. The different evolution of the Antarctic polar vortex in March 1997 compared to March 2003 (compare Figures 1 and 14) results in different tracer-tracer correlations for the two incipient vortices. The smaller vortex in March 1997 than in March 2003 results in a strong variability of ozone mixing ratios in March 1997 at equivalent latitudes equatorward of 80°S. Although the maximum ∇PV · v is not very large (most of the time below 22.8 PVU/degrees*m/s), the inner vortex in March 1997 at very high latitudes seems to be isolated at this time of the year because of a compact O3/N2O relationship very close to the pole (see Figure 15a). In both years, HNO3/N2O (see Figures 10a and Figure 16 (left), black line) (NO2/N2O only for 1997) scatter plots are compact within the entire polar region in March.
 In the first part of April, a partially isolated inner vortex core exists inside the vortex during both years. O3/N2O correlations indicate significantly lower ozone concentrations inside the inner polar vortex compared to the outer vortex. A maximum deviation from the incipient reference function of up to ≈0.7 ppmv is found at this time for both years. The mixing ratios of HNO3 during April and May are slightly different in 1997 and 2003. HNO3 increases less at altitudes above ≈540 K in 1997 compared to 2003. In May, HNO3 does not reach mixing ratios of more than 15 ppbv inside the vortex core, as was observed at the end of May 2003 (not shown for 1997).
 Further, the O3/N2O correlations in June 1997 show larger ozone mixing ratios above the 150 N2O level compared to June 2003. These larger ozone mixing ratios in 1997 are apparent within the vortex by May. A likely reason for this is a weaker transport barrier at the outer vortex edge during the setup phase in winter 1997 than in winter 2003, compare Figures 1 and 14. This results in a stronger increase of ozone in 1997 than in 2003 due to mixing of extra vortex air masses into the vortex before June. Thus the shape of the tracer-tracer relationship in June is not the same for these two years because it depends on the dynamical development of the incipient vortex.
4. Tracer/N2O Development During the Setup Phase of the Early Arctic Vortex 2003
 For the Arctic winter 2003–2004, observations during the setup phase of the vortex are available from ILAS-II. Figure 17 shows that in 2003 the Arctic vortex started forming at the end of September at altitudes above 550 K. At this time, a very weak vortex edge is determined for equivalent latitudes poleward of 70°N. Since October 2003, the pattern of ∇PV · v indicates the existence of both an inner and an outer transport barrier at 650 K at 60°N and 70°N equivalent latitude and partly at lower altitudes at the beginning of November. Since November, a vortex edge is observed almost continuously between 550 and 650 K at ≈60°N equivalent latitude.
 Although no ILAS-II measurements are available after 24 October 2003, the setup phase of the Arctic vortex can be investigated. During the end of September and October 2003 (corresponding to March and April in the SH), measurements were obtained inside the inner vortex edge or entire vortex edge at altitudes between 550 and 650 K. ILAS-II made measurements at ≈70°N, and the early vortex was located poleward of 75°N. Therefore only a few measurements were obtained within the Arctic incipient vortex.
 A large variability in O3/N2O correlations is observed in September 2003 between the 50 to 150 ppbv N2O level (≈500–700 K) (see Figure 18a), as in the Antarctic vortex (Figure 15a). At the beginning of October, profiles measured within the inner vortex core poleward of 75°N show a compact O3/N2O relationship that is used to derive an incipient vortex reference function (see Figures 18a–18c, black line), although the transport barrier at the inner vortex edge seems to be weak (Figure 17). As for the Antarctic, a separation between inner and outer vortex profiles is noticeable especially at altitudes above the 80 ppmv N2O level, in accordance with the existence of two vortex transport barriers above 550 K at this time. A minimum of ozone occurs poleward of 80°N at the 130 ppbv N2O level.
 Low ozone concentrations of the incipient Antarctic vortex 2003 and 1997 (during April) occur exactly at the corresponding time of the Arctic early winter 2003 (during October). As described for the setup phase of the 1999 Arctic vortex, chemical NOx driven ozone depletion reduces ozone concentrations at this time of the year [Kawa et al., 2002]. The large variability of inner vortex profiles indicates the inhomogeneous distribution of ozone mixing ratios with the incipient vortex. During 21–24 October 2003, profiles within the inner vortex are more compact than during the first half of October (Figure 18c). At this time, the inner vortex edge is stronger and has developed over the entire altitude range between 475 and 550 K.
 The tracer-tracer correlations of HNO3/N2O at the end of September 2003 in high northern latitudes are compact, similar to the tracer-tracer correlations at high southern latitudes at the end of March. The derived reference for late September 2003 and the range of uncertainty, described by the standard deviation of the profiles, within the vortex is shown in Figure 19, black line. The increase in HNO3 between the end of September and early October (black line compared to colored diamonds in Figure 19 (left)) in the Arctic is significantly smaller than the increase observed in the Antarctic. Further, there is less variation between profiles inside the vortex core and the outer vortex in the Arctic than in the Antarctic. This is in agreement with the fact that the sunlit hours per day differ by up to 8 hours per day between 90°S and 55°S equivalent latitude for the 2003 Antarctic winter with very few sunlit hours per day close to the pole (see Table 1). For the Arctic, the variability of sunlit hours for different equivalent latitudes is less strong and there are a greater number of sunlit hours per day, probably because the Arctic vortex moves more often toward lower latitudes during this time of the year compared to the Antarctic vortex. At the end of October/April the conditions in both hemispheres are more similar to each other.
 Ozone concentrations in the incipient Antarctic and Arctic polar vortex show large variations (between 2 and 4 ppmv). Meteorological analyses and tracer-tracer correlations indicate a separation of the incipient polar vortex into an inner vortex that is noticeably isolated from an outer vortex by a transport barrier. During the setup phase of the vortex, the outer vortex is influenced by mixing with air from midlatitudes, which increases ozone mixing ratios. Inside the isolated inner vortex ozone mixing ratios show a variability of ≈0.5 ppmv within the same equivalent latitude bin. Patches of very low ozone are already observed by MIPAS (ENVISAT) in March. The lowest ozone mixing ratios observed by ILAS and ILAS-II are during early April, 1.7 ppmv (Antarctic), and the first half of October, 1.8 ppmv (Arctic) in the inner vortex at 500–550 K. This corresponds exactly to the time when the vortex becomes isolated at these altitudes and when the solar illumination decreases. In accordance with the mechanism described by Kawa et al. , we find net ozone destruction during the first weeks of April (Antarctic) and October (Arctic), owing to decreasing solar illumination.
 CLaMS box model simulations indicate that moderate chemical ozone loss during the times considered here is mainly caused by NOx-catalyzed cycles. The conversion between NOx and HNO3 observed in ILAS measurements is in good agreement with box model simulations. The rough estimation of chemical ozone loss using ILAS-II observations during the first week of April in the Antarctic winter 2003 is also in agreement with box model simulations.
 In the Antarctic at the end of April no ozone loss is detected within the inner vortex corresponding to the lack of sunlight in high polar latitudes (poleward of 80°S). From that time onward, the transport barrier between the inner and outer vortex vanishes and ozone becomes uniform inside the entire vortex through mixing. During June, the ozone-tracer correlations become uniform for profiles located within the entire vortex and ozone loss stops within the entire vortex, because of the lack of sunlight. At this time, an early winter reference function is derived to calculate chemical ozone loss in spring [Tilmes et al., 2006a]. The exact shape of the early winter reference function depends on the extent of vortex isolation during the setup phase of the polar vortex.
 We gratefully acknowledge the contribution of all members of the science team of the Improved Limb Atmospheric Spectrometer (ILAS and ILAS-II) led by Y. Sasano and H. Nakajima for processing ILAS (V6.0) and ILAS-II (V1.4) data. ILAS and ILAS-II were developed by the Ministry of the Environment (MOE), Japan, and were aboard the ADEOS and ADEOS-II satellite launched by the Japan Aerospace Exploration Agency (JAXA). ILAS and ILAS-II data were processed at the ILAS and ILAS-II Data Handling Facility, National Institute for Environmental Studies (NIES). Thanks are also due to the European Space Agency for providing MIPAS data and the UK Meteorological Office for providing meteorological analyses. We further thank Janet Carter-Sigglow for a grammatical and stylistic revision of the manuscript and S. Tilmes thanks the Deutsche Akademie der Naturforscher Leopoldia and the Bundesministerium für Bildung und Forschung for supporting this study.