In this study, ILAS-II (Improved Limb Atmospheric Spectrometer) measurements were used to analyze chemical ozone loss during the entire Antarctic winter 2003, using the tracer-tracer correlation technique. The temporal evolution of both the accumulated local chemical ozone loss and the loss in column ozone in the lower stratosphere is in step with increasing solar illumination. Half of the entire loss in column ozone of 157 DU occurred during September 2003. By the end of September 2003, almost the total amount of ozone was destroyed between 380 and 470 K. Further, ozone loss rates increased strongly during September for the entire lower stratosphere. The values of accumulated ozone loss and ozone loss rates are strongly dependent on altitude. Once ozone loss is saturated during September, especially at latitudes between 380 and 420 K, ozone loss rates decrease, and accumulated ozone loss can no longer increase. Moreover, at altitudes above 470 K, accumulated ozone loss depends on the amount of PSCs occurring during winter and spring. During September, ozone mixing ratios show a large day to day variation. Box model simulations by the Chemical Lagrangian Model of the Stratosphere (CLaMS) show that this is a result of the different histories of the observed air masses. Further, the box model supports the general evolution of ozone loss values during September as a result of the strong increase of halogen catalyzed ozone destruction.
 Chemical ozone loss in the lower polar stratosphere has been a research focus since the discovery of the Antarctic ozone hole in the mid-eighties [Farman et al., 1985]. Severe chemical ozone destruction during the winter period in the Antarctic and, in recent cold winters, also in the Arctic is a result of anthropogenic emissions of long-lived halogen compounds (in particular CFCs and halons) into the atmosphere. Chlorine reservoir species (in particular HCl and ClONO2) are activated on the surface of the polar stratospheric clouds (PSCs) that exist in a sufficiently cold winter polar stratosphere. The resulting photo-labile forms of chlorine are photolyzed upon exposure to sunlight. In spring, even with low Sun, effective ozone-destroying catalytic cycles [Molina and Molina, 1987; McElroy et al., 1986; Solomon et al., 1986] cause significant ozone loss.
 In contrast, since the early nineties, the annually recurring Antarctic ozone hole has displayed much less variability in size and extent than ozone loss in the Arctic. Therefore Antarctic ozone loss has frequently been derived from measurements of the total column. In this way, however, the impact of transport on the total column ozone cannot be investigated.
 Various model studies have been performed to investigate Antarctic winter ozone loss. Brasseur et al.  reported that their model does reproduce the Antarctic total column ozone depletion as observed by TOMS. There, the maximum of observed total ozone depletion over the Antarctic is in agreement with satellite measurements. However, the occurrence of Antarctic ozone loss and ozone loss rates simulated by models may not agree in altitude and time [e.g., Chipperfield et al., 1996]. A very detailed comparison with measurements is necessary as recently performed by Feng et al.  for a comparison of the Antarctic winters 2000 and 2002.
 However, apart from continuous total column ozone measurements (e.g., by the TOMS or GOME instrument), no continuous data sets of mixing ratios of different species, including ozone as well as long-lived tracers, were available over an entire Antarctic winter to compare model results. Other data sets available either do not provide ozone and a long-lived tracer measured at the same time (e.g., POAM, SAGE), which is necessary to apply the tracer-tracer correlation technique, or do not measure continuously over an entire winter and spring within the Antarctic polar vortex (e.g., HALOE and CLAES). Only very recently first results have been reported from the AURA/MLS and the Envisat/MIPAS instrument [Santee et al., 2005; Glatthor et al., 2005].
 Using measurements of O3 and N2O by the ILAS-II instrument aboard the ADEOS-II satellite, it is possible for the first time to deduce chemical ozone loss continuously over an entire Antarctic winter. The tracer-tracer correlation technique (TRAC) that was applied successfully to the Arctic is used here to quantify Antarctic chemical ozone loss. The temporal development of local accumulated ozone loss will be analyzed for different altitudes, as well as the accumulated loss in column ozone. A comparison with meteorological conditions will be conducted using the UK Met Office meteorological analysis [Swinbank and O'Neill, 1994] for different altitude ranges.
 The continuous ILAS-II data set further allows the derivation of daily ozone loss rates within the polar vortex at different altitude intervals. ILAS-II observations during September 2003, and the derived daily ozone loss rates will be compared with results of a box model simulation of the Chemical Lagrangian Model of the Stratosphere (CLaMS) [McKenna et al., 2002b, 2002a] to scrutinize the large ozone loss values derived for September 2003. Here, we use only the CLaMS modules for trajectory and chemistry calculations. Further, the ILAS-II results reported here can be used as a comparison for other model and measurement results.
2. ILAS-II Measurements and Methodology
 The ILAS-II (Improved Limb Atmospheric Spectrometer) instrument aboard ADEOS-II (Advanced Earth Observing Satellite) observed the entire Antarctic winter 2003 continuously from 2 April to 24 October 2003 [Nakajima et al., 2006]. The occultation satellite instrument measured sunrise and sunset data up to 14 times per day. The sunset mode of the instrument covers the area of the polar vortex very well during the entire austral winter and spring (Figure 1) and is therefore well suited to observe chemical ozone loss. Poleward of the latitude of the measurement location (Figure 1, shaded area) the earth was not illuminated between April and mid-September.
 The ILAS-II version 1.4 data set (the first public release) includes O3, HNO3, N2O, CH4, and the aerosol extinction coefficient at 780 nm from cloud top up to 70 km (T. Yokota et al., IImproved Limb Atmospheric Spectrometer II (ILAS-II) version 1.4 algorithm for retrieval of gas and aerosol profiles in the stratosphere, manuscript in preparation, 2005). Vertical profiles of other atmospheric trace gases, such as NO2, H2O, ClONO2 and N2O5, were also retrieved during the entire measurement period (not yet validated). The accuracy of the ozone is estimated to be better than ±10% [Sugita et al., 2006].
 In this study, the tracer-tracer correlation method [e.g., Proffitt et al., 1990; Müller et al., 2001; Tilmes et al., 2003, 2004] is used to calculate chemical ozone loss. Using this method, deviations from an early winter reference function derived for chemically unperturbed conditions in an established vortex are used to identify chemical ozone loss. Here, N2O can be used as the long-lived tracer. Although there is an offset of N2O ILAS-II version 1.4 data compared to ODIN/SMR (v1.2) (Sub-Millimeter Radiomenter) for N2O mixing ratios less than 100 ppbv, this offset does not show any seasonal change [Ejiri et al., 2006; Urban et al., 2005]. A constant offset of the long-lived tracer used will not affect the results of the tracer-tracer correlation analysis.
3. Meteorology of the Antarctic Winter 2003
 The Antarctic vortex started forming during March 2003 [Tilmes et al., 2006], at the time when an edge of the vortex started to come into existence according to the definition of [Nash et al., 1996]. During the setup phase of the vortex, two relative maxima of the gradient of potential vorticity exist using UK meteorological analysis. The vortex was partly separated into two regions, one region was located within equivalent latitudes equatorward of 70°S and another within the region poleward of 70°S equivalent latitude (inner vortex) [Tilmes et al., 2006]. On the basis of that study, the polar vortex edge was calculated for these two separate regions as well. During April to October 2003, a continuous polar vortex edge was determined equatorward of 70°S at altitudes between the 475 and 650 K potential temperature level (Figure 2). The potential vorticity at the edge of the vortex increases until mid-July (Figure 2, solid lines). An indication of a disturbance of the vortex is noticeable during the second part of July because the potential temperature of the edge of the vortex significantly decreases at 650 K. However, at altitudes below 650 K the potential vorticity at the edge of the vortex still increases until October 2003, and no minor warming was observed in this winter (M. Streibel, personal communication, 2004). In addition to the vortex edge, an inner vortex edge was determined poleward of 70°S, for certain time intervals. An inner edge was found for the setup phase of the vortex [Tilmes et al., 2006], at the time of a slightly disturbed vortex in July and at the end of September and in October 2003, at altitudes of 650 K. During November, only an inner vortex edge could be determined using the Nash et al.  criterion at altitudes above 550 K. At this time the vortex begins to break down although vortex remnants continue to exist until the end of November.
 The area of possible PSC existence (APSC) deduced from analyzed stratospheric temperatures and sunlight hours per day have an influence on chemical ozone loss within the lower stratosphere [Tilmes et al., 2004]. In this study, we will use these values for the interpretation of chemical ozone loss values.
 The PSC threshold temperature was calculated with averaged mixing ratios of ILAS-II HNO3 and H2O measurements. Using a HNO3 mixing ratio of 10 ppbv and a H2O mixing ratio of 5 ppmv, as is done for Arctic conditions [Tilmes et al., 2004], APSC is 28% larger than that deduced using ILAS-II measurements. Thus the decrease of HNO3 and H2O mixing ratios caused by the impact of denitrification and dehydration, which is assumed to be permanent, on the threshold temperature for PSCs should be included in the calculation of APSC for Antarctic conditions, as done in this study.
 Additionally, in the Antarctic winter 2003, measurements of daily mean PSC cloud top heights by the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) aboard ENVISAT [Spang et al., 2005] are available (Figure 3, black pluses). In addition to APSC, these measurements provide precise information about PSC occurrence.
 Calculated APSC and actual detected mean PSC cloud top heights by MIPAS are in general agreement. Stratospheric temperatures reached the threshold for PSC existence from mid-May until the end of September 2003 (Figure 3). First sulfuric ternary solution (STS) particles were measured at the end of May by MIPAS and the first nitric acid trihydrate (NAT) particles on 10–12 June 2003 at altitudes below 24 km (≈650 K). Since 20 July, mean cloud top heights detected by MIPAS were below 580 K. With the descent of the vortex air masses, mean PSC cloud top heights decreased significantly from 570 K to 400 K during August. Throughout September 2003, an APSC was calculated at altitudes below 500 K and MIPAS mean cloud top heights of PSC were below 470 K.
4. Tracer-Tracer Correlations
 Using the tracer-tracer correlation technique, an early winter reference function for the ozone-tracer relation in the established polar vortex is derived. Deviations from this reference function can be attributed to chemical ozone loss, because transport processes can be excluded using tracer-tracer correlations within an isolated vortex [Müller et al., 2001, 2005; Tilmes et al., 2003].
 The evolution of the incipient Antarctic polar vortex between March and June 2003 is discussed in detail by Tilmes et al. . In that study it is shown that tracer-tracer correlations throughout the entire vortex are compact and constant by mid-June 2003, and that the vortex core is isolated at that time. At this time of the year, during the polar night, no chemical ozone loss is expected to occur. Therefore a reference for chemically unperturbed conditions is derived for 11–20 June 2003 from ILAS-II measurements located in the Antarctic polar vortex core (see Figure 4a, black line).
 Although ILAS-II profiles are located at rather low geographical latitudes up to 65°S (see Figure 1), the equivalent latitudes of these measurements range between 65 and 90°S (see Figure 4a). This is the case because the vortex core is not steadily located above the geographic South pole but is sometimes shifted toward lower geographical latitudes. All these profiles are used to derive the early winter reference function.
 To allow precise comparison of ILAS-II measurements with the synoptic meteorological data, the position of ILAS-II profiles are converted to noon time using trajectory calculations based on UKMO wind data [Tilmes et al., 2003].
 Using N2O as a long-lived tracer (mixing ratios in ppbv) and O3 (mixing ratios in ppmv), the early winter reference function (valid for range 10 ppbv < N2O < 300 ppbv) is derived as:
The scatter of profiles inside the polar vortex measured by the standard deviation is estimated to be σ = 0.2 ppmv.
 Ozone mixing ratios of profiles located poleward of an equivalent latitude of 80°S scatter up to 0.5 ppmv below the derived reference function for altitudes above the 100 ppbv N2O level (Figure 4a, green profiles) possibly as a result of a slight isolation between a previously existing inner vortex within the entire vortex [Tilmes et al., 2006]. An overestimation of chemical ozone loss of profiles located poleward of an equivalent latitude of 80°S up to 0.5 ppmv is therefore possible, but has no impact on derived ozone loss rates. Note, however, that the air mass poleward of 80°S constitutes only about 25% of the air mass poleward of 70°S and has little impact on the averaged accumulated ozone loss inside the vortex core.
 By mid-July 2003, significant deviations from the early winter reference function occur (not shown) and at the beginning of August all profiles measured inside the vortex core scatter below the early winter reference function (Figure 4b); including profiles located poleward of equivalent latitude of 80°S.
 During August and September, the signature of chemical ozone loss in the evolution of tracer-tracer profiles becomes clearly noticeable (Figures 4b–4g). During August, the strongest deviations from the reference occur for profiles located at lower equivalent latitudes, between 70°S and 80°S (red and blue profiles). This is because these profiles are exposed longer to sunlight, which results in larger ozone loss values. During 20–30 September, all the profiles inside the polar vortex core rather suddenly indicated a very strong decrease of ozone mixing ratios toward zero (see Figure 4g). The temporal evolution of ozone loss during the second part of September will be discussed in detail below in section 5.2. During October 2003, very low ozone mixing ratios were reached, between 0.0 and 0.4 ppmv between the 80–240 ppbv N2O level, for all profiles measured inside the Antarctic polar vortex.
5. Accumulated Ozone Loss
 The accumulated local ozone loss is ozone loss that occurred in the period between the time of the early winter reference function and the date considered for derivation the ozone loss. Ozone loss is derived over an altitude interval between 350 and 600 K potential temperature. In order to interpret a large scatter of ozone loss profiles, daily variations of ozone mixing ratios will be discussed as well. Additionally, we show the temporal evolution of local ozone loss averaged within certain altitude ranges, smoothed over 10 days. The results will be discussed and compared with meteorological analyses. Further, the temporal evolution of loss in column ozone derived between 350 and 600 K will be shown.
5.1. Local Ozone Loss Profiles
 The temporal evolution of accumulated local ozone loss profiles can be followed during the entire Antarctic winter 2003 using ILAS-II measurements. In Figure 5, vertical ozone profiles measured within the vortex core (red lines) and estimated ozone loss profiles (black lines) are shown for a period of ten days in each panel to give an overview.
 The early winter reference function was derived for 11–20 June 2003. At this time, the ozone loss profiles do not show any ozone loss, as expected. Only two profiles that are located poleward of the equivalent latitude of 80°S show an ozone loss (ΔO3) of ≈0.5 ppmv for altitudes above about 500 K. As described in section 4, this observation should not be attributed to chemical ozone loss.
 Beginning in August, significant local ozone loss occurred (Figure 5, black lines) at altitudes between 350 and 600 K potential temperature. At the end of August, the average local ozone loss between 400 and 550 K is 0.79 ppmv with a standard deviation of 0.31 ppmv. During mid-August, a maximum accumulated ozone loss of 1.5 ppmv occurred at altitudes between 460 and 520 K potential temperature.
 During September, ozone loss profiles show a much stronger scatter compared to August 2003. The standard deviation of local ozone loss profiles in the first half of September reaches 0.37 ppmv. In the second half of September, loss profiles scatter less. However, inside the vortex core a separation of ozone loss profiles is obvious between 400 and 500 K during 21–30 September 2003 (Figure 5). On the one hand, maximum ozone loss values less than ≈2.3 ppmv were measured before 25 September, and, on the other hand maximum local ozone loss larger than 2.3 ppmv was measured after 25 September. We will discuss this in detail in section 5.2. The averaged local ozone loss between 400 and 550 K is 2.3 ppmv. At the end of September, a maximum local ozone loss of 2.8 ppmv was reached at ≈475 K potential temperature with an average of 2.5 ppmv between 400 and 550 K. Since the beginning of October, chemical local ozone loss was very consistent for all profiles measured by ILAS-II and the maximum and averaged local ozone loss did not increase after the end of September. However, the standard deviation decreased toward 0.08 ppmv. Ozone mixing ratios did not decrease any more during October.
 The evolution of ozone loss derived in this study is furthermore of interest considering the previous Antarctic winter 2002. In 2002, a major warming occurred on 21 September 2003 [Newman and Nash, 2004] and ozone loss came to a halt [Hoppel et al., 2003]. Ozone depletion rates in the polar vortex region rapidly decreased to zero [Grooß et al., 2005]. If such an event had happened in winter 2003, 82% of the entire loss in column ozone in 350–600 K would have been reached. In 380–420 K most of the ozone was already destroyed by 21 September 2003.
5.2. Daily Variations of Ozone Mixing Ratios
 In this section, the strong variability of vertical ozone loss profiles during September (Figure 5) will be discussed. Figure 6 (bottom) shows the ozone mixing ratios for two potential temperature levels, at 500 K and 550 K, between 5 and 10 September, observed within the vortex core. Strong daily variations of ozone mixing ratios of about 1 ppmv are obvious and cannot be explained by differences in equivalent latitude. To investigate this finding in more detail, backward trajectory calculations were performed starting at the location of the observations.
 Some of the 20-day back trajectories of observed air masses did stay close to the pole while others originated from lower latitudes. Figure 6 (top) shows the fraction of time at which each air parcel was in sunlight (solar zenith angle <95°). The sunlight fraction along the back trajectory ending at the locations of the observations varies between 8% and 45%. Indeed, it is obvious that air masses with a high sunlight fraction, which obtained more sunlight, correspond to observations of low ozone mixing ratios and vice versa.
 Further, the amount of illumination controls the amount of ClONO2 mixing ratio. If large ClO mixing ratios prevail, ClONO2 production is controlled by the production of NO2 from photolysis (and reaction with OH) of HNO3. ClONO2 mixing ratios were larger if the observed air mass was illuminated more strongly during 20 days before the measurement time (not shown). This is in agreement with the assumption that as long as some amount of chlorine is still activated, larger amounts of ClONO2 correspond to greater ozone destruction and therefore smaller ozone mixing ratios. Therefore, at this time of the year, the large variation of ozone mixing ratios and ozone loss values is a result of the different histories of the observed air masses.
 Up to in 24 September 2003, the variability of the sunlight time fraction of backward trajectories decreased (Figure 7, top), since all vortex air was exposed to a significant amount of sunlight. Correspondingly, the variability of ozone mixing ratios and therefore the variability of ozone loss values decreases as well.
 Back trajectory analyses further indicate that between 24 and 26 September, the characteristics of air mass history changed. The latitude over the previous 20 days of the observed air parcels at 500 K potential temperature changed from 85°S ± 3°S on 24 September, to 75°S ± 4°S on 26 September (Figure 7). Therefore air masses observed before 24 September and after 26 September are not necessarily comparable and the day-to-day variations in ozone should not be interpreted as chemical change between 24 and 26 September 2003. Furthermore, larger variations in sunlight time between 24 and 26 September becomes obvious compared to some days previously (Figure 7). However, at this time, the air masses indicating the lowest ozone mixing ratios received the smallest number of sunlight hours for 20 days before the measurement. This is in contrast to the anticorrelation between ozone mixing ratios and sunlight time that was found during the first part of September. In late September, the stronger illumination of air masses possibly resulted in an almost complete chlorine deactivation and therefore lower ozone loss values.
 From 26 September onward, the history of the air masses changed compared to 24 September (Figure 7). Ozone mixing ratios are rather uniform and increase slightly with decreasing averaged latitudes for 20-day backward trajectories and increasing solar illumination between 26 and 30 September.
 To remove the effect of the explained short-term variability in ozone mixing ratios and therefore local ozone loss values, in the following analysis local ozone loss will be smoothed. However, during 24–27 September (depending on altitude) local ozone loss values and deduced ozone loss rates may be still influenced by the observations of noncomparable air masses.
5.3. General Evolution of Local Ozone Loss
 To describe the general temporal evolution of local ozone loss, the daily averaged ozone loss values in the vortex core are smoothed over a period of 10 days. This was likewise done for meteorological values as the solar illumination on the area of PSC existence (APSC) and the possible area of PSC existence (APSC). Four different altitude ranges, 380–420 K, 430–470 K, 520–560 K, and 580–620 K, are considered separately (Figure 8).
 Chemical ozone loss started in July 2003 for all altitudes considered with beginning illumination, in accordance with the current understanding of ozone destruction. Further, chemical ozone loss is expected to correlate with APSC. The largest APSC was calculated for 380–420 K and 420–470 K and, correspondingly, the largest ozone loss occurred in these altitude regions. Ozone loss of 1.2 ± 0.2 ppmv was calculated up to the beginning of September 2003 for these two altitude intervals. Up to 22–23 September 2003, the ozone loss reached 2.3 ± 0.2 ppmv below 470 K potential temperature and ozone mixing ratios were below 0.3 ppmv. After this date, the remaining ozone was almost completely destroyed before the beginning of October 2003 (Figure 8, top, red line). At altitudes of 430–470 K, ozone loss increased before the beginning of October up to 2.8 ± 0.2 ppmv before ozone mixing ratios were nearly zero. Thus almost the total amount of ozone between 380 K and 470 K was destroyed before the end of September.
 For altitude intervals 520–560 K and above, accumulated ozone loss is less, possibly because the APSC decreases from July onward. Further, mean PSC cloud top height detected by MIPAS strongly decreases during August (section 3). Since mid-September, PSC probability is nearly zero as derived using UKMO analysis at these altitudes and no PSC events were detected by MIPAS during the whole of September. Up to 1 September 2003, 0.8 ± 0.2 ppmv ozone loss was reached in 520–560 K. During the second part of September 2003, a strong increase of accumulated ozone loss occurred, up to 1.9 ± 0.2 ppmv until the beginning of October 2003. This increase of accumulated ozone loss is more rapid than at altitudes below. At 520–560 K, ozone is not completely destroyed so that the accumulated ozone loss is not limited by the amount of initially prevailing ozone as is the case at lower altitudes (see Figure 8, red lines). Further, the slope of accumulated ozone loss, and therefore ozone loss rates, increases with increasing solar illumination during this time of the year (see section 7, for further discussion). However, as described in section 5.2, the increase of accumulated ozone loss is possibly enhanced by the changing air masses observed during the second half of September, and therefore estimated ozone loss rates (section 6) are possibly overestimated.
 At 580–620 K, a large APSC was derived using UK meteorological analysis up to the end of August. However, cloud top heights of PSCs measured by MIPAS do not indicate the existence of PSCs at altitudes above 580 K after mid-July 2003 (Figure 2). At these altitudes, accumulated ozone loss is almost zero (within the range of uncertainty) before September 2003. Only during September did some ozone loss occur that increased up to 0.4 ± 0.2 ppmv at the beginning of October 2003.
5.4. Accumulated Ozone Loss in Column Ozone
 Accumulated chemical ozone loss in column ozone is derived by integrating the ozone loss profiles over a certain altitude range [e.g., Tilmes et al., 2004; Salawitch et al., 2002]. Here, the altitude range of 350 and 600 K potential temperature is used, this is the altitude range over which the halogen catalyzed polar ozone loss occurs (as discussed above). The temporal evolution of chemical loss of column ozone inside the vortex core is shown in Figure 9. Accumulated ozone loss is smoothed here over 20 days to eliminate any short-term variability.
 In addition to ozone loss (black line), the area of possible PSC existence is shown (blue line), averaged over the same altitude range (350–600 K). First STSs were already detected by MIPAS in mid-May. The volume of possible PSC existence increases up to its maximum at the beginning of August and decreases thereafter. Solar illumination at PSC area increases after the beginning of July. Only with increasing solar illumination, did chemical ozone loss start in July 2003 as expected from the current understanding of polar ozone destruction mechanisms. At the beginning of September 2003, 79 ± 17 DU ozone were destroyed. This is half of the entire ozone loss of 157 ± 17 DU that occurred up to the beginning of October 2003. In step with increasing Sun hours per day, ozone loss increases during the entire winter and spring until October. About 88% of the proxy ozone (ozone for chemically unperturbed conditions) is destroyed in 350–600 K at the beginning of October. As discussed in section 5.3, the most effective ozone destruction occurred at altitudes below 470 K.
 Loss in column ozone calculated over the altitude range of 380–550 K (115 ± 15 DU) can be compared with ozone loss derived for very cold Arctic winters. For example, in the cold winter 1999–2000 with a vortex located close to the pole, column ozone loss of 83 ± 6 DU inside the vortex core was derived using HF as a long-lived tracer [Tilmes et al., 2004]. Another cold Arctic winter, 1992–1993, with less PSC but with an increased burden of aerosols reached 100 ± 25 DU [Tilmes et al., 2004]. Interestingly, the difference between column ozone loss in cold Arctic winters and in the Antarctic winter 2003 is not very large in 380–550 K (S. Tilmes et al., Chemical ozone loss in the Arctic and Antarctic stratosphere between 1992 and 2005, manuscript to be submitted, 2006).
6. Ozone Loss Rates
 In order to calculate ozone loss rates local ozone loss was smoothed over 10 days. Ozone loss rates were estimated by calculating the difference between local ozone loss averaged over the entire vortex core of two subsequent days, as discussed above. Further, the resulting ozone loss rates were again smoothed over ten days, to reduce day-to-day variations. The resulting ozone loss rates (black line) and the corresponding standard deviation (dotted black lines) are shown in Figure 10.
 Ozone loss rates are greatest during mid-September between 380 and 550 K. In 380–420 K ozone loss rates indicate a small maximum during mid-August of ≈35 ± 10 ppbv per day and, further, a strong increase before mid-September up to 58 ± 15 ppbv per day. At the beginning of October, ozone loss rates are nearly zero at 380–420 K, because no ozone is left that could be destroyed.
 At 430–470 K, ozone loss rates are similar to those at 380–420 K. However, slightly smaller ozone loss rates are deduced for the end of August compared to some weeks before, which may be caused by strongly decreasing APSC. Correspondingly, the mean cloud top height detected by MIPAS is below 450 K at this time. With increasing solar illumination, ozone loss rates increase as well up to 82 ± 10 ppbv per day. The occurrence of PSC events detected by MIPAS in mid-September at these altitudes may further enhance ozone loss rates (Figure 3). As in 380–420 K, ozone loss rates are nearly zero at the beginning of October when ozone is almost completely destroyed.
 At 520–560 K, during July and August, ozone loss rates are more variable and smaller compared to altitudes below, in correspondence with a smaller APSC and the resulting lower chlorine activation. This is in agreement with larger ClONO2 mixing ratios compared to lower altitudes, indicated from preliminary ClONO2 ILAS-II observations. However, a strong increase of ozone loss rates of up to 90 ± 15 ppbv per day on 24 September occurred in mid-September. These ozone loss rates are the strongest reported in this study. As discussed in detail in section 5.2, the history of observed air masses changed between 24 and 27 September 2003. At altitudes below, this effect seems to be less significant, because the strongest ozone loss rates were derived before 24 September. However this potential problem for all altitudes should be kept in mind when comparing these results with other studies.
 At 580–620 K the results of tracer-tracer correlations indicate very low ozone loss rates until September 2003. At the beginning of September ozone loss rates of 20 ± 5 ppbv per day and from mid-September ≈25 ± 5 ppbv per day were deduced. This corresponds to the very low chlorine activation assumed from rather large mixing ratios of the preliminary ClONO2 measurements.
7. Comparison With Box Model Simulations
 Ozone loss rates derived using tracer-tracer correlations (see section 6) depend on the time and altitudes considered. Especially during 10–30 September 2003, a strong increase of ozone loss rates occurs within the altitude range of 380–560 K.
 Box model simulations with the Chemical Lagrangian Model of the Stratosphere (CLaMS) [McKenna et al., 2002b, 2002a] were performed to investigate chemical ozone loss occurring during 10–30 September 2003. The box model includes all currently know chemical reactions of stratospheric relevance using the JPL2003 reaction rates [Sander et al., 2003].
 Starting at three isentropic levels, 450, 500 and 550 K, two vortex air parcel trajectories per level as an example were chosen in such a way that the average latitude of the trajectories is about 80°S with the standard deviation of the latitude being smallest and largest, respectively. Therefore one trajectory stays close to 80°S whereas the other shows excursions to lower-latitude regions. Along these trajectories chemical ozone loss is simulated by the CLaMS chemistry module, whereby the initialization is compiled from ILAS-II data for equivalent latitudes poleward 75°S. Further, total inorganic chlorine was derived from a relation with CH4 [Grooß et al., 2002] and total inorganic nitrogen from a correlation with N2O [Grooß et al., 2004]. Total inorganic bromine was set to 22 pptv [Sinnhuber et al., 2002]. These simulations were performed to investigate how well the observed ozone depletion rate can be reproduced.
 The simulated ozone mixing ratios and the corresponding ILAS-II observations are shown in Figure 11. As explained above, on a specific day, observed ozone mixing ratios vary because of the different histories of the observed air masses. However, the general amount of ozone depletion is well reproduced by the simulations between 10 and 30 September 2003, for the three considered theta levels (Figure 11). The simulations show that the large ozone loss rates are a result of long sunlight time and persistent low temperatures causing very efficient chlorine-catalyzed ozone loss.
 Between 24 and 26 September 2003, the large variations of ozone mixing ratios are a result of the changing history of the air masses (section 5.2) and are not caused by chemical effects. This effect seems to be most significant at altitudes above 500 K. However, the model simulated ozone loss rates of ≈65 ppbv/day at 450 K between 10 and 30 September, which is in good agreement with ILAS-II results at 430–470 K (Figure 10, second panel). The results are also in general agreement with model simulations of the Antarctic ozone hole by Brasseur et al. . In that study, ozone loss rates of 80 ppbv per day were derived during September.
 Further, the simulated NO2 and ClONO2 mixing ratios (not shown) are in general agreement with the (not yet validated) ILAS-II measurements. As for ozone, variations of the mixing ratios of NO2 and ClONO2 owing to the different histories of air masses inside the vortex, as described in section 5.2, are not represented in box model results.
 Chemical ozone loss was derived using tracer-tracer correlations based on ILAS-II observations during the entire Antarctic winter 2003. We consider ozone loss inside the polar vortex core only.
 Ozone mixing ratios indicate a strong variation, with day-to-day variations larger than 1 ppmv during the first half of September. Backward trajectory calculations have shown that these variations are a result of the different histories of the observed air masses, that is in particular a strong variation of the amount of sunlight time the air parcel received before the measurement time. In October, local ozone loss became very uniform and reached 2.5 ppmv between 400 and 550 K with a standard deviation of 0.08 ppmv.
 The temporal evolution of local ozone loss and accumulated loss in column ozone during July and the end of September are in step with increasing solar illumination on activated air masses. Half of the accumulated loss in column ozone over the winter (157 DU in 350–600 K) occurs during September 2003. About 88% of the proxy ozone was destroyed within this altitude range.
 In the present study, it is shown that accumulated ozone loss and ozone loss rates are strongly dependent on the altitude considered. Although the APSC is largest at altitudes between 380 and 420 K, accumulated local ozone loss does not exceed 2.3 ± 0.2 ppmv until 22–23 September 2003, because all the ozone was already destroyed within this altitude interval. At 430–470 K 2.8 ± 0.2 ppmv ozone were destroyed up to October, 2003. At altitudes above, less ozone loss occurred (1.9 ± 0.2 ppmv in 520–560 K and only 0.4 ± 0.2 ppmv in 580–620 K).
 Ozone loss rates in 380–420 K have a maximum of 58 ± 15 ppb per day around 18 September 2003. In 430–470 K, 82 ± 10 ppbv per day were reached around 19 September 2003. Since ozone mixing ratios approach zero, ozone loss rates will decrease. At 520–560 K and 580–620 K, ozone loss is much less compared to altitudes below and ozone was not destroyed completely, which is in accordance with smaller APSC and therefore possibly less strong chlorine activation.
 Box model results using the CLaMS model reproduce the general temporal development of ozone mixing ratios between 10 and 30 September 2003. It is shown that large ozone loss rates at the second half of September are a result of halogen-catalyzed destruction enhanced by the strong increase of solar illumination of the vortex area during this time of the year.
 We gratefully acknowledge all members of the science team of the Improved Limb Atmospheric Spectrometer (ILAS-II) led by Y. Sasano and H. Nakajima for processing ILAS-II version 1.4 data. ILAS-II was developed by the Ministry of the Environment, Japan (MOE), and was on board the ADEOS-II satellite launched by the Japan Aerospace Exploration Agency (JAXA). ILAS-II data were processed at the ILAS-II Data Handling Facility, National Institute for Environmental Studies (NIES). Thanks are also due to the European Space Agency for providing MIPAS near-real-time level 1b data and the UK Meteorological Office for providing meteorological analyses.