We have used a three-dimensional chemical transport model to quantify Arctic ozone loss in 2004/2005 and compare it to other winters through 2006. Relative to Arctic stratospheric variability, 2004/05 was a very cold winter with large regions of possible NAT (nitric acid trihydrate) and ice polar stratospheric cloud formation. These areas were the largest during the past 12 years in January and the vortex area was similarly the largest in early March. Accordingly, the model produces strong denitrification, extensive chlorine activation and large chemical ozone loss of up 75% locally and ∼140 DU in the vortex-averaged column, which slightly overestimates that derived from observations. Compared with similar calculations for recent years, 2004/05 compares with 1999/2000 as one of maximum modelled loss. Sensitivity experiments show that small regions of extreme ozone loss, near 100% at some altitudes, could have happened if the winter of 2004/05 was followed by a spring like 1997 with a long-lasting cold polar vortex.
 Arctic winter/spring chemical ozone loss has been extensively studied over the past 15 years and we now have a good understanding of the chemical processes which drive the loss under given meteorological conditions. The focus has now switched to monitoring, and predicting, how the meteorology of the Arctic may change and what impacts this will have on O3 depletion in the next decade or so.
 The magnitude of ozone loss is very sensitive to stratospheric temperature during the winter [e.g., Goutail et al., 2005] and the largest local losses of 70% so far observed occurred in the cold winter 1999/2000 [e.g., Sinnhuber et al., 2000]. Observations during the Arctic winter 2004/05 indicated that this was also a cold winter with, for example, ice PSCs (polar stratospheric clouds) observed by the groundbased lidar at Ny-Ålesund (79°N, 12°E) in late January 2005 for the first time (M. Maturilli, personal communication, 2005). Estimates of ozone loss for winter 2004/05 have been presented by Manney et al. , Rex et al. , and Singleton et al. . Comparisons with our study are given in section 3 below. von Hobe et al.  used the in situ M55 Geophysica data and CLaMS model simulations to determine strong chlorine activation, denitrification and ozone loss for this winter.
 Following a situation in which models tended to underestimate Arctic O3 loss, recent studies with the latest version of the SLIMCAT 3D CTM have shown much better agreement between the model and the observations in the lower stratosphere (LS) (i.e. winters 1999/2000, 2002/03, and 2003/04; see Feng et al.  and Singleton et al. [2005, 2007]). Overall the model also appears able to reproduce the past climate sensitivity of Arctic ozone depletion on temperature [Chipperfield et al., 2005]. In this paper we use SLIMCAT to quantify the amount of chemical ozone loss in the cold Arctic winter/spring of 2004/05 and compare it with the other 11 Arctic winters under different meteorological conditions, including the recent, warmer winter of 2005/06. Sensitivity experiments are also performed to see whether an “Arctic ozone hole” could have happened if the winter of 2004/05 had remained cold with a long-lasting polar vortex, such as in 1997.
2. Model and Experiments
 SLIMCAT is an off-line 3-D CTM described in detail by Chipperfield . The model uses a hybrid σ-θ vertical coordinate [Chipperfield, 2006] and here horizontal winds and temperatures are specified using European Centre for Medium-Range Weather Forecasts (ECMWF) meteorological analyses (ERA40 reanalyses [Uppala et al., 2005] until 1999; operational analyses thereafter of 60 vertical levels to 0.1 hPa until 1 February 2006, then 91 levels to 0.01 hPa). Vertical advection in the θ-level domain is calculated from diabatic heating rates using a radiation scheme which gives a better representation of vertical transport and age-of-air [Chipperfield, 2006]. The model contains a detailed stratospheric chemistry scheme and a simple NAT-based denitrification scheme. The details are as described by Feng et al. .
 In this study a series of higher resolution (2.8° × 2.8°) runs were initialised from a low resolution (7.5° × 7.5°) multiannual run, which started in 1977, on December 1 of each year. All runs used 24 σ-θ levels from the surface to ∼55 km (resolution of ∼2 km in the LS) with pure θ surfaces above 350 K. The surface values of tropospheric source gases (CH4, N2O, halocarbons) were specified from World Meteorological Organization (WMO)  and an extra 100 pptv of chlorine was assumed to reach the stratosphere from short-lived Cl source gases [WMO, 2003]. The model includes an extra 6 pptv of bromine reaching the stratosphere from short-lived species [Feng et al., 2005], giving a total loading of about 21 pptv in 2004. Singleton et al.  also used the same model version to infer ozone loss for the Arctic winters 1999/2000 and 2004/05. Two sensitivity runs were performed where the model run for 2004/05 was continued from February 28, 2005, using meteorology from either 1997 or 2000.
Figure 1a shows the minimum temperatures (Tmin) north of 50°N from December 2004 to April 2005 from the ECMWF analyses. Clearly, low temperatures (Tmin ≤ 195 K) occurred persistently above 450 K during the winter and descended with time, even reaching below 350 K in late January 2005. As a result large areas of PSCs could exist. Figure 1b shows the diagnosed area of the possible NAT (nitric acid trihydrate) PSC existence using the expression of Hanson and Mauersberger  and modelled H2O and HNO3. NAT PSCs could have formed around 500 K from early December 2004, then gradually extended down to the LS after mid-December and even down to 350 K after mid-January 2005. NAT PSCs could mainly form between 400–600 K, and the largest area of potential NAT PSC existence reached over 14 × 106 km2 in late January 2005 but decreased to less than 2 × 106 km2 by late February. A small area of possible NAT PSCs occurred below 493 K in early March 2005.
Figures 1c and 1d show the time series of potential NAT and ice PSC areas at 456 K for the winters from 1994/95 to 2005/06. The ice PSC existence calculation used the modelled H2O and the expression of Marti and Mauersberger . Figure 1c shows that the largest potential NAT PSC area at 456 K reached ∼15×106 km2 in late January 2005 while winter 1998/99, along with 2001/02, had the smallest potential PSC area (no more than 2 × 106 km2). NAT PSCs formed late in 1996/97 but lasted longer, until late March 1997, due to the very low temperatures in the long-lasting vortex in that year. Lidar observations confirmed that PSCs existed as late as March 18, 1997 [Donovan et al., 1997]. Figure 1c also shows that very early NAT PSCs could have formed in 1994/95 and 2002/03, although higher up at 493 K just 2002/03 stands out with the earliest PSC formation (not shown). Compared with the previous 10 years, we can see clearly that very low temperatures in 2004/05 caused the largest area of possible PSC I and II formation after mid January, which indicates a strong potential for O3 loss. In contrast, winter 2005/06 was very warm with very limited potential for occurrence of NAT PSCs. Figure 1e shows the area of the polar vortex, defined as the area enclosed by the 36 PVU (1 PVU = 10−6 Km−2 kg−1 s−1) modified potential vorticity (MPV, referenced to 475 K) [Lait, 1994] contour. At 456 K the polar vortex in 2004/05 had already formed by December and dissipated in April. 2004/05 had the largest area of polar vortex with maximum ∼30 × 106 km2, while 2005/06 had the smallest. 2005/06 had the shortest-lived polar vortex which ended late January while 1996/97 had the longest-lived polar vortex in the data record [Coy et al., 1997].
3.2. Chlorine Activation and Denitrification
 The extent of chlorine activation plays a key role in polar ozone destruction, and denitrification delays the deactivation process. The modelled chlorine activation and denitrification is shown in the auxiliary material. In summary, the low temperatures led to strong denitrification from early January, which was larger than that modelled for 1999/2000. There was large activation until early March which was similar to 1999/2000. After this higher temperatures led to relatively rapid deactivation in 2004/05.
3.3. Ozone Loss
 The meteorological conditions (temperature and polar vortex) in 2004/05 were extreme (Figure 1) and by some measures colder than other recent Arctic winters [Manney et al., 2006]. We now investigate whether the model also gives a good simulation of ozone observations in this winter. Ozone profile comparisons at Ny-Ålesund and Sodankylä for selected dates in 2004/05 (around day 22 of each month at Ny-Ålesund and around day 10 at Sodankylä) are shown in Figure S1 in the auxiliary material. The model simulates the profiles very well, though O3 is slightly overestimated around 100–300 hPa, likely due to the coarse vertical resolution in the model below 350 K. Figure S1 also shows the modelled chemical ozone loss (diagnosed as the difference with respect to a passive O3 tracer initialised on December 1) on the same profile days. This shows that the majority of the ozone loss occurred between 400–650 K with the largest loss around 450 K.
Figure S1 also shows time series comparisons with sonde observations at 460 K. Further comparisons are included in Figures S2 and S3 in the auxiliary material. Ny-Ålesund and Sodankylä were well inside the Arctic vortex in February and late March. Overall, the model gives a good simulation of observed O3. However, the model overestimates the observed O3 at the start of the winter at Ny-Ålesund (by ∼0.3 ppmv) but agrees well at Sodankylä. The model also underestimates the observations around days 70–80 (the maximum difference reaches ∼0.5 ppmv) (see also Figure S2, auxiliary material). This is clearly a tendency for the model to overestimate ozone loss in mid March 2005. The model also successfully reproduces the increase of December ozone, due to the descent of higher ozone mixing ratios, and the large decline from January to mid-March inside the polar vortex. The maximum modelled ozone loss was 2.0–2.3 ppmv inside the polar vortex, diagnosed from days when the vortex is over the stations. As the vortex moves the diagnosed loss above a station can decrease due to transport effects.
Manney et al.  used Microwave Limb Sounder (MLS) O3 and N2O observations to estimate chemical O3 loss for the Arctic winter 2004/05. They derived a rough estimate for the vortex-averaged loss of 1.2–1.5 ppmv near 450 K, but noted that the winter was dynamically disturbed and local losses in the outer vortex were as large as ∼2 ppmv. Our modelled ozone loss is larger than their vortex-averaged values but more consistent, though slightly larger, with their maximum local losses. Manney et al.  deduced that the peak O3 loss near 18 km for 2004/05 was not as large as some previous colder winters. Rex et al.  quantified the ozone loss from sonde, SAGE III and POAM III data and deduced a column loss of 121 ± 20 DU and noted that this 2004/05 loss was the largest so far observed. The equivalent modelled column loss over the same period is 134 DU (see auxiliary material). Singleton et al.  also quantified ozone loss in Arctic 2004/05 winter using various satellite observations and compared it with the same version of the SLIMCAT chemical transport model (CTM) used here. The model reproduced the magnitude and timing of the loss although the maximum loss was 10–15% larger than observed. Vortex-averaged profile comparisons in the auxiliary material (Figure S3) confirm that the model appears to overestimate the observed loss between mid January and mid March. It is worth reiterating that we no longer have the situation where the model significantly underestimates ozone loss and that this version of the model gave reasonable comparisons in earlier winters [e.g., Feng et al., 2005]. It is not possible to isolate the specific reason(s) for this apparent overestimate without comprehensive comparisons and the winter 2004/05 was not extensively observed as in the winters 1999/2000 and 2002/03. Comparisons with the limited in-situ aircraft data show that the model does overestimate the tracer descent in this winter (Figure S4, auxiliary material). This would increase inorganic chlorine (Cly) in the lower stratosphere leading to increased O3 loss. Also, the simple model NAT-based denitrification scheme was based on detailed comparisons for the winter 1999/2000. It produces strong denitrification in 2004/05 but this may be an overestimate for the different meteorology (e.g. the assumed large NAT particle size may not be appropriate for the later winter). Resolving this issue will probably require detailed studies of earlier winters with similar meteorology for which extensive data is available (e.g., 1999/2000) or the use of new satellite date for recent years.
3.4. Comparison With Previous Winters
Figure 2 shows time series of vortex-averaged chemical ozone loss at 456 K and summed over the partial LS (380–550 K) column for 12 winters. The zonal mean chemical ozone loss over the latitude band 65°N–90°N is used if the value of MPV is less than 36 PVU. This definition tends to exclude more midlatitude air masses below 400 K than one based on the 65°N equivalent latitude contour [e.g., Feng et al., 2005]. At 456 K there is an average ∼60% chemical O3 loss by the end of March for 2004/05 which is very close to the largest modelled loss of 1999/2000. (Figure 3 and Animation S2 in the auxiliary material show that the maximum local loss inside the vortex reached 75%.) There was also large polar ozone loss in 1995/96 (∼50% by the end of March). The modelled ozone losses at 456 K were similar in 1994/95, 1996/97 and 2002/03 (∼35% by the end of March). In the warm winters (e.g., 1998/99, 2000/01, 2001/02 and 2005/06) there was less ozone loss (no more than 15–20%) due to the short period of PSCs and chlorine activation and the disturbed polar vortex (Figure 1).
 The LS partial column ozone losses show very similar variations to 456 K in terms of timing and year-to-year variations although some differences exist. Clearly, 1999/2000 and 2004/05 have the largest partial column ozone loss which reached about 140 DU by the end of March, followed by 1995/96 and 1994/95, while 1998/99 and 2005/06 had the smallest column ozone loss. Although the ozone loss at 456 K in 1994/95 was less than 1995/96 the LS column ozone loss was very similar by the end of March. Winter 1997/98 was also cold, but the ozone loss started later, from mid-January, in line with the possible PSC formation. In 2002/03, the chemical loss started earlier than in the other winters due to extremely low temperatures, early chlorine activation and vortex distortions leading to sunlight exposure [e.g., Tilmes et al., 2003; Feng et al., 2005; Goutail et al., 2005; Raffalski et al., 2005].
Figure 2 also shows how the availability of sunlight limits O3 loss during this period. The shape of the envelope of maximum loss (i.e. conditions of full activation) on any day tends to follow the shape of the curve of accumulated polar insolation (plotted on an arbitrary scale). Interannual variability in the midwinter vortex location, as in 2002/03, can lead to relatively strong losses early on, but the increasing daylight determines the overall increase until late March. Even in the coldest winters, the modelled loss rate decreases after mid March, due to deactivation associated with vortex warming and breakdown and we have so far avoided extreme O3 loss. Whether such extreme loss can occur over the next decade or so, while halogen levels remain high, will require the occurrence of an exceptionally long-lived and cold vortex. Figure 2 includes the result of the one sensitivity experiment where the run for 2004/05 was extended using meteorology for 1997. The 1997 meteorology extends the loss beyond that modelled for other years, producing a vortex-averaged column loss of nearly 170 DU and some regions inside the vortex of 100% loss (see Figure 3 and auxiliary material). Figure S7 (in auxiliary material) compares the modelled profile of vortex-averaged O3 loss for the 7 relatively coldest years. Based on the model, 2004/05, along with 1999/2000, shows the largest average loss in the LS of around 2.2 ppmv at 430 K.
3.5. Sensitivity Experiments
 It is interesting to explore what would have happened if the large ozone loss in Arctic 2004/05 winter was followed by a spring with a long-lasting cold polar vortex. To do this we have performed two sensitivity experiments where the model run for 2004/05 was repeated from late February onwards using the meteorology of either 1997 or 2000. The meteorology was changed on February 28, 2005, corresponding to the time when winter 1997 became colder than 2005 in the lower stratosphere. On the day when the meteorology was changed the modelled chemical fields were binned as a function of equivalent PV latitude based and θ. Using the meteorology for 28 February 1997 or 29 February 2000 the model fields were then remapped in equivalent latitude - θ space to ensure the tracers were consistent with the atmospheric circulation for the different years. The model run was then continued using 1997 or 2000 meteorology.
Figure 3 (top) shows the minimum temperature (K) at 456 K from March to April for 2005, 2000 and 1997 from ECMWF analyses. Figures 3 (middle) and 3 (bottom) compare the maximum modelled local ozone loss and minimum total column ozone in the polar region when 2004/05 winter is followed by 2000 and 1997 meteorology. Temperature increased and was above 195 K from around March 10 in 2005 and 2000, while in 1997 is remained low until late March. There would have been more ozone loss if the Arctic winter 2004/05 after late February was followed by 2000 meteorological conditions. The local ozone loss would reach ∼90% (also see in the animation). Interestingly, Arctic ozone loss would have been even more severe and complete loss would have occurred around late March if the winter of 2004/05 was followed by 1997 conditions which had a record long-lasting cold polar vortex. Figure S7c shows the total column ozone comparison. TOMS data shows that there was no “Arctic ozone hole” (total column ozone was above 220 DU) except a minihole occurred Scandinavia around 5 March 2005. SLIMCAT overestimates the observed total column ozone largely due to too much O3 in the lowermost stratosphere (see profile comparison in Figures S1 and S3). However, the model has very similar variations with observations. There would be no ozone hole structure when the Arctic winter 2004/05 was followed by 2000 meteorological conditions even the ozone loss would reach ∼90%. Interestingly, the modelled total column ozone would be below 220 DU after 19 March to early April when the winter of 2004/05 was followed by a spring like 1997 with a long-lasting cold polar vortex. However, the “Arctic ozone hole” area has relative small horizontal scale with respect to the wider Antarctic polar ozone hole.
 The meteorological conditions in 2004/05 were extreme: Very low temperatures occurred persistently throughout the winter polar lower stratosphere and even extended below 350 K by the end of January 2005. Such low temperatures caused the largest areas of possible PSC formation at 456 K (∼18 km) in recent years. The 2004/05 Arctic winter also exhibited one of the largest areas for the polar vortex until mid-March.
 Given these cold conditions the modelled ozone loss in 2004/05 was correspondingly large. The local maximum modelled ozone loss in 2004/05 reached 75% over a limited area of the vortex (see auxiliary material). The overall degree of modelled ozone loss inside the polar vortex in 2004/05 Arctic winter/spring at 456 K has a very similar value as the record loss observed in the year 1999/2000. However, comparisons with in-situ data show that the model does overestimate the tracer descent in early March 2005 which would increase Cly in the lower stratosphere leading to increased O3 loss. This is clearly also a tendency for the model to overestimate ozone loss in mid March 2005. These discrepancies in the transport (tracer diabatic descent and mixing etc.) and chemical (chlorine and denitrification etc.) contributions to the ozone changes, which vary with year studied, require further investigation. Based on the model results, before 1 March 2005, during conditions of strong chlorine activation, O3 loss was limited by the availability of sunlight, and the modelled loss was near the maximum limit of interannual variability. However, less chlorine activation in March 2005 slowed down further chemical ozone loss in the later year and prevented severe loss. Model sensitivity experiments showed that Arctic ozone loss would have continued and small regions of severe loss could have occurred if the winter of 2004/05 had been followed by a spring with a long-lasting cold polar vortex.
 This work was supported by the EU SCOUT-O3 project and also by the U.K. NERC. We are grateful to NASA for the use of the TOMS data and BADC for ECMWF analyses. We thank the reviewers for the helpful comments.