Quantitative chemical ozone loss rates at the 475 K isentropic surface inside the Arctic polar vortex are evaluated for six winters (January through March) using a satellite-based Match technique. Satellite observational data are taken from the Polar Ozone and Aerosol Measurement (POAM) II for 1994–1996, the Improved Limb Atmospheric Spectrometer (ILAS) for 1997, and the POAM III for 1999–2000. The largest ozone loss rates were observed in the end of January 1995 (∼50 ± 20 ppbv d−1), February 1996 (∼40–50 ± 15 ppbv d−1), February 1997 (∼40 ± 8 ppbv d−1), January 2000 (∼60 ± 30 ppbv d−1), and early March 2000 (∼40 ± 10 ppbv d−1). The probability of polar stratospheric cloud (PSC) existence is estimated using aerosol extinction coefficient data from POAM II/III and ILAS. Ozone loss and the PSC probability are strongly correlated and an absolute increase of 10% in the PSC probability is found to amplify the chemical ozone loss rate during Arctic winter by approximately 25 ± 6 ppbv per day or 3.2 ± 0.7 ppbv per sunlit hour. Relationships between average Arctic winter ozone loss rates and various PSC- and temperature-related indices are investigated, including the area of polar vortex that is colder than the threshold temperature for PSC existence (APSC), the PSC formation potential (PFP), and the potential for activation of chlorine (PACl). Of these three, PACl provides the best proxy representation of interannual variability in Arctic ozone loss at the 475 K level. Large ozone loss occurred primarily for air masses that experienced low temperatures between 187 K and 195 K within the previous 10 days and the ozone loss rates clearly increase with decreasing the minimum temperature. The particularly large ozone losses of ∼9 ± 3 ppbv per sunlit hour in February 1996 and January 2000 were associated with low minimum temperatures of 187–189 K, simultaneously with high PSC probabilities.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Stratospheric ozone in the polar region during winter and spring varies substantially from year to year. This interannual variability (IAV) is due to both chemical processes (chlorine- and bromine-catalyzed ozone destruction) and dynamical processes (transport by the residual circulation and mixing across the edge of the polar vortex). The monthly averages of total column ozone during early spring (March in the Arctic and October in the Antarctic) observed from satellite instruments, such as the Total Ozone Mapping Spectrometer (TOMS) or the Ozone Monitoring Instrument (OMI), provide useful measures of the IAV in stratospheric ozone depletion; however, these observational representations of IAV cannot differentiate between dynamical and chemical variations. In the Arctic, the IAV in planetary wave activity affects both transport into the polar vortex and chemical ozone loss, thereby contributing significantly to the IAV in stratospheric ozone [Tegtmeier et al., 2008a]. This strong dependence of Arctic ozone on dynamics makes it challenging to diagnose the chemical loss rate of ozone from observations that also include the effects of transport and mixing [World Meteorological Organization (WMO), 2011].
 Several methods have been proposed to quantitatively evaluate the chemical loss amounts of ozone. These methods include: the ozone-tracer correlation method (first used by Proffitt et al. ), the vortex-average method [Manney et al., 1994; Braathen et al., 1994], the Match method [von der Gathen et al., 1995], the passive subtraction method [Lefèvre et al., 1998], and the Lagrangian transport calculation method [Manney et al., 1997]. The details and references for each of these methods have been summarized by WMO [2003, 2007]. Arctic ozone losses derived using different methods or data generally agree quantitatively. Examples of this agreement have been published for the winters of 1994/1995, 1995/1996, and 1996/1997 [Harris et al., 2002], 1999/2000 [Newman et al., 2002], and 2004/2005 [WMO, 2007]. These comparative studies have only discussed accumulated ozone loss either on monthly time scales or for the whole winter–spring, and have not examined local ozone loss rates. Of the aforementioned methods, the Match method is the only one that can derive local ozone loss rates over short periods.
 The Match analysis method requires intensive measurements of ozone with a high vertical resolution throughout the polar stratosphere. These requirements are met by solar occultation sensors onboard sun-synchronous polar-orbiting satellites, such as the Improved Limb Atmospheric Spectrometer (ILAS), the Polar Ozone and Aerosol Measurement (POAM) II, and the POAM III. Among satellite-borne instruments, these solar occultation sensors are capable of the highest vertical resolution measurements because they use the bright sun as a light source. With the sensor in sun-synchronous polar orbit, approximately 14 ozone profiles per day are obtained over high-latitude regions in both the Northern Hemisphere (NH) and Southern Hemisphere. The data quality of the satellite-borne sensor measurements is spatially and temporally homogeneous, which represents an advantage over ozonesonde measurements.
 We have previously applied the Match technique to ILAS ozone profile data to successfully analyze quantitative chemical ozone loss rates and amounts in the Arctic polar vortex for the winter and spring of 1997 [Sasano et al., 2000; Terao et al., 2002]. To compensate for the weaknesses of the satellite sensor data (i.e., a lower vertical resolution and a larger volume of air sampled than ozonesonde data), the analysis method has been refined by adding multiple trajectories with very strict criteria [Terao et al., 2002]. This modification allows the method to more accurately identify double-sounded air masses. Terao  extended the satellite-Match analysis to POAM II and POAM III data, and presented preliminary estimates of ozone loss rates in the Arctic polar stratosphere for six winters between 1994 and 2000. The satellite-Match technique has also been applied to POAM III measurements in the Antarctic [Hoppel et al., 2005]. In this paper, we present and discuss comprehensive evaluations of Arctic chemical ozone loss rates derived from POAM/ILAS-Match analysis and analyze these results in the context of recent findings regarding polar stratospheric cloud (PSC) and temperature-related indices.
 The relationship between chemical ozone loss and temperature should be quantitatively established in regard to possible future changes in stratospheric temperatures [e.g., Stolarski et al., 2010]. PSCs may form at temperatures below approximately 195 K in the polar stratosphere. PSCs are variously composed of nitric acid trihydrate (NAT; known as Type Ia PSC), supercooled ternary (H2SO4/HNO3/H2O) solutions (STS; known as Type Ib PSC), and ice particles (known as Type II PSC) [e.g., Peter, 1997; Lowe and MacKenzie, 2008]. The saturation temperature for NAT (or threshold temperature for NAT existence, TNAT) can be calculated from stratospheric concentrations of water vapor (H2O) and nitric acid (HNO3) [Hanson and Mauersberger, 1988]. The threshold temperature for STS, which is several Kelvins lower than TNAT, can be similarly parameterized [Tabazadeh et al., 1994; Carslaw et al., 1995]. The geographical area or volume of vortex air for which the temperature is lower than TNAT (APSC or VPSC) is conventionally used as a measure of chlorine activation due to PSCs, and hence ozone losses in the polar region [Rex et al., 2004, 2006; Tilmes et al., 2004].
 Liquid binary (H2SO4/H2O) solutions could also play an important role for chlorine activation in the winter polar vortices through heterogeneous reactions, even in the presence of moderate amounts of volcanic aerosol that followed the eruption of Mt. Pinatubo in 1991 [e.g., Kawa et al., 1997; Engel et al., 2000; Massie et al., 2000]. The threshold temperature for activation of chlorine (TACl) has therefore also been suggested as a useful proxy for the chemical loss of ozone [Drdla, 2005; Drdla and Müller, 2010]. TACl is defined as the temperature at which active chlorine (ClOx = ClO + 2 × ClOOCl) increases by an amount equivalent to 10% of the total inorganic chlorine (Cly) over a period of 1 day. Chlorine activation of 10% is equivalent in the current atmosphere to about 0.35 ppbv of ClO [Drdla and Müller, 2010]. TACl has been used in several studies, particularly with regard to geoengineering schemes for enhancing stratospheric sulfate aerosols and possible changes in stratospheric water vapor concentrations [Tilmes et al., 2007, 2008; Feck et al., 2008].
 Previous studies have shown that VPSC correlates well with observed column Arctic ozone losses accumulated over the entire winter [Rex et al., 2004; Tilmes et al., 2004]. The observed correlation has been reproduced by a three-dimensional (3-D) chemical transport model [WMO, 2007]. The actual occurrence of PSCs does not necessarily correspond to VPSC, however. Observations by a satellite-borne aerosol lidar, the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), show that the vertically integrated area of observed PSCs is generally smaller than VPSC in both the Arctic and Antarctic [Pitts et al., 2007, 2009; WMO, 2011]. These results suggest that estimates of PSC occurrence derived from TNAT-based VPSC are likely to be biased high. Tilmes et al. [2006, 2007, 2008] proposed two different indices, the TNAT-based PSC formation potential (PFP) and the TACl-based potential for activation of chlorine (PACl), and reported that column Arctic ozone losses correlate better with PACl than PFP. Some chemistry climate models (CCM) qualitatively reproduced correlations between ozone loss and PACl but the observed and simulated correlations quantitatively differed [SPARC CCMVal, 2010]. Comprehensive observational studies are thus required to better quantify the empirical relationships among ozone loss rates, observed PSCs, and temperature.
 This study analyzes observational evidence of relationships among ozone loss rates, PSC occurrence, and temperature during the Arctic winter, also considering PSC- and temperature-related indices (APSC, PFP, and PACl). Previous studies have shown empirical correlations between the IAVs of ozone loss and these indices [e.g., Rex et al., 2004; Tilmes et al., 2004, 2006, 2008; Müller et al., 2008; Harris et al., 2010]; however, these studies have not shown the IAV of PSC occurrence directly. The ILAS and POAM II/III provide concurrent observations of both ozone mixing ratio and aerosol extinction coefficient (AEC). The occurrence of PSCs can be estimated by statistical analysis of the AEC data. The POAM/ILAS Match analysis provides unique and important information on relationships among observed ozone loss rates and PSC existence.
 We derive chemical ozone loss rates in the Arctic for the six winters of 1993/1994, 1994/1995, 1995/1996, 1996/1997, 1998/1999, and 1999/2000 on the 475 K isentropic surface (approximately 19 km in altitude), which serves as a representative level for the lower stratosphere. Previous studies have reported large ozone losses within this analysis period, particularly during the winters of 1995/1996, 1996/1997, and 1999/2000: the accumulated ozone loss around the 475 K level was 0.68–0.75 ppmv (parts per million volume) between 29 January and 3 March 1996; 0.9–1.1 ppmv between 30 January and 21 March 1997 [Harris et al., 2002]; and 1.7 ppmv (±20%) between 20 January and 12 March 2000 [Newman et al., 2002]. The ozone loss was small during the winters of 1997/1998 and 1998/1999 [Schulz et al., 2001]. The satellite measurements from the Microwave Limb Sounder (MLS) [Manney et al., 2003] and Halogen Occultation Experiment (HALOE) [Tilmes et al., 2004] also showed the large ozone losses in the Arctic winters of 1995/1996 and 1999/2000.
 The data and analysis method are described in section 2. The time series of ozone loss rates is presented in section 3.1. The time series of PSC occurrence and relationships between ozone loss and PSC occurrence, APSC, PFP, and PACl are shown in section 3.2. The minimum temperatures experienced during the preceding 10 days by the corresponding air parcels are examined in section 3.3. The IAV in TACl during 1994–2000, when moderate amounts of sulfate aerosols had existed, is discussed in section 4, and the results are compared with those of ozonesonde-Match analyses. The conclusions of the study are summarized in section 5.
2. Data and Methods
Figure 1a shows the zonal and temporal coverage of POAM II, ILAS, and POAM III measurements in the NH. Each sensor covers the latitude region of 55°N–71°N. The measurement periods extend from October 1993 to November 1996 for POAM II, November 1996 to June 1997 for ILAS, and April 1998 to December 2005 for POAM III. The latitudes of POAM II/III and ILAS measurements during the winter–spring period of January, February, and March (herein JFM) range from 64°N at the beginning of January to 68°N (POAM II/III) or 70°N (ILAS) at the beginning of March (Figure 1b). The difference in latitudes between POAM II/III and ILAS measurements is small, with a maximum difference of 2 degrees in March. The POAM II/III and ILAS measurements were made over a limited range of latitudes but covered all longitudinal regions (Figure 1c).
2.1. POAM II/III Data
 POAM II and POAM III were developed by the United States Naval Research Laboratory. POAM II was launched onboard the Satellite Pour l'Observation de la Terre (SPOT) 3 on 26 September 1993 [Glaccum et al., 1996]. POAM III, which succeeded POAM II, was launched onboard SPOT 4 on 23 March 1998 into the same orbit as SPOT 3 [Lucke et al., 1999]. POAM II and POAM III instruments were solar occultation sensors with a 9 channel photometer. The photometer on POAM II covered wavelengths between 352 nm and 1060 nm with a spectral resolution of 2.0–16.7 nm and the photometer on POAM III covered wavelengths between 354 nm and 1018 nm with a spectral resolution of 2.1–17.7 nm. The instantaneous field-of-view (IFOV) of POAM II/III in the vertical direction was approximately 0.8 km. POAM II provided vertical profiles of ozone, nitrogen dioxide, and AECs at six wavelengths (352.3, 441.6, 601.4, 781.0, 921.0, and 1060 nm), and POAM III provided vertical profiles of ozone, nitrogen dioxide, water vapor, and AECs at six wavelengths (353.4, 442.3, 603.4, 779.4, 922.4, and 1018 nm). The SPOT 3 and SPOT 4 satellites were launched into a sun-synchronous subrecurrent orbit at 833 km altitude with an inclination of 98.7 degrees and a period of 101 min. The local mean Sun time at the descending node was 10:30.
 The present study uses vertical profiles of ozone mixing ratio and AEC at 1060 nm for POAM II and at 1018 nm for POAM III. The data were retrieved using the POAM II V6 [Lumpe et al., 1997] and POAM III V3 [Lumpe et al., 2002] retrieval algorithms. Error estimates for ozone retrievals using POAM II V6 are the same as for those using the V5 retrieval algorithm. The POAM II V5 ozone retrievals were validated by comparisons with ozonesonde measurements [Deniel et al., 1997] and with satellite measurements from MLS, HALOE, and Stratospheric Aerosol and Gas Experiment (SAGE) II instruments [Rusch et al., 1997]. Random and systematic errors in the POAM II V5 ozone retrievals are both approximately 5% between 10 km and 50 km altitude. A comparative study of ozone profiles measured by seven satellite instruments, including the POAM II V6 data, reported that these observations of ozone generally agree to within 0.25 ppmv in the lower stratosphere [Manney et al., 2001]. Another comprehensive comparison of multiple satellite-borne sensors revealed a negative bias of up to 0.4 ppmv in POAM II V6 ozone in the lower stratosphere [Danilin et al., 2002].
 The POAM III V3 ozone retrievals were validated by comparison with observations from balloon-borne, aircraft-borne, and ground-based instruments made during the SAGE III Ozone Loss and Validation Experiment and Third European Stratospheric Experiment on Ozone (SOLVE/THESEO) 2000 campaign [Lumpe et al., 2003]. The POAM III ozone data agree with the SOLVE/THESEO data to within 7–10% with no detectable bias between 14 km and 30 km altitude. Randall et al.  compared POAM III V3 ozone measurements with both ozonesonde measurements and satellite observations from the HALOE and SAGE II instruments, and showed that agreement among these data sets is generally within 5% from 13 km to 60 km in the NH. The latest POAM III retrieval is V4; however, the V4 ozone data have changed very little from the V3 data, and the results are not sensitive to the use of V3 rather than V4.
 The AEC data from POAM II V6 at 1060 nm and POAM III V3 at 1018 nm have been compared with SAGE II data at 1020 nm. For POAM II, differences relative to SAGE II are within ± 10% between 12 km and 27 km and random errors are approximately 10% below 25 km [Randall et al., 2000]. For POAM III, differences relative to SAGE II are within ±30% between 10 km and 22 km and random errors are less than 20% below 22 km [Randall et al., 2001]. During the lifetime of POAM III, there has been some instrument degradation that affects the retrieval of AEC (POAM III V4 description); however, the effect of this degradation is small through 1998–2000 before increasing significantly in early 2001.
Figure 2 shows daily variations in the number of POAM II/III and ILAS measurements (open columns) and in the number of measurements inside the polar vortex at the 475 K isentropic surface (green) in the NH from 1 January to 31 March in 1994, 1995, 1996, 1997, 1999, and 2000. In this study, we focus on ozone loss rates and PSC occurrence inside the polar vortex. The vortex edge and the vortex boundary region were defined for each day using a method similar to that of Nash et al.  in which potential vorticity (PV) gradient and wind speed were used. We followed the Nash et al. method, except that the location of the vortex edge and boundary region was constrained by the derivative of the product of PV gradient by wind speed to reduce multiple peaks of PV gradient [Ninomiya and Nakane, 1998].
 Throughout the JFM period, POAM II/III and ILAS made observations both inside and outside the polar vortex during all analyzed years. In 1994, the number of POAM II measurements was smaller by half when compared with measurements in the following years both because data collection was compromised by an alternating day on/day off schedule and because of data loss during early mission operation. The number of measurements inside the vortex is smaller in 1995 than in the other winters because the vortex area was small during February and March (shown later). In 2000, some episodes of missing POAM III data occur during early January and mid-February. The numbers of total measurements and measurements inside the vortex during the JFM period are summarized in Table 1. We did not analyze POAM III data after 2000 because the number of POAM III measurements was reduced by half after mid-March in 2000. The total numbers of POAM III measurements in the NH during JFM are 644 in 2001, 600 in 2002, 1072 in 2003, 650 in 2004, and 790 in 2005. These numbers are similar to POAM II in 1994, which produced very few double-sounded air parcels (Table 1). Although we have only used data from 1999 and 2000, the POAM III measurements are crucial to this study because they provide insight into episodes of no ozone loss in 1999 and huge ozone loss in 2000.
Table 1. Total Number of Measurements, Number of Measurements Made Inside the Vortex, and Number of Matching Pairs Inside the Vortex on the 475 K Isentropic Surface Observed by POAM II/III and ILAS for 1 January Through 31 March of the Specified Year
Inside the Vortex
2.2. ILAS Data
 ILAS was developed by the Ministry of the Environment, Japan (formerly Environment Agency of Japan), and launched onboard the Advanced Earth Observing Satellite (ADEOS) on 17 August 1996. ILAS was a solar occultation sensor that consisted of a light-collecting telescope, a sun-tracking device, a sun-edge sensor, electrical circles, and two grating spectrometers [Sasano et al., 1999]. ILAS provided vertical profiles of gaseous species (ozone, nitric acid, nitrogen dioxide, methane, nitrous oxide, and water vapor) using a 44 channel infrared spectrometer that covered wavelengths between 6.21 μm and 11.77 μm with a spectral resolution of 0.13 μm, and AEC, temperature, and pressure using a 1024 channel visible spectrometer covering wavelengths between 753 nm and 784 nm. The IFOV at tangent height was 1.6 km in the vertical direction and 13 km in the horizontal direction. ADEOS was launched into a sun-synchronous subrecurrent orbit at approximately 800 km altitude with an inclination of 98.6 degrees and a period of 101 min. The local mean Sun time at the descending node was 10:40.
 We use vertical profiles of ozone mixing ratio and AEC at 780 nm obtained by the ILAS V5.20 retrieval algorithm [Yokota et al., 2002]. The average estimated root-sum-square total uncertainties in the ozone retrievals are 14% at 15 km, 9% at 20 km, and 5% at 25 km in the NH between November 1996 and June 1997. The measurement repeatability is estimated to be 7% at 15 km and 3% at 20 km. The gas mixing ratios derived from ILAS may be biased for measurements that were collocated in space and time with PSCs [Yokota et al., 2002]; we have therefore eliminated ozone data that may have been measured in the presence of PSCs. The number of PSC-reduced ILAS profiles is shown in Table 1.
 The V5.20 ILAS ozone data have been validated against ozone data obtained by ozonesondes, by balloon-borne and aircraft-borne instruments, by ground-based instruments, and by the satellite sensors HALOE, SAGE II, and POAM II [Sugita et al., 2002]. The agreement between ILAS ozone data and these other data is generally within ±10%. Focusing on the wintertime NH, comparison with the ozonesonde data yields relative differences of 14% at 11–15 km and 8% at 16–20 km during February, and of 7% at 11–15 km and 5% at 16–20 km during March. The V5.20 ILAS AEC data have been compared with SAGE II data that were converted to ILAS wavelength (780 nm) by log linearly interpolated using AEC at 525 nm and 1020 nm [Saitoh et al., 2002]. Differences between ILAS and SAGE II retrievals of AEC are generally within 10–20% for AEC values that exceed 1 × 10−5 km−1.
2.3. Satellite-Match Analysis
 The analysis method that is used for calculating ozone loss rates in this study is nearly identical to that used by Terao et al. , with the exception of the diabatic descent rates and coordinates for the trajectory analysis. Here we briefly summarize the method and describe changes; additional details of the satellite-Match method are provided by Terao et al. .
 POAM II/III and ILAS data are provided as functions of geometrical altitude with 1 km intervals. The vertical coordinate is transformed to a potential temperature coordinate using United Kingdom Meteorological Office (UKMO) assimilation data [Swinbank and O'Neill, 1994], and the ozone mixing ratios and AECs are interpolated onto the 475 K potential temperature surface for analysis. The UKMO assimilation data are also used for meteorological analysis and trajectory calculations. The UKMO data are provided daily at 12:00 universal time (UT) on a 2.5° × 3.75° (latitude × longitude) grid on the 22 Upper Atmosphere Research Satellite (UARS) standard pressure levels.
 Ten day forward trajectories are used to search for air parcels that were sounded twice by POAM II/III or ILAS at different locations and at different times (double-sounded parcels, or match pairs). The trajectory analysis is conducted using an isentropic Lagrangian trajectory model supplied by the UKMO [Terao et al., 2002.] We calculate a set of forward trajectories starting from/around the first measurement point, and a set of backward trajectories starting from/around the matching second measurement point. Several criteria are applied to evaluate trajectory dispersions and check the accuracy of each match pair using a cluster of both forward and backward trajectories. Terao et al.  developed the coordinates for these trajectory clusters to account for the volume of the air mass sampled by the ILAS measurement. The volume of an air mass sampled by a POAM II/III measurement is different from that sampled by an ILAS measurement, however: POAM II/III has a larger IFOV in the horizontal direction than ILAS. We therefore introduce alternative coordinates for the trajectory clusters. Four trajectories are initialized 100 km to the north, south, east, and west of the measurement point at the 475 K level, and two trajectories are initialized at the measurement point location but at ±15 K in potential temperature. In total, 14 trajectories (one forward trajectory from the first measurement point, a cluster of six forward trajectories surrounding the first measurement point, one backward trajectory from the second measurement point, and a cluster of six backward trajectories surrounding the second measurement point) are calculated for each matching pair. The distance criteria for selecting double-sounded air parcels are unchanged from those used by Terao et al. . The ILAS data are reanalyzed using the trajectory coordinates defined for this study, rather than those defined by Terao et al. .
 The calculation of the change in ozone mixing ratio between two observations of a double-sounded air parcel accounts for changes in potential temperature due to diabatic effects. The ozone mixing ratio of the second measurement is modified to account for any potential temperature changes along the trajectory. We use diabatic cooling rates calculated using SLIMCAT 3-D chemical transport model [Chipperfield, 1999].
 A statistical treatment is applied to calculate rates of ozone change for each day. We identify a subset of double-sounded air parcels that are gathered within 7 days prior to and following the target day. Assuming that ozone changes are linearly proportional to the sunlit time along the trajectory, we calculate a proportional coefficient (ozone change against sunlit time) from each subset using the least squares method. The rate of ozone change in ppbv (parts per billion volume) per sunlit hour (ppbv sunlit h−1) can be converted to a rate of ozone change per day (ppbv d−1) by multiplying by the average sunlit time (in hours) for each day. We also calculate ozone change rates using a regrouped data set related to minimum temperatures along 10 day backward trajectories from double-sounded air parcels in section 3.3: A subset of double-sounded air parcels obtained in the minimum temperature range of ±1 K (from 185–187 K to 201–203 K) is used for regression analysis for each month and for each year.
2.4. Detection of PSC
 Observations of AEC made by satellite-borne sensors have previously been used to detect PSCs. For example, Poole and Pitts  proposed an approach to identifying PSCs and identified a trend in the frequency of PSCs using the Stratospheric Aerosol Measurement II (SAM II) instrument. In their approach, they first calculated the median value and median deviation, referred to as the background value and deviation, respectively, of AEC for every 10 day period at each altitude level inside the vortex. Only observations for which the temperature exceeded 200 K were used to calculate the background value and deviation. They then defined the threshold value used to identify PSCs as three times the median deviation above the median value. Values of AEC larger than this threshold were regarded as PSCs provided that the temperature was lower than 200 K.
Fromm et al.  applied a similar method to POAM II AEC at 1060 nm to develop a 3 year Antarctic PSC climatology. This POAM II PSC detection algorithm was later revised by Fromm et al.  and adapted to POAM III AEC by Bevilacqua et al. . The threshold value for these studies was determined as 2.7 standard deviations above the mean. Hayashida et al.  and Saitoh et al.  employed similar PSC detection methods for ILAS AEC at 780 nm, and determined the threshold value as five standard deviations above the mean.
 The present study detects PSCs using POAM II/III and ILAS AEC data at the 475 K isentropic level. The background reference extinction value and its deviation are calculated in the same way as in these previous studies (i.e., the mean and standard deviation are calculated for each 10 d period from all AEC data inside the vortex for which the temperature exceeds 200 K). The threshold value is defined as three standard deviations above the mean, which is between the values defined by Fromm et al.  for POAM and Hayashida et al.  for ILAS.
3.1. Ozone Loss Rates
Figure 3 shows ozone mixing ratios observed by POAM II (1994, 1995, and 1996), ILAS (1997), and POAM III (1999 and 2000) at the 475 K isentropic surface in the Arctic region for the JFM period. Measurements made inside the polar vortex are marked by solid black circles. Figure 3 roughly illustrates the evolution of ozone mixing ratios inside the vortex during JFM. During JFM 1994, ozone mixing ratio decreased slightly from 3 ppmv (mid-January) to 2.5 ppmv (end of March). Two ozone depletion episodes occurred during JFM 1995: in early February and in mid-March. During 1996, ozone mixing ratios decreased from 2.5 to 3.0 ppmv in January to 1.5–2.0 ppmv in March. JFM 1997 is characterized by continuous ozone depletion from the end of January, when ozone mixing ratios inside the vortex are greater than 3 ppmv, to the end of March, when the minimum ozone mixing ratios reach 1.5 ppmv. The ozone mixing ratios inside the vortex vary substantially in March 1997 (observed values ranging from 1.5 ppmv to 3.4 ppmv in the end of March), indicating that the magnitude of chemical ozone loss is strongly dependent on location relative to the vortex [Terao et al., 2002]. Ozone concentrations increased throughout the winter of 1999; this is the natural evolution of polar ozone during winters when chemical ozone depletion is weak. Large ozone depletions occurred during JFM 2000, with ozone mixing ratios inside the vortex decreasing from 3.5 ppmv at the beginning of January to near 1 ppmv in mid-March.
Figure 4 shows the locations of double-sounded air parcels obtained inside the vortex for each Arctic winter as a function of equivalent latitude on the 475 K level. Equivalent latitude is based on the area enclosed by contours of PV. The double-sounded air parcels are distributed in space throughout the polar vortex, from the center of the vortex to the vortex edge, especially in 1996, 1997, and 2000. We therefore define the ozone change rates derived from the match pairs within the vortex as “vortex-averaged” values.
Figure 4 also shows the winter–spring evolutions of the vortex edge and the vortex boundary region for each year. A warming event was observed in February 1995, and the vortex area was relatively small (north of 70°–73°N in equivalent latitude) during February and March of that year. In 1996, the stratospheric final warming began in early March, and the area of the vortex started to diminish from the beginning of March. In 1997, the polar vortex formation occurred relatively late in the season, and the vortex edge was not clear before 9 January. The analysis is therefore only performed for observations made after 10 January 1997. Once formed, the polar vortex in 1997 was both very strong and symmetric around the North Pole, and it persisted until late spring [Coy et al., 1997]. During JFM 1999 the polar vortex was very weak and determination of the vortex edge using PV is only successful for February. We define the vortex edge as 65°N equivalent latitude from 1 January to 30 January. This is a typical location for the vortex edge in the other winters, and double-sounded air parcels at equivalent latitudes higher than 65°N are considered to be inside the vortex for the analysis. The breakup of the vortex occurred in early March in 1999, so the analysis for that year was terminated on 1 March. The vortex breakup in 2000 also occurred relatively early in the season, in late March, so the analysis was terminated on 20 March.
Table 1 shows the number of double-sounded air parcels observed inside the vortex by ILAS and POAM II/III during JFM. In the satellite-Match analysis, matching pairs are identified by chance within the existing data set inside the vortex. This differs from the intentional creation of match pairs in real time during an ozonesonde-Match campaign, in which ozonesonde launches are coordinated with trajectory forecasts. The number of matching pairs in this study therefore varies interannually, and depends on the shape and location of the polar vortex relative to the POAM II/III and ILAS measurement locations. The number of double-sounded air parcels is very small for 1994 because very few observations were made inside the vortex. The statistical significance of detected ozone change rates in 1994 was therefore lower than those in other years. The number of double-sounded air parcels identified for 1995 is larger than for 1994 but still smaller than for the following years. We obtain more than a hundred matching pairs during JFM in 1996, 1997, 1999, and 2000. The largest number of matching pairs is identified in ILAS observations from 1997.
 Ozone change rates (in ppbv d−1) derived from the POAM/ILAS-Match analysis are shown in Figure 5. In 1994, the local maximum ozone loss rates were 25 ± 25 ppbv d−1 in mid-January and 50 ± 20 ppbv d−1 in late February; however, the statistical significance of the ozone change rates derived for 1994 is small. Less than ten double-sounded air parcels were included in the calculation of each ozone change rate, resulting very large error bars. The number of double-sounded air parcels is also relatively small for 1995, and ozone change rates in February and March were calculated from less than ten matching events. A relatively larger number of double-sounded air parcels were identified during January 1995. The local maximum ozone loss rate was 83 ± 29 ppbv d−1 in the end of January (∼50 ± 20 ppbv d−1 for a running average). A second local maximum ozone loss rate of 52 ± 76 ppbv d−1 occurred during mid-March, but the uncertainty in this value is quite large. In 1996, which was a very cold winter in the Arctic stratosphere [Manney et al., 1996], the maximum ozone loss rates occurred at the beginning of January (∼50 ± 20 ppbv d−1), and prolonged significant ozone loss rates (∼40–50 ± 15 ppbv d−1) occurred throughout February. Ozone losses were moderate from early March 1996. The statistical errors for ozone change rates derived for 1996 are smaller than those derived for 1994 or 1995.
 The winter of 1997 is characterized by continuous ozone loss at rates of 20–46 ppbv d−1 from the end of January through the end March. The maximum ozone loss rate is 46 ± 8 ppbv d−1 between late February and early March. A small peak in ozone loss occurred in late January. The ILAS measurements for 1997 provided the largest number of double-sounded air parcels of the six analyzed winters. ILAS therefore produced ozone change rates with high accuracy and the small statistical uncertainty. Terao et al.  presented an in-depth analysis and discussion of ozone loss in 1997.
 The largest ozone loss of the six analyzed winters occurred in 2000, following minimal ozone loss in 1999. The local maximum ozone loss rates during 2000 were approximately 70 ± 30 ppbv d−1 (∼60 ± 30 ppbv d−1 for a running average) during mid-January and approximately 60 ± 30 ppbv d−1 (∼40 ± 10 ppbv d−1 for a running average) during late February and early March.
3.2. PSCs and Ozone Loss Rates
 The number of PSC events per day (red column) and the number of measurements with temperatures below 195 K (blue column) at the 475 K isentropic surface are shown in Figure 2. POAM II, ILAS, and POAM III observed PSCs in every winter but 1999, when no PSCs were detected. Most of the PSCs were observed when temperature was below 195 K. PSCs were observed for two periods in 1994: in mid-January, and at the end of February and beginning of March. In 1995, PSCs were observed from January to early February, along with an isolated PSC event in mid-March (although temperature for this event exceeded 195 K). In 1996, PSCs were observed continuously throughout January and February into early March. Intermittent PSC events were observed in 1997, both in mid-January and from mid-February through mid-March. A large number of PSCs were observed in 2000, particularly in January.
 The total number of measurements and the number of measurements inside the polar vortex vary from day to day and from year to year. Thus, the count of PSC events does not serve as an equivalent index for PSC activity in different portions of the JFM season, nor can it be used to compare and contrast PSC activity during different years. Instead, we introduce a probability of PSC occurrence, which is defined as the ratio (%) of the number of PSCs detected to the number of measurements inside the vortex [Fromm et al., 1999].
 The PSC probability for each day is shown along with the ozone change rates in Figure 5 (red columns). The highest PSC probabilities (∼50%) were observed in January and mid-February of 1996, in January of 1997, and in January of 2000. The PSC probability was 100% on 11 February 1996; however, this value is questionable because only two observations were made inside the vortex on that day. PSC probabilities were consistently smaller during February and March than during January.
 The patterns of PSC probabilities well correspond to the geographical areas where temperatures have been below the threshold temperature for PSC formation (APSC) at 475 K (blue curve in Figure 5; adopted from WMO [2003, Figures 3–30]). The coefficient of determination (r2) between PSC probability and APSC is 0.76 for JFM, 0.71 for January, 0.84 for February, and 0.34 for March. This indicates that the PSC probability is a reasonable diagnostic for the whole vortex, even though the POAM II/III and ILAS measurements are made in narrow latitude regions. However, the small sample of the vortex might have missed several PSC events, resulting in discontinuity of observed PSC probabilities.
 The patterns of high PSC probabilities (and APSC) are very similar to those of high ozone loss rates, although they are not always contemporaneous. The highest ozone loss rates either occur simultaneously with or slightly lag spikes in the PSC probability (i.e., in January and March in 1995, in the beginning of January and February in 1996, between mid-February and early March in 1997, and in January and between mid-February and mid-March in 2000). Chlorine activation occurs because of low temperatures and/or large surface area densities of aerosols and PSCs. Ozone losses subsequently occur in the presence of active chlorine during sunlit times. It is therefore not surprising that large ozone losses do not always immediately take place when PSC probabilities are high. We therefore examine the relationship between ozone losses and PSC probabilities from the perspective of longer time scales (monthly and JFM).
Figure 6 shows relationships between the PSC probability and the averaged ozone change rate (in ppbv d−1 and in ppbv sunlit h−1) for the JFM seasons and the individual months of January, February, and March. We consider the vortex time period for calculations hereafter: day 1–90 for 1994, 1995, and 1996; day 10–90 for 1997; day 1–60 for 1999; day1–80 for 2000. The PSC probability shown here was calculated directly for the corresponding period in each year (for individual months or for JFM). Errors in the PSC probability are estimated based on the square of the number of samples [Fromm et al., 1999]. The average ozone change rate is calculated for each year as the mean of daily ozone change rates for the corresponding period, and errors in the average ozone change rate represent one standard deviation of the distribution of daily change rates.
 The ozone change rate and the PSC probability are negatively correlated, except for March for which no correlation is found. The r2 value between PSC probability and ozone change rate per day is 0.56 for JFM, 0.66 for January, 0.85 for February, and 0.04 for March (Figures 6a–6d). The slope of the reduced major axis (RMA) regression equation is −2.5 ± 0.6, −1.7 ± 0.4, and −2.8 ± 0.4 (ppbv d−1%−1) for JFM, January, and February, respectively. If March 1996 is removed, however, an obvious correlation can be identified for March 1995, 1997, and 2000 (r2 is 0.97 and the regression coefficient is −4.4 ± 0.6 ppbv d−1%−1). The ozone change rates during March are discussed in further detail in section 4. The correlation is highest and the slope steepest for February (and March if 1996 is removed) than for January. The result for JFM suggests that an absolute increase of 10% in the probability of PSC occurrence during Arctic winter typically leads to an increase of 25 ± 6 ppbv d−1 in the ozone loss rate.
 The seasonal increase in slope of the correlation from January to February/March is very small for ozone change rate per sunlit hour (Figures 6e–6h). The regression coefficient is −0.32 ± 0.07, −0.28 ± 0.06, and −0.33 ± 0.04 (ppbv sunlit h−1%−1) for JFM, January, and February, respectively. This indicates that the relationship between the PSC probability and ozone loss rate per sunlit hour is almost constant during the winter. More sunlight in the late winter leads to the increase in slope of the relationship in February and March for ozone loss rate per day. The r2 between PSC probability and ozone change rate per sunlit hour is similar to that for ozone change rate per day: 0.62 for JFM, 0.64 for January, 0.88 for February, and 0.07 for March (0.91 if 1996 removed). Against sunlight, the result for JFM suggests that an absolute increase of 10% in the PSC probability leads to an increase of 3.2 ± 0.7 ppbv sunlit h−1 in the ozone loss rate.
 The ozone loss rates are highly sensitive to the PSC probability during January and February, suggesting that IAV in PSC probability is a good measure for IAV in ozone loss rates during January and February. For JFM and January, the ozone loss rate in 1996 was approximately half that in 2000, although the PSC probability in 1996 was roughly the same as (January) or larger than (JFM) that in 2000. The derived ozone change rates per sunlit hour for 1996 and 2000 differ in the same way. The differences in January ozone loss between 1996 and 2000 substantially reduce the calculated correlation between ozone loss rate and PSC probability. These differences are discussed further in section 4.
Figure 7 shows the relationship between averaged ozone change rates (in ppbv sunlit h−1) during JFM and various PSC- and temperature-related indices: JFM-averaged APSC, and PSC formation potential (PFP) [Tilmes et al., 2006] and potential for activation of chlorine (PACl) [Tilmes et al., 2007]. APSC shown here is the area of the vortex air for which the temperature is lower than TNAT at the 475 K level, whereas VPSC is the volume of the vortex area at temperatures below TNAT between 400 K and 550 K [Rex et al., 2004]. PFP is calculated as VPSC divided by the volume of the vortex for each day that the vortex can be identified between 400 K and 550 K. The daily values of PFP are integrated for each winter and then divided by the length of the period considered (from mid-December to the end of March). PFP extends the concept of VPSC by taking into account the differences in vortex conditions from year to year (i.e., IAV in the duration and volume of the vortex). PACl is calculated similarly to PFP, but uses the volume of the vortex area at temperatures below TACl (VACl) rather than VPSC. PACl is scaled to match the trend of effective equivalent stratospheric chlorine (EESC). Here we use PFP and PACl values at the 475 K level (S. Tilmes, personal communications, 2011) for consistently comparing with ozone loss rates observed at 475 K. The meteorological data and vortex definition are different between our analysis and Tilmes (personal communications, 2011).
 Correlations with ozone change rate are found for all three indices, with r2 values of 0.90 for APSC, 0.89 for PFP, and 0.94 for PACl. The correlation with ozone change rate is highest for PACl rather than APSC or PFP, indicating that PACl is a better indicator of ozone losses at 475 K during JFM. The correlation between ozone change rate and APSC/PFP is substantially degraded by the large value of APSC/PFP in 1996 and the small value of APSC/PFP in 1995. Our results agree with the results of Tilmes et al. , who showed that PACl provided the best correlation with column ozone losses between 400 K and 550 K.
 The r2 values between ozone loss rates and temperature-based quantities (APSC, PFP, and PACl) are higher than that for the PSC probability (Figures 6a and 6e). The largest difference between the PSC probability and three temperature-based quantities is observed in 1996: The PSC probabilities during JFM in 1996 and 2000 are similar, but APSC, PFP, and PACl are larger in 2000 than in 1996. The PSC probability derived from POAM and ILAS AEC data is a direct measure of PSC occurrence, however, the samplings are limited for the latitudes of 64°N–70°N thus the PSC probability might be a less good measure as a representation for the whole vortex. On the other hand, APSC, PFP, and PACl are a good measure for the whole vortex but these temperature-based quantities are likely to be biased high on estimates of PSC occurrence [Pitts et al., 2007, 2009; WMO, 2011]. The advantage of this study is to have shown the quantitative relationship between the observed PSC occurrence and observed vortex-averaged ozone loss rates.
3.3. Temperature History of Air Mass and Ozone Loss Rates
 The temperature history of an air mass can be used to evaluate the likelihood of PSC formation and associated phase transitions [e.g., Larsen et al., 2004]. Figure 8 shows the seasonal evolution of minimum temperatures along forward trajectories between two measurements of a double-sounded air parcel (open circles) and along 10 day backward trajectories of a double-sounded air parcel (solid squares). The former represents the temperature history of the air parcel while the ozone loss is observed, while the latter represents the temperature history of the air parcel before the ozone loss occurs [Terao et al., 2002]. The double-sounded air masses experienced particularly low temperatures (less than 195 K) inside the vortex in mid-January and early March of 1994; mid-January and mid-March of 1995; January through early March of 1996; mid-January and February of 1997; and throughout January to mid-February and into early March of 2000. Minimum temperatures near 188 K were observed in mid-January and mid-February of 1996, and January of 2000. The winter of 1999 was very warm. Other than in mid-February, air masses were not exposed to temperatures below 195 K.
 UKMO temperatures around TNAT during the winter of 1994/1995 have been reported to be higher than temperatures obtained from radiosonde measurements by up to 1.7 K at 475 K, with a standard deviation of about 2 K [Pullen and Jones, 1997]. Another experimental study showed that UKMO temperatures between 30 and 146 hPa were higher on average than temperatures obtained from balloon-borne measurements by 0.44 ± 1.25 K in 1997, 1.36 ± 2.11 K in 1999, and 0.41 ± 1.07 K in 2000 [Knudsen et al., 2002]. These warm biases may represent a source of uncertainty in the present analysis, but their influence was minor in 1997 and 2000.
Figure 9 shows relationships between the minimum temperature experienced in the preceding 10 days and ozone change rates per sunlit hour for corresponding double-sounded air parcels. Ozone change rates are calculated by regression analysis for regrouped double-sounded air parcels obtained in the temperature range of ±1 K for each month and for each year. Ozone losses occurred primarily for minimum temperatures between 187 K and 195 K. The ozone change rates derived from all data set (JFM 1994−2000; thick circles with line) indicated that the ozone loss rates clearly increases with decreasing the minimum temperature: −5.5 ± 0.9 ppbv sunlit h−1 (188 K), −3.8 ± 0.8 ppbv sunlit h−1 (190 K), −2.7 ± 0.4 ppbv sunlit h−1 (192 K), −2.7 ± 0.4 ppbv sunlit h−1 (194 K), −1.8 ± 0.5 ppbv sunlit h−1 (196 K), −1.5 ± 0.7 ppbv sunlit h−1 (198 K), −1.5 ± 0.6 ppbv sunlit h−1 (200 K), and −0.8 ± 0.5 ppbv sunlit h−1 (202 K).
 The particularly large ozone losses of ∼9 ± 3 ppbv sunlit h−1 in February 1996 and January 2000 were associated with very low minimum temperature of 187−189 K. The highest PSC probability also observed during these periods (∼16% in January 2000 and ∼11% in February 1996, Figures 6b−6c). Low minimum temperatures do not necessarily lead to ozone losses, as instances of small or zero ozone loss are also found to be associated with low minimum temperatures; however, these small ozone loss rates are statistically insignificant. It is unlikely that ice PSCs contributed to the observed large ozone losses since the minimum temperatures rarely dipped below TICE (∼188 K). However, the temperature history using the UKMO meteorological data could not resolve mountain wave effects that could induce NAT particles in the latitude range we observed. Felton et al.  showed that NAT crystals down wind from mountain wave-induced ice clouds make up 11% of the clouds observed during SOLVE campaign in the winter 2000 in the NH.
 Our results indicate that the TACl-based PACl provides the best proxy representation of IAV in Arctic ozone loss at the 475 K isentropic surface, as similar to results from partial column ozone losses between 400 K and 550 K [Tilmes et al., 2008]. Here we analyze the IAV in TACl at 475 K during Arctic winters from 1994 to 2000. Trends in background aerosol loading in the stratosphere affect the potential for chlorine activation [Tilmes et al., 2008]. Background concentrations of stratospheric aerosols were still large in 1994 following the eruption of Mt. Pinatubo in 1991, and gradually decreased to a near-constant level after 2000 [Thomason et al., 2008].
 TACl depends on the mixing ratio of H2O, the surface area density (SAD) of background aerosols, and altitude [Drdla and Müller, 2010]. SAD is calculated from AEC measurements by POAM II/III and ILAS: We select AEC data inside the vortex at temperatures above 210 K and average them during JFM for each year. SAD is then calculated using the equations derived by Steele et al.  for POAM II/III (1000 nm) and by Hervig and Deshler  for ILAS (780 nm). For H2O, we used HALOE V19 data [Harries et al., 1996; Kley et al., 2000] because POAM II did not measure H2O (although ILAS and POAM III did). The HALOE data were averaged over the area inside the vortex during JFM. The number of HALOE observations inside the Arctic vortex during JFM each year (∼50–115) is small compared to the number of POAM or ILAS observations but large enough to estimate the average value of H2O during JFM. HALOE also provides SAD data [Hervig and Deshler, 1998], so for validation purposes we calculate TACl derived from HALOE SAD along with TACl derived from POAM/ILAS SAD.
Figure 10 shows JFM averages of H2O (HALOE), AEC (POAM II/III and ILAS), SAD (POAM II/III, ILAS, and HALOE), and derived TACl at the 475 K isentropic surface for each year. The two calculations of TACl (derived from POAM/ILAS SAD and HALOE SAD) agree within uncertainty ranges. The following discussion is based on TACl derived from POAM/ILAS SAD. Decreasing trends are identified in both AEC and SAD between 1994 and 1996, though the trends are rather weak after 1997: The larger values are deduced for 1997 (and 1998) than for 1996. AEC decreases with increasing altitude above the 475 K level, so that AEC and SAD values at the 475 K level are controlled by the diabatic descent of air. The polar vortex was very tight in 1997, resulting in a small diabatic descent [e.g., Tegtmeier et al., 2008b] and consequently large AEC and SAD. Values of TACl were higher in 1994 (194.6 ± 0.6 K) and 1995 (194.4 ± 0.8 K) because SAD is larger because of the remnant of Pinatubo aerosols. These values and the value in 1997 (194.8 ± 0.5 K) approached TNAT (195 K). The values of TACl were low in 1996 (193.6 ± 0.9 K), 1999 (193.3 ± 0.6 K), and 2000 (193.3 ± 0.8 K).
 The IAV in TACl during 1994−2000 is small (within 1.5 K) but varies the correlation between JFM ozone loss and temperature-related indices. By comparing PFP (Figure 7b) and PACl (Figure 7c), the largest difference was seen in 1996: The decrease in PACl with regard to PFP (−0.06) significantly contributed to the higher correlation between JFM ozone loss and PACl rather than PFP. The PACl decrease in 1996 might be caused by the low TACl (i.e., small vortex area below TACl).
 In January, the IAV in ozone loss is correlated with PSC probability; however, different ozone losses occurred in 1996 and 2000 despite similar PSC probabilities (Figures 6b and 6f). The result may indicate that the predominant type of PSCs (i.e., the reaction probability and the magnitude of denitrification) and/or the surface area density of PSCs and background aerosols differ between January 1996 and January 2000. Differences in denitrification, PSC types, and/or surface area density will lead to differences in the magnitudes of chlorine activation through heterogeneous reactions. Observational evidence compiled by previous studies indicates that daily maximum slant-column densities of OClO measured by the Global Ozone Monitoring Experiment (GOME) were larger in 2000 than in 1996 during January [Wagner et al., 2002; Kühl et al., 2004].
 Denitrification occurs through sedimentation of HNO3-containing PSC particles, and hence controls the amount of HNO3. Chlorine deactivation by ClO + NO2 + M → ClONO2 + M, which depends on the photolysis of HNO3, controls the ozone change rate. Larger ozone loss rates have been observed in airborne measurements of more denitrified air in the winter of 2000 [Gao et al., 2001]. A large denitrification event due to NAT PSCs was observed in JFM 2000 [Popp et al., 2001]. The magnitude of denitrification in January 2000 was larger than that in January 1996 [Hintsa et al., 1998]. Lidar measurements at Ny-Ålesund indicate that more liquid PSCs than NAT PSCs occurred during January 1996, whereas liquid and NAT PSCs were more evenly partitioned during January 2000 [Massoli et al., 2006]. Recent measurements from CALIOP strongly suggest that the dominant mode of PSCs in the Arctic winter stratosphere is an external mixture of liquid and solid particles [Pitts et al., 2011]. The relative partitioning between liquid and solid phases is critically important for determining chlorine reactivity. The reaction probability of NAT has been the subject of considerable debate, with results variously suggesting that the reaction probability for NAT is comparable to or significantly lower than that for STS [Carslaw and Peter, 1997; Carslaw et al., 1997; Peter 1997; Hanisco et al., 2002]. HNO3-containing large particles, so-called NAT-rock particles, were also observed in the Arctic stratosphere during the winter of 2000 [Fahey et al., 2001], but POAM III measurements could not capture extinction caused by these particles because of their low number density. This inability to observe NAT-rock particles may lead to a partial underestimation of the magnitude of PSC probability in January 2000.
 Large denitrification was observed during the winters of 1995, 1996, 1997, and 2000 [e.g., Hintsa et al., 1998; Sugita et al., 1998; Irie et al., 2001; Popp et al., 2001], and the MLS observations [Santee et al., 2000] and a model simulation [Mann et al., 2003] suggest that the largest denitrification during these years occurred in 2000. GOME OClO column measurements [Kühl et al., 2004] and the MLS ClO measurements [Santee et al., 2003] are largely consistent with these findings, with larger values observed in March of 1996, 1997, and 2000 than in other years (although the observed amount of active chlorine was moderate in 1995). Suppression of chlorine deactivation due to denitrification may contribute to the IAV in ozone losses during March.
 These observational evidences, larger denitrification and ClO concentrations in 2000 than in 1996 throughout JFM, are consistent with the differences in calculated ozone change rates between 1996 and 2000 shown in Figure 6. However, the magnitudes of ozone losses in January and March 1996 that we deduce in this study are significantly smaller than those calculated by ozonesonde-Match analysis. The ozone loss rate for March 1996 deduced by ozonesonde-Match was ∼20 ppbv d−1, four times larger than our result (∼5 ppbv d−1). The lack of correlation between ozone loss rates and PSC probabilities in March is due to the small ozone loss rate calculated for 1996.
 We compare local ozone loss rates with those calculated by ozonesonde-Match analyses [Rex et al., 1997, 1999, 2002; Schulz et al., 2000, 2001]. The ozone loss rates derived from ozonesonde-Match analyses at 475 K for various winters from 1992 to 2001 were collected and summarized by WMO [2003, Figures 3–30]. In 1995, 1997, 1999, and 2000, our running means of ozone loss rates quantitatively agree well with ozonesonde-Match results. Specific points of agreement include the large ozone losses at around day 30 (∼50 ppbv d−1) and day 80 (∼30 ppbv d−1) in 1995, day 60 in 1997 (∼40 ppbv d−1), and day 60 in 2000 (∼40 ppbv d−1), and the lack of ozone loss in 1999. The large ozone loss rate that we derive for January 2000 (∼60 ppbv d−1) is larger than that derived by the ozonesonde analysis (∼40 ppbv d−1), but the largest discrepancy is in 1996. The ozonesonde-Match results for 1996 indicate that ozone loss took place throughout JFM at nearly constant rates (30–40 ppbv d−1 during January and 20 ppbv d−1 during February and March), whereas our results show no significant ozone loss in mid-January or late March. One possible cause for the difference in the derived ozone loss rates for January is that the number of ozonesonde measurements from January 1996 is relatively small [Rex et al., 2003], as is the number of POAM II match pairs for that period. The 3-D chemistry transport model did not simulate large ozone losses in January 1996 [Kilbane-Dawe et al., 2001]; the simulated ozone loss rates are similar to our satellite-Match results. Differences in measurement locations relative to the vortex may also contribute to the differences between satellite-Match and ozonesonde-Match results for 1996, both in mid-January and in late March.
 We have evaluated chemical ozone loss rates (per day and per sunlit hour) in the Arctic at the 475 K isentropic surface using a satellite-Match technique and observational data from POAM II for the winters of 1993/1994, 1994/1995, and 1995/1996, ILAS for the winter of 1996/1997, and POAM III for winters of 1998/1999 and 1999/2000. The largest ozone loss rates occurred in January 1995 (∼50 ppbv d−1), February 1996 (∼40–50 ppbv d−1), February 1997 (∼40 ppbv d−1), January 2000 (∼60 ppbv d−1), and early March 2000 (∼40 ppbv d−1). The evolutions and magnitudes of the derived ozone loss rates quantitatively agree with those derived by ozonesonde-Match analyses [Rex et al., 1997, 1999, 2002; Schulz et al., 2000, 2001; WMO, 2003], except for 1996 during mid-January and late March.
 We have quantified PSC probabilities at the 475 K level using AEC data from POAM II/III and ILAS. The results show that patterns of high PSC probabilities are very similar to those of high ozone loss rates. The highest positive correlations between monthly PSC probabilities and ozone loss rates per sunlit hour occurred in February (r2 = 0.88) and January (r2 = 0.64). No correlation is found for March; however, this lack of correlation is solely due to small ozone loss rates in 1996. A correlation is found for March when considering only 1995, 1997, and 2000 (r2 = 0.91). This relationship demonstrates that IAV in the magnitude of ozone losses is tied to IAV in PSC probabilities. The regression coefficient between the PSC probabilities and ozone loss rates per sunlit hour was almost constant throughout JFM, however, for ozone loss rates per day, the seasonal increase from January to February/March in the regression coefficient was found because of more sunlight in late winter to spring. Regression analysis for the full JFM season indicates that, on average, an absolute increase of 10% in the PSC probability contributes an additional ozone loss of approximately 25 ± 6 ppbv d−1 or 3.2 ± 0.7 ppbv sunlit h−1. Relationships between JFM-averaged ozone loss rates per sunlit hour and PSC- and temperature-related indices, APSC [e.g., Rex et al., 2004], PFP and PACl [Tilmes et al., 2008], indicate that the PACl (r2 = 0.94) is a better proxy representation for Arctic ozone losses at 475 K than APSC (r2 = 0.90) or PFP (r2 = 0.89).
 Large derived ozone loss rates were typically associated with air masses that experienced low temperatures (187–195 K) within the previous 10 days. The ozone loss rates clearly increase with decreasing the minimum temperature: ozone change rates averaged for JFM and for 1994−2000 are from −5.5 ± 0.9 ppbv sunlit h−1 at 187−189 K to −0.8 ± 0.5 ppbv sunlit h−1 at 201−203 K. The particularly large ozone losses of ∼9 ± 3 ppbv sunlit h−1 in February 1996 and January 2000 were associated with very low minimum temperatures of 187−189 K, simultaneously with high PSC probabilities.
 We used HALOE H2O and POAM/ILAS AEC to calculate the IAV in TACl. TACl decreased from 194.6 ± 0.6 K during the winter of 1994 to 193.3 ± 0.8 K during the winter of 2000. This decrease is associated with reductions in background sulfate aerosols. The decrease is interrupted in 1997 and 1998, when weak diabatic descent inside the vortex led to higher SAD, which in turn led to higher TACl. The low value of TACl in 1996 (193.6 ± 0.9 K) caused the decrease in PACl, which significantly contributed to the higher correlation between ozone loss and PACl rather than PFP. Our observational evidence shows the high correlation between Arctic ozone loss and TACl/PACl, as well as Tilmes et al. . The injection of sulfate aerosols into the stratosphere either by huge volcanic eruptions or by “geoengineering” schemes to counteract global warming would lead to increases in TACl, probably resulting in enhanced ozone loss in the Arctic winter stratosphere.
 We are grateful to the ILAS/ILAS-II Data Handling Facility at the National Institute for Environmental Studies for processing and providing the ILAS data; the NASA Langley Research Center, Atmospheric Sciences Data Center for the POAM II and POAM III data; and the NASA Langley Research Center and the NASA Langley Chemistry and Dynamics Branch for the HALOE data. We also thank Richard Swinbank for supplying the UKMO assimilation data and trajectory program, Martyn P. Chipperfield for providing the SLIMCAT diabatic cooling rate data (via Hideaki Nakane), Simone Tilmes for providing PFP and PACl values at the 475 K level, and Peter von der Gathen and Hideaki Nakajima for their valuable comments.