During the Stratospheric Aerosol and Gas Experiment (SAGE) III Ozone Loss and Validation Experiment (SOLVE)/Third European Stratospheric Experiment on Ozone (THESEO) campaign, Polar Ozone and Aerosol Measurement (POAM) III sampled in the vortex core, on the vortex edge, and outside the vortex on a near-daily basis from December 1999 through mid-March 2000. During this period, POAM observed a substantial amount of ozone decline. For example, ozone mixing ratios in the core of the vortex dropped from about 3.5 ppmv in mid-January to about 2 ppmv by mid-March at 500 K. The ozone chemical loss indicated by these measurements is assessed using two methodologies. First, the POAM data is used to construct vortex-averaged ozone profiles, which are advected downward using vortex average descent rates. The maximum ozone loss (1 January to 15 March) is found to be about 1.8 ppmv. In a second approach, the REPROBUS 3-D CTM is used to specify the passive ozone distribution throughout the winter. The chemical loss in the vortex is estimated by performing a point-by-point subtraction of the POAM measurements inside the vortex from the model passive ozone evaluated at the time and location of the POAM measurements. Both ozone loss estimates are in general agreement and they agree well with published loss estimates from ER2 and ozonesonde measurements.
 It is now well established that significant photochemical ozone loss occurs during the coldest Northern Hemisphere winters [see Manney and Sabutis, 2000, and references therein]. It has been difficult for state-of-the-art photochemical models to reproduce observed ozone loss rates with high accuracy [Lefèvre et al., 1994; Chipperfield et al., 1996a, 1996b; Becker et al., 1998; Deniel et al., 1998; Goutail et al., 1999]. This suggests that our quantitative understanding of polar ozone loss is still not complete. On the other hand, deriving ozone loss from observations is difficult because it requires quantifying the dynamical contribution to the observed ozone evolution. It is possible that measurement/model discrepancies are the result of inaccuracies in the estimation of ozone photochemical loss from measurements. This is one of the issues that the Stratospheric Aerosol and Gas Experiment (SAGE) III Ozone Loss and Validation Experiment (SOLVE) and the Third European Stratospheric Experiment on Ozone (THESEO) campaigns were designed to address. The SOLVE/THESEO campaign, based in Kiruna, Sweden during the 1999/2000 Arctic winter, was a joint NASA/ESA mission designed to improve our understanding of polar ozone loss. The mission included coordinated, multiplatform (balloon, airplane, ground-based, and satellite) measurements and extensive analysis efforts.
 The Naval Research Laboratory's Polar Ozone and Aerosol Measurement (POAM III) satellite instrument provided continuous, high resolution (1 km) stratospheric ozone measurements during the SOLVE/THESEO campaign. POAM measurements have been used in several previous studies examining the morphology of wintertime polar ozone and estimating photochemical loss [Randall et al., 1995; Bevilacqua et al., 1997; Deniel et al., 1998]. During the SOLVE/THESEO campaign, POAM was the only satellite instrument providing high resolution ozone profiles over the Arctic on a continuous basis. The POAM data therefore provides a useful context for interpreting the more detailed but also more localized balloon, airborne, and ground-based measurements. While missions of the scope of the SOLVE/THESEO campaign are not conducted every year, satellite instruments such as POAM provide multiyear data sets. It is therefore important to evaluate and refine our ability to extract ozone loss information from satellite measurements. This is particularly important because a solar occultation instrument such as POAM produces a relatively sparse data set (15 measurements per day in each hemisphere) compared to that obtained with atmospheric emission measurements. Also, solar occultation measurements require sunlight (90 degree zenith angle) so the measurement sampling will tend to be biased toward the sunlit portion of the vortex. Interpretation of the POAM data is also made more difficult because water vapor, the only tracer measured by POAM, is not well suited for helping quantify ozone loss. The SOLVE/THESEO mission, therefore, provides an excellent opportunity for evaluating the importance of these potential difficulties in estimating ozone loss using POAM data.
 In this paper we begin by giving a brief overview of the POAM III instrument, measurement sampling, and ozone retrievals. A synopsis of the ozone measurements during the 1999/2000 winter is shown and is contrasted to the 1998/1999 winter and to ozone in the Southern Hemisphere winter. Next, the POAM measurements are used to infer ozone photochemical loss. This is done using two approaches to estimating the dynamical contribution to the ozone time evolution: vortex average diabatic descent, and passive ozone fields derived from the REPROBUS 3D chemical transport model (CTM) simulation. Finally, the REPROBUS full chemistry simulation is used to estimate the effects of POAM vortex sampling biases.
2. POAM Instrument
 The Polar Ozone and Aerosol Measurement (POAM III) instrument is a satellite-based nine-channel, visible/near-infrared photometer that uses the solar occultation technique to measure ozone, aerosol extinction, water vapor, and nitrogen dioxide in the polar stratosphere [Lucke et al., 1999]. Launched in March 1998, POAM III makes 14–15 measurements per day in each hemisphere around a circle of latitude. The POAM III measurement latitude varies slowly, ranging from 55°N to 73°N in the Northern Hemisphere and from 63°S to 88°S in the Southern Hemisphere, with identical coverage each year. During the SOLVE campaign the POAM III (hereafter denoted as POAM) measurement latitude was between about 64°N and 67.7°N, Since the latitude of Kiruna is 67°N, the POAM measurement sampling was well suited to supporting the SOLVE campaign.
 In order to illustrate the POAM measurement sampling with respect to the vortex, Figure 1 shows the equivalent latitude of each POAM measurement point obtained during the SOLVE campaign along with the equivalent latitude of the center (green line), and inner and outer edges (red lines) of the polar vortex. The vortex boundaries were calculated using the UK Meteorological Office (UKMO) analysis and the vortex discrimination algorithm given by Nash et al. . In general, POAM samples outside, on the edge, and inside of the polar vortex on a daily basis throughout the winter. From mid-December to mid-January the vortex was centered near the pole so that POAM (measuring at about 65°N) did not sample the very highest equivalent latitudes. However, from about the beginning of February through mid-March POAM did sample near the vortex core on a regular basis. Throughout this paper we will refer to measurements that are on or inside the inner vortex edge as “in-vortex” data. Measurements that are on or outside of the outer vortex edge are labeled as “out-vortex.”
 The POAM ozone retrievals presented in this paper are version 3.0 (the currently operational version), which is publicly available. The version 3.0 ozone retrievals have been validated by comparison with the Halogen Occultation Experiment (HALOE) and the SAGE II measurements by Randall et al. (Comparison of ozone measurements from POAM III, SAGE II and HALOE, in preparation for submission to Journal of Geophysical Research). In the Northern Hemisphere the agreement is generally within 5% from 13 to 60 km. The situation is not as clear-cut below 13 km because of increasing variability in the comparisons, but it appears likely that the POAM measurements are biased increasingly high with decreasing altitude below 13 km to about 25% at 10 km. Also, POAM measurements have been compared extensively with coincident ER2, DC8 lidar and balloon borne ozone measurements made during the SOLVE campaign [Lumpe et al., this issue]. The results are very good, with POAM agreement with the coincident measurements being well within the uncertainty levels determined from the POAM error analysis [Lumpe et al., 2002] and our previous comparisons.
3. POAM Ozone Measurements
Figure 2 shows an overview of the POAM measurements during the 1999/2000 winter at the indicated potential temperature levels both in-vortex and out-vortex. Each time series has been smoothed with a 13 point moving average. The factors important in producing in/out of the vortex ozone differences are discussed by Manney et al. [1995b] and Randall et al. . Generally speaking, the most important dynamical mechanisms are diabatic descent inside the vortex and mixing between low and midlatitude air outside the vortex. Below about 500 K, where the ozone vertical gradient is very large, diabatic descent of ozone enriched air dominates, producing larger in-vortex ozone mixing ratios. Above about 500 K, horizontal advection and mixing of ozone-rich low latitude air begin to dominate, and the sign of the cross-vortex ozone gradient is reversed (higher out-vortex mixing ratios). In the early part of the winter, the in-vortex and out-vortex ozone mixing ratios shown in Figure 2 appear consistent with this simple picture. In addition to dynamical influences, however, ozone chemical loss occurring primarily in the vortex needs to be considered. The steep ozone decreases observed beginning in January at 500 K and in February at 400 K are the manifestation of the chemical loss that has been well documented during the cold SOLVE winter [Rex et al., 2002; Schoeberl et al., 2002]. There are also large variations observed in the out-vortex measurements at all potential temperature levels in Figure 2. Although a discussion of these is outside the scope of this paper, we do point out that outside the vortex there is a large variation in the equivalent latitude sampling of POAM. These equivalent latitude variations drive the out-vortex ozone variations exhibited in Figure 2.
 In Figure 3a, the in-vortex ozone mixing ratio time series at 500 K shown in Figure 2 is compared with the corresponding time series for the 1998/1999 winter. The 1998/1999 winter experienced a rare mid-December stratospheric warming and the lower stratospheric temperatures remained high until the end of January [Manney et al., 1999]. POAM detected only 11 polar stratospheric clouds (PSCs) in early December before the major warming, and a few in early February. With warm temperatures and few PSC's, no significant chemical ozone loss was found during the 1998/1999 winter [Schulz et al., 2001]. The contrast in the behavior of ozone in the two years exhibited in Figure 3a is striking. In early December the mixing ratios are about the same in the two years. However, in 1998/1999, after the initial warming in December, the mixing ratios increase fairly steadily throughout the winter as expected from diabatic descent. Mixing ratios increase initially in 1999/2000 as well, but reach a peak and decline precipitously after the beginning of January, such that by early March mixing ratios are smaller by nearly a factor of 2 than they were in the previous winter. It is not possible to quantitatively estimate the chemical change from Figure 3a because the details of the dynamics are different each winter. However, we do note that the difference between the 1998/1999 and 1999/2000 in-vortex ozone at the end of the winter is very close to the 1999/2000 ozone loss estimate at 500 K developed in the next section.
 To give a different perspective to the loss observed in the 1999/2000 winter, Figure 3b compares the in-vortex ozone at 500 K with that of the Southern Hemisphere for the 1999 Antarctic winter. The ozone time series are remarkably similar during the winter months (NH: Dec–Feb; SH: Jun–Aug). The apparent ozone loss rate seen in Figure 3b is influenced by the POAM vortex sampling and the average solar exposure in the vortex, both of which are substantially different in the Arctic and Antarctic. From the POAM data (accounting only for diabatic descent), we find that the average Arctic loss rate in February 2000 (see section 4) at 480 K is 25 ppbv/day (0.9%/day). From an analysis of 5 years of POAM measurements of Antarctic ozone [Bevilacqua et al., 1997; Hoppel et al., 2000] we find an average loss rate during the months of August and September at 480 K of about 13 ppbv/day (0.5q%/day) and 67 ppbv/day (3.1%/day) respectively. Although it would be more appropriate to compare the ozone loss rates per solar exposure, the POAM data does suggest that the Arctic ozone loss in February and early March 2000 was similar to the equivalent time period in the Antarctic.
 However, as Figure 3b also shows, the Antarctic ozone loss continues in September, while the Arctic ozone loss subsides. Although POAM no longer sampled the Arctic polar vortex after the middle of March, an analysis of ozonesonde data [Rex et al., 2002] have shown that some ozone loss continued to the end of March. However, the vortex split into two pieces after 15 March and the minimum temperatures rose above 200 K by 20 March as the vortex was dissipating. In contrast, the Antarctic vortex persists until early November with temperatures generally remaining cold enough for PSC formation through mid-October [Fromm et al., 1997]. The period of steepest ozone decline is in September when in-vortex air is exposed to a large amount of sunlight. In the equivalent time period in the Northern Hemisphere (early March) the vortex begins to dissipate, slowing ozone loss. This figure suggests that to achieve ozone loss in the Arctic of the magnitude that is routinely observed in the Antarctic would require a cold, strong vortex which persists until at least the beginning of April.
4. Ozone Loss
 In this section we estimate the vortex average photochemical ozone loss for the 1999/2000 winter from the POAM measurements shown in Figure 2. In order to do this, the dynamical contribution to the ozone change must be determined. Several techniques have been used to estimate this dynamical contribution. Müller et al.  used a tracer (methane) correlation approach to estimate ozone loss from HALOE measurements. Manney et al. [1995a, 1996] used a 3-D trajectory model to estimate ozone loss from MLS measurements. Goutail et al.  used the REPROBUS 3-D chemical transport model (CTM) to derive ozone chemical loss from SAOZ total column ozone measurements, and Deniel et al.  used the same approach to derive ozone loss as a function of potential temperature from POAM II data. Rex et al. [1997, 1999, 2002] have used diabatic trajectories and ECC sonde data (the Match technique) to derive ozone loss. Bevilacqua et al.  used vortex average diabatic descent to account for the dynamical contribution in estimating ozone loss in the Antarctic ozone hole from POAM II data. Recently, Richard et al.  have used multiple tracer measurements to account for both descent and mixing across the vortex edge. A detailed comparison of ozone loss calculated from different techniques can be found in Harris et al. .
 POAM does not measure a tracer species that is well suited for estimating ozone loss. Therefore, the dynamical component must be obtained from other sources. Two methods are utilized for doing this. In the first method, diabatic heating calculations are used to infer the vortex average descent, and then the descent is used to remove the dynamical component from the vortex average ozone in a manner similar to that described by Bevilacqua et al. . In addition, the REPROBUS CTM is used to specify the dynamical change in ozone for each of the POAM measurements as described by Deniel et al. . The implementation of these two methods to calculate photochemical ozone loss for the 1999/2000 winter from POAM measurements is described in detail in the next two sections.
4.1. Vortex Averaged Descent
 The vortex average descent method of inferring ozone photochemical loss assumes that the dynamical contribution to in-vortex ozone change is dominated by diabatic descent. In other words, other dynamical processes, such as in/out-vortex mixing, are neglected. It also assumes that diabatic descent can be adequately represented in a vortex average sense. Under these assumptions, the accuracy of the method is determined by the accuracy with which the vortex average ozone and average descent can be determined. The vortex average ozone is estimated by averaging all the POAM in-vortex measurements on an isentropic grid for each day, and then applying a 9-day running mean filter.
 In order to derive the vortex average diabatic descent rates necessary to implement this method, we have used heating rates from the radiative transfer model of Rosenfield et al. , calculated using the UKMO temperature analysis. The heating rates were used to specify the vertical velocity in a 3-D diabatic trajectory model which also uses the UKMO horizontal winds. The trajectory calculation was initialized using an ensemble of air parcels distributed uniformly throughout the vortex on each level of an isentropic grid. The vortex average descent curves were then taken to be the time evolution of the average potential temperature of each ensemble of trajectories. We will refer to these descent curves as the GSFC/NRL descent.
 As a check on our derived descent rates, we have also used the vortex average descent calculated by Rex et al. . The Rex et al. descent rates were calculated using heating rates from the SLIMCAT 3D CTM, which uses a modified version of the MIDRAD radiative transfer model [Chipperfield, 1999]. The heating rates were vortex averaged into 1-D heating rate profiles on an isentropic grid. Next, the rates were integrated in time to produce a potential temperature time series, or descent curve, for each initial potential temperature level for the January through March time span. We will refer to these descent curves as the SLIMCAT/Rex descent.
 The descent curves are used to estimate what the vortex average ozone would be in the absence of chemical change. We call this quantity the passive ozone, because it behaves as a passive tracer. We begin with an initial ozone mixing ratio profile on an isentropic grid which characterizes the vortex prior to chemical ozone destruction. The passive ozone for each day is then constructed from the initial profile by simply adjusting the initial isentropic grid by an amount determined from the descent curves. The vortex average ozone for 1 January has been chosen as the initial profile. The chemical ozone change is the difference between the measured profile and the passive ozone profile. Implementation of this approach is illustrated in Figure 4, which shows the initial ozone on 1 January along with the final ozone and passive ozone on 15 March. The dynamical contribution to the ozone change (initial ozone - passive ozone) is about 0.5 ppmv at 475 K, which is a significant fraction of the chemical ozone change.
 The results of the vortex average descent method are summarized in Figure 5. Figure 5a gives the results obtained using the GSFC/NRL descent, and Figure 5b gives the results calculated in an identical way for the SLIMCAT/Rex descent rates. In both cases, ozone loss begins in early January between about 450 and 570 K, and peaks at about 475 K on 15 March. 15 March has been used as the final date because POAM did not sample inside the vortex after this date. Other measurements have shown continued ozone loss to about the end of March [Rex et al., 2002]. The average loss rate from 1 February to 15 March is 27 ppbv/day, or 1.0 %/day.
 Because both sets of loss calculations use the same ozone measurements and averaging scheme, all differences can be directly attributed to differences in the descent curves, which are shown as dotted lines on the contour plots. Above about 480 K the GSFC/NRL and SLIMCAT/Rex derived ozone changes are virtually identical. Between 450 K and 480 K the SLIMACT/Rex values are slightly larger than the GSFC/NRL values (<10%), while below about 430 K the GSFC/NRL values become increasingly larger as a result of faster descent rates. However, it is difficult to estimate the ozone loss using the vortex average approach below 430 K for several reasons. In the lower stratosphere the net heating rates are smaller while the uncertainties caused by clouds and aerosols are larger. This may lead to larger relative errors in the heating rate calculation. Also, the loss calculation is more sensitive to errors in the descent calculations because the vertical gradient of the ozone mixing ratio is large below 450 K. Finally, there can be significant amounts of mixing across the vortex edge below 450 K [Dahlberg and Bowman, 1994; Manney et al., 1994] which renders the descended initial ozone profile inadequate for representing the effects of transport. Below 500 K, the ozone abundance is lower outside of the vortex (except after substantial in-vortex loss) so that mixing would lower the in-vortex ozone and lead to an overestimate of chemical loss.
 The ozone loss calculation was started on 1 January because the vortex was mainly in darkness during December, when chemical loss is expected to be small [Schoeberl et al., 2002]. However, it is interesting to examine how robust the results are to the choice of the initialization date. One would expect that beginning the calculation as late as possible would minimize the effect of errors in the descent rates and provide the best initial vortex average simply because it is closest in time. Figure 6 shows the cumulative ozone change on 15 March, calculated for the GSFC/NRL descent, with initialization dates of 10 December, 20 December, and 1 January. Above 430 K there is very good agreement between the calculations initialized on 1 January and 10 December. However, at and above 430 K, more loss (by about 0.2–0.3 ppmv) is seen for the calculation initialized on 20 December. The reason for the greater loss can be traced to the initial ozone profile. Careful examination of Figure 2 shows that the 20 December average ozone is larger than on 10 December and 1 January at 600 K and 700 K. This larger initial ozone leads to a larger estimate of ozone loss. The origin of the higher ozone mixing ratios on 20 December is a POAM sampling bias. Although all of the POAM measurements used in the averaging are within the inner edge of the vortex (using the Nash criteria), Figure 1 shows that POAM samples fewer values of high equivalent latitude for the week near 20 December than for the week immediately before or after. Ozone versus potential vorticity (PV) plots [Randall et al., 2002] show that ozone abundances at 650 K are higher near the edge than inside, at this time, resulting in the higher 20 December initialization profile. This illustrates an important weakness of this (and in fact many other) chemical loss calculations performed on relatively sparse solar occultation satellite data: obtaining an unbiased vortex average initial profile. One possible improvement would be to use the correlation between ozone and PV to reconstruct a vortex ozone average based on PV-area weighting rather than using a simple average. This is the method used to initialize the REPROBUS model at high latitudes as discussed in the next section.
4.2. Error Analysis
 There are many sources of uncertainty in the vortex average descent method of calculating ozone loss. These include the degree of isolation of the vortex, measurement error, initialization error, sampling biases, and errors in the descent rates. This method inherently assumes that the vortex is isolated, and that the delineation of the “inner edge” of the vortex corresponds to the isolated region. We do not attempt to quantify the uncertainty in these assumptions, except to note that for the 1999/2000 winter, Richard et al. (submitted manuscript, 2001) have used tracer measurements to argue that the inner vortex was largely isolated during the winter. The uncertainty due to measurement error is much easier to quantify. The random error in the individual POAM measurements is less than 10%. The averaging of 9 days of vortex data results in an total measurement error of less than a few percent. From Figure 6, we estimate the ozone initialization error to be roughly ±0.15 ppmv. In section 5, the uncertainty due to the POAM sampling bias is explored, yielding a rough error estimate of ±0.1 ppmv. The uncertainty in the vortex average descent calculation is a significant error source. In order to quantify this error, we assumed that the heating rates were in error by a constant value of ±0.1 K/day. This error was then propagated through the ozone loss calculation. This results in an altitude dependent error in the ozone loss on 15 March with a maximum value of 0.4 ppmv at 420 K. The RMS average of the 3 largest error sources (descent rates, ozone initialization, and sampling bias) is used for the error bars shown in Figure 8 for the vortex average descent results.
4.3. Passive Ozone from CTM
 As discussed above, the vortex average descent method of calculating ozone loss neglects in/out-vortex horizontal mixing, and assumes unbiased vortex-averages are obtained. Some of these limitations are avoided by using the passive ozone field from the REPROBUS chemistry transport model (CTM). The REPROBUS model is a three dimensional CTM forced by the winds and temperature from the European Center for Medium-range Weather Forecasts (ECMWF) [Lefèvre et al., 1994]. The model run used here is a recently improved version (run #824). The ozone field was initialized on 1 December 1999 by using the ozone-PV correlation derived from POAM III (high latitudes) and HALOE (low latitudes) data from 12 November to 5 December. The model was then run for 10 days in order to remove the effects of any noise in the initial ozone field. On 11 December, the passive ozone was set equal to the model (chemical) ozone. For the remainder of the winter, the passive ozone was advected as a passive tracer in order to simulate what the ozone would be in the absence of chemical change. The passive ozone field was then sampled at the time and location of each POAM measurement. The ozone chemical change is found by subtracting the passive ozone from the measured ozone, analogous to the methodology used by Deniel et al. . The ozone change calculated in this manner is then vortex averaged as in the previous analysis. It is important to note that although we are again examining the vortex average ozone change, this ozone change calculation (POAM ozone - passive ozone) was performed on a point by point basis rather than as the difference between two average profiles as in the previous analysis.
 The resulting time evolution of chemical ozone change is shown in Figure 7. The descent curves (dotted lines) overlaid on Figure 7 are the GSFC/NRL rates shown in Figure 5b. While we do not have a measure of the vortex average descent from the model, we can expect it to be similar to the descent from other models. In general, below 550 K the time evolution of the ozone change calculated with the REPROBUS model is similar to the vortex average descent loss in Figure 5. Note that the horizontal and vertical scales in Figure 7 are larger than in Figure 5 because the initialization date is earlier and total descent is greater. In both cases the loss begins in early January and maximizes near 475 K with total cumulative loss of about 1.9 ppmv. This suggests that at the altitude of maximum ozone loss, in/out-vortex mixing was not a significant factor in determining the in-vortex ozone mixing ratios. However, it is important to point out that this is true for the 1999/2000 winter, during which the vortex was unusually cold. It may not apply to the more active Northern Hemisphere winters, which are more typical.
 At about 600 K and above, Figure 7 shows a small amount of ozone loss at the starting date of 11 December. Since we do not expect chemical ozone loss at this time, this initial loss is likely the result of inaccuracies in the initial ozone-PV regression, and possible differences between the UKMO PV field used for the regression and the ECMWF PV field used by the model. This initial ozone change is equivalent to about a 5–10% discrepancy between the POAM data and the initial passive ozone. Of greater significance is the large ozone change seen between 580 K and 720 K during January and early February. This loss of almost 1.1 ppmv is not seen in the vortex average descent analysis (Figure 5), nor is it seen in any ozonesonde analysis [Rex et al., 2002; Schoeberl et al., 2002]. The source of this apparent loss can be traced to unusually high values of passive ozone in the model near the inner edge of the vortex. In the model, a strong gradient in the passive ozone exists between the middle and inner edges of the vortex above 600 K. From early January to early February, the high ozone gradient moves slightly inward, increasing the ozone near the inner edge where many of the POAM measurements are located. This feature may be the result of unrealistically large mixing across the vortex edge within the model or model descent rates being too large. These issues are still under investigation.
Figure 8 compares the cumulative loss on 15 March for the three calculations presented here and for the Match technique [Rex et al., 2002], which is discussed in section 6. Below 450 K all of the calculations agree reasonably well. Above 480 K, the calculation using the REPROBUS passive ozone yields greater ozone loss than the others, with possible explanations having been discussed above.
5. POAM Vortex Sampling Biases
 In the previous analysis, POAM data was used to estimate the vortex average ozone change by simply calculating the change in the average of the in-vortex ozone measurements. The average of the in-vortex measurements may be biased relative to the true vortex average due to spatial inhomogeneities in the ozone abundance. Inhomogeneities may arise from several mechanisms including nonuniform descent, mixing across the vortex boundary, or variations in the ozone loss rate. The latter mechanism is certain to be a factor because chemical ozone loss requires sunlight, leading to increased loss near the vortex edge during times when the vortex is centered near the pole. This feature can be seen during January 2000 in the model simulations of both the SLIMCAT and REPROBUS models (see THESEO 2000 web site). Thus it is possible that, since the POAM measurement requires sunlight, this sampling bias could lead to an ozone loss estimate that is larger than the true vortex ozone average. In order to explore the effect of POAM sampling, we use the REPROBUS run with full chemistry to represent the true ozone field within the polar vortex. In order to simulate the POAM sampling, the in-vortex REPROBUS ozone loss was averaged only within the 64°–66°N latitude band. The latitude grid spacing of the model is 2°, so this band averaging includes 2 latitude grid lines. This roughly simulates the POAM sampling pattern during the winter, although it does not account for the exact longitude and local time of the POAM measurement. This average is then compared with the complete vortex average, calculated using the appropriate area weighting of the grid points. Figure 9a shows the time series of the ozone change for both averages using the REPROBUS output sampled every 10 days. Both averaging methods clearly show the evolution of the ozone loss. Figure 9b shows the difference in the ozone change obtained from the two averaging methods. As suspected, the ozone loss from the POAM-like sampling is systematically larger during the month of January. The differences are generally less then 0.1 ppmv throughout the winter. On a relative basis, the differences maximize at 0–25% in January when the ozone loss is small, and then decrease as the cumulative ozone loss increases throughout the winter. To the extent that this simulation accurately reflects the ozone distribution and POAM sampling, it suggests that the error due to sampling biases on the cumulative ozone loss estimate is small compared to the variation in loss estimates shown in Figure 8. However, sampling biases could lead to significant errors when calculating ozone loss rates based on time periods of less than a month. Schoeberl et al.  have identified a positive correlation during the month of January between the calculated ozone loss from POAM data and the solar exposure of the air parcels being measured. When restricting the POAM measurements to equivalent latitudes of greater than 75 degrees, the maximum ozone loss on 15 January was found to decrease by about 0.2 ppmv. Although the Schoeberl et al. analysis does not attempt to reference the differences to a “true” vortex average, it does indicate that the size of the uncertainty in the January ozone loss due to sampling issues may be as large as 0.2 ppmv.
6. Comparison With Other Loss Estimates
 There were many measurements made during the SOLVE campaign and many other estimates of ozone loss. In this section we compare the ozone loss values derived in this paper with several other estimates. Richard et al.  have calculated ozone loss rates using ozone and tracer measurements from the ER2 and balloon flights. This analysis leads to an ozone loss estimate for 3 February to 13 March of about 1.7 ppmv at 450 K, decreasing to about 0.76 ppmv at 400 K. While this agrees very well with the loss calculated for 1 January to 15 March in this work, we find that by the beginning of February significant loss has already occurred (.5 ppmv at 450 K). In contrast, Richard et al. finds little evidence for ozone loss before 3 February in the ozone-tracer relations. Rex et al.  have derived ozone loss from ozonesonde data using the Match technique, and using vortex average descent. They find good agreement between these two loss estimates. The ozone change (5 January to 15 March) from the Match analysis is shown in Figure 8, and agrees well with the vortex average descent results except at the peak of the loss (460–480 K) where the Match loss is nearly 0.4 ppmv larger. The Match technique is unique in that it measures ozone loss rate directly, and then integrates the loss rate to obtain the total ozone loss. Rex et al. also finds good agreement between ozone loss inferred from ozonesonde data and from POAM data using the SLIMCAT/Rex descent (same calculation as Figure 5b). Schoeberl et al.  have used ozonesonde, balloon, and POAM data to estimate ozone loss using two methods. The first method is a regression analysis that uses vortex average descent, while the second method employs a diabatic trajectory model. The cumulative ozone loss from 1 December to 15 March is found to maximize at 1.5–2 ppmv near 470 K and decrease to about 1 ppmv at 400 K and to 0 ppmv at 600 K. This again agrees very well with the vortex average descent results shown here. In conclusion, there seems to be general agreement among most ozone loss estimates, with loss peaking at about 1.5–2 ppmv near 475 K. However, there are differences in the timing of the loss, especially in January, and various loss estimates can differ by as much as 0.5 ppmv in the range of 400 K to 600 K. These differences may be related to the fact that the “average” loss reported by each study represents a slightly different type of average. For example, both the vortex boundary and the actual area sampled in each study is slightly different. In addition, the impact of mixing across the vortex boundary will differ for each study. The Match technique is fairly immune to mixing effects because the ozone loss rate is found from short duration trajectories. The vortex average descent method cannot account for any mixing. Methods employing long duration diabatic trajectories [i.e., Schoeberl et al., 2002] rely on the accuracy of the wind fields and may be sensitive to how trajectories entering or exiting the vortex are handled.
7. Summary and Discussion
 We have presented an overview of POAM stratospheric ozone measurements obtained in the 1999/2000 Northern Hemisphere winter. We have also derived photochemical ozone loss from these measurements using two approaches to estimate the dynamical contribution: vortex average descent (using two sets of descent values), and using the REPROBUS passive ozone model simulation. Above 430 K, the cumulative losses calculated with the two different descent rates are virtually identical, and these loss values agree very well with those determined using the REPROBUS passive ozone calculation below about 500 K. In all cases, the ozone loss maximizes at about 475 K on 15 March with a value of about 1.8 ppmv.
 The agreement between the ozone loss below 500 K, calculated from vortex average descent and from the REPROBUS passive ozone, indicates in/out-vortex mixing was not a very important contributor to dynamical ozone change during the 1999/00 winter. However, there is a systematic discrepancy between the average descent and the REPROBUS values above 500 K, with the REPROBUS loss values systematically larger by about 0.5 ppmv. This discrepancy is likely due to mixing at the vortex edge within the model.
 We have also used the REPROBUS full model (including the chemistry module) to estimate possible biases in the ozone loss values resulting from the POAM sampling. This effect is found to be small compared to the total loss values and no larger than the other uncertainties in the calculation. The small magnitude of these biases, combined with the good agreement between the POAM data and other data sets show that POAM can provide an accurate picture of vortex average chemical ozone depletion down to the 400 K level.
 In conclusion, it is clear from papers in this issue, as well as several other papers [Santee et al., 2000; Sinnhuber et al., 2000], that the 1999/2000 Arctic winter experienced very large photochemical ozone loss. Chlorine was highly activated, and Figure 3b suggests that, through early March, the chemical loss of ozone was very similar to that obtained in the Antarctic. However, in the 1999/2000 winter the loss began to slow in mid-March as the vortex began to dissipate. In the Southern Hemisphere, the vortex remains stable through October, and the loss rates maximize in September in response to increasing sunlight. This suggests that a crucial factor in determining cumulative ozone loss is the longevity of the vortex, during which time chlorine can remain activated from continued PSC processing or from previous denitrification.
 The authors thank Bjoern-Martin Sinnhuber and the SLIMCAT team for providing heating rate data, and Gloria Manney for helpful discussions. The Version 3.0 POAM III data can be obtained from the NASA Langley Research Center EOSDIS Distributed Active Archive Center.