The Brewer-Mast ozonesonde was used at Canadian stations from 1966 until 1980, when the Canadian network switched to the electrochemical concentration cell sonde. While the sondes appear to agree relatively well in the stratosphere, there is an evident discrepancy of 10–20% in tropospheric measurements [e.g., Tarasick et al., 1995, Figure 4]. Comparison of Brewer-Mast sondes with a calibrated ozone source yields some interesting insight into this discrepancy. Sonde response is strongly dependent on the preflight preparation procedures employed. Although sondes prepared via procedures introduced in the 1980s [Claude et al., 1987] perform quite well, when prepared according to the procedures used in Canada in the 1970s, Brewer-Mast sondes indicate 10–30% lower ozone than the calibrator. The following points are noted in particular: (1) a new Brewer-Mast sonde shows a large (∼15%) increase in sensitivity between successive experiments; (2) especially at low (<100 ppb) O3 levels, the response even of previously flown sondes increases slowly with time; and (3) sondes show an additional slow increase of response with time that is apparently caused by ozone reactions with the phosphate buffer. The overall response curve indicated by 1, 2, and 3 implies that after correction to the observed total ozone, the earlier part of a flight would yield values that are too low, while the latter part would be too high. By applying a varying ozone input, simulating the typical variation in absolute ozone concentration experienced by a sonde in flight, we show that this can explain both the average correction factor (1.255) for the Canadian Brewer-Mast record and the 10–20% discrepancy in tropospheric measurements.
 The longest continuous record (since January 1966) of vertical profile measurements of ozone is for Resolute Bay, Canada (74.7°N, 95.0°W). However, the Brewer-Mast ozonesonde was used at Canadian stations until 1980, when the Canadian network switched to the electrochemical concentration cell (ECC) sonde. Three other Canadian stations have Brewer-Mast records of significant length as well: Edmonton (53.6°N, 114.1°W), Goose Bay (53.3°N, 60.3°W), and Churchill (58.8°N, 94.1°W). While the sondes appear to agree relatively well in the stratosphere, attempts to combine the two data sets into a homogeneous record are hampered by a discrepancy of 10–20% in the tropospheric values measured by the two types. This apparent difference in tropospheric response between the ECC and Brewer-Mast sonde results in discontinuities in the mean values for these four stations at 1980, which can give a positive trend for the combined record even when both data sets individually have negative trends [e.g., Tarasick et al., 1995, Figure 4]. (Tarasick et al.'s Figure 4 is reproduced here as the top panel of Figure 8; see also Wang et al.  and London and Liu .) If the new Brewer-Mast pump correction of Steinbrecht et al.  is applied to these data, then the tropospheric values decrease by a further 5–7%, increasing this discrepancy.
 The Brewer-Mast sonde [Brewer and Milford, 1960] is a galvanic cell consisting of a platinum cathode and a silver anode in a buffered 0.1% KI solution. It requires an external applied potential of 0.41 V to balance its internal emf. The operating principle is the well-known reaction of potassium iodide with ozone:
Thus for each molecule of ozone, two electrons are produced and an equivalent amount of current flows through the external circuit. The measurement is therefore, in principle, absolute; however, there may be losses of ozone in the pump and of ozone or iodine to the walls of the sensor chamber (the chamber is constructed of acrylic plastic, and the pump piston is steel and, more importantly, is oiled). In addition, there appear to be slow side reactions with the phosphate buffer that can increase the stoichiometry of reaction (1) [Saltzman and Gilbert, 1959; Flamm, 1977].
 We have previously reported work comparing the response of both new and older models of Brewer-Mast and ECC sondes [Tarasick et al., 1996], which found no evidence of a change in response of either sonde. No evidence was found to suggest that ECC sondes, of either older or newer designs, might have overestimated tropospheric ozone, but Brewer-Mast sondes did appear to have underestimated it. An important issue is the preflight preparation of the sonde. Brewer-Mast sondes flown in Canada were prepared according to procedures described by Mueller  (which are still distributed by the manufacturer). Currently operational Brewer-Mast stations employ some version of the procedures developed by Claude et al. , which are generally acknowledged to be superior but were not published until after the Canadian program had switched to ECC sondes. The significant difference is that Claude et al. recommend disassembly and a complex cleaning of the sonde before use. In addition, more recent practice at both Payerne and Hohenpeissenberg is to charge the sonde with sensing solution at least 24 hours before use.
 Unfortunately, Canada did not participate in either of the two World Meteorological Organization international ozonesonde intercomparisons that took place in the 1970–1980 period [Attmannspacher and Dütsch, 1970, 1981]. However, the original documentation of the preparation procedures for Brewer-Mast sondes exists (and is essentially Mueller ); so it is possible to examine in the laboratory the response to ozone of sondes prepared in this way.
 Experiments were conducted at laboratory temperature and pressure, using either a National Institute of Standards and Technology (NIST)-traceable TEI ozone calibrator (model 49C-PS) or a NIST-traceable Dasibi ozone calibrator (model 1008-PS). The expected precision of either instrument was 1 ppb. The air output from the ozone calibrator was allowed to flow through Teflon tubing which vented to the laboratory, while the sondes sampled air from the tube. Tubing lengths were kept short (2 m) in order to avoid ozone loss. Except as otherwise specified, sondes were prepared according to standard procedures used in Canada [Mueller, 1976]: that is to say, the sondes were not disassembled and cleaned, nor was the pump re-oiled; after removal from the plastic shipping bag, the pump rate was measured, then the sonde was ozonized by exposure to a high-ozone source, and then, following several minutes of exposure to ozone-free air, filled with sensing solution.
 The electronics supplied with the sonde were not used. Instead, the required 0.41-V bias was supplied by a battery and resistive voltage divider; a variable resistor allowed this to be adjusted. Bias voltage was adjusted before each experimental run and checked occasionally during the run. A simple operational amplifier electrometer circuit was used to convert current to voltage, which was read by an HP34401A digital multimeter. Both the ozone calibrator reading and the voltmeter output were recorded digitally via RS-232 interfaces to a PC. Ozone partial pressure was then calculated by using the standard formula
where i is the measured current in microamperes, TB (normally the sonde box temperature) is the laboratory temperature in kelvins, and t is the time in seconds to pump 100 mL of air, measured via the standard soap bubble apparatus. Partial pressures were converted to mixing ratio by dividing by the measured atmospheric pressure.
 As the ozone calibrator output is in ppb, it was only necessary to correct for the zero offset (measured before each run) and for the measured calibration correction to the NIST transfer standard. Although in our experiments the sondes produced a measurable background current in the great majority of cases, this does not appear to have been the case in early experiments [Brewer and Milford, 1960; Griggs, 1961], and standard practice is not to subtract it. This is of some significance to the application of our results, and is discussed below. In all of the results that follow, background current was subtracted except where it was negligible (less than 1 ppb).
3. Sonde Response
Figure 1 shows the results of a typical experiment, in which a new Brewer-Mast sonde, prepared via Mueller , was exposed to ozone repeatedly, at intervals of several hours. It illustrates several features of the response of Brewer-Mast sondes. First, and most evident, is that the sondes always indicate less ozone than the calibrator, typically by about 15–20%. Although response improved with repeated calibrations, as described below, it never exceeded 90% in the tests described here, even for sondes prepared via Claude et al. . This less-than-integral response is well known and is the reason that ozone soundings with Brewer-Mast sondes are corrected to agree (after integration) with the total ozone measured by a Dobson or Brewer spectrophotometer. Typical correction factors for well-prepared sondes at Hohenpeissenberg average about 1.1 [Claude et al., 1996] (implying an average response of 90%), while for the Canadian Brewer-Mast record they average 1.25 (implying an average response of 80%).
 Perhaps surprisingly, the reason for this does not appear to be incomplete absorption of ozone from the air stream. It is a simple experiment to vary the depth of solution in a sensor cell and observe the ozonesonde response: no change, even for depth variations of 50%, was observed, as long as the solution completely covered the cathode gauze. However, replacing the sensing solution with ECC (1% KI) solution immediately increased the response to 100%. This was investigated more thoroughly by Powell and Simmons , who found an increase of as much as 30% when the solution strength was varied between 0.1% and 1.0% KI. Unfortunately, it is not clear whether the buffer concentration was also varied (see the discussion below).
 These observations indicate that all the ozone in the air stream is being absorbed into the solution, as otherwise the stronger KI solutions would also give a less-than-integral response. Since reactions (1) and (2) are independent of KI concentration, the difference (for well-conditioned sondes) is probably due to evaporation of iodine from the weaker KI solution. Consideration of the reaction
which has an equilibrium constant of 768 M−1 [Davies and Gwynne, 1952], shows that in a 1.0% KI solution, 98% of the iodine will be in the form of tri-iodide, while in a 0.1% KI solution, only 82% will be in the ionic form. The possibility of iodine evaporation was also noted by Brewer and Milford .
 A second feature that is present in all the response curves is a slow increase in response with time, even after 20 min on a constant ozone input. This also resulted in a large (5–10 ppb) residual signal after the ozone input was decreased to zero. This residual decayed to zero with a time constant of about 20 min. However, it disappeared immediately if the solution was changed. A similar effect is observed with ECC sondes and is believed to be caused by the reaction of ozone either with, or catalyzed by, the phosphate buffer. It appears to be well known in the literature about the reaction of ozone with neutral buffered KI [Saltzman and Gilbert, 1959; Flamm, 1977] and may have been responsible for at least some of the controversy in the 1970s about the stoichiometry of the neutral buffered KI method [see Pitts et al., 1976; Bergshoeff et al., 1980, and references therein]. In the experiments illustrated in Figures 1–3, the second reaction was a small effect, but clearly evident, accounting for typically 4–6% of the total response.
 More evidently from Figure 1, there is a large increase in response (∼10%) between the first and second trials and much less between the second and third. In numerous calibrations of new Brewer-Mast sondes prepared according to Mueller , we found this effect to be typical: it was observed with each sonde tested and varied in magnitude from about 5% to more than 20%.
 In contrast, it was found that older, previously flown Brewer-Mast sondes (taken from a stock of recovered sondes, flown in the late 1970s) show much smaller or negligible changes in response between successive sensitivity trials. The salient difference between the new sondes and the used sondes is evidently previous exposure to sensing solution and ozone. Figure 2 shows the response to ozone of a new Brewer-Mast sonde which, after conditioning with high ozone in the normal manner, was exposed to solution for 9 days before its first use in a sensitivity experiment. The results in this case are similar to those for the previously flown sondes, with only a small increase in response during the second test.
 Similarly, no large increase in response between successive sensitivity trials was seen in new sondes prepared via Claude et al.  (where the sondes were charged with sensing solution at least 24 hours before use, as in current practice at Hohenpeissenberg and Payerne). An exception to this was noted in the case of a sonde that was subjected to partial preparation via Claude et al. , that is, with the cleaning of the silver anode via KCN but without the heating of the cathode or the ultrasound bath.
 Taken together, the above observations suggest that the large increase in response between the first and second test of Figure 1 is related to (lack of) iodine conditioning. An iodine conditioning effect has been noted for glass [Bergshoeff et al., 1980; Saltzman and Gilbert, 1959], and iodine conditioning effects on several plastics, including the acrylic plastic of which the Brewer-Mast sensor is constructed, were studied by Griggs . However, the sensitivity change is apparently permanent: it was not observed with previously used sondes that had been in storage for 20 years. This suggests that it is related to iodine conditioning of the Pt cathode, as both iodine and iodide are known to adsorb strongly on platinum surfaces, and the adsorption is thermodynamically irreversible; i.e., the iodine is not removed by rinsing with solvent. Saturation coverage is as high as one I per Pt in the surface layer [Lane and Hubbard, 1975]. Iodine produced by the reaction of ozone with KI but adsorbed onto the Pt surface would not accept electrons from the cathode via (2) and hence would not contribute to the sensor output current. An estimate of the cathode surface area yields about 3 × 1016 atoms of Pt in the surface layer, while the number of atoms of I produced in the sensor by 60 min of exposure to ozone at 70 ppb is about 4 × 1016, so that such an effect could reasonably account for a 10% loss in the latter (at least on arithmetical grounds). The heating of the cathode in the Claude et al. pretreatment may render it more susceptible to rapid chemisorption of iodide from the sensing solution, such that the effect is minimal after 24 hours of exposure to iodide in the sensing solution.
 In addition to the initial change just discussed, the response of sondes over several calibrations was observed to vary systematically in other ways. Figure 3 illustrates this behavior: while the second run is actually more than 20% higher than the first (both on 12 January), the third, taken 3 days later, is about 4% lower at its midpoint. The fourth, again 3 days later, is slightly lower again (although it should be noted that this is at a lower ozone input), while the fifth run, on the same day, is almost exactly the same as the second run. The sensing solution was changed before each run, but ozonization of the dry sonde was performed only before the first run. This behavior was observed frequently; sonde response was generally somewhat lower for a calibration repeated after an interval of several days, but was always slightly higher for a repeat calibration after an interval of less than 24 hours. It will be noted that, except for the first, the response curves approach each other closely after 30 min. Thus the differences are primarily in terms of response time rather than ultimate response. The differences also appear to depend on the amount of ozone input: the fourth run, at 20 ppb, is lower than the third, at 40 ppb, although the time interval since the previous run was the same.
 The differences of response time in these curves is not in the rapid first-order response, which has a time constant of about 20 s [Brewer and Milford, 1960; De Muer and Malcorps, 1984]. Rather, the differences of about 5% after 10 min imply a time constant of the order of a few minutes. A frequency response of this timescale was also observed by De Muer and Malcorps  (their ω3) and interpreted as a response in the air-sampling system (i.e., the Brewer-Mast pump). The effect illustrated here, however, is clearly related to the history of the sonde being tested. Although this does not rule out wall effects in the air-sampling system, a more plausible explanation is that it is related to conditioning of the sensor walls by iodine adsorption: this would also be consistent with the observations of Griggs  and the fact that the effect appears more marked at lower ozone input levels, as well as that the effect disappeared when the sensing solution was replaced with ECC (1% KI) solution.
4. Background Current
 As was noted earlier, in the majority of our experiments the sondes produced a measurable background current, generally of the order of one to several ppb. This was true even of sondes that were prepared via Claude et al. . However, this appears to have not been the case in early experiments [Brewer and Milford, 1960; Griggs, 1961], and standard practice is not to subtract it. More importantly, examination of a representative sample of original chart recordings for which background current was measured incidentally as part of the preflight preparation (Figure 4) shows clearly that background currents for Canadian Brewer-Mast soundings were generally negligible.
 We also found low or negligible background currents for repeat experiments with sondes, as long as the ozonization step was not repeated. This suggests that ozonization is the cause of the higher background current, where there is not sufficient time for it to decay away. The source of high ozone used to condition sondes in the 1970s was an ozone conditioning cabinet containing several UV lamps. Unfortunately, none still exist, but it is possible to estimate the concentration of ozone produced in it as between 200 and 500 ppb, about 100 times less than the type of ozone generator in current use in Canada, which was used for most of the experiments described here. A similar increase in average background current for Brewer-Mast sondes is also reported by DeBacker  after the Uccle program switched from an ozone cabinet to a higher-intensity ozone conditioning source. Further evidence is found in three experiments with sondes for which the ozonization step was accidentally omitted from the preparation: all showed negligible background current.
 In the experiments described in the next section, two sondes (B14804 and B14805) were ozonized using the high-ozone source; three (B14794, B14795, and B14796) were ozonized at 500 ppb, and two (B13235 and B13069) were ionized at 200 ppb. Four showed significant background currents (2.2–2.6 ppb), while the remaining three (B14794, B13235, and B13069) showed none. Subtraction of the background current from B14795, B14796, B14804, and B14805 in the simulated flights described below gives response curves similar to those for the other three sondes. If this is not done, then the response curves show much less coherence in the troposphere, and the average response below 300 hPa varies between 0 and −20%, with an average deficiency of 5–10%, rather than the 10–20% we find below.
5. Simulated Flights
 The several time-dependent effects on sonde response discussed above are difficult to separate, as they show significant sonde-to-sonde variability and they depend on the ozone input and the sonde history as well as time. However, in the Canadian sonde program, few sondes are recovered (most are lost in the Arctic or Atlantic Oceans), so most sondes are new instruments. Their ozone exposure is also somewhat predictable, being quite low (2–4 mPa) for the first 40 min or so, in the troposphere, then much higher. This suggests the possibility of measuring an overall response curve that will be a function of time, a combination of the time-dependent effects described above. As the individual time dependencies each induce an increase of response with time, their combined effect will certainly be to produce a lower response in the troposphere than later in the flight.
 Experiments were therefore conducted, at surface pressure, to investigate the response of Brewer-Mast sondes during simulated “flights.” Sondes were prepared according to standard procedures used at Canadian stations in the 1970s. Specifically, after removal from the plastic shipping bag, the pump rate was measured, then the sonde was ozonized by exposure to a high-ozone source for 30 min, and then, following several minutes of exposure to ozone-free air, filled with 2.5 cm3 of sensing solution. During the next hour the sonde was run intermittently, as a decay test was performed and the background current was measured. After 1 hour the “flight” was begun by setting the ozone calibrator to produce the amount of ozone appropriate to the tropospheric part of the profile. The flight was simulated by changing the ozone input at the appropriate time (i.e., using a hypothetical ascent rate for the sonde), as indicated in the idealized profiles shown in Figure 5. Five of the sondes in the experiments shown in Figures 6 and 7 were new instruments, while two (B13235 and B13069) had been previously exposed to sensing solution and ozone.
 Evidently this method does not simulate the pressure and temperature changes of an actual balloon flight. The temperature change experienced by the sonde is fairly small [Claude et al., 1987]. Pressure should produce important differences above 100 hPa (where pump efficiency becomes an issue) but may also have unknown effects on the bubbling action and hence on ozone absorption efficiency and on I2 transport to the cathode (J. Davies et al., paper in preparation, 2002). This is also suggested by intercomparison results [DeBacker et al., 1998], indicating pressure-dependent differences in response between Brewer-Mast and ECC sondes at pressures below 100 hPa.
 However, to the extent that these effects can be neglected, the response curves shown in Figure 6 should be representative of those achieved during actual flights in Canada in the 1970s. Two quite different profiles were used, in order to verify that the sonde response was not strongly dependent on the details of the ozone profile. The first profile in Figure 5 is an idealization of the “standard midlatitude profile” used in the JOSIE 1996 intercomparison [Smit et al., 1996], while the second is taken from an actual flight at Resolute on 26 February 2000. Different balloon ascent rates were used as well, resulting in somewhat different time intervals for the ozone input steps. For four of the sondes, using the standard midlatitude profile, pressure profiles and ascent rates typical of Edmonton were used (14794, 14796, 14804, and 14805), while for the fifth (14795) the pressure profile and ascent rate for the actual Resolute flight was used. Despite these differences, the results (Figure 7) show considerable similarity.
 Using these pressure profiles, it was possible to integrate both the calibrator output and that of the sonde and so calculate a correction factor for the “flight” (Table 1). This also requires an assumption about the time interval between starting the sonde pump (so that the sonde begins to respond to the surface ozone concentration) and actual balloon release. This was estimated on the basis of current practice, and so the first 10 min, representing the largest variation in response in Figure 6, has been omitted. The estimate of 10 min may be excessive, as the interval may have been as little as 3 min; in this case, the tropospheric response would be even lower. Figure 7 shows the result of applying the derived correction factors to the response curves of Figure 6 (plotted against the assumed pressure profile in each case). Response profiles for Brewer-Mast and ECC soundings during JOSIE 1996 [Smit et al., 1996] are also indicated. For these seven simulated flights the average response below 300 hPa is 10–20% too low in absolute terms, and even more when compared with the JOSIE profiles. The correction factors are similar to those found in the Canadian record, averaging 1.25 for the seven sondes.
Table 1. Ozone Input Profile, Hypothetical Pressure Profile and Balloon Ascent Rate, Shown Along With Calculated Parameters for Each Simulated Sounding
Ascent Rate, m s−1
Sonde Integral, DU
Calibrator Integral, DU
Flight Duration, min
Except that tropospheric part was 2 mPa, rather than as shown in Figure 5.
Except that tropospheric part was 3 mPa, rather than as shown in Figure 5.
 The average response curve in Figure 7 is similar in form to that deduced by De Muer [1980, 1984] from a comparison of the ascent and descent portions of soundings made at Uccle, Belgium. The correction derived by De Muer is nearly twice as large, however, as he assumed that the descent values were correct. In fact, descent values in the troposphere will be strongly affected by the second, slow reaction discussed above, as the sonde descends rapidly from the stratosphere where ozone partial pressures are much greater. If the slow reaction represents 5% of the total response at 20 mPa, it will be 20% of the tropospheric response at 5 mPa.
Figure 8 is an example of a preliminary attempt to correct for these deficiencies in tropospheric response. The top panel is reproduced from Tarasick et al.  and shows trends calculated from Brewer-Mast and ECC data for the 700–500 hPa layer at Edmonton. The data are as recorded in the World Ozone and Ultraviolet Data Centre; that is, the Brewer-Mast soundings have all been linearly scaled to agree with a total ozone measurement. The lower panel shows the same data, except that each Brewer-Mast sounding has been corrected, in the troposphere, by the average response for 700–500 hPa from Figure 7 and the sounding has been rescaled to agree with the total ozone measurement. The correction seems to remove the discontinuity in the tropospheric time series, and the overall trend is now similar to the trends for the separate data sets.
 The low level of ozone concentration that we have inferred above for the ozonization of sondes by the conditioning cabinet raises another potential source of poor tropospheric response. Three sondes for which the ozonization step was omitted showed a pattern of increased response with time that was similar to that in Figure 7, except that the overall response was lower: averaging between 61% and 77%. Treating these runs as simulated flights gave correction factors between 1.303 and 1.636, with “tropospheric” responses between 7% and 30% low. This suggests that inadequate ozonization might lead to further deterioration of tropospheric response, and examination of the Canadian record shows that there is indeed a weak positive correlation, at all four stations at all levels, between high correction factors and low tropospheric ozone values.
 Brewer-Mast sondes were employed in the Canadian ozone sounding network during the period 1966–1979. When prepared according to the procedures then in use in Canada, they show an average response of about 80% compared with a NIST-traceable ozone calibrator, which is consistent with the average total ozone correction factor for ozonesonde flights during this period of 1.255. At no time did any sonde give a reading higher than the calibrator. Possible loss processes include evaporation of iodine from solution, losses of ozone in the pump and to the sensor walls, and of iodine by adsorption to the Pt cathode and possibly the walls as well. In general, except for iodine evaporation, these loss processes are strongly dependent on the details of the preflight preparation procedures employed. For sondes prepared according to the procedures introduced in the 1980s [Claude et al., 1987], they are much reduced or eliminated.
 Two loss processes are observed to diminish with time, such that the response of the sonde will improve during a flight. The first, and largest, causes new Brewer-Mast sondes to show a large (∼15%) increase in sensitivity after first exposure to sensing solution and ozone. This effect is much reduced with previously flown sondes and in sondes that are conditioned by being exposed to sensing solution for several days before calibration. It is also quasi-permanent. We suggest, therefore, that it is related to conditioning of the Pt cathode by iodine adsorption. The second process may to be related to iodine conditioning of the sensor walls and consequent changes in iodine or ozone loss there, as it is also reduced by exposure to sensing solution and ozone, but the improvement is temporary.
 Sondes show an additional slow increase of response with time that is apparently caused by secondary reactions with the phosphate buffer.
 The above mentioned processes, both individually and in combination, will cause tropospheric values for ozone soundings to be too low in relation to stratospheric measurements and therefore too low overall once the total ozone correction factor is applied. By exposing Brewer-Mast sondes, prepared via the procedures used in Canada, to ozone inputs similar to actual atmospheric profiles, we have shown that it is possible to reproduce both the low tropospheric response and the typical total ozone correction factors found in the Canadian Brewer-Mast record. The average correction factor for the sondes in Figure 7 is 1.247, while the average for the Canadian Brewer-Mast record is 1.255.
 While noting that our experiments have been limited to surface temperature and pressure, we suggest that the correction to tropospheric ozone values in the 1970s at Canadian stations that is indicated by Figure 7 is sufficient to explain the low tropospheric values for Brewer-Mast ozone soundings in Canada. We intend to continue these preliminary experiments with corresponding work in the field and in a pressure–temperature chamber and to apply these results to a reevaluation of the Canadian Brewer-Mast sounding record, which will then be submitted to the World Ozone and UV-B Data Centre.
 Finally, we note that field intercomparisons of sonde types and of preparation procedures may produce ambiguous results unless the effects of differences in sonde preparation are properly understood. (This is likely to be equally true for ECC sondes.) For example, such apparently unimportant (and generally unrecorded) factors as the length of time between filling the sonde with solution and launch, or between calibration and launch, may seriously affect the tropospheric results in particular. The potential biases that may result from such recent improvements, or others in the past, should also be considered in future analyses of trends derived from ozone sounding data.
 We are grateful to J. Easson and D. Anderson of CSIRO, Australia, for having kindly provided some of the sondes used in this study. We are indebted to I. A. Asbridge for assistance in interpreting the older data records.