This study documents the history of the Payerne (Switzerland) ozone series obtained with the Brewer-Mast sonde from the end of 1966 until the change to the electrochemical concentration cell (ECC) sonde in autumn 2002, as well as the reevaluation of the original data. Several corrections were made in order to improve the homogeneity and the quality of the time series. We furthermore derived long-term trends for the reevaluated time series using atmospheric variables in a stepwise regression model. In the stratosphere, trends over the 1970–2002 period remain nearly the same as over periods ending a few years earlier. For tropospheric ozone trends, a hockey stick model allowing for a change in trend in 1990 was used and a sensitivity analysis with different data sets was carried out. Besides the standard World Meteorological Organization (WMO) data evaluation procedure, we used alternative data sets (1) accounting for the preflight laboratory calibrations, or (2) ignoring the total ozone normalization, (3) as well as correcting for chemical interference with SO2. With all data sets, tropospheric trends were strongly positive in all seasons over the 1967–1989 period. In the 1990–2002 period, winter trends remained positive over the whole troposphere with all data sets, whereas in the other seasons, trends were generally negative near the ground and shifted to zero or positive values with increasing altitude in the troposphere. The alternative evaluation procedures strongly affect the derived tropospheric trends in the 1990–2002 period and their uncertainties.
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.
 The stratospheric ozone layer has been undergoing a significant decrease since the 1970s, also over midlatitudes [Stolarski et al., 1992; Harris et al., 1997; Solomon, 1999; World Meteorological Organization (WMO), 1999; Staehelin et al., 2001] mainly attributable to anthropogenic ozone depletion. However, the Montreal Protocol and the subsequent amendments led to an important decrease in the emissions of ozone depleting substances (ODS). In the troposphere, the low background ozone concentrations have increased by more than a factor two since World War II in the planetary boundary layer over Europe most probably because of anthropogenic emissions of ozone precursors [e.g., Staehelin et al., 1994]. In order to limit photochemical smog pollution industrialized countries reduced these emissions in the last decades.
 Ozonesonde records provide a unique data source to derive ozone profile trends. A few records started more than a decade earlier than ozone satellite measurements. They have been widely used for long-term trend analysis providing information from the planetary boundary layer up to approximately 30 km asl. However, the quality assessment of these data sets is a difficult task [SPARC/IOC/GAW, 1998], particularly with respect to the troposphere, as well as to the 1970s since almost no redundant information is available for this period. Swiss ozone profile measurements with ozonesondes started at the end of 1966 and have been widely used in trend analyses [e.g., Dütsch et al., 1991; Staehelin and Schmid, 1991; Stolarski et al., 1992; Logan, 1985; Logan et al., 1999; Weiss et al., 2001; Koch et al., 2002; Naja et al., 2003; Delcloo and De Backer, 2005]. The Payerne ozone time series is one of the longest in the world and probably contains the largest number of ascents. In the first part of this paper we report the results of the reevaluation of this series. In the second part we present results of statistical trend analyses based on the reevaluated series. In order to evaluate the reliability of tropospheric ozone trend results we also present alternative data sets, e.g., attempting to account for preflight laboratory calibration and SO2 interference.
2. Measurements, History, and Data Reevaluation of the Payerne Ozonesonde Series
2.1. Principle of Measurements
 The measurement of the vertical ozone profiles has been performed at Payerne using a small sonde carried aloft by a meteorological balloon that bursts at an altitude of typically 30–33 km. The Brewer-Mast (BM) ozone sensor [Brewer and Milford, 1960] has been continuously used from November 1966 to August 2002. The ozone partial pressure PO3 (in nbar) is determined from the electrochemical current through the ozone sensor (i in μA) by
where t is the time for 100 cm3 of air to pass through the miniature pump (in s), and Tp is the air temperature inside the pump (in K). In equation (1), the number of ozone molecules pumped through the cell is converted to the ozone volume concentration in the ambient air. For this conversion, information on the decrease in the pump efficiency with altitude is needed, as well as on the air temperature inside the miniature pump. As these measurements are not performed in the operational soundings, standard values are recommended by the WMO procedure [Claude et al., 1987]. The electrochemical efficiency of the BM sonde reduces the measured values (wall losses in the inlet and collection efficiency of the bubbler) leading to additional uncertainties. The WMO standard procedure includes linear scaling of the ozone profile with the total ozone measured by a Dobson spectrophotometer applying the correction factor CF. Claude et al.  includes further recommendations concerning the sonde preparation, the preflight procedure and the data evaluation.
2.2. Historical Changes
 Regular ozone soundings have been carried out in Switzerland since November 1966. Initially, the balloons were launched from Thalwil near Zürich, then since mid-August 1968 from the MeteoSwiss aerological station at Payerne (46.80°N, 6.95°E, 491 m asl) in the centre of Swiss Plateau. Thalwil is located 140 km ENE from Payerne. Since the beginning of the monitoring program, three soundings have been planned every week, i.e., on Mondays, Wednesdays, and Fridays. The launch time changed several times before 1982. The main features and changes of the soundings are summarized in the first lines of Table 1 and in Table S1 in the auxiliary material.
 A very careful preflight preparation and conditioning of the Brewer-Mast sonde is required for reliable and reproducible measurements (e.g., before launch the sonde has to be exposed to high ozone concentrations). However, in the earlier years this practice was performed only qualitatively. Three main improvements in the preflight procedures occurred in 1976, 1983 and in 1993. In 1976 the sonde preparation was moved from Zürich to Payerne and improved according to the preflight protocol used in Hohenpeissenberg. In January 1983, a laboratory calibration with a reference ozone UV photometer was added. The results were used to document the quality of the preflight procedure and to exclude low-quality sensors. The bubble meter for the preflight measurements of the pump airflow was replaced by an electronic Hastings flowmeter in mid-December 1983. In 1993 the procedures were fully reexamined and improved in several points, including complete disassembly and meticulous cleaning of every new sonde followed by repeated calibration with fresh sensing solution.
Table 1. Main Features and Changes of the Ozone Series of Payernea
Nov 1966 to Jul 1968
Aug 1968 to Oct 1980
Nov 1980 to Mar 1990
Apr 1990 to Dec 1998
Jan 1999 to Aug 2002
Normalization (CF) is based on the homogenized time series of total ozone measured at Arosa. In ozone processing, every ozone profile was checked by visual inspection and reprocessed in an uniform way. Preflight calibration is only taken into account in the alternative time series (Jan 1983: laboratory calibration unit put in operation). SO2 interferences are only taken into account in an additional processing. Rejection criteria include soundings not reaching 26 hPa and with CF outside the range 0.9–1.4, with additional rejection of soundings with bad preflight calibration results and without or with doubtful temperature measurements (normal data set). More restrictive criteria are applied for special data sets.
1976 BM 730-8
special styrofoam box
Swiss sonde CH
Swiss sonde SRS
since 1982: 1100 UTC
Pump efficiency correction
WMO standard, Tp (pump air temperature) = 300 K
WMO standard, Tp = 300 K
WMO standard, Tp = 280 K (Dec 1983: change of laboratory flowmeter)
 From 1966 to November 1980 the Brewer-Mast ozone sensor was connected to the US VIZ meteorological sonde. During this period the Brewer-Mast sensor was placed in the original Brewer-Mast Styrofoam box. From November 1980 to March 1990 the Brewer-Mast sonde flew together with the mechanical Swiss radiosonde (CH). The ozonesonde was integrated in a separate compartment of the radiosonde Styrofoam box and got a new electronic interface to the electronics of the radiosonde. The new electronic Swiss Radiosonde (SRS) was introduced in April 1990. The ozonesonde was packed jointly with a new electronic interface in a separate compartment of the Styrofoam box containing the meteorological sensors and the electronics.
 Additional small changes of instrumentation and operating procedures possibly influencing the data quality of the measurements [see SPARC/IOC/GAW, 1998] are documented in the detailed (unpublished) station history. In autumn 2002, the Brewer-Mast ozonesonde was replaced by the ECC sonde.
2.3. Data Reevaluation
 In this section we summarize the reprocessing and homogenization procedures applied to the Payerne series (see Table 1, for more details see Stübi et al.  and Favaro et al. ). In the first step a visual plausibility check was performed on each sounding. Suspect drops of the measured ozone values most likely caused by the freezing of the iodide solution at high sounding altitudes were removed. A comparison with the ozone profiles from Hohenpeissenberg (Germany) of the same day was performed before removing a single ascent, leading to the removal of a few percent of the profiles, primarily of the 1970s.
2.3.1. Pump Flow and Pump Temperature Correction
 Time t in equation (1) is measured in the laboratory for every BM ozonesonde prior to the launch. The change of the bubble meter by an electronic Hastings flowmeter introduced a rupture in the time series of the airflow rate through the pump (+3–4%). This was not corrected for, but was automatically (at least partially) compensated by an opposite reduction of the CF factor (linear scaling). The standard pump efficiency correction recommended by Claude et al.  was applied. Other groups proposed new pump efficiency functions [De Backer, 1999; Steinbrecht et al., 1998b; SPARC/IOC/GAW, 1998; Lehmann and Easson, 2003]. However, twin soundings with BM and ECC ozonesondes performed at Payerne showed a limited influence of a change of this function for the Payerne BM ozone time series [Stübi et al., 1998; R. Stübi et al., manuscript in preparation, 2007].
 The temperature Tp inside the pump also directly affects the calculated ozone concentration (see equation (1)). The constant value of 300 K for Tp recommended by Claude et al.  has been used for the original BM-VIZ package until November 1980. It causes an uncorrected distortion of the ozone profile of a few percents, as the pump temperature decreases by several degrees during the sounding. The special BM packages designed for the two Swiss meteorological sondes (BM-CH and BM-SRS) did not thermally protect the BM pump in the same way as the original BM-VIZ package and a constant value of 280 K has been applied since December 1980. This lowers the ozone measurements by 7% (equation (1)). According to recent measurements with the Payerne BM sonde design and battery, a mean value between 285 and 290 K would be closer to the reality [see also Steinbrecht et al., 1998b]. The CF correction is expected to linearly adjust this scaling to the total ozone. However, the change in the vertical profile of the air pump temperature between the standard BM box and the new box are not taken into account by the standard evaluation procedure and can be responsible for discontinuities of a few percent in the ozone time series at different altitudes. Note that the errors in the pump efficiency function and in the pump temperature introduced by the standard WMO procedure have an opposite influence on the final ozone values. However, other effects like the evaporation of the solution are ignored by the selected approach.
2.3.2. Corrections for Problems in the Early 1990s
 An unexpected ozone break occurred in April 1990 in conjunction with the change of the meteorological sonde and electronic interface (see section 2.2 and Figures S1 to S3), which could not be properly explained by laboratory experiments. It was partially removed with a statistical correction. The comparison with Alpine surface ozone monitoring stations (Zugspitze, Jungfraujoch) also confirmed the anomaly of the Payerne ozone soundings between April 1990 and March 1993. The statistical model of Tiao [Tiao et al., 1986; Bojkov et al., 1990] was adapted by using a rectangular step function for that period. In the troposphere, the derived corrections amounted to almost 20% of the raw ozone values and were highly significant. At 10 hPa and above, the corrections were also highly significant. Comparisons with the ozone soundings at Hohenpeissenberg, before and after completion of this homogenization, confirmed the large improvement of the reevaluated Payerne time series, but still showed some anomalies (e.g., at 15 hPa).
2.3.3. Corrections for Launch Time Changes
 The original time series shows a strong decrease of the measured ozone concentrations at 925 hPa during the years 1977–1981 (Figure S1), most probably caused by launch time changes. Similar changes took place between 1966 and 1981 (Table S1). We refined the corrections proposed by Staehelin and Schmid  using diurnal ozone cycles measured at surface stations at different altitudes that were averaged on a monthly basis. For more details see Favaro et al. , Figures S1–S3 and section 3 below.
2.3.4. Corrections Due to the Changes of Meteorological Sondes
 The heterogeneities in the meteorological parameters are mainly due to the changes of the meteorological radiosonde (Tables 1 and S1). Pressure measurement errors have the largest impact on the ozone profiles in the middle stratosphere (see the pressure corrections on ozone in Figure S3). At temperate latitudes, a pressure sensor giving too high pressure values in the stratosphere, e.g., +0.5 hPa from 100 hPa up, causes ozone concentrations being 6% too low at altitudes above 20 hPa [Claude et al., 2000]. At Payerne, three meteorological radiosondes have been used with the ozonesonde (see section 2.2). The procedures used for the correction of their pressure and temperature measurements are given by Favaro et al. . Table 1 simply quotes which parameters were corrected for each of the three radiosondes and Figures S1 to S3 illustrate the effects of these corrections.
2.3.5. Linear Scaling of Profile to Total Ozone (CF)
 For the application of CF, the total ozone amount of the profile is first determined including ozone amount above the balloon burst level, by assuming there a constant ozone mixing ratio. For soundings with a balloon burst height above the pressure level of 17 hPa, the mean ozone value over the last 2 hPa (but not above the 8 hPa level) was vertically extrapolated assuming a constant volume mixing ratio. The use of the extrapolation procedure at altitudes above the ozone volume mixing ratio maximum implies an overestimation of the herewith computed residual ozone values, which is expected to result in an overestimation of the equivalent total ozone value. However, in the case of the Payerne BM soundings, Calisesi et al.  showed by comparison with microwave remote sensing data that the residual ozone overestimation is largely balanced by the too low correction applied to the profile for the loss of the sensor pump efficiency above 20 hPa, resulting in a small bias for CF. For soundings with balloon burst height between 30 and 17 hPa, a SBUV climatology for northern latitudes between 40 and 50 degrees is applied (Solar Backscatter UV Radiometer, on board NOAA satellites [McPeters et al., 1997]).
 The total ozone columns required for the CF computation were taken as daily averages of the Dobson spectrophotometer measurements performed at Arosa (1860 m asl, 200 km to the east of Payerne) using the homogenized Dobson101AD series [Staehelin et al., 1998]. The height difference between Payerne and Arosa is taken into account by integrating the ozone soundings only above the height of Arosa. Strong horizontal gradients of the ozone distribution in the distance between Arosa and Payerne occasionally occur, which, however, are expected to cancel out in mean values. On days when Dobson values were missing, values were estimated before 1978 [Dütsch, 1974], then TOMS vs. 7 values were used taking into account an adjustment based on the regression with the Dobson series, and since 1994 new satellite instruments have been used.
 The CF allows a valuable quality improvement of the sounding profiles measured with the Brewer-Mast sensor at the altitudes with large ozone concentrations, partially compensating for instrumental uncertainties. Comparisons with SBUV(/2) confirmed that correction by total ozone measurements noticeably reduces ozonesonde uncertainties in the stratosphere [Fioletov et al., 2006]. However, the use of CF for tropospheric ozone from BM sondes was questioned [SPARC/IOC/GAW, 1998; Beekmann et al., 1995; De Backer et al., 1998; Thouret et al., 1998]. Furthermore, the JOSIE experiment of 1996 concluded that the ozonesonde background current is another key issue of uncertainty in the troposphere [Smit and Kley, 1998]. Consequently, alternative or additional time series are introduced in this study, in order to take into account shortcomings of the standard WMO procedure.
2.3.6. Additional Corrections Using Preflight Laboratory Calibration
 Since 1983 every sonde was calibrated in the laboratory before launch yielding slope and intercept versus a known ozone concentration enabling to evaluate the BM sonde background current and to calculate an alternative ozone time series, in which the calibration slope and intercept are applied before the normalization with total ozone (intercept: see Figure S4). Between 1985 and 1993, the measured offsets were mostly well over 3 nbar, while they were on average close to 3 nbar in 1984 and after 1993 [Favaro et al., 2002]. This value was assumed for the years prior to 1983, when laboratory calibrations were not available.
 A systematic application of both slope and offset of preflight calibration presupposes that the laboratory calibration is valid for the entire profile, which is questionable. We chose a version of this correction that can be considered as valid both in the troposphere and in the stratosphere: calibration offset and slope are applied before total ozone normalization. In the stratosphere, this procedure leads to ozone values similar to the standard series. In the troposphere, where the ozone concentrations are low, the influence of the calibration offset is large and this series noticeably differs from the standard one. This procedure leads to different CF factors.
2.3.7. Additional Corrections for SO2 Interference
 SO2 causes a negative interference with ozone measured by the Potassium Iodine method, with a stoichiometry equal to one [Schenkel and Broder, 1982]. Indeed some Payerne soundings registered zero or negative ozone values in the first winters and the SO2 interference possibly introduced an artificial ozone trend in the planetary boundary layer. Before 1980, the winter surface SO2 concentrations reached on average 10 ppb at Payerne, while only around 1 ppb was measured after 2000 [Lövblad et al., 2004]. At the Jungfraujoch station (3580 m asl), the SO2 mean values amount to approx. 20% of the Payerne values, but exhibit a smoother annual cycle.
 On the basis of the continuous SO2 measurements at Payerne and Jungfraujoch (since 1969, unpublished BUWAL-EMPA data), an additional correction for the monthly ozone time series in the low troposphere was developed. The monthly SO2 surface measurements at Payerne (960 hPa) were used to drive an exponential decrease approximately fitting the Jungfraujoch measurements at 650 hPa (only four seasonal exponential constants over the years 1967–2002). The resulting SO2 concentrations were added to the ozone values in all tropospheric levels. By this approach the possible effect of the SO2 accumulation as HSO3 ions in the sonde bubbler solution has not been accounted for, i.e., the possible poisoning of the bubbler solution in the free troposphere has been neglected in this study. However, we found that, in most cases, the very low raw ozone measurements quickly increased above elevated temperature inversions in winter time, suggesting a limited poisoning effect.
2.4. Data Sets Produced in This Study
Table 2 provides an overview of the different data sets produced for this study, with their specifications and acronyms. We distinguish two categories of data sets: (1) the normal data set and its subsets, which are all based on the standard WMO evaluation procedure, but defined by different CF ranges used as quality criterion, and (2) the alternative data sets including additional corrections aiming at further improvements of the intrinsic measurement quality.
Table 2. Ozonesonde Data Sets Used in This Study
Number of Ascents
Number is depending on the CF used for the selection.
Normal Data Set and Subsets of This Data Set (All Are Based on the Standard WMO Data Evaluation)
CF between 0.9 and 1.4 and sonde reaching 26 hPa (normal data set)
CF between 0.9 and 1.4 and sonde reaching 17 hPa
CF between 1.0 and 1.2 and sonde reaching 17 hPa (reduced higher-quality data set)
Alternative Data Sets for Sensitivity Runs in the Troposphere (With Additional Corrections)
DS-N-Cal + SO2
similar selection criteria as DS-Normal, but with preflight calibration (before 1983: offset set to 3 nBar) applied prior to total ozone normalization (CF) and correction for SO2 interference in the low troposphere
 Within the first category, the selection criteria influencing data quality and number size are presented in Table 2. The range of CF of 0.9–1.4 has been retained as normal selection criteria (DS-Normal) and the range of 1.0–1.2 for a reduced higher-quality data set (DS-Reduced). Note that the higher-quality data set exhibits half the CF decrease compared to the normal data set, but this data set only includes 50% of the ascents (see Table 2).
 In the first alternative data set with additional corrections of Table 2, the preflight laboratory calibrations (slope and intercept) were applied before normalization with total ozone and the SO2 correction completed the procedure (DS-N-Cal + SO2). Data sets with only one of both corrections are named DS-N-Cal, respectively DS-N + SO2. In the alternative data sets without total ozone normalization the CF factors are solely used to reject soundings of poor quality. This leads to data series DS-N-NoCF (corresponding to DS-Normal) or DS-R-NoCF (corresponding to DS-Reduced) of Table 2. The additional SO2 correction is specified by the acronym +SO2.
3. Discussion of Quality of the Reevaluated Payerne Series
 The total ozone normalization factor (CF) provides general information on the data quality of ozonesondes. Figure 1 shows a decrease of the CF between 1967 and 2002, indicating a general improvement of the data quality throughout that period. The obvious breaks coincide with changes in the preflight procedures and in the instruments (see section 2 and Table 1). The break in 1976 is clearly linked to first improvements in the sonde preparation. This strongly reduced the number of soundings with a CF above 1.4. The CF increase between 1980 and 1981 suggests that the air temperature change caused by the change of the styrofoam box (280 K instead of 300 K) was less than assumed. The sharp CF decrease in 1984 corresponds to the change of the laboratory flowmeter and is slightly larger than the expected influence of the airflow measurement change. The CF increase in the early 1990s is due to the problem explained in section 2.3.2. Finally the improved preparation procedure introduced in 1993 further decreased the CFs. Not only the yearly median values show improvements, but also their distributions. The yearly median CF decreased from roughly 1.25 to 1.1 over the whole measurement period and the interquartile range dropped under 0.1 in 1994. The yearly soundings numbers range from 35 (1972) to 167 (1979); nine years have less than 100 soundings (1968–1975, 1990).
 The use of CF to linearly scale the ozone profiles is expected to noticeably correct the stratospheric ozone measurements for changes in time of the linear components of t and Tp in equation (1). The reevaluation of the Payerne ozone time series 1967–2002 also improved its homogeneity (see auxiliary material). Claude et al.  estimated the errors for the BM sonde and summarized the results of past intercomparison campaigns, but mainly for the stratosphere. In JOSIE, all operational types of ozonesondes were compared in a controlled environmental chamber capable of simulating real flight conditions [Smit and Kley, 1998]. The original BM sonde processed with CF correction yielded a precision of about ±10%, a bias of −3% and an accuracy of ±(10–13)% in the tropospheric simulation. SPARC/IOC/GAW  summarized the results of several intercomparisons with other profiling techniques. The Brewer-Mast sonde precision was estimated to be in the range of ±10–20% in the troposphere, but with a bias not larger than ±5%. For altitudes above 28 km, a systematic underestimation with altitude was found (−15% at 30 km), while in the low to middle stratosphere, the total ozone normalization was allowing a precision of ±3%, and systematic biases to other ozone sensing techniques were smaller than ±5%. Crude estimates of the final errors on trend calculations yielded potential drift errors for a 25-year time series of 2.3% per decade at 5 km altitude, 0.7% per decade at 20 km, and 1.2% per decade at 27 km. In the following, we will focus on the troposphere where the ozone measurement quality is most challenging and the use of CF is controversial.
 The two other main European ozonesonde series (Hohenpeissenberg [Köhler, 1995; Steinbrecht et al., 1997] and Uccle [De Backer, 1999; Lemoine and De Backer, 2001]) were used for a homogeneity analysis based on different statistical tests [Favaro et al., 2002]. In the years 1981–1983, they revealed noticeable differences between the tropospheric values of Payerne and the other two stations. Figure 2 compares the tropospheric ozone time series of Payerne and Hohenpeissenberg. At the 850 hPa level, the lower Hohenpeissenberg values can be explained by the early morning launch at that station. At the higher levels, the differences are rather small before 1976 and after 1990, but they are particularly large in 1981–1983 and opposite jumps in the Payerne series delimit this period in the middle and upper troposphere. Payerne and Hohenpeissenberg behave opposite in this period. The technical changes introduced at Payerne in 1981 and 1983 may have been responsible for too low tropospheric ozone measurements (3–5 nbar) in this period. Further similar comparisons pointed out to other periods with remaining uncertainties left after the reevaluation of the Payerne series (see Table 3).
Table 3. Periods of the Payerne Time Series With Main Remaining Uncertainties After the Present Reevaluation
Parameter and Altitude Range
First years–whole period
ozone: in the planetary boundary layer (troposphere)
ozone: in the upper troposphere and tropopause region; meteorological parameters: temperature and geopotential altitude too high in the stratosphere
ozone: in the whole troposphere (values too low)
ozone: residual anomalies in the upper stratospheric levels as well as in the troposphere
 Besides the comparison with the Hohenpeissenberg ozone soundings (BM ozonesonde, standard WMO evaluation), Figure 2 presents two alternative data sets for Payerne: (1) without considering the total ozone normalization and (2) accounting for the preflight calibration before applying this normalization. Both procedures lower the ozone values compared to the standard series. The time series without total ozone normalization (DS-N-NoCF) increases the temporal heterogeneity (see, e.g., the 700–600 hPa levels in the first half of the 1980s). The preflight calibration also increases the differences with the standard Hohenpeissenberg time series, but the time series DS-N-Cal best overlies the measurements of the Zugspitze surface GAW station at 3 km altitude in South Germany near Hohenpeissenberg [Scheel et al., 2003; Oltmans et al., 2006], which is drawn on the 700 hPa panel of Figure 2. This suggests that this time series represents a better choice than the normal time series without CF as alternative to the normal time series with CF for long-term tropospheric trend calculations for the Payerne series. The effect of the SO2 correction is illustrated in Figure 2 on the example of the time series without total ozone normalization whose two versions DS-N-NoCF and DS-N-NoCF + SO2 are represented at the six pressure levels.
4. Ozone Trend Analyses
 This chapter is divided into stratospheric (section 4.1) and tropospheric (section 4.2) trend analyses making use of the different data sets of the reevaluated ozone BM series of Payerne (1967–2002). Comparisons with other monitoring measurements are presented in section 4.3.
4.1. Stratospheric Trend Analysis
 This section is mainly devoted to explore the effect of the reevaluation of the BM ozone series of Payerne and no effort was made to document the effectiveness of the Montreal Protocol [Newchurch et al., 2003; Steinbrecht et al., 2004, 2006]. Long-term ozone trends are usually determined by multiple linear regression models accounting for natural variability with explanatory variables (usually Quasi Biennial Oscillation (QBO) and the 11 year solar cycle) and using an autocorrelated residual error [see, e.g., WMO, 2003; Staehelin et al., 2001]. Here we followed the multiple linear regression approach used by Weiss et al.  that has already been applied to the Payerne series. A stepwise regression is performed independently for selected pressure levels, as well as for annual, seasonal or monthly means. Cp statistics, a measure related to the Alaike's information criterion, is used to determine the statistical significance of the considered terms (trend slopes and explanatory variables) in order to exclude nonsignificant parameters. The atmospheric variables used in this study are presented in Table 4. The tropopause pressure is the most dominant atmospheric variable for the Payerne ozone soundings [Weiss, 2000; see also Steinbrecht et al., 1998a; Hoinka et al., 1996]. As the ozonesondes only slightly outreach 30 km and our analysis ends in 2002, no trend change was introduced in the stratosphere and the start of the single linear ramp was fixed on the 01.01.1970.
Table 4. Choice of Trend Slopes and Atmospheric Variables for the Different Ozone Trend Models
 In a first step we determined simple linear trends in order to test the effect of different samples (i.e., larger sample size but lower quality or smaller sample size but higher quality). Very similar annual and seasonal trends were found for the different CF ranges (DS-Normal, DS-Norm17 and DS-Reduced) provided (auxiliary material, Figure S5). The data set DS-Normal was regarded to be more appropriate for trend analyses than DS-Norm17 and DS-Reduced as its larger number size suggested better meteorological representativity.
 In a second step, we calculated stratospheric ozone trends with the multiple regression model using the data set DS-Normal. Figure 3 shows the comparison of the fitted annual values with the measurements (six top plots with stratospheric levels). The influence of the solar cycle is significant in the four upper levels (15–50 hPa, or 28–20 km), where the model yields a solar influence of an amplitude between 4 and 5 nbar. At the 50 hPa level, the model also shows the effect of the two important volcanic eruptions (El Chichon and Pinatubo). At 70 and 100 hPa (18 and 16 km), the solar cycle is not any more significant, unlike the volcanic eruptions. This analysis highlights the major influence of the solar cycle peak of 2002 on the apparent trend decreasing between 1995 and 2002 at the altitude of the ozone maximum and above up to 15 hPa (28 km). The negative impact of the Pinatubo volcanic eruption on ozone in the low stratosphere between 1991 and 1995 amplified the solar signal as described earlier [Steinbrecht et al., 2004].
Figure 4 shows vertical trend profiles over the period 1970–2002 in comparison with other recent studies. One trend profile applies to the reevaluated normal data set (DS-Normal), another profile is based on the same model, but applied to the alternative data set accounting for the preflight ozone calibration (DS-N-Cal). The results labeled by MCH 2000 refer to results of our previous study [Favaro et al., 2002], which were based on a very similar data set, but ending in 2000. The last profile (WMO 1998 LM) reproduces results of the Ozone Assessment of 1998. This analysis used an earlier version of our data set ending in 1996 that slightly differs from our reevaluated series. The results shown here are those of the Harvard University model [Logan et al., 1999]. Figure 4 also includes the uncertainties at the 95% confidence level (thinner curves). In the stratosphere between 100 and 7 hPa, where a unique trend slope proves to be significant over the entire period from 1970 to 2002, all calculations lead to very similar trends: between −4% and −5% per decade at the 100 hPa level (12 km), a weak relative minimum of −2.5% at 20 hPa (27 km) and an absolute negative minimum of −7% at 7 hPa (33 km). The correction accounting for preflight calibration has a small impact on the trend results at the level of the ozone maximum and below it. The very last years before 2002 do not significantly influence the trend magnitude. Hence the calculated stratospheric trends are robust in relation to the choice of model and data set and their uncertainties are rather small.
4.2. Tropospheric Trend Analysis
 Different to the stratospheric trend analysis, a change in trend was added in our tropospheric model, adopting the “hockey stick” model of Reinsel et al. . The time evolution of the emissions of the ozone precursors in the industrialized world supports such an assumption, as well as the European ozone time series [Oltmans et al., 2006]. Moreover, a large trend change can be seen in the lower levels of the Payerne soundings around 1990 (Figure 2). Considering that the anthropogenic emissions of nitrogen oxides strongly increased in the 1960s in Western Europe and in Switzerland, peaked between 1980 and 1990, and then decreased considerably in the 1990s [Lövblad et al., 2004], our tropospheric trend model has a first trend slope over the whole period 1967–2002 and a second one starting on the 01.01.1990. The tropospheric proxys are presented in Table 4.
 This section is mainly devoted to explore the sensitivity of this tropospheric trend model to the different data sets of Table 2. Extending the first results of the stratospheric analysis, we found that the data selections based on different CF ranges also provide similar trends in the troposphere and detect as well in most seasons equal trend breaks. We compared then the fitted annual values with the measurements provided by data set DS-Normal (Figure 3 (bottom) with tropospheric levels). The model yields significant atmospheric proxys in the upper troposphere, but not at the pressure levels between 400 and 700 hPa; however, the change in the trend slope introduced at the beginning of 1990 is significant, although the resulting trend slope of the 1990s itself is not significantly different from zero. The model again reveals significant atmospheric proxys in the boundary layer (not presented in Figure 3). The positive linear trend simulated over the years 1970–1989 fits rather well the measurements at the 300 hPa level (9 km) if we remind that the measurements were recognized being a little too low in the years 1981–1983 (see section 3). At the 500 and 700 hPa levels (5.5 and 3 km), even if the measurement problem would be corrected for, the fitted linear positive trend rather corresponds to a flat evolution before 1980 and after 1990, with a strong increase in between.
Figures 5a and 5b extend the tropospheric results shown in Figure 3 to the alternative data sets DS-N-Cal + SO2 (Figure 5a) and DS-N-NoCF + SO2 (Figure 5b). In Figures 5a and 5b, the SO2 corrections are added to the ozone values and the corrected ozone is drawn with thin dashed lines. The ozone values that were corrected by the calibration are noticeably lower compared to those provided by the standard WMO evaluation procedure and the SO2 correction has a slight opposite effect at the 500 and 700 hPa levels before 1990 (stronger effect in the boundary layer). At the 300 and 500 hPa levels, the model does not provide evidence for a change in ozone trend in the year 1990 (Figure 5a). The strong stepwise ozone evolution found in the standard time series is noticeably weakened. The ozone increase over the years 1967–2002 roughly amounts to 30–40% of the start values (or to 5–8 nbar compared to start values between 15 and 20 nbar). At the 700 hPa level, the model still reveals a change in ozone trend in 1990, but the slope remains ascending after that year. Instead of a single change in trend, a close look at the corrected time series at 700 hPa would suggest a flat evolution before 1980, then a steep increase until 1995 followed by a period with a flattening course.
 The ozone values without total ozone normalization (Figure 5b) are strongly lowered in the first years characterized by high CF factors. As a result, the trends are significantly different in the whole troposphere. The heterogeneities suspected in the standard series in the 1980s are amplified in that series without CF. This finding corroborates the analysis of Figure 2 (section 3).
 This analysis shows that the choice of the ozone evaluation procedure has a larger influence on trend calculations in the troposphere than the data selection only based on CF range. Furthermore, the sensitivity of trends to these corrections is strongly enhanced in the 1990s compared to the previous decades. The alternative data sets with additional corrections allow better evaluating the uncertainty of the calculated trends. Among the three data sets that are presented in Figures 3 and 5, DS-N-Cal + SO2 (with preflight calibration and SO2 correction) depicts the smoothest time evolution and is most closely related to the surface ozone measurements at the mountain site Zugspitze (Figure 2) and best supports the linear or bilinear hypothesis.
Figures 6 and 7extend the analyses of the different data sets to seasonal trends. The ozone trends over the period 1967–1989 are all strongly positive and statistically significant (Figure 6). The trend magnitudes of the calculated trends greatly differ between the different data sets, particularly for summer, but the confidence intervals of trends strongly overlap. They amount to 20% per decade in winter and spring, and lie between 10 and 20% per decade in summer and autumn. Their statistical uncertainties are varying between ±2 and ±8% per decade at the 95% level. Their vertical profiles generally exhibit a rather flat form. The trends from the first two data sets (DS-Normal and DS-N-Cal + SO2) compare rather well, those from DS-N-NoCF + SO2 are generally more positive. The SO2 correction has a noticeable influence on trends in the winter planetary boundary layer. Because of the SO2 decrease prior to 1990, the corrected ozone values increased less than the uncorrected ones. This effect is hardly visible in the other seasons. The absence of heavy industry around Payerne always kept the summer SO2 concentrations at rather low level. In that season, the model calculations provide slightly larger ozone trends in the planetary boundary layer than in the free troposphere. The enhanced ozone increase in the planetary boundary layer is consistent with the European wide increase in ozone precursors prior to 1990.
Figure 7 (similar to Figure 6) is related to the second period 1990–2002. The most striking feature is the very large statistical uncertainty of the calculation with a change in the trend slope when based on the normal data set. In addition, the trends derived from the three data sets are noticeably different from each other. In the middle and upper troposphere, the alternative data set DS-N-Cal + SO2 supports a unique slope over the full period 1967–2002 in most seasons. This is largely linked to the high laboratory calibration offsets measured in the years 1985–1993. Furthermore, the measurements at the early 1990s are less reliable because of technical problems (see section 2.3.2).
 Winter trends remain positive (10–15% per decade) and unchanged during the whole 1967–2002 period when based on data set DS-N-Cal + SO2. The two alternative data sets provide very similar winter trends in both periods. In the three other seasons, DS-Normal and DS-N-Cal + SO2 provide negative trends in the planetary boundary layer, but only summer trends are statistically significant. On the contrary, data set DS-N-NoCF + SO2 maintains positive (but not significant) trends almost down to ground in spring and autumn. In the middle and upper troposphere, these three seasons are either characterized by an absence of trend (DS-Normal) or by a positive trend (DS-N-Cal + SO2 and DS-N-NoCF + SO2). These two new series do not exhibit such large changes in trend in 1990 as the normal one.
 In order to analyze the sensitivity of the tropospheric trend calculations to all possible corrections of the Payerne soundings, we repeated the trend calculations with different combinations of the corrections presented in Table 2 (preflight ozone calibration only, no CF, SO2 interference in the low troposphere). They confirmed that the selected SO2 correction in the low troposphere only has a noticeable influence on winter trends, in both periods (1967–1989, 1990–2002). We simulated some SO2 poisoning of the ozonesonde with a SO2 correction extending further into the middle troposphere (slower vertical SO2 decay than measured between Payerne and Jungfraujoch). This did not strongly change the trend results in the middle troposphere. On the contrary, considering preflight calibration or neglecting total ozone normalization has a strong influence on trends after 1990. Moreover, latter correction also noticeably influences trends in the first period.
Figure 8 reduces Figure 7 to annual trends in order to point out on some more general features. In the case of the normal data set SD-Normal, the model determines a second trend slope since 1990, which is added to the first one and forms a resultant trend for the period 1990–2002. According to Reinsel et al. , the statistical errors on both trend slopes as well as their covariance have to be combined. In our case, the correlation ranges between −0.7 and −0.8 and the error on the second slope term (1990–2002) is more than twice the error on the first slope term (1967–2002). Consequently, despite the strong negative correlation, the statistical errors on the resultant tropospheric trends over the years 1990–2002 are much larger than the errors on the first period. Furthermore, these trends are not any more statistically different from zero (95% confidence level) although the trend changes in 1990 are significant. As already shown on a seasonal basis, data sets DS-N-Cal + SO2 and DS-N-NoCF + SO2 behave differently than the normal data set and keep the trends positive in the free troposphere after 1990 (diamonds with dotted lines and triangles with dashed lines in Figure 8). These corrected data sets provide trends that noticeably differ from those of the normal data set in the troposphere over the 1990–2002 period. However, in the planetary boundary layer, the different data sets and models exhibit similar negative trends. In the free troposphere, the differences between the first one and the two other data sets increase with altitude. Whereas the trends based on the normal data set oscillate around zero with a very large uncertainty, the trends based on DS-N-Cal + SO2 increase to values around 10% per decade with a smaller statistical uncertainty (no trend change in 1990). Although DS-N-Cal + SO2 and DS-N-NoCF + SO2 provide similar vertical trend profiles, the large uncertainty associated with the latter one makes the trends very uncertain.
 As already shown by the seasonal analysis, the results of the different trend calculations in the troposphere are more coherent over the first period than over the second one. Over the years 1967–1989, the normal time series is characterized by an averaged trend of about 18% per decade between 800 and 300 hPa. DS-N-Cal + SO2 delivers about 4% lower values and DS-N-NoCF + SO2 about 4% higher ones (figure not shown). Over the shorter following period 1990–2002, the differences in trends obtained on the basis of these three data sets are only occasionally smaller than 5% per decade, often lie between 5 and 10% per decade, and can be even larger in the free troposphere. If we associate these trend differences with potential drift errors, we obtain values that are well above the SPARC/IOC/GAW estimate of 2.3% per decade at 5 km altitude for a 25 year time series, partly due to the splitting in two periods.
4.3. Comparison With Other Time Series in the Troposphere
 We performed comparable trend calculations for the Hohenpeissenberg ozone soundings with a data set downloaded from the World Ozone and Ultraviolet Data Centre (WOUDC). For both stations, annual trends in the troposphere based on the standard WMO evaluation procedure compare well within their statistical error in the 1970–1989 period. At the 700 hPa level (free troposphere), the Hohenpeissenberg ozone values show a strong increase between 1970 and 1987, which was maximal in the 1980s (see Figure 2, DWD Ozonbulletin No. 75 and 82 (http://www.dwd.de/de/FundE/Observator/MOHP/hp2/ozon/bulletin.htm), and Oltmans et al. ). This behavior was less pronounced in the Payerne soundings, but the ozone stepping up in the 1980s appeared in both time series, although the period with steep increase started earlier and the maximum values also occurred earlier compared to Payerne. In the second period 1990–2002, the tropospheric ozone evolution is noticeably different in the two data sets. Contrary to Payerne, Hohenpeissenberg provided negative trends in the whole troposphere. In the middle and high troposphere, the Hohenpeissenberg soundings already reached their ozone maximum in the late 1980s, and exhibited later on a slight but nearly continuous decrease. Nevertheless, this comparison shows that both ozone series underwent a rather similar time evolution in the troposphere with a change in trend around 1990 (Figure 2).
 Surface ozone measurements benefit from a higher accuracy standard and a better temporal representativity than ozonesondes and can be used for trend comparison. Such measurements have been available since 1991–1992 at a few Swiss stations within a 100 km distance from Payerne, up to 3580 m asl (Jungfraujoch). At all these stations, the annual mean of ozone concentration increased between 1991 and 1998 and stabilized afterward until 2002 [Lövblad et al., 2004]. Ordóñez et al.  analyzed daily maxima of 11 surface ozone stations sites of the Swiss Plateau for the 1992–2002 period using a multiple regression model for meteorological adjustment. They found significant upward trends for winter ozone maxima, similar to the results of planetary boundary layer of the reevaluated Payerne ozone series (Figure 7). For the other seasons they did not find significant downward trends; however, they analyzed daily ozone maxima and not sonde ascents available only three times per week. Ozone in the planetary boundary layer is influenced on local scales by the emission of NO because of titration and the decreasing NO emissions partially explain the winter trends. Ordóñez et al.  also analyzed Ox (= O3 + NO2) which account for titration, showing that the decreasing NO emissions only partially explain the increasing ozone winter trends, which is consistent with the results of numerical studies of the EMEP model [Jonson et al., 2006].
 At Zugspitze (3 km asl), whose time series is presented in Figure 2, the ozone increase between 1978 and 1990 corresponds to 25% per decade. Ozone values leveled off near 1999, but the annual trend still amounted to 5% per decade over the years 1990 to 2002. It has been mainly explained by an increase of the low ozone values in the winter months, with a similar behavior at Jungfraujoch [Beilke et al., 2002] analogous to the results of the Payerne soundings (normal data set yielding a winter trend of 5% at the respective altitude, Figure 7). However, the Payerne soundings exhibit at 700 hPa an annual trend of −1.5 ± 6% per decade over the years 1990–2002 (Figure 8, normal data set). If the preflight ozone calibration and the SO2 interference are taken into account, the annual trend changes its sign and amounts to +3 ± 5% per decade (Figure 8). The Payerne data sets DS-N-Cal + SO2 and DS-NoCF + SO2 depict a positive annual, but nonsignificant, trend similar to that of Zugspitze. The series with additional corrections confirm the flattening of the ozone trends in the low troposphere after 1990 shown by the ozone surface monitoring stations, but strongly enlarge the uncertainty in the tropospheric trends of the Payerne time series over the period 1990–2002.
 In a recent work, Ordóñez  analyzed ozone trends of three European alpine stations of the period 1992–2002 confirming the significant increase reported earlier for Jungfraujoch [Brönnimann et al., 2002]. This increase is similar in magnitude as observed in Mace Head, a coastal site in Western Ireland [Simmonds et al., 2004] and as derived from measurements from the regular aircraft program MOZAIC (Measurements of Ozone and Water Vapour by Airbus In Service Aircrafts, available since 1994 [Zbinden et al., 2006]). However, these trends are different from those of the free troposphere derived from the Payerne ozone sondes when using the normal data set for a very similar period (see Figures 7 and 8).
 The high alpine site Jungfraujoch is usually exposed to the free troposphere in winter and autumn, while trace gas composition in spring and summer is more often distributed by planetary boundary layer air [Zellweger et al., 2003] and therefore the representativity of Jungfraujoch measurements for free tropospheric ozone trends can be questioned. However, careful comparison with ozone MOZAIC measurements from landings and starts from the airport of Frankfurt (Germany) yielded coherent variability and similar trends (1994–2002) with surface ozone measurements at Jungfraujoch, suggesting good representativity of the free tropospheric ozone trends deduced from high alpine site. The laboratory calibrated ozone sonde trends (Figures 7 and 8) for the period 1990–2002 are in closer agreement with the Jungfraujoch trends, suggesting again that this data set is more realistic for tropospheric ozone than the normal data set.
 Within the framework of the reevaluation of the ozonesonde record of Payerne, several data sets have been produced either on the basis of the standard WMO evaluation procedure or introducing additional corrections in the troposphere where ozonesonde data quality is most challenging. These data sets provide a good basis for trend calculations allowing for a change in trend, as well as for sensitivity analysis.
 In the stratosphere, the standard WMO data evaluation procedure is well accepted and a single linear trend still appears appropriate for the Payerne time series ending in 2002. Between 100 and 10 hPa (16–31 km), the ozone trends since 1970 span a range between −1.5% and −6% per decade depending on the altitude and the season, consistent with earlier studies [e.g., Logan et al., 1999]. The extension of the Payerne time series until 2002 does not basically change the past findings related to the stratospheric ozone decline. The results are also in a good agreement with the long-term trends of the German soundings at Hohenpeissenberg. Solar cycle represents an important explanatory variable to understand the stratospheric ozone time evolution between 1990 and 2002 at the levels between ozone maximum and balloon burst altitude, as well as stratospheric aerosols in lower stratospheric layers.
 For tropospheric ozone trends, the hockey stick model allowing for a change in the magnitude of the trends at the beginning of the year 1990 fits rather well the measurements. The choice of the ozone evaluation procedure proved to have a much larger influence on trend calculations in the troposphere than data selections only based on CF ranges. Compared to the case of a single slope model, the positive ozone trends over the 1967–1989 period became stronger with the hockey stick model. They amounted to values between +10% and over +20% per decade depending on altitude, season and data set. Depending on the data set choice, the model provided significant or nonsignificant changes of the trend slopes in 1990, but the trends in the 1990–2002 period turned out to be generally characterized by a large uncertainty. However, all data sets agreed with winter trends remaining positive in the whole troposphere over this second period. In the other seasons trends were generally negative near the ground and shifted to zero or positive slopes with increasing altitude. The change in trend found in the Payerne soundings also appeared in the Hohenpeissenberg soundings in the late 1980s. However, surface ozone monitoring stations within the Alps maintained a positive ozone trend during the 1990s, with a level off near 2000, similar to the results of the MOZAIC measurements for the altitude of Jungfraujoch.
 The different alternative data sets document the large influence of the possible corrections to the standard WMO evaluation procedure for the Payerne series. The corresponding potential drift error on tropospheric ozone trend calculations amounts to a few percents per decade over the 1967–1989 period, but up to 10% per decade over the shorter next period 1990–2002, partly due to the changes in trend slopes in 1990 allowed by the hockey stick model. On the basis of a comparison with other ozone series and Ordóñez , the corrections for preflight ozone calibration and SO2 interference in the low troposphere seem to provide the most coherent Payerne ozone time series in the troposphere.
 This study was performed within the framework of the Swiss ozone monitoring and research program contributing to the Global Atmosphere Watch (GAW) [Müller and Viatte, 2005]. We acknowledge the work of previous project contributors which has been supported through a special project fund. We are very grateful for the continuous efforts of the persons responsible for the ozone soundings in Switzerland since the end of 1966. We would like to thank all members of the technical staff of Payerne for the careful performance of the prelaunch procedures and ascents with the Brewer-Mast sonde, as well as for the data handling. We are grateful to H. Claude and his colleagues (DWD) for the Hohenpeissenberg ozone soundings, to H. De Baker and his colleagues (IRMB) for the Uccle ozone soundings, as well as to H. Scheel (DLR) for the Zugspitze ozone series. SO2 data sets of Payerne and Jungfraujoch were provided by the Swiss Agency for the environment, forests and landscape. Most data sets were accessed through the WOUDC and WDCGG world data centers. Finally, we acknowledge the three anonymous reviewers' constructive comments to improve the paper.