Description of the long-term ozone data series obtained from different instrumental techniques at a single location: the Observatoire de Haute-Provence (43.9°N, 5.7°E)

A description of the long-term data series of stratospheric ozone at the Observatoire de Haute-Provence is presented. At this station, data sets with temporal length of a decade or more are provided in the framework of the Network for Detection of Stratospheric Change by ground-based experiments: Dobson spectrophotometer (in both column and Umkehr mode), lidar and ozonesondes. The data time series obtained from these various instruments operated simultaneously at a single site and complemented by SAGE II spaceborne measurements are first described with respect to instrumental uncertainties, sampling rate and time evolution. These data series are then compared to each other in terms of sampling rate and average vertical profiles. The difference between the mean ozone profiles of the data sets can partly be explained by the different sampling rate of the instruments. Using the overlap and the complementarity of the various data sets, a preliminary estimate of the long-term evolution of ozone over the last decade over Southern France is conducted. Trend values for both total column and vertical profiles are derived using the multi-regression statistical model AMOUNTS O3. In the 25/40 km altitude range, a similar ozone decrease from −4%/decade to −10%/decade is observed from lidar, Umkehr and SAGE II data series in good agreement with previous estimates. In the lower stratosphere (15/25 km), large negative trends in the ozone vertical profile are observed. In addition, the negative trend of −5.4%/decade in total ozone inferred from the Dobson measurements over the period 1983/1995 is in good agreement with the integrated trend profile.


Introduction
Satellite and ground-based measurements have demonstrated that ozone depletion has been occurring for almost two decades on a global scale [Stolarski et al., 1992;Gleason et al., 1993;World Meteorological Organization (WMO), 1994]. Of particular interest are the trends observed in the northern hemisphere midlatitudes which occur over highly populated areas. Ozone trends inferred from the longest total ozone series available, Dobson measurements at Arosa (47 N) between 1926 and 1996, reach À2:3% AE 0:6%/decade (Staehelin et al., 1998). Analysis of the long-term series of ozone pro®les reveals that the largest decrease is observed in the lower stratosphere. It reaches À8%/ decade at 15 km, À2%/decade at 30 km and À7%/ decade at 40 km for measurements over northern midlatitudes between 1980±1996 (SPARC-IOC, 1998. Though the understanding of the chemical, dynamical and radiative processes leading to those trends has been greatly improved (WMO, 1998), it is still dicult to accurately quantify the natural variability and the ozone trends as a function of altitude, as each experimental technique presents some speci®c instrumental uncertainties (SPARC-IOC, 1998). Only a few comparisons of ground-based data sets measuring at the same location, continuously over a long time period, are available, as experiments rarely operate at the same site. At the Observatoire de Haute-Provence (OHP,43.9 N,5.7 E), several instruments (Dobson, lidar and ozonesondes) have been operating routinely for more than a decade, providing independent measurements of total ozone content and vertical pro®le. The combination of these existing data sets oers an unique opportunity to compare dierent techniques and to validate the concept of a multi-instrument monitoring station as proposed by the Network for Detection of Stratospheric Change (NDSC). In this paper, we present an extensive comparison of the dierent ozone data sets obtained at OHP in the framework of the NDSC. This will allow to assess the reliability of such measurements before they are made available to the full scienti®c community in the near future. Long-term changes will also be investigated as they bring additional information on the complementarities and the capabilities of the dierent instruments operated at OHP to insure a long-term geophysical monitoring free of any instrumental drift. Section 2 is devoted to the description of the observational methodologies used for the ozone measurements at OHP: lidar, ozonesondes, Dobson, Umkehr and the satellite borne instrument SAGE (Stratospheric Aerosol and Gas Experiment) II. All the data sets will be limited to the end of 1995, to take into account the limits of validity of some of them. In Sect. 3, the main characteristics of the database of ozone vertical pro®les available at the station are discussed, based on several intercomparison studies between the dierent data sets. The analysis used to quantify the long-term evolution of the ozone total content and vertical pro®le is presented in Sect. 4. Finally, although the time lengths of some of the databases are still short, preliminary results on trends in the ozone vertical pro®le are presented together with a ®rst quantitative estimate of the impact of the 11-year solar cycle and the volcanic aerosols on the ozone changes.

Instruments and data at OHP
The Observatoire de Haute-Provence is located in southern France (43.9 N, 5.7 E). Measurements have been made there regularly for more than 15 years and have already been widely used for the study of ozone and temperature variability and for various satellite validation (Pelon and MeÂ gie, 1982;Pelon et al., 1986;de La NoeÈ et al., 1987;Pommereau and Goutail, 1988;Attmannspacher et al., 1989;Hauchecorne et al., 1991;Lacoste, 1994;Keckhut et al., 1995;Guirlet, 1997). Since 1991, the station has been part of the Primary Alpine station of the NDSC [Network for Detection of Stratospheric Change, (Kurylo and Solomon, 1990)] and is representative of the northern hemisphere midlatitudes. The coordination with other European complementary stations of the NDSC (intercomparison of experimental techniques, validation of measurements) has been mainly performed in the frame of the EC project ESMOS (European Stratospheric Monitoring Stations) (Simon et al., 1997).

Presentation of the instruments
2.1.1 Dobson and SAOZ instruments. Total ozone measurements have been performed routinely at OHP since September 1983 (Table 1) using a Dobson spectrophotometer as part of the Automated Dobson Network (Komhyr et al., 1989). The method is described in detail in Dobson (1931Dobson ( , 1957. The aerosol eect is estimated to be lower than 1% to 2% in aerosol background conditions (WMO, 1988). The errors due to Mount Pinatubo aerosols do not exceed AE1% (WMO, 1994). The precision of the long-term ozone measurement for annual means is estimated to be AE1% (WMO, 1988). The accuracy is strongly dependent on the quality of the calibration of the instrument and of its operation (WMO, 1988). Several intercomparison campaigns (Arosa, 1990;Hradec Kralove, 1993;Izana, 1994) allowed to produce a high quality long-term total ozone record at OHP (WMO, 1994), and one can thus consider that the Dobson data are reliable measurements for the detection of a long-term evolution of total ozone.
Total ozone measurements are also provided at OHP by the UV-visible SAOZ spectrometer since June 1992 (Pommereau and Goutail, 1988). Due to the short time length of the record, we will use in the frame of this study only the total ozone data provided by the Dobson spectrophotometer.
2.1.2 Ozonesondes. Since the beginning of the ozonesonde measurements at OHP, many instrumental changes have been made. The most important one is the change in the type of sondes in 1991. The Brewer-Mast sonde was used from October 1984 until April 1991 and the operation of ECC sonde started in January 1991 and was continued until now (Table 1). With the shift of type of sondes, the number of pro®les per year increased stepwise in 1991. The altitude range of the measurements is between the ground and about 25 km to 30 km altitude. The vertical resolution is about 200 m. The instrumentation method, the operating procedure and the main errors of the measurement are detailed in previous papers (Beekmann et al., 1994;  1995; Ancellet and Beekmann, 1997;Vialle, 1997). The precision of the Brewer sonde ranges from about AE10 to AE15%, while the ECC sonde has a better precision of about AE5%. Their accuracy is between AE10 and AE20% (Smit and Kley 1998). The precision of the stratospheric measurements is improved by normalizing the sonde pro®le to the total ozone measured by the Dobson spectrometer (WMO, 1982). This normalization reduces the bias between the Brewer-Mast measurements and the ECC measurements at altitude levels lower than 23 km, and intercomparison campaigns showed a negative bias of À5% between ECC and Brewer-Mast sondes if this correction factor is applied (Hilsenrath et al., 1986;Beekmann et al., 1994). This bias is taken into account in the present study.  (1985), Pelon et al. (1986), Godin (1987), Godin et al. (1989). In presence of high aerosol loading, the ozone pro®le retrieved from the Rayleigh signals is locally perturbed at the altitude of the aerosol cloud. Thus, after the Mount Pinatubo eruption in June 1991, the lidar ozone pro®les were available only above the volcanic cloud, so that the minimum altitude of the measurements gradually decreased from 30 km in December 1991 to 20 km in the second part of 1993. There were several changes on the instrumental con®guration between 1986 and 1994 to improve the quality of the measurements. The main one was the implementation of new optical and electronic detecting systems on the ozone lidar experimental set-up at the beginning of 1994. Two so-called``Rayleigh'' pro®les corresponding to the atmospheric elastic scattering provide an ozone pro®le between 18 and 45 km. Another so-called``Raman'' pro®le obtained simultaneously from the ®rst Stokes vibrational Raman scattering of the emitted wavelengths by molecular nitrogen (McGee et al., 1993) provides an ozone pro®le unperturbed by the aerosols, between the tropopause and 25 km. The ®nal ozone pro®le is then computed from the combination of the``Raman'' and the``Rayleigh'' pro®les. This improvement of the lidar system led to a substantial increase of the number of pro®les per year (from about 40 in 1987 to about 100 in 1995).
The vertical resolution of the ozone pro®le is altitude dependent. Since 1994, the vertical resolution is about 1.5 km at 15 km (``Raman'' pro®le). It increases to 0.5 km at 20 km (``Rayleigh'' pro®le), decreasing up to 2 km at 30 km and 4 km at 40 km.
The accuracy of the ozone lidar measurements is estimated to be about AE3% in the case of background aerosol loading (Godin et al., 1989). The precision of the measurement decreases from about AE2% below 20 km to more than AE10% above 40 km (Godin et al., 2000). The improvement of the signal to noise ratio over time may have induced a time dependent bias in the upper part of the pro®le. In terms of trends assessment, and because of the lack of measurements after the Mount Pinatubo eruption due to aerosol contamination, the lidar measurements could not be used below 25 km as the record was interrupted between 1991 and 1993 (Table 1). Above 40 km, the precision of the measurements during the ®rst years of the record may be too low to detect a reliable trend. Therefore, use of lidar measurements for trends assessment over the period 1984±1995 will be restricted to the altitude range 25± 40 km, complementing the ozonesonde records at higher altitude.
2.1.4 Umkehr method. The Umkehr measurements have been made routinely at OHP since September 1983 (Table 1). Ozone partial pressure is obtained in 12 atmospheric adjacent layers, of 5 km width each, between the ground and the 60 km altitude level (GoÈ tz et al., 1934;Mateer and DeLuisi, 1992;SPARC-IOC, 1998). The vertical resolution of the Umkehr measurements is equal to 12.5 km (WMO, 1988). The measurement errors are about 10% in layers 4 through 8 which correspond to the only layers where independent information could be retrieved (WMO, 1988). The partial pressure in each layer is further converted to ozone concentration using the National Centre for Environmental Prediction (NCEP).
The Umkehr measurements are perturbed by stratospheric aerosols which induce a negative bias in the ozone partial pressures derived in the stratosphere. A ®rst estimate of the aerosol eect on the Umkehr measurements according to the optical depth has been derived from about 2000 SAGE II/Umkehr intercomparisons (Newchurch and Cunnold, 1994;Newchurch et al., 1995). We have used in this study aerosolcorrected data provided by M. Newchurch (private communication). Because of the large error of the measurements due to the volcanic cloud after the Mount Pinatubo eruption, no data is available at OHP between September 1991 and June 1993.
Due to this lack of measurements for about two years, the Umkehr measurements may not give very reliable values of the ozone trend at the OHP in the lower stratosphere.
2.1.5 Satellite instruments. SAGE II. The satellite experiment SAGE II was launched on the ERBS satellite and has provided ozone pro®les since October 1984 (Table 1) by measurement in the solar occultation mode. Details on the measurement method are given in previous papers (Chu and McCormick, 1979;WMO, 1988;Chu et al., 1989;SPARC-IOC, 1998). The version of the data used here is version 5.93 which was the last public version of the data available at the time of this study.
The vertical resolution is about 1 km. For the time periods without high stratospheric aerosol load, the statistical error is larger than AE10% for the altitude levels lower than 15 km. It ranges from AE5 to AE10% between 15 and 20 km, is equal to AE5% between 24 and 45 km, and increases above until the upper altitude of the vertical pro®le (Attmanspacher et al., 1989;Cunnold et al., 1989;Newchurch et al., 1995). For the altitudes lower than 12 km, the uncertainties on the ozone measurements become very large (up to 15%) because of the low ozone quantities and the large aerosol correction which is required, due to the aerosol interference in the ozone spectral channel at 600 nm.
No ozone value is available at the altitude levels of the volcanic aerosol layer after the Mount Pinatubo eruption due to the high contamination of the 600 nm channel by the volcanic aerosols (Table 1). Thus, the SAGE II measurements should be useful for the detection of the ozone trend mainly in the upper stratosphere.
In order to compare the ground-based measurements and the satellite measurements, for each available day, the selected SAGE II ozone pro®le corresponds to the closest pro®le to the OHP station, within a radius of 1000 km. The SAGE II data have been chosen for this evaluation because they provide the best vertical resolution comparable with the lidar one. For instance, the SBUV measurements have a much lower vertical resolution than the SAGE II measurements (WMO, 1994;SPARC-IOC, 1998) and are also sensitive to the presence of aerosols (McPeters et al., 1994).
TOMS. The Nimbus 7 TOMS provides total ozone measurements on nearly every day from November 1978 until May 1993 (Stolarski et al., 1991;Gleason et al., 1993). The TOMS measurements are used mainly for validation purposes. We use in this study version 6.0 of the TOMS data. Some previous studies showed that trends derived from version 7 TOMS data (McPeters et al., 1996) are very similar (within AE1 to 2% per decade) to the ones derived from the earlier version 6 data (Stolarski et al., 1991(Stolarski et al., , 1992.

Instrument time series
Due to the dierences between the technical constraints of each experiment, the ozone databases available at OHP are heterogeneous. As previously outlined, the vertical resolution and the uncertainties on the ozone measurements are very dierent according to the type of experiment. The same comment also applies to the sampling rate which shows large variations as a function of time, and according to the type of experiment.
2.2.1 Sampling rate. As the ozone trends have been shown to be seasonally dependent (Stolarski et al., 1991;WMO, 1994;Bojkov and Fioletov, 1995;Chandra et al., 1996;Harris et al., 1997), the sampling rate throughout the year should be as uniform as possible. Indeed, dierences in the long-term sampling between the various instruments operated at OHP could result from the dierent observation conditions required for each instrument. For example, the sampling throughout the year for the Umkehr, Dobson and ozonesonde measurements is rather regular, and the variation of the number of measurements per month does not exceed 35%. On the contrary, there is a larger variation in the monthly number of lidar measurements (Fig. 1). This number is lower in May and in September (for the second part of the year) over the whole period of measurements. This can be related to the frequent cloudy skies observed during these months over OHP, preventing lidar operation, as this requires clear sky conditions.
Similarly, a large variation is also observed in the monthly number of SAGE II measurements with lower values during the summer months, particularly in August (Fig. 1). This is merely due to the variation in the satellite orbit which overpasses the northern midlatitudes more frequently in winter than in summer (Cunnold et al., 1989).
Beyond a given limit of several measurements per week, the increase in the sampling rate of the measurements does not bring increased con®dence for trend estimates, as soon as two successive measurements are not independent with reference to the ozone ®eld variability due to the frequency of geophysical processes sampled at OHP. This degree of dependence between two successive measurements is given by the autocorrelation (see Appendix A). However, such measurements can be averaged to reduce uncertainty in the natural variability, which in turn impacts the con®dence level in the trend assessment. Considering the data sampling available from measurements at OHP and the autocorrelation of two successive measurements, the best compromise between the dependence of successive measurements and the reduction of the uncertainty is obtained by monthly averaging of the data. In this study, monthly means have been considered for all data series.
2.2.2 Altitude dependence of the number of measurements. Due to the evolution of the characteristics of the instruments and to the varying observation conditions as reported previously, the total number of available measurements can also be altitude dependent. This altitude dependence is illustrated by the variation of the percentage ratio between the number of monthly means available for each instrument and the total number of months over the whole period (Fig. 2). This ratio is uniform in all layers and close to 80% for the Umkehr measurements. For the SAGE II and the lidar measurements, the ratio is lower at the altitude levels below 25 km, due to the perturbation in the lower stratosphere by the high aerosol loading after the eruption of Mount Pinatubo. At the altitude levels higher than 40 km, the number of lidar measurements available is also decreasing due to experimental limitations, likewise the number of ozonesonde measurements above 25 km. This ratio is only indicative and should be considered carefully, as even for the higher values close to 70±80%, it does not preclude that large temporal gaps could exist in the data record (see Sect. 2.1).

Long-term data intercomparison
In this section, we will compare the ozone mean values by the dierent instruments. This comparison may highlight possible biases of instrumental origin between the dierent techniques. Such bias is often associated with the instrumental con®guration and thus may change with the successive improvements of the instruments, inducing drift and arti®cial trends. Thus, its quanti®cation would be of prime importance for the assessment of long-term data set continuity. In a second part, the main characteristics of the ozone seasonal variability as recorded by the various instruments will be compared, further to check the consistency of the data sets, and the ability of the dierent techniques to record atmospheric variability on a monthly basis.
Comparisons of instruments at OHP have been made at several occasions in the frame of the NDSC (Godin, 1987;Beekmann et al., 1994;Mc Gee et al., 1994;WMO, 1994;Smit et al., 1998). Although such campaigns are of prime importance for instrument validation, they can eciently be complemented by continuous comparisons on longer time scales between collocated instruments. This is achieved in this paper by comparing the average ozone values obtained from the dierent instruments during the time periods given in Table 1.
The mean dierence between TOMS and Dobson measurements at OHP over the period 1983±1993 is equal to 3:4 AE 2:7 DU. This dierence might be due to the slight dierence of location between TOMS and Dobson measurements.
The dierences between the average vertical ozone pro®les obtained from lidar, SAGE II, ozonesonde and Umkehr measurements (Fig. 3) are not signi®cant at the 1 r level, and one can note the good agreement between Umkehr and the values derived from other instruments, throughout the whole altitude range, despite the coarse resolution of these measurements.
However, dierences can still be observed and related to the operating mode and characteristics of the instruments. Below 25 km altitude, the average pro®le derived from the SAGE II measurements shows higher mean values than the average pro®les of the groundbased experiments. The values of the relative dierence between lidar and SAGE II measurements [(lidar-SAGE II)/SAGE II in %] varies from À21:2% at 15 km to À0:2% at 25 km and remains then roughly constant up to 42 km (Fig. 4), before increasing rapidly to large positive values up to 50 km. Similarly, the values of the relative dierence between ozonesonde and SAGE II measurements [(ozonesonde-SAGE II)/SAGE II in %] extend from À13:6% at 15 km to 1.0% at 25 km. When the dierences between the mean pro®les are calculated only with the measurements in time coincidence (measurement times not dierent from more than 24 h, Fig. 5), one can ®rst observe that the amplitude of the dierences between ozonesonde and SAGE II measurements is greatly reduced between 20 and 30 km, independently of the sign of the bias previously observed. This certainly emphasizes the eect of the Fig. 2. Altitude dependence of the percentage ratio between the number of monthly means available for each instrument and the total number of months over the whole period considered in this study (see Table 1) sampling rate especially for measurements such as the ozonesondes which are performed at a rather low sampling rate. Whereas large positive values of the dierence between lidar and SAGE II measurements are still observed in the upper altitude range, the agreement is better than 1% from 25 to 40 km when only measurements in time coincidence are considered. The maximum relative dierence above 25 km altitude is then equal to À0:7% at 32 km as compared to 6.6% when using the whole data set. However, at the altitude levels lower than 25 km, the dierences between SAGE II and lidar measurements are still largely negative reaching À14:2% at 15 km. And the dierence between SAGE II and ozonesonde measurements reaches even larger values after selection of time coincident data (À18:9% instead of À13:6% at 15 km). This overestimation of the ozone values below 25 km by the SAGE II instrument as compared to ground-based instruments is probably due to an insucient correction of the aerosolinduced eect on the SAGE II derived ozone concen-trations at lower altitudes (Cunnold et al., 1996). The use of the SAGE II version 5.96 data which were not available at the time of this study, might lead to an improvement to this problem (SPARC-IOC, 1998).
The values of the relative dierence between lidar and ozonesonde measurements [(lidar-sonde)/sonde in %] Fig. 4. Altitude dependence of the relative dierence between the mean vertical pro®les of lidar and SAGE II measurements between 10 km and 50 km plotted with the two r standard deviation limits (top); same ®gure for ozonesonde and SAGE II measurements (middle); same ®gure for lidar and ozonesonde measurements (bottom) Fig. 3. Mean vertical ozone pro®les (top) and corresponding standard deviation (1 r) (bottom) between 10 km and 50 km for SAGE II, lidar, ozonosondes and Umkehr measurements at OHP vary from À8:8% at 15.5 km to À1:2% at 20.5 km (Fig. 4). Between 21 and 28 km, the relative dierences of the pro®les can partly be explained by the dierent sampling rate of the instruments, as shown by the smaller values of the relative dierence calculated with only measurements in temporal coincidence (between À3:4% at 28 km and 0.1% at 24 km) (Fig. 5). Concerning the other altitudes, one should keep in mind that there is a low number of lidar measurements available in the altitude range 15±20 km before 1994 and a low number of ozonesonde measurements available in the altitude range 25±30 km for the whole period.
The vertical pro®les of the standard deviation of the mean pro®les (Fig. 3) show similar values for lidar, SAGE II and ozonesonde measurements above 25 km. The standard deviation determined from Umkehr measurements is lower at all altitudes. This could be related to the lower vertical resolution of these measurements, which thus integrate part of the variability. Similarly, the standard deviation derived from lidar measurements is lower below 20 km as compared to SAGE II and ozonesonde measurements, re¯ecting again the lower resolution of lidar measurements in this altitude range (see Sect. 2.1).

Long-term evolution of ozone vertical distribution at OHP
The assessment of trends will be restricted to altitude ranges where the measurements are valid and available without temporal gaps. Therefore, based on the results of Sects. 2.1, 2.2 and 3, the trend calculation will be limited to speci®c time periods and altitude ranges for each instrument. This selection will also ensure that the number of ozone measurements available will be similar for the various instruments. As such, ozonesonde measurements will be used only between 10 and 25 km, lidar data only between 25 and 40 km and SAGE II data only between 22 and 50 km. The Umkehr measurements will be used only in layers 4 to 8 which are the only ones providing independent information (WMO, 1988).
All time series used in this study are restricted to the end of 1995 (Table 1). The length of the data sets is thus short compared to the time scales of some of the processes that drive the variability of the ozone layer, such as the occurrence of volcanic eruptions or the 11year solar cycle. Nevertheless, the complementarity of the dierent experiments providing measurements in the same place over the same period of time could lead to preliminary conclusions on the long-term evolution of the ozone vertical distribution at OHP.

Method of analysis
The method of analysis is based on a multiparameter least-squared ®t of the monthly means to extract the trend from other variations of the ozone vertical distribution (Press et al., 1989). As the ozone data sets used for this study are not exactly of the same time length, it is necessary to remove the main components of the interannual variability for each of the data sets, in order to compare and analyse the trends inferred from each of them. The regression model has been developed and adapted from the so-called AMOUNTS temperature trend model of Hauchecorne and Keckhut (Hauchecorne et al., 1991;Keckhut et al., 1995Keckhut et al., , 1999. The statistical parameters associated with this ®tting method  Fig. 4, with data sets being restricted to only measurements in temporal coincidence are described in Appendix A. The components of the ozone variability taken into account include the seasonal variations of ozone concentrations, and the response to the quasi-biennial oscillation (QBO), to the 11-year solar cycle, and to the stratospheric aerosol forcing. The seasonal cycles are parameterised using a constant term and a sine and cosine annual and semi-annual functions. The ozone concentration at time t and altitude z is represented by: where: a A 1 z; . . . ; a T z are the coecients computed by the model for the dierent forcings; c A 1 t; . . . ; c T t describe the temporal evolution of the forcings; O H 3 t; z is the deviation term from the best ®t.
The quasi-biennial oscillation (QBO) in¯uences the circulation at mid-latitudes and consequently the ozone distribution in those regions (Randel and Wu, 1996;Sitnov, 1996). The con®dence interval of the further derived trend component decreases when the QBO signal is removed from the time series, according to previous results (Keating et al., 1994a). Thus, to infer ozone trends as accurately as possible, the QBO signal is removed in all of the following studies. However, the amplitude of this ozone response to the QBO has been shown to be rather low (Logan, 1994), and our analyses also show that this contribution to the long-term evolution of the ozone at OHP is generally not significant. The QBO proxy used in this study corresponds to the wind velocities at 45 hPa measured by radiosoundings at Singapore (1.22 N,103.55 E): and provided by B. Naujokat (Meteorologisches Institut, Frei UniversitaÈ t Berlin, Germany). The solar induced variability comes from variations of the solar activity (Hood et al., 1993;Keating et al., 1994b;Hood, 1997) especially at wavelengths lower than 200 nm, for which the solar¯ux plays a dominant role in the dissociation process of molecular oxygen. The solar index used in this study is the He I index, which corresponds to the variation in the linewidth of the He I absorption transition at 1083 nm: Data are available from observations at the Kitt Peak Observatory (Arizona, United States) over the time period of ozone measurements at OHP and are provided by the Solar Index Data Base from the Observatoire de Meudon (France). The largely enhanced aerosol loading which occurred in the stratosphere following the eruption of Mount Pinatubo in June 1991 increased the surface areas available for heterogeneous chemical reactions leading to increased destruction of ozone below 25 km (Tie et al., 1994;Solomon et al., 1996Solomon et al., , 1998. This eect has been estimated to correspond, at northern mid-latitudes, to a decline ranging between 3 and 12% for the total ozone, (WMO, 1991) and about 15% in the 8±24 km altitude range (Angell, 1998).
To take into account the chemical depletion in ozone concentration in the lower stratosphere, and following the approach of Portmann et al. (1996), a proxy index is considered which corresponds to the product: where St; z is the aerosol surface area density as calculated by Jackman et al. (1996) and Thomason and Poole (1997) from the SAGE measurements for the 40± 50 N latitude band; and DClt is the variation of the stratospheric chlorine loading which is considered as having increased linearly by 36% between October 1984and December 1995(WMO, 1994Solomon et al., 1996). As pointed out by Solomon et al. (1996), it is worth noting that a possible confusion can be made between the aerosol loading and the 11-year solar cycle signals on ozone changes. Both peak in 1991, with recovery (volcanic aerosols) and minimum (solar activity) occurring simultaneously in 1995 which corresponds to the end of the time period of this study. However, these two processes do not aect the stratospheric ozone concentration in the same altitude ranges. Therefore, the study of the long-term evolution of the ozone vertical distribution as performed here could allow to dierentiate between the solar cycle and aerosol loading induced ozone changes. The model also includes a trend function which is linear as a function of time to account for the long-term ozone depletion.
The model will be referred hereafter as the AMOUNTS O 3 statistical model. A validation exercise has been completed by comparing the trends estimated by this statistical model on a reference time series, namely the TOMS total ozone measurements for the 40± 45 N latitude band between 01/1979 and 05/1991, with those calculated using validated statistical models (WMO, 1994). For the seasonal trends and the yearround values, there is a very good agreement between the values retrieved using the AMOUNTS O 3 model with those published in WMO (1994), with a maximum dierence of 0.4%/decade.

Seasonal variations of the ozone vertical distribution
The variation of ozone at mid-latitudes is controlled by both chemical and physical processes, leading to annual and semi-annual altitude dependent cycles (Brasseur and Solomon, 1986;WMO, 1994;Guirlet, 1997). The main objective of this section is to assess the ability of the various instruments to monitor the atmospheric interannual variability, before investigating the in¯uence of other processes on the long-term evolution of ozone. The de®nition of the amplitude and phase of the annual and semi-annual variations is given in Appendix A.
Concerning the amplitude of the seasonal variation for both annual and semi-annual cycles, there is a good agreement between the results obtained from the dierent instruments (Fig. 6). For all of them, the amplitude of the annual variations is maximum at 20 km (between 8:1 Â 10 11 mol Á cm À3 for the lidar and 9:9 Â 10 11 mol Á cm À3 for SAGE II; 8.1 DU for Umkehr) and minimum at 25 km (between 4:0 Â 10 10 mol Á cm À3 for the ozonesondes and 1:5 Â 10 11 mol Á cm À3 for the lidar; 2.1 DU for Umkehr). A secondary maximum is derived at 30 km from SAGE II (4:5 Â 10 11 mol Á cm À3 ), lidar (5:8 Â 10 11 mol Á cm À3 ) and Umkehr (5.0 DU) measurements. At altitudes below 20 km, the amplitude of the annual variation as derived from lidar, SAGE II and ozonesondes decreases to similar values at 15 km altitude (5:2 Â 10 11 mol Á cm À3 , 7:2 Â 10 11 mol Á cm À3 , 6:4 Â 10 11 mol Á cm À3 respectively). The maximum observed at altitudes below 15 km from the ozonesonde record could be related to the higher variability of the ozone vertical distribution at the tropopause level due to stratosphere±troposphere exchange processes and the variation of tropopause altitude (see also the phase of this variation, Fig. 7). For all measurements, the amplitude of the semi-annual variation is much lower than the annual cycle and decreases with altitude above 15±18 km. All four instruments show a very low level of variability in the ozone pro®le at 25 km which corresponds to the upper part of the ozone maximum.
As for their amplitudes, the values of the phase of the seasonal variations show a good coherence between the dierent data sets (Fig. 7). For the four data sets, the amplitude of the annual variation is maximum between February and April at the altitude levels lower than 25 km, and between June and August at the altitude levels higher than 25 km. One should note that at altitude levels higher than 30 km, the value of the phase of the annual variation of ozone concentration is similar to the one of temperature as observed from lidar measurements at OHP (Hauchecorne et al., 1991). The amplitude of the semi-annual variation is maximum in spring and in autumn for the altitude levels lower than 20 km, and in winter and in summer for the altitude levels higher than 25 km.
These main characteristics of the ozone seasonal variations are in agreement with 2D-model calculations (Brasseur et al., 1990) of ozone variations at northern hemisphere mid-latitudes (Guirlet, 1997). As such, the very obvious change in the amplitude and phase observed at the 25 km altitude level from the four measurement techniques is consistent with the change in the processes that control ozone vertical distribution. These include dynamical control for altitudes lower than the altitude level of the maximum ozone concentration, and chemical control above 30 km.
In conclusion, there is a rather good agreement between the results of the various experimental data sets showing their coherence in retrieving the seasonal variations throughout the whole period of time considered in this study. This coherence will allow us to subtract the seasonal variations for all experiments with a high degree of con®dence, before calculating the in¯uence of other processes in the next sections.

Quanti®cation of the long-term evolution of stratospheric ozone at OHP
In this section, a ®rst estimate of the long-term evolution is given for both total ozone and ozone vertical Fig. 6. Altitude dependence of the amplitudes of the annual (top row) and semi-annual (bottom row) ozone variations as derived from SAGE II, lidar, ozonosonde (left column) and Umkehr measurements (right column). For Umkehr measurements, the altitude of reference is the centre altitude of the Umkehr layer distribution. This calculation takes into account the seasonal variations, the QBO and the 11-year solar cycle. The eect of the aerosols will be discussed afterwards with reference to ozonesonde measurements. All uncertainties reported hereafter correspond to 1 r standard deviation.
4.3.1 Trends in total ozone. The trends in total ozone are calculated using the Dobson measurements at OHP for both seasonal values (with winter extending from the 1st of December to the 1st of March) and year-round values.
The annual trend at OHP calculated using the AMOUNTS O 3 model is À5:4% AE 1:7% per decade over the period 1983±1995 (Fig. 8). It has to be compared with the trend value of À4:4% AE 1:15% per decade reported in (WMO, 1998) for the same station over the period 1983±1997 which includes the recovery period after the Mount Pinatubo eruption. According to the WMO (1998), the dierence in the northern hemisphere (35 N, 60 N) between trends calculated until the end of 1995 and end of 1997 is of the order of 1%, giving thus con®dence in the total ozone trend values derived at OHP.
As for most of the stations in the northern hemisphere mid-latitudes, the trends are more negative in winter and spring (Fig. 8). Indeed no signi®cant trends are found in summer and autumn at the 1 r con®dence level (WMO, 1998). The seasonal trend values obtained at OHP over the periods 1983±1995 and 1983±1997 are shown on Fig. 8 and compared to the trends derived from TOMS measurements from 1979 and 1997 and averaged over the latitude band 40±45 N (WMO, 1998). The seasonal eect is much more marked for the OHP data. Similarly increased values for the springtime trends are also observed for the period between 1979 and 1997 in nearby European stations such as Hohenpeissenberg [À7:2% AE 1:55% per decade (WMO, 1998)] and Vigna di Valle [À7:6% AE 1:8% per decade (WMO, 1998)]. Though the starting date of the series strongly in¯uences the trend value, the dierence between the trend values could also partly be due to a longitude dependent eect already shown by several authors (Stolarski et al., 1992;Hood and Za, 1995;Chandra Fig. 8. Ozone trends derived from Dobson measurements at OHP between September 1983and December 1995and between September 1983and December 1997(WMO, 1998; ozone trends derived from TOMS measurements between January 1979and December 1997(WMO, 1998. Trends are given in % per decade with a 1 r uncertainty limit for the four seasons winter, spring, summer, autumn (W , Sp, Su, A) and for the whole year (Y ). The seasonal variations, the QBO and the 11-year solar cycle have been removed  Bojkov et al., 1998). Further analysis would certainly be required to con®rm this eect.
4.3.2 Trends in stratospheric ozone vertical distribution. The reduction in total ozone is associated with a negative trend of the ozone pro®le in the stratosphere, as revealed by the analyses of the ozone pro®le measurements (Fig. 9). The SAGE II derived altitude dependent trend decreases rather regularly as a function of altitude above 25 km up to 40 km where it reaches À8:5% AE 1:9%/decade over the period 1984±1995. In the same altitude range and taking into account the coarse vertical resolution of the Umkehr measurements, the Umkehr derived trends in layers 6 to 8 are in rather good agreement with the SAGE II derived values. No signi®cant trend is observed from Umkehr measurements in layers 4 and 5.
Although the trend values derived from lidar measurements are more negative than the SAGE II values between 28 and 34 km, the dierences are not signi®cant at the 1 r con®dence level. Indeed, there is a rather good agreement between the altitude dependent trends derived from these two instruments at lower (between 25 and 28 km) and upper altitudes (up to 38 km). At the altitudes higher than 38 km, the dierences between the trend values inferred from lidar and from SAGE II measurements may have the same origin as the bias between lidar and SAGE II measurements ( Fig. 5; Sect. 3), con®rming that trend estimates inferred from lidar measurements at those altitudes are not signi®cant.
This picture of altitude dependent trends as observed at OHP is roughly consistent with what could be expected based on previous studies using longer term records in the northern hemisphere mid-latitudes (SPARC-IOC, 1998). The upper altitude trends (35± 40 km) are linked to ozone destruction by gas phase chlorine catalytical cycles and the observed values at OHP from SAGE II and Umkehr measurements are in agreement with model calculations (WMO, 1994). In the intermediate altitude levels between 25 and 35 km, signi®cant negative trends are also observed from other data records (SPARC-IOC, 1998), but the absolute values inferred from measurements at OHP seem rather large and do not agree with present model calculations. This might be related to an insucient length of the data record, especially for the lidar measurements.
In the lower stratosphere, where no results other than those derived from ozonesonde measurements are available, the trends are largely negative (Fig. 9). The maximum value of the absolute amplitude is obtained at the 17 km altitude level (À19:9% AE 7:8%/decade over the period 1984±1995). Large negative values are computed up to 25 km altitude from the ozonesonde measurements. Those values are larger than those calculated from simulation models which include heterogeneous processes on sulfate aerosols and transport from higher latitudes where larger ozone losses are observed (WMO, 1998). At this stage of the analysis, we can not dismiss the facts that: (1) the changes in the operational procedure may have altered the long-term continuity of the measurements; (2) the large decrease observed after the Mount Pinatubo eruption in this altitude range might have also impacted on the trend values calculated until only 1995. Nevertheless, it is worth noting that the integration of the vertical pro®le of ozone trends derived from ozonesonde measurements between 10 and 25 km and of the vertical pro®le of ozone trends derived from SAGE II measurements between 25 and 50 km, leads to a decrease of À6:4%/decade, which is in reasonable agreement with the value derived from year-round Dobson measurements (À5:4%/decade) over the same period. As the upper altitude trends inferred from the SAGE II measurements are in agreement with model calculations, there might be a slight overestimation of the amplitude of trends inferred from ozonesondes as compared to the trends inferred from Dobson measurements, taking into account the fact that, as previously discussed, the ozonesondes may not provide ideally reliable measurements in the lower stratosphere. Nevertheless, because of the agreement between trends in the upper stratosphere and the model calculations, the large negative values inferred from both ozonesonde measurements in the lower stratosphere and from Dobson measurements in winter and spring are likely to result from a time evolution of the ozone pro®le rather than from an instrumental eect only. 4.3.3 In¯uence of the 11-year solar cycle and of the volcanic aerosol loading. Considering the altitude dependence of the 11-year solar cycle contribution, the values as derived from the SAGE II measurements show a positive eect (from solar minimum to solar maximum) between 35 and 45 km, reaching a maximum of 4.4% per solar cycle at 41 km (Fig. 10). This is in agreement with previously reported results, as well as low negative values observed around 30 km (Miller et al., 1996). The trend estimates presented up to this point were calculated without taking into account the eects of stratospheric aerosols. Including those could be made by using the methodology described in Sect. 4.1. The vertical pro®les of trends derived from ozonesonde measurements with and without an aerosol term are presented on Fig. 11. This comparison shows that taking into account the volcanic aerosol eect leads to a reduction of the amplitude of the trend, with a maximum eect of 4%/decade between 15 and 18 km.

Concluding remarks
The present comparative studies have shown that the lidar is the only ground-based instrument that can provide reliable measurements of the ozone vertical pro®le in the whole altitude range between 15 and 40 km. However, this is achieved only with up-to-date systems which include all technological improvements developed over the last decade. These include the reduction in signal induced noise at higher altitudes by the use of a mechanical chopper, the adaptation to the high signal dynamics by the use of several optical receiving channels, and the due account of aerosol interferences in the lower stratosphere by using the DIAL Raman scattering technique as recommended by the NDSC protocol. These improvements have been made gradually between 1986 and 1994 at OHP, which explains why the present data record cannot be considered with a high level of con®dence throughout the whole stratosphere. Indeed, the statistical analysis of the long-term lidar data set demonstrates that reliable trends can only be derived in a restricted altitude range from 25 to 38 km. Based on a simple analysis which takes into account the observed variability in the ozone ®eld and present instrumental uncertainties (see Appendix B), one can calculate the earliest dates at which reliable trends will be derived at OHP as a function of altitude ( Table 2). The results show that reliable trend estimates could be made within the next few years and thus before the turn over of stratospheric chlorine loading which is due to occur in the beginning of this century (WMO, 1998).
With those comparative studies, we also demonstrated the requirement for continuous long-term operation, over at least two or three decades, with appropriate continuous instrument evaluation at OHP and any other  (1) is the same as in Fig. 9, i.e. without considering the aerosol eect; curve (2) corresponds to estimated trends when removing the eect of aerosol forcing. Dierence between curve (2) and curve (1) as a function of altitude (right panel ) Table 2. Requirement on the length of a given time series for detection of a signi®cant trend at the 95% con®dence level from lidar measurements at OHP. n 0 is the timelength in months for each altitude level. The last column gives the expected year of achievement considering the starting date of the measurement series at OHP (in brackets) Altitude r r (%) t(z) (%) Dt  1998 (1986) NDSC station. This was part of the objectives of the Network for Detection of Stratospheric Change implemented when mid-latitude trends were ®rst observed 10 years ago, with the aim to produce high quality longterm data sets for ozone assessment. According to the present study, this must be based on the complementarity and the overlap of the data sets, as they appear to be needed highly for a reliable monitoring of long-term evolution of ozone.