S P A R C Stratospheric Processes And their Role in Climate A project of the World Climate Research Programme Home Initiatives Organisation Publications Meetings Acronyms and Abbreviations Useful Links

3.2. Description of data sets

Chapter 1 of this report describes the instruments used for profile measurements in detail. The data from several instruments are inter-compared in Chapter 2. In this section we give a brief description of the data sets used, emphasising those issues most important to the trend analyses. The four data sets used are ozone sondes, Umkehr, SAGE, and SBUV. TOMS data is used to check for consistency of the altitude profile of trends with those in total column ozone.

The SPARC/IOC committee re-evaluated the data at several stations as shown in Table 3.1. Data for the Canadian stations are those available from the World Ozone and Ultraviolet Radiation Data Center (WOUDC) in July 1997. This activity did not include re-evaluation of the Canadian data. There are concerns about the quality of the Brewer Mast data for these stations, as discussed in Chapter 1. Most of the sonde data used here for trend analysis are scaled to ozone column measurements on the Bass-Paur scale (Bass and Paur, 1985; Paur and Bass, 1985) by use of a correction factor. The Canadian Brewer-Mast sonde data are multiplied by 0.9743 to put them on the Bass-Paur scale. The Lauder data has no correction factor applied for total ozone. The data for Wallops Island from the WOUDC were reprocessed to scale each profile to re-evaluated ozone column data (Oltmans et al., 1997). The data for Payerne are undergoing further analysis, and an interim data set was provided for the analyses here.

The sonde data were provided in a variety of formats. To facilitate trend analysis, these were processed into a common format. This format gives the column of ozone in Dobson units for 33 equally spaced layers of log pressure ranging from 1000 to 6.3 hPa (30 layers up to 10 hPa). The vertical resolution, ~1 km, was chosen to be similar to that of SAGE data. The WOUDC data and the data reanalysed for this report were put into this format.

Two groups conducted trend analyses of the sonde data for this report, Logan and Megretskaia from Harvard University, and Tiao, Choi, and Zhang from the University of Chicago. These groups treated the sonde data differently using selection criteria which depended on the correction factor, CF (the factor scaling the sonde data to the total ozone column). The Harvard group used the same criteria for the CFs as Logan (1994), 0.9-1.35 for BM sondes except for Hohenpeissenberg, where 0.9-1.2 was used, and 0.8-1.2 for ECC and KC sondes. They analysed the trends for 33 layers and for 11 layers obtained by summing 3 consecutive layers; only the latter results are shown here. The Chicago group used the CF criteria of 0.9-1.2 for BM and 0.9-1.15 for ECC and KC sondes as in their earlier work (Miller et al., 1995). They required a total ozone column reading for the day of the sounding, and a balloon burst occurring above 16 hPa. They aggregate the 33 layers into 15 Umkehr layers that they have used in their previous analyses. Furthermore they removed the scaling to the total ozone column from the sonde data by dividing each profile by the correction factor. If there is a trend in the CF this can result in different trends being derived for the ozone profile (e.g., Logan, 1985, 1994; Miller et al., 1995). The fractions of soundings that met various criteria for each method (BM and ECC) are given in Table 3.2.

The stricter requirements used by the Chicago group result in a significantly larger fraction of the soundings being omitted, particularly for the BM stations, as shown in Table 3.2. The second column gives the fraction of soundings retained in the Harvard analysis, and the last column the fraction retained in the Chicago analysis. Hohenpeissenberg is the exception, since both groups use the same CF criteria, 0.9-1.2. In the worst cases, the Canadian BM data and Payerne, only 25-44% of the soundings meet the Chicago group’s criteria compared to 70-90% that meet the Harvard group’s criteria. For all of the BM stations, a larger fraction of soundings are omitted from the earlier part of the record than from the later part. This results from the downward trend in the correction factors (see Chapter 1). For the Canadian ECC soundings, Wallops Island, and the Japanese stations, the Chicago group retain about 45-65% of the soundings, while the Harvard group retain about 85-95%. Trends in the correction factor using the Chicago data selection criteria are given in Table 3.3. There are small but significant trends in the CF at several stations, mostly in the range -1 to -4%/decade. Trends were computed for 1970 (or the beginning of the record if after 1970) to 1996 by both groups and for 1980-96 by the Harvard group.

There is a significant difference in tropospheric ozone values measured by BM and ECC sondes that must be accounted for when deriving trends for the Canadian stations. Intercomparisons in the 1970s and early 1980s showed that ECC sondes measured more ozone in the troposphere than BM sondes by about 15 to 20% (with a range of 7-38%) (Logan, 1994), as discussed in Chapter 1. Tiao et al. (1986) used an intervention term in their statistical trend model at the time when BM sondes were replaced with ECC sondes, and this approach was adopted here by both groups. The magnitude of the intervention term is similar to the differences between EEC and BM sondes found in the intercomparisons, and varies among stations (Tiao et al., 1986). For trends starting in 1980 (September 1980 for Goose Bay), the Canadian data are obtained exclusively with ECC sondes, so no intervention term is necessary.

a) Brewer Mast data

b) ECC type data

The models used for trend analysis are described in the Appendix (section 3.9). The Harvard and Chicago models are similar in that they fit 12 monthly means and 4 seasonal trends, and allow for the dependence of ozone on the QBO and solar flux. The major differences between the models are the inclusion of autoregression, the assumption of zero trend before 1970 in the Chicago model and the fact that the Chicago group remove outliers from the model fit. Eight of the stations have data before 1970 (Table 3.1). Vertical profiles of trends are best compared as relative trends (e.g., percent per decade), since each group used different layers; unfortunately percentage trends were computed relative to a different reference. The Harvard trends are given relative to the mean of the time series for which the trend is calculated. The Chicago trends relative to the seasonal intercept in 1970 (or beginning of series if later) adjusted for solar effects and intervention if used. Column trends in the troposphere (1000-250 hPa) and stratosphere (250-16 hPa) are shown, so that the results can be compared in the same units (DU) for the same columns.

The Dobson/Umkehr data used in this study resulted from a concerted effort by the World Ozone and Ultraviolet Radiation Data Center (WOUDC) to insure that all Umkehr data were in the database, were on the Bass-Paur scale, and that the most current calibrations were used. A significant number of previously unavailable Dobson/Umkehr observations entered the database as a result of this effort and were made available for this report. Because of limited availability, no Brewer/Umkehr data were used in this analysis. However, Umkehr records from the Brewer network would provide a significant enhancement to the Dobson/Umkehr records and are highly desirable for future trend analyses.

The Umkehr data used in this analysis were current in the WOUDC in October 1997 where they were inverted from N-values using the Umkehr[92] inversion algorithm (Mateer and DeLuisi, 1992) employing the uniform S_x error covariance matrix. Previously, some Umkehr records had been inverted at individual stations (not at WOUDC) introducing some uncertainty in the uniformity of the inversion algorithm. All the Umkehr records in this report resulted from a single inversion algorithm with uniform characteristics.

The Chapter 2 regression analysis of the time series of SAGE-Umkehr differences for 15 stations indicated that not all of those records were reliable for ozone trend estimation. Difficulties with record length, missing data periods, and level offsets reduced the number of stations used for trend analysis to two Southern Hemisphere stations, Perth and Lauder; and six northern hemisphere stations, New Delhi, Cairo, Tateno, Boulder, Belsk, and Haute Provence. The Arosa data were not available from WOUDC at the time of this study and, therefore, were not included in the Chapter 2 analysis. However, one of the research groups doing trend analyses for this Chapter used data obtained directly from the Arosa station for analysis.

In concert with Chapter 2 recommendations, we analysed Umkehr trends in layers 1+2+3+4 (termed layer 4^-), individual layers 4 through 8, and layers 8+9+10 (termed layer 8⁺). We also report trends in total ozone for the selected stations. All analysed Dobson/Umkehr records were corrected for aerosol effects either before the trend analysis using the correction factors of Mateer and DeLuisi (1992) or with an explanatory variable in the statistical trend model. In both approaches, periods of very high stratospheric aerosol loading (~1 year) following the eruptions of El Chichon (April 1982) and Mt Pinatubo (June 1991) were omitted. While several Dobson/Umkehr stations report observations prior to 1977, all analyses in this chapter consider either the period 1979-1996 (Newchurch and Yang) or 1977-1996 (Reinsel) for trend evaluation.

SAGE I ozone measurements were almost all made at sunset. Stratospheric aerosol loading in 1979-1981 was exceptionally low and therefore the impact of aerosols on the ozone measurements is expected to have been small or non-existent. An important uncertainty in the SAGE I measurement concerns the reference altitude for each profile. The ad hoc correction for this error (Wang et al., 1996) is an attempt to remove this systematic error source above an altitude of approximately 20 km. Ozone trends will be reported separately for the SAGE I/II time series and for the SAGE II time series alone.

The SAGE II data used in this report are from retrieval version 5.96 that was released in February 1997. There was no comparable recent re-retrieval of the SAGE I data. SAGE II v5.96 possesses a non-physical separation of sunrise and sunset ozone values of approximately 10% above 45 km altitude. This separation is larger than in v5.93. Chapter 2 studies indicate that this separation is relatively systematic and should not affect the ozone trend results. Removal of aerosol influences from the ozone retrievals has been improved in v5.96 but SAGE ozone values below an altitude of approximately 22 km were still contaminated for approximately 2 years after the Pinatubo eruption. The SAGE II error bars reflect this contamination and only SAGE data with less than 12% error bars are used in the trend analyses. At higher altitudes, the SAGE II data set is more complete and extends from October 1984 to December 1996. All SAGE ozone trends are reported on altitude levels (see Chapter 2).

The TOMS total ozone record from the Nimbus 7 instrument extended from November 1979 until May 1993. The Meteor 3 instrument was launched in 1991 and lasted until the end of 1994. These data were combined into a single time series following McPeters et al., 1996. There is then a gap in the TOMS record until the Earth Probe (EP) instrument that was launched in August, 1996. A final calibration has yet to be established for the EP TOMS data. Therefore total ozone trends are calculated from November 1978 through October 1994.

The Version 6.0 NIMBUS-7/SBUV data and the Version 6.1.2 NOAA-11/SBUV2 data are used in this report. Both data sets are processed with the Version 6 BUV algorithm described in Chapter 1 and Bhartia et al. (1996). The instruments make a series of measurements approximately every 2 degrees of latitude over the sunlit portions of 14 orbits separated by 27 degrees of longitude. These are used in a profile retrieval algorithm and reported as ozone profiles in DU for 12 Umkehr layers (~5 km thick each) although, as noted in Chapters 1 and 2, the actual vertical resolution is poorer, particularly in the lower stratosphere.

The SBUV data are available for November 1978 to May 1990. The current Nimbus 7 SBUV data include an updated calibration to correct for time-dependent instrument changes. Bhartia et al. (1996) have estimated the drift errors in the SBUV profile to be ± 5%/decade at 1 hPa and ± 2%/decade at 10 hPa. The data after February 1987 are affected by an out-of-synchronisation condition and have been corrected by a "scene-stabilisation" method described in Gleason and McPeters (1995). As recommended in Chapter 2, SBUV data for 1990 are not used in either the SBUV to SBUV2 adjustment or the trend calculations.

The SBUV2 data are available for January 1989 to October 1994 with decreasing coverage of the southern hemisphere over the instrument lifetime due to precession of the NOAA-11 equator crossing times. The changing orbit results in a loss of coverage as far north as 20° S in July of 1994. The NOAA-11 SBUV2 data have recently been reprocessed using updated calibrations and instrument behaviour characterisations and an algorithm change to correct for grating position errors in the latter part of the record to produce the Version 6.1.2 data set. The NOAA 11 absolute calibration was adjusted to match the Shuttle SBUV (SSBUV) calibration (Hilsenrath et al., 1995), while the time-dependent calibration was maintained using an on-board calibration lamp system and verified though comparisons with SSBUV measurements (Ahmad et al., 1994). However, known errors in the data set remain. The Version 6.1.2 algorithm is summarised in Chapters 1 and 2.

The NIMBUS 7 version 6 SBUV data through December 1989 and the NOAA-11 version 6.1.2 SBUV2 data from January 1989 through October 1994 are used to create a combined data set. The 1990 N7 SBUV data are disregarded. The SBUV and SBUV2 data are joined by adjusting the SBUV data by the average difference between the SBUV and SBUV2 data over 1989. These empirical adjustment factors are functions of both latitude and layer and range from -15% to 12% for the layer from 32 hPa to the ground (layers 4-0) with smaller adjustments in the upper layers. The adjustments in layer 8 (2-4 hPa), for example, range from 0.6 to 3.8%. The adjusted SBUV and SBUV2 data are then averaged during the overlap period to create a consistent time series. The SBUV2 daily average, 5° zonal means are filtered by average solar zenith angle and average latitude to eliminate data taken at extreme angles as a result of the NOAA-11 drifting orbit. On any given day, if the average solar zenith angle for a latitude zone is greater than 80° or the average latitude of the measurements in the bin is greater than 1° off the centre latitude of the bin, the daily average value for that zone is ignored. A 5° zonal average requires ~70% of the measurements to be present, and both 5° averages are required to create a daily 10° zonal mean. Ten daily, zonal average values are required to create the monthly average value for that bin. These filters are consistent with Chapter 2 recommendations. Only "Error Code 0" data (Fleig et al., 1990) have been used. These are data from the ascending part of the orbit for which no error flags occurred. The data have also been screened for measurement contamination from volcanic aerosols (Torres and Bhartia, 1995) by deleting all data equatorward of 40^o for one year following the April 1982 El Chichon and June 1991 Mt Pinatubo eruptions.