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

2.2. General Issues

There are now several stratospheric ozone profile data sets which overlap the long-term data sets for a period of several years. We used intercomparisons among these various data sets to partially test the validity of trends determined from the longer data records. Some limited comparisons are also included from more restricted campaigns (spatially and temporally), to present additional information about possible systematic errors.

Given two data sets covering a few years of data, we do not try here to deduce a linear trend after allowing for modelled (fitted) variations such as seasonal and solar cycle effects (see WMO Report of the International Ozone Trends Panel: 1988). This is because we mainly want to check how the data sets track each other over time, without trying rigorously to take into account the variations in terms of real "trends." Also, certain phenomena like solar and QBO effects cannot be removed reliably, based on a few years of data alone. Chapter 3 deals with the trend issues for the long-term datasets.

One component of the data quality analysis will rest upon simple linear regression (least-squares) fits for the time series, or difference time series for coincident data sets, to assess possible drifts with time in the measurement systems. For global datasets such as those from SAGE II and UARS, comparisons are given for rates of change from coincident measurement sets as well as from zonally-binned datasets. Ground-based measurements, such as lidars or sondes, are also used to check rates of change by comparison with coincident satellite data sets (e.g. versus SAGE II). Temporal sampling issues can be analysed by comparing these values against comparisons of the rates of change inferred from each "full" dataset. The standard error in these slopes are provided but it must be kept in mind that these are generally underestimates of the true error because the analyses ignore the temporal autocorrelations in the data (WMO Report of the International Ozone Trends Panel: 1988). Also, zero error is generally assumed for the time-varying component of systematic error, which is an idealised assumption that could be a root cause for differences between datasets. Trends of differences (i.e. drifts) between data sets are used to address this issue.

Significant changes in tracking or consistency between datasets with altitude (or latitude), can provide clues about possible reasons for differences. In particular, newer data sets such as UARS data should be checked against other existing datasets in the upper stratosphere, since there is good documented consistency between SAGE, SBUV, and Umkehr trends in that region (e.g. WMO Scientific Assessment of Stratospheric Ozone: 1994). This necessary (but not sufficient) check (for SAGE II data, in particular), provides added confidence in the mid- to lower stratosphere intercomparisons. Another useful analysis is to compare datasets under different conditions of aerosol loading (e.g., use a time interval shortly after the eruption of Mt Pinatubo, as well as one from a more quiescent time); this issue comes up because there is evidence that SAGE ozone retrievals can be contaminated by severe aerosol loading (see Cunnold et al., 1996a). This chapter will give recommendations for screening the SAGE data so that improved data quality exists for trend determination, even if this comes at the cost of somewhat more limited time coverage.

SAGE II and other limb-viewing satellite instruments (including HALOE and MLS) have nominal instantaneous vertical fields of view (FOV) ranging from 0.5 to about 5 km. On the other hand, both ozonesondes and ozone lidar have considerably higher vertical resolution. Since ozone profiles may exhibit strong vertical structure, comparisons of ozone measurements by different instruments should reflect those differences. In the lower stratosphere, ozone decreases rapidly below the ozone peak (~21 km). Since the IFOV vertically averages the measured slant path transmission (or emission), considerable biases in the ozone profile derived from limb-viewing instruments are possible (depending on the slope of the gradient and the field of view of the instrument). For example, Figure 2-1 shows the impact of the IFOV on the retrieval of a species with steep gradients both above and below a peak (much like ozone) based on convolving the FOV with the natural weighting function but without measurement noise and algorithm smoothing. For larger IFOVs, there are subtle problems around the peak but a significant bias is obvious below the peak. To properly interpret differences between measurement systems, comparisons of ozone (both spaceborne and in situ) should reproduce the vertical smoothing scheme inherent in SAGE II as well as HALOE and MLS (including both geometry and IFOV). It should be noted that, simple vertical smoothing of retrieved profiles does not fully correct for the field-of-view effect (though such smoothing may be adequate for narrow fields of view) but it is the approach which has been adopted for most of the comparisons performed in this chapter.

Ozone measurements made by SAGE or Lidar are recorded at geometric altitudes but many other ozone measurements (e.g. SBUV, MLS and ozonesondes) have pressure as their natural reference. A comparison of ozone trends and inferred solar cycle effects shows that there is better consistency between SAGE and the other satellite measurements and long term self-consistency in the SAGE measurements if the trends and solar cycle effects are reported on altitude levels (Cunnold and Wang, 1997). This better consistency is probably related to the presence of long term trends in tropical temperatures in the upper stratosphere in the NWS data that are being used to convert SAGE altitudes to pressures (Cunnold and Wang, 1997). These trends exceed those predicted by models and result in geopotential height changes of approximately 300 m between 1988 and 1994 in the tropics at 1 hPa. This produces a trend difference at that location over the 1988 to 1994 period between SAGE trends on altitude surfaces versus those on pressure surfaces of approximately 1%year^-1. At ~30 km altitude and above in the mid-latitudes the effect of temperature uncertainties on this scale conversion is likely to impact ozone trends by ~0.2%year^-1 over as short a period as 6 years. It is therefore recommended that ozone trends be reported in the natural vertical coordinate of the individual measurement technique. This would avoid the use of temperatures with trends which have their own inherent uncertainties. If this is not done, direct comparisons between measurements would then require an assumption that temperature changes were insignificant. A comparison of SAGE ozone trends on altitude and pressure levels would however probably bound the effects of temperature uncertainties.

There are also some altitude to pressure conversion issues to be considered in use of SBUV2 data. The SBUV2 retrieval algorithm assumes a constant pressure gradient/mass density relationship which ignores the inverse dependence of gravitational force on distance squared. Equation (2.2) provides a simple relationship which can be used to correct for this effect where P_c is the corrected pressure, P_i is the initial pressure, Z is altitude, and R_e is the radius of the Earth.

In comparing measurements from different instruments the comparison process itself introduces uncertainty because of space and time sampling questions. There are really three questions for each comparison: are the measurements at the same location, are the measurements at the same time, and are the instruments measuring the same thing (i.e. high vertical resolution point measurement versus a low vertical resolution measurement over a broad area)? The answers vary considerably for different measurement systems, limiting the quality of comparisons that can be done.

Space/time sampling ranges from sporadic point measurements (balloon sondes, Umkehr) to full global coverage on a daily basis (TOMS). Lidar, balloon sondes, and Umkehr data sets comprise point measurements made at a few locations around the globe and must be regarded as both spatially and temporally sparse. Lidar measurements are generally made at night, require clear conditions, and the data records are only a few years long (since 1988 at Table Mountain). Umkehr measurements are made near sunrise and near sunset and also require clear conditions. These measurements have been made for many years and are of interest for extending the ozone record back in time, but even in the early 1980's the record is sporadic and difficult to use to systematically evaluate ozone profile changes. Balloon sonde measurements have the advantage that they can be made day or night and under cloudy conditions, but at many sites balloons are flown only monthly, or weekly at best. Ground-based microwave measurements also have the advantage that they can be made day and night, but again the available data records are short.

The advantage of satellite measurements is that coverage is much better, but there are still limitations depending on the platform and on the measurement technique. Moreover, a space-based measurement is usually an average over some IFOV which covers a finite area on the ground, an issue when doing a comparison with a point measurement. Many of the comparisons in this report are relative to SAGE II. Because SAGE uses the solar occultation technique, it makes only two profile measurements per orbit, about 30 measurements per day. SAGE II is in a precessing orbit such that the sunrise and sunset latitudes shift from high northern latitudes to high southern latitudes approximately monthly. Coverage is fairly good between 60°N and 60°S but is sporadic in the polar regions. SAGE has good altitude resolution, ~1km, but it samples a horizontal volume of about 300 km along the line of sight at the tangent point. HALOE, on the Upper Atmosphere Research Satellite (UARS), is also an occultation experiment and has similar spatial and temporal coverage. The MLS instrument, also on UARS, measures limb emission and thus has continuous coverage along the orbital track; but it has similar horizontal averaging characteristics as SAGE and HALOE. The view geometry gives coverage from about 30 degree latitude in one hemisphere to about 80 degree latitude in the other hemisphere, with the prime hemisphere changing every 34 days as the orbit precesses. The SBUV and SBUV/2 instruments measure ozone using backscattered ultraviolet sunlight. Since both Nimbus 7 and NOAA 11 were in polar orbits and the instruments are nadir viewing, they produce coverage from 80°S to 80°N on a daily basis, but only along the orbital track. No measurements are possible at high latitudes in winter (polar night). The IFOV on the Earth surface is a 200 km square.

Analysis of the TOMS total ozone data shows that at spatial separations between measurements greater than about 5^o great arc distance, the correlation begins to drop. This is most evident at mid to high latitudes in winter and early spring, when the ozone field is highly variable. The effect depends on altitude - ozone in the upper stratosphere is much less variable and a match within <20^o will be highly correlated. As the time separation increases, the effect is equivalent to an increasing spatial separation. An air parcel measured by a balloon sonde today may have been a thousand kilometres away two days prior. Two approaches to improving the number and quality of matches will be discussed in section 2.2.5 - trajectory mapping and potential temperature-potential vorticity interpolation.

Various data sets are compared in this report, with SAGE as the prime data set acting as a transfer standard. If, for example, a single SAGE profile is compared with an individual balloon sonde profile at a site like Hohenpeissenberg, the number of matches will be quite small. In general, comparisons will be more significant statistically if monthly zonal means are compared. In the 30°-50°N zone there are enough balloon sonde stations (or Umkehr stations) for a good comparison. Station-to-station biases can be a problem if the data records from the different stations are non-uniform - i.e. if a station with a data record only in the late 80's has a bias relative to a station that has data mostly in the early 80's and both are included in the long term comparison.

The comparison of the lidar stations is possibly the most difficult because of the small number of sites and the short data record length. Any altitude dependent biases can be determined, limited by the number of matches that go into the comparison, but the time dependence will be very uncertain because of the short length of record.

The satellite-to-satellite comparisons will generally have many more matches, though the UARS orbit was such that HALOE still has only limited direct matches with SAGE II. Here again good comparisons of zonal means can be made. Some of the best comparisons can be made of SBUV2 with SAGE II, because there will be an SBUV-IFOV that can be matched with each SAGE-IFOV every day. A limiting factor here is the differing vertical resolution - 1 km for SAGE versus about 8 km for the SBUV experiment.

Given the space and time sampling of the various instruments, we conclude that a good job can be done on determining the altitude dependent biases, but that the ability to determine time dependence of differences will be limited because of the shortness and/or non-uniformity of many of the data records

The most frequently used methods for ozone data validation in the past have involved intercomparison either of individual pairs of "coincident" measurements or of zonally-averaged data. These techniques generally provide an effective means of validating the observations from instruments achieving essentially global coverage on a daily basis, but this is not the case for instruments which sample the atmosphere more sparsely (in either space or time), such as the satellite-borne solar occultation instruments and the ground-based instruments. For these latter instruments, the "coincidence" approach relies upon extensive data validation campaigns that provide a relatively small subset of "coincident" measurements, where the definition of "coincidence" often implies separations of hundreds of kilometres in space and/or many hours in time. This approach most often produces a statistically unsatisfying comparison of only a small subset of the observations, which must then be assumed representative of the larger data set. Alternatively, zonal averaging of the individual data sets and subsequent intercomparison of zonal mean analyses generally provides statistically significant results, but effectively eliminates the possibility of examining zonal variations. In both cases, normal meteorological variability adds considerably to the statistical uncertainty associated with the validation results.

Recently, two alternative approaches to data validation have been developed: coordinate mapping (or constituent reconstruction) and trajectory mapping. Each approach takes advantage of a number of quasi-conserved properties of the atmospheric flow, both to enable relaxation of the data coincidence constraints, and to reduce the effects of the short period and large amplitude dynamical fluctuations which are ubiquitous in the atmosphere.

In the lower stratosphere where the greatest uncertainty remains with respect to both ozone measurement accuracy and long-term trends, ozone acts as a relatively long-lived tracer in the absence of heterogeneous chemical processes. Ozone mixing ratio tends therefore to be well conserved in a lagrangian or "along the flow" sense. Furthermore the potential temperature (PT) of an air parcel tends also to be well-conserved along the flow, as does an air parcel's potential vorticity. Coordinate mapping and trajectory mapping exploit two or more of these three quasi-conserved properties of the stratospheric flow to enhance our ability to effectively intercompare ozone data sets.

Trajectory mapping, by employing O₃ and PT conservation assumptions, allows a stricter definition of "coincidence" to be used than in the traditional technique, increases the number of coincident data pairs available for comparison, and at the same time enables a decrease in statistical uncertainty to be achieved. Coordinate mapping employs all three of the above conservation assumptions, to enable virtually the entire data set from each instrument to be used in the intercomparison, while greatly reducing the contribution to uncertainty attributable to atmospheric variability. The details of each of these two approaches and their application to an intercomparison between SAGE II and HALOE data are provided in Section 2.4.6 below.