SECTION III
INDEPENDENT SOURCES FOR COMPARISON
ALAN LAPWORTH
RUMEN D. BOJKOV
This section describes the use of independent external sources of data to evaluate consistency and credibility of Dobson- instrument data. This evaluation is in addition to the primary state of the instruments considerations based on actual instrument calibrations as described in the previous sections and is best applicable to long-term ozone records.
3.1 Use of Independent Data Sources
Initial evaluation of Dobson-station ozone data will usually be made without reference to other measurements, but eventually the data should always be re-evaluated by intercomparison with another source(s). It is only since this procedure has become routine (WMO Ozone Project Report #18, 1988) that major improvements in the quality assurance of the long-term total ozone data base have become possible. It is extremely difficult to maintain accuracy of an isolated series of measurements without feedback from an external reference.
The most useful aspect of comparing Dobson-station data with some reference is that of checking for changes in published stations ozone values at times when the instrument has been recalibrated and when the log book shows that an adjustment or modification may have been made to the instrument. As will be seen below, there are often problems in comparing station data with an external reference, but even when the external data do not provide absolute values, changes in the relation between the Dobson-ozone and the comparison time series can often be noted and could be very instructive. Correctable errors in Dobson data generally fall in the range between 1.0% and 5%. Errors less than 1.0% are difficult to detect by the intercomparison methods covered below, while errors greater than around 5% are very seldom and result in substantial doubt about the quality of the given measurements.
3.2 Check for Internal Consistency
Before comparisons with independent sources are made, the station time series should be examined for internal consistency. In this case the comparison is between measurements made at one period with earlier or later measurements. The implicit assumption made is that the annual course of ozone from one year are similar to another, and although this is not always true, it is adequate for re-evaluation work.
In checking for internal consistency, it is useful to remove as much of the normal variation as possible. The main variability is the annual ozone cycle, which results from the wintertime transport of ozone from its production zone high above the equator, downwards and towards the poles. The resulting changes in total column ozone values are particularly marked at latitudes greater than about 30 degrees. The transport across the mean geostrophic flow is a secondary circulation and is enhanced by the eddies resulting from Rossby wave changes in winter.
The annual variation can be removed from the time series as follows. Simply by subtracting the mean annual variation over the period from each individual year. However, it is best to first remove the high frequency daily variations from the data by block averaging into monthly means. A mean is then taken for each month over the total number of years in the period. Then, the resulting monthly means are subtracted from the individual months in the series to give a difference curve.
An alternative method of removing the annual variation is to subtract a twelve month running mean. The running mean is performed on the time series of monthly averages, and a cosine filter is used with a half width of six months. This procedure has the disadvantage that instrument calibration changes are not so clearly defined, but on the other hand, natural fluctuations are not so intrusive. This method also reduces the problem with the annual average method that the annual mean is affected by any badly calibrated data.
This difference curve will show large variations during the winter and small during the summer months. The effect of these variations can be removed by dividing each monthly difference by the monthly standard deviation. The monthly standard deviation can be calculated using the daily ozone values within the month concerned, averaged over all the years in the period. The high variability in winter is partly due to the intensity and variability of the transport mechanisms operating during these months and partly due to the associated variations in tropopause height.
The resulting time series of differences normalized by standard deviations can then be examined for significant deviations from the zero line. There will be both real deviations in total column ozone amounts present and also those resulting from bad calibrations or instrumental problems. (An example is shown in Figure III-1). Comparisons of the differences with the instrument logbook, taking particular note of intercomparison dates, will often distinguish between these possibilities. It should be noted that many fluctuations in the difference curve will be of meteorological origin, and the removal of these will be discussed in the next section, when reference to external data is considered. Considerable scatter in the monthly means will also be noticed in months during which the number of daily observations is sparse (in general a monthly mean is considered not representative if constructed from less than 13 daily values).
Before using the difference time series, it is advantageous to smooth the series with some type of running mean filter. The filter period may extend over many months but 12 months is generally used. Although such filtering removes the sharpness of any steps in the series, significant changes that persist for a long period often become more clearly defined in relation to the noise of the shorter fluctuations. In particular, the nearly two year period of the Quasi-Biennial Oscillation (QBO) often becomes clearly visible (see Figure III-2).
A further check of internal consistency should be made by examining the monthly standard deviations over the period. These may be normalized by long-term-mean monthly standard deviations in a way similar to the difference curve. Any period of abnormally high standard deviation will need to be examined more closely.
Figure 1: The normalized monthly deviations reveal strong discrepancies in the New Delhi ozone record around 1961 and again between 1968 and 1975 (Bojkov, 1987)
Figure 2: Monthly normalized deviations smoothed by 12-month running means for total ozone (continuous) and 100 hPa temperatures (dotted) show well pronounced QBO. The arrows are indicating the time of the easterly phase of the stratosphere equatorial wind (Bojkov, 1987).
3.3 Reference to External Data Sources
There are limitations to internal consistency checks for total ozone time series, the main one being the difficulty of distinguishing instrumental from real ozone fluctuations. Three main sources of external data are available. The first of these is an extension of the internal check by taking account of ozone fluctuations known to be related to particular meteorological quantities. This technique has the advantage of being available over the whole period since hemispheric coverage by balloon upper air soundings became available and were centrally archived, this is since 1957 for the Northern Hemisphere and since 1978 for global data. The method has the disadvantage that not all ozone fluctuations are readily correlated with meteorological data, so that there could still be residual non-instrumental data variations in the time series. It also requires good understanding of the relationships and some judgement in the application of some manipulation of the data.
The other two methods are more direct and more reliable. The second method is that of comparison with another total ozone data set from a ground station in the same synoptic circulation regime as the first station (see Figure III-3). This has the advantage of simplicity in application. Also if the comparison station is well chosen and correctly calibrated, there will be very few of non- instrumental fluctuations. The main problems are in choosing an appropriate station, i.e., one with a well calibrated instrument and with a sufficiently long record. In some areas there may be several good comparison stations, while for others, there may be none.
The third method uses direct comparison with TOMS satellite data (Version 6 or later). This method is undoubtedly the best of the comparison data sources available, but it only covers the period between November 1978 and May 1993. There is also the problem that TOMS data are valued for their relative independence from ground station data, and there is a risk of introducing spurious correlations between data sets by using one to correct the other. Therefore, they are generally used to only pinpoint the discrepancies and identify their magnitude.
.
Figure 3: The 12-month smoothed total ozone at Arosa (continuous) and Mt. Louis (dotted) show clearly the start of major discrepancies in the records around 1971 (Bojkov, 1987).
3.3.1. Meteorological quantities
The time series of total column ozone above a station is significantly affected by various disturbances of meteorological origin, and this results in well known correlations between the ozone series and time series of several meteorological variables. These correlations can be used to remove fluctuations from the normalized ozone difference series, derived as described in section 3.2 above to reduce the amount of fluctuation of non- instrumental origin.
The most important is the relation with the 100 hPa temperatures, which give fluctuations from mean values of around 15%. This correlation is caused mainly by vertical motions in the stratosphere, which lead to a raising or lowering of the temperature on constant pressure surfaces simultaneously associated with an increase or decrease of ozone amount. Also horizontal advection can further enhance the physical relationship which may yield a correlation coefficient of 0.6 or more at mid latitudes. The reason for the correlation with horizontal advective motions appears to be that such motions advect both ozone and warm air in the stratosphere, although the relation is not a straightforward one. Such processes account for a high correlation over timescales of the order of weeks or even months.
Over a whole year however, a phase difference can be apparent between the two series, the peak values of total ozone occurring about three months before those of 100 hPa temperature. What makes the correlation extremely valuable from the point of view of re-evaluation work is that there is still some relatively low correlation over timescales of the order of several years. At latitudes of less than about 30 degrees the correlation becomes insignificant but so also do the fluctuations.
The 100 hPa level is generally used because it has the highest correlation coefficient, also it suffers relatively little from seasonal variations, such as those caused by stratospheric warmings at higher levels. The 100 hPa level is just below the ozone maximum, and is the level at which vertical motions have their greatest affect on local ozone concentrations At latitudes higher than about 70 degrees (in both hemispheres), however, sometimes the 50 hPa level temperature is better correlated with ozone.
High correlations (around 0.75 or more depending on the period) occur between total ozone and either potential vorticity or the vertical component of absolute vorticity. Both of these variables, though require more than single station data to calculate, and in the case of potential vorticity, require data generally not available in upper air data sets from earlier years. It is possible that some combination of 100 hPa temperatures and wind velocities may be significantly better correlated for some stations than 100 hPa temperatures alone.
There are two methods of identifying potential instrument problems in the examination of the fluctuations and the correlations of the ozone time series and the 100 hPa temperatures. The first of these is a graphical method and simply consists of preparing a normalized difference series of monthly means of the 100 hPa temperatures (in an exactly similar method to that used for the total ozone time series) and then comparing the two series by plotting them on the same graph (see examples on Figure III-2, 6, 10). The use of difference series avoids problems caused by the fact, that although the two series are well correlated on a daily basis, they have rather different annual variations, the ozone maximum being in spring, while that of the temperature series occurs during the summer. It is important that the annual mean curves and standard deviation values for both parameters (ozone and temperature) should be taken over identical time periods, and that only days for which both types of data are available be used in calculating monthly means. This method has the advantage of simplicity, and in particular, all fluctuations in both series are immediately obvious to the person performing the re-evaluation and there are no hidden anomalies.
To compare two partially correlated time series in this manner is however, not always easy and a second technique may be used. This consists of calculating the regression coefficients relating the two series and then applying these coefficients to the 100 hPa series to adjust the ozone series. Only one series remains, which has some of its fluctuations removed. The regression is performed as follows. First, the annual variations of both ozone and temperature time series are removed by subtracting either the annual mean series taken over more than ten years or by subtracting a running mean. The running mean is constructed using a cosine filter with a half-height width of around 60 days. This filter length has been found to be long enough to eliminate short-term ozone variations while maintaining the longer term annual variation. The coefficients of a simple linear regression between the two resulting difference series are then calculated by standard methods. The time series of daily ozone values is then used as a basis, from which is subtracted the difference between the individual daily temperature values and the annual mean temperature (converted to an ozone difference by application of the regression coefficients). The resulting curve is the ozone series modified by removing the features that correlate well with the temperature time series. It is best studied by averaging into monthly mean values and dividing each monthly value by the standard deviation for the daily values within that month. The standard deviation should be calculated from the adjusted ozone values.
Values for 100 hPa temperatures for any station can be obtained from National Climatic Data Center (Federal Building, Asheville, North Carolina 28801, U.S.A. or from Roy Jenne, National Center for Atmospheric Research (P.O. Box 3000, Boulder, Colorado 80302, USA). Monthly values are available globally since around 1978, and for the Northern Hemisphere since 1954. Upper air data for 50 hPa and above extends back to only 1964. Daily-charts of hemispheric temperature and geopotential distribution at standard upper air levels have been routinely prepared by the Department of Meteorology, Free University of Berlin, Carl- Heinrich-Becker Weg 6-10, Berlin 12165, Germany. They are distributed to many libraries and could be obtained on request from Prof. K. Labitzke. It is very much desired that monthly mean total ozone and 100hPa temperatures be centrally prepared for selection of stations with long-term observational record and deposited at the WMO World Ozone Data Center (W03DC). Thereafter annual updating in the annual catalogue of the W03DC will be extremely useful.
The ozone time series is also influenced by and correlated with the equatorial stratospheric wind Quasi-Biennial Oscillation (QBO), which has a period of about two years, with El Ninol Southern Oscillation (ENSO), which has a period of around four years, and with the eleven year solar cycle. The influence of the solar cycle is fairly small (around 1%) but the other two cycles are both noticeable at stronger than 3% level. A signal related to the QBO can be easily detected in the 100 hPa time series. The monthly values of 30 hPa winds over Singapore (representing the QBO although the 50 hPa winds are also useful), the ENSO index and the sunspot cycle (representing the solar cycle), are given in the Appendix.
Unfortunately the effects of the QBO and ENSO are not simple but depend on their phase relative to the ozone annual cycle. In addition, the effects of the QBO for middle latitude stations are not in phase with the Singapore 30 or 50 hPa winds, as the phase log varies with latitude. As a result it is probably best to simply compare the monthly values of the meteorological parameters representing these two time series with either the monthly ozone difference time series or the ozone series adjusted for 100 hPa temperatures as described earlier in this section. once correlations with these four cycles have been identified, the correlations can be used to identify and possibly eliminate deviations in the ozone difference series on the grounds that such deviations are most probably caused by real fluctuations in total ozone values, rather than by changes in instrument calibration.
An additional check can be made on residual deviations. The charts of hemispheric 50 hPa or 100 hPa winds and temperatures for months that show large ozone deviations can be compared with similar charts for the same month in other years, to determine if there are any large anomalies in the general circulation. Large ozone deviations can be associated with such anomalies in the circulation.
3.3.2 Comparison with another ground station
Comparison with ozone data from another ground station is comparatively easy, the only proviso being that the comparison station should be in the same synoptic atmospheric circulation regime. Such a station may not always be available, especially in the Southern Hemisphere. Monthly averages for each station are then derived, and either the two series can be compared visually or a time series of differences between them can be calculated. In constructing the monthly means it is important that only days when data are available from both stations be used. The comparison can be repeated if more than one suitably positioned neighboring ozone station is available. The main disadvantage is that the comparison station may have errors of its own, and there may be some real differences between the two stations. Also data may not be available over the full period. Nevertheless if the second station is well positioned, it should give better results than in comparison only with 100 hPa temperatures.
As with the use of meteorological correlations, this method has the advantage that it can be extended back prior to the ozone-satellite era (i.e., before 1978). Ozone values for neighboring stations can be obtained from "Total Ozone Data for the World" (the "Red Book") published by the WMO World Ozone Data Centre (W03DC) operated by the Canadian Atmospheric Environment Service (AES, 4905 Dufferin Street, Downsview Ontario M3H 5T4, Canada) . When searching the volumes for station data note that the station order has changed over the years and also that stations often delay submission of their data for publication. A full check has to be made for corrections to the once published data submitted by the stations, which usually replaces the previously submitted data on the W03DC computer memory and are republished in the "Red Books".
3.3.3 Satellite data
Satellite data which covers the period from November 1978 to May 1993 provides the best available data for Dobson data evaluation. The instrument used is the Total Ozone Mapping Spectrometer (TOMS) on the Nimbus 7 satellite. Use of these data is similar to that for comparison with a ground station: monthly means are derived, and the differences between satellite and ground station Dobson data calculated and plotted in a time series. In constructing the monthly means it is important to use only the days when simultaneous data are available.
TOMS data are available for any Dobson station; therefore, there should be no large fluctuations in the difference series of monthly means that are not due to ground station calibration errors. Only current TOMS data, (now identified as Version 6 data) should be used, as previous versions were affected by different algorithms used and by diffuser plate degradation.
Although TOMS Version 6 data are of very high quality they are not perfect and the following problems should be noted. TOMS does not measure tropospheric ozone very well particularly where there are clouds. The quality of data at high solar zenith angles, such as occur over the polar regions, may be reduced. Volcanic dust, particularly that due to the 1982 El Chichon and the 1991 Pinatubo eruption, affects for a few months the TOMS data, noticeably in the equatorial regions, and correction algorithms may not be fully effective. In recent years a number of problems occurred with the satellite, which has operated far beyond its design life, all were however identified and corrective measures were taken by the NASA TOMS data processing team. Chief among these problems is a degradation of the orbit, consequently recent (1992) measurements were made not exactly at but sometime before local noon. Finally, since February 1992, a problem involving shadowing of a calibrator for a certain time has occurred, which will result in some degradation of the absolute calibration.
A second TOMS instrument is at present in use aboard a Russian Meteor-3 satellite. Because of its orbit however, only limited coverage is available. Additional TOMS instruments should be launched in the future. TOMS Version 6 data are available on floppy diskettes, compact discs or through electronic mail from Rich McPeters, Code 916, (NASA Goddard Space Flight Centre, Greenbelt, Maryland 20771, USA) . Complementary satellite data series are available from the NASA SBUV (Nimbus 7) and NOAA-9 and -11 SBUV/2 instruments derived from nadir observations. These data are currently being reprocessed and will be archived and available in the near future.
It is known that a bias exists between the World Primary Standard Dobson Instrument #83 and TOMS ozone values of about 4.5%, and a bias between it and the SBUV of about 1. 5%. These biases could stem from uncertainties in the calibrations of the three types of instruments.
Figure 4 and 5: The drifting of the Oslo instrument expressed as step changes, and big µ dependence in the Hobart record in the early 1980's until the proper calibration was made in 1985, are clearly demonstrated in the monthly differences between the ground based and the TOMS instruments (Bojkov, 1990).
3.3.4. Ozone sondes
A time series of total ozone from ozone sondes flights integration may be available at some Dobson stations. These sonde data however, usually do not have the sufficient calibration accuracy necessary for long-term stability of total ozone measurements. In fact, according to long standing practices recommended by the ozone Commission and WMO, the sondes need to be calibrated with reference to local Dobson (or Brewer) instrument total ozone measurements. In general such sonde data are not suitable for re-evaluation work.
3.4 'Mu' Dependence
When comparing with a source of external data, a regular annual variation is sometimes apparent in the series of differences between locally measured ozone values and ozone values measured by an external data source (see Figure III-5). The most common reason for this is a variation with the solar zenith angle (i.e. , a µ dependence - where µ is the secant of the zenith angle). The cause is usually a calibration error and/or scattered light, which have an effect that will vary as the µ values vary with the time of year. In general, annual variations of this type should disappear when the data are correctly calibrated with respect to the external source. If calibrations of the given Dobson instrument are not available but the µ dependence is clearly appearing in the record, one could derive empirical corrections for each month of the year and to apply it during the re evaluation. Nevertheless, there may be other causes for such annual variations in the difference series. For instance, pollution is often greater in winter, due to the increased emissions by fossil-fuel power stations, and the resulting sulphur dioxide if emitted in proximity of a Dobson station may affect the pollution sensitive CD wavelength observations in use at this time of the year by a few DU.
3.5 Comparison by Observation Type
When total ozone data from a station are being compared with other data, it is best to make separate comparisons of certain data types. There are several subdivisions that can be made, but it is of particular importance to isolate the direct sun observations. In general, it is useful to construct two time series, one for all observation types, and one for direct sun measurements only. The number of direct sun measurements may be limited at any station, particularly during certain seasons, and this will give rise to statistical fluctuations when comparing with external data. Because of this, months in which observations are available for significantly less than half the number of days should be removed from the intercomparison.
It is important to use only the same days in each month for which both direct sun ozone data and comparison data are available when calculating the monthly means for the direct sun comparisons. The data can also be usefully subdivided according to wavelength type, the primary subdivision being to separate AD from CD observations. Such division of the data often makes the necessary calibration changes much clearer, by eliminating the scatter in observation types that are of fundamentally lower quality. In addition, the particular wavelength for which recalibration isnecessary may be clearly identified. Another possible subdivision is by µ range. However the range of µ values available will have a seasonal dependence, and some ranges will not be available throughout the year. The observations made at lower µ values-in the range of 1.5 to 2.5 should give better quality data.
Figure 6 and 7: Monthly normalised deviations smoothed by 12-month running means at Hradec Kralove for total ozone (continuous) and 100 hPa temperatures (dotted) , show major discrepancies in the record of the initially published ozone data and very good concurrence after the ozone calibrations have been incorporated (Bojkov, 1987).
3.6 Provisional Calibrations
One of the main problems of attempting to re-evaluate ozone measurements is that of missing data - both intercomparison data and station data. In addition, problems arise with intercomparisons that are recognized in retrospect as having significant errors. In these cases, reference data can permit some provisional calibrations.
3.6.1 Missing intercomparisons
For missing or bad intercomparisons two remedies are possible. The first is to use standard lamp data to transfer a calibration between the same instrument at two different periods. Whether this is possible depends on what modifications may have been made to the instrument between the two periods. For some types of modification (such as a wedge change) this will not be possible. It should be noted that the transfer of a calibration by a standard lamp between two different Dobson instruments can result in considerable errors if there are even small differences in the optical alignment of the two instruments, which will affect the results.
A second possible remedy is to use external reference data to transfer the calibration from one period when the calibration is considered to be correct to another period when it is unknown. Such external data reference could be the differences with TOMS and/or differences with nearby stations to remain constant, although this second remedy causes dependence on an external source of data, the dependence is limited to assuming constancy in the reference data source over the time between the two periods, which may be only a few months. Absolute accuracy of the reference is not necessary unless both periods of data are uncalibrated, in which case any use of external data should be noted in Line 5 of the W03DC 'Red Book' format and the data marked as provisional with an indication of the date (month, year) of submission.
3.6.2 Missing Raw Data
This is the most difficult situation to deal with, and the best that can be done is to calculate approximate IN' values (IN' is the logarithm of the intensity ratio) using the published ozone values together with a mean µ value for the observation hour given. These IN' values may then be adjusted where necessary from corrected intercomparison data or, as above, using the reference data as a transfer standard, and the ozone values recalculated. Such a transfer is more useful for observations made at small µ values.
Figure 8 and 9: The 12-month running mean of total ozone at Reykjavik (continuous) and Lerwick (dotted) show numerous discrepancies between the two records however, after the re- evaluation of the Reykjavik data in line with a 1977 calibration the two stations show good concurrence (Bojkov, 1990).
3.7 Examples
The above is a short summary of the techniques used for examining ozone station data. Sometimes it is very difficult for someone new to this type of work, to gain an appreciation for the way in which the discrepancies and features being investigated will manifest themselves above the general noise levels of the time series involved, and an appreciation for the relative merits of the different methods used. The best way of demonstrating the principles involved is by using examples from some real cases of station-data review and these will be described below. Several of these cases are illustrated with figures provided by R. Bojkov who has been the pioneer in this type of work and prepared the first "provisionally re-evaluated" data used in the WMO Ozone Report #18.
The Figures (III-6 and III-7) use ozone data from the station of Hradec Kralove and 100 hPa temperature values from the upper air station at Wroclaw, about 150 km to the north-east. The ozone data series from this station was re-evaluated by K. Vanicek in 1991 by applying the R-N tables, derived from an intercomparison with the world standard in 1986 to observations made since 1962 but without considering the intercomparisons at Shiofock (1964). Both figures show time series of normalized monthly deviations of total ozone (full lines) and 100 hPa temperatures (dotted lines) , which have been smoothed by 12-months running mean filter. The Figure III-6 shows the total ozone time series as it was originally published in the 'Total Ozone Data for the World, (Red Book), and the second Figure III-7 shows it after the 1991 re-evaluation of the data by Vanicek. A very good correlation between the two series over all timescales is very noticeable in the Figure III-7. However, great discrepancies are apparent in the uncorrected ozone data (Figure III-6). It should also be noted that some of the larger errors in the ozone values are already apparent in the normalized ozone difference series in the mid-sixties values fall well below minus one sigma, and in the early-eighties the values exceed plus one sigma. variations with a period of the QBO are also apparent in these and succeeding records.
Figures III-8 and III-9 show a comparison between two ozone stations close to the polar circle and in the same atmospheric circulation region - those of Reykjavik in Iceland and Lerwick, in the Shetland Isles. The stations are about 950 km apart. In both figures, the monthly total ozone values have simply been smoothed with a twelve month running filter to remove the annual variations. The Reykjavik ozone values are connected by a solid line the Lerwick values by a dotted line. The Reykjavik ozone data series has been corrected for an instrument calibration shift in 1977, by reference to the continuity of the 100 hPa temperature series. Figure III-8 shows the Reykjavik data as published in the "Red Book", and Figure III-9 shows it after re evaluation. The effect of the 1977 calibration on the correlation between the two series is very marked.
The next four figures III-10 to 13 refer to the station at Potsdam near Berlin. The Figure III-10 shows the normalized differences of ozone and 100 hPa temperatures smoothed over a period of several months. It will be seen that during the sixties the difference series for total ozone was significantly greater than 1.5 sigma. In addition the difference series for ozone in the early seventies is much lower than the corresponding 100 hPa series. The next Figure III-11 repeats the Potsdam series and compares it with a similar ozone series for the station of Belsk in Poland, which is about 200 kms north-east of Potsdam. This comparison shows the same discrepancy features as the comparison with the 100 hPa temperatures. The next Figure III-12 repeats the comparison between Potsdam and Belsk but in a different way, using monthly ozone values smoothed with a twelve month filter to remove seasonal effects, and the same discrepancies are obvious - Potsdam data are too high in the 1960's and too low in the early 1970's. Finally, Figure III-13 shows another comparison using twelve month running mean filtered data, this time between Potsdam and Hradec Kralove. This station is only about 300 kms south-east of Potsdam, and once again clearly shows similar features of the discrepancies as identified in the previous figures.
Figure 10 and 11: Monthly normalised deviations smoothed by 12-month running means for total ozone at Potsdam (continuous) and 100 hPa temperatures (dotted) and at Belsk (dotted on the lower frame) show similar discrepancies which disappear after re-evaluation (Bojkov, 1987).
Figure 12 and 13: The 12-month running mean smoothed total ozone at Potsdam (continuous) compared with Belsk and Hradec Kralove (dotted) show very similar discrepancies obviously due to faulty Potsdam record (Bojkov, 1987).
Figure 14: The normalized deviations of the re-evaluated direct sun ozone record of Potsdam compared with the 100 hPa temperatures at Berlin (similarly to Figure 10) shows good agreement (Feister, 1993).
Figure 15: The monthly mean differences between the Dobson at Singapore and the overpressing TOMS show great discrepancies resulting from faulty calibrations in 1982 and 1986 (Lapworth, 1992)
Recent re-evaluation of the Potsdam data by U. Feister taking into consideration the instrument calibrations have brought the ozone record of Potsdam in agreement with the well maintained nearby stations of Belsk and Hradec Kralove (see Figures III-14 which are similar to III-10 but without discrepancies). This demonstrates the usefulness of the comparisons described earlier in this chapter for identifying the time of occurrence and the magnitude of the discrepancies.
The remaining figures show time series that, unlike the previous examples have not been filtered with 12-month running means. It will be seen that this lack of filtering has the advantage that sudden changes due to erroneous intercomparisons or other errors are well defined. on the other hand, there is the disadvantage that it is often difficult to distinguish long-term changes amongst the short term variability. Figure III-9 shows the time series of differences between TOMS Version 6 data and the published in the W03DC "Red Books" data for the Singapore station over the period 1979 to 1990. This clearly shows the effects of erroneously done calibrations made in 1982 and 1986. A careful re-evaluation of these two intercomparisons showed that the original calculations had biased the intercomparison results too far towards high µ value data. It was found that the ground-station satellite differences were greatly reduced in magnitude if the results of these two intercomparisons were ignored and a constant calibration over the period assumed. Thorough calibration and review of the Singapore record by WMO sponsored expert during the past two years re established a series of reliable ozone data which were replaced in the WO3DC.
The next three figures (III-16, 17, & 18) also use differences between TOMS Version 6 and Dobson data at the ground station in Lerwick in the Shetland Isles. In all three figures, circled monthly values connected by a full line denote total ozone values taken only from direct sun measurements. Dotted monthly values connected by a dashed line denote total ozone values taken from all types of measurements. Monthly values are only plotted where at least five measurements of the type indicated were made in the month and thus some points are not representative as monthly mean but only as differences between two assembles of data. The aim of these illustrations is to show the effect of the re- evaluation of ozone data in reducing the differences with TOMS and not to have absolute monthly mean comparisons.
Figure 16, 17 and 18: The monthly mean differences between the Dobson at Lerwick and the overpassing TOMS direct sun (continuous) , all observations (dotted) of the originally published record (16) ; of the re-evaluated record without cloud correction (17); and the best agreement with the re-evaluated data corrected also for cloudy zenith measurements (18), (Lapworth, 1992).
The Figure III-16 shows the data as originally published in the "Red Book". Notice the enormous bias in the direct sun measurements and the scatter in the total values exceeding ± 5%. The Figure III-17 shows the data after their re-evaluation calibration by A. Lapworth in 1991. The monthly averages from all type of measurements still show considerable scatter. The direct-sunlight difference values however, show a much reduced deviation and less µ-than those in the previous figure. The Figure III-18 shows the recalculated results after a separate calibration of the cloudy zenith measurements. For this, the cloudy zenith observations are calibrated in terms of their difference from equivalent blue sky zenith measurements, the difference being given in the form of a polynomial. The polynomial was derived using ground based measurements, and it is noticeable that the magnitude of the differences with the satellite measurements is significantly reduced. The station Lerwick has a very low number of sunshine hours and the majority of observations are of the zenith cloudy type.
The remaining Figures (III-19 through 25) all relate to the total ozone measurement station of Bracknell in southern England. Here, these data are combined with data from Camborne, which is about 340 km to the west of Bracknell. The instrument was transferred from Bracknell to Camborne when the Bracknell station was closed in mid 1989. However, both stations are subject to similar atmospheric circulation and for purposes of this illustration the change in location is irrelevant.
The first three of these Figures (III-19 to 21) are all of TOMS ground-station differences plotted using the same conventions as for Lerwick above. The Figure III-19 shows the data as originally submitted to the WO3DC and published in the "Red Book". Notice the enormous negative deviations in the early 1980s, and the extremely large positive deviations in the late 1980s. These were already identified previously in the paper by Bojkov et.al. JGR 1988. Figure III-20 shows the TOMS-ground station differences after re evaluation by A. Lapworth in 1991. This data series has been corrected by replacing R-N tables from invalid (erroneous) calibrations by others from previous intercomparisons or standard lamp series. Use was made of TOMS data in identifying bad comparisons. Although there are no large deviations, the overall scatter is still fairly large. The µ dependence of the monthly differences, where the deviations were high during the late eighties, are reduced from those of the previous figure. The Figure III-21 shows the re evaluated data using AD wavelengths only, and this has a much reduced scatter of about ±2.5%. A possible cause of the large scatter at CD wavelengths might be the effect of pollution at the Bracknell site from fossil fuel power stations in winter, when only CD measurements are made. However, there are no actual S02 concentration measurements and this is only a speculation. It is more reasonable to assume that there is a systematic effect due to bad calibration of the instrument which was one of the oldest Dobson still in use.
The remaining four Figures (III-22 to 25) are plotted to show how much of the calibration errors could be detected using normalized difference series both with and without 100 hPa temperature data. The temperature data are obtained from the neighboring site of Crawley (about 50 km away) in the case of Bracknell. Camborne is itself an upper air station.The first two figures (III-22 and 23) show the normalized difference series for all observation types before (III-22) and after (III-23) the re-evaluation. In general, high positive differences of the late eighties are in evidence, but as expected the errors are not so clear as the comparisons with the satellite data. The last two figures repeat these series using data that have been linearly adjusted for the correlation with 100 hPa temperature measurements. It will be seen that these series are far less noisy, and are in many ways similar to the satellite difference series. The negative deviations of the early eighties and the positive deviations of the late eighties in the originally submitted data published in the W03DC "Red Book" are significantly clearer as are the improvements in the re-evaluated data. The main problem with the use of normalized difference series however, is also clearly in evidence in the form of some negative differences in 1990 and 1992. Part of these could be due to real anomalies in the total ozone time series versus the 100 hPa temperatures and do not appear so strongly in the satellite differences (compare Figures III-21 with III-25).
Figure 19, 20 and 21: The monthly mean differences between the Dobson at Bracknell-Camborne and the overpassing TOMS. Continuous lines are direct sun, dotted are all observations. The originally published record (19) shows great discrepancies and a step jump about 1984; the re-evaluated data (20) and the re-evaluated using AD wavelength only (21) are providing a more coherent record (Lapworth, 1992).
Figure 22 and 23: The monthly ozone normalized deviations before (22) and after re-evaluation (23) of the Bracknell-Camborne record show noticeable improvement (see also Figure 24 and 25) (Lapworth, 1992)
Figure 24 and 25: Same as Figure 22 and 23 but when the ozone record is corrected using the 100 hPa relationship. The original data (24) and the re-evaluated data (25) both show improvements compared with Figure 22 and 23 (Lapworth, 1992)
The cases illustrated above are not intended to be definitive but have all been chosen from cases available at the time of writing to give some idea of the large number of different techniques that can be used, and some of the strengths and weaknesses of the different approaches. Further examples can be found in the papers by R. Bojkov quoted in the reference section and the numerous figures in chapter 2 of the Ozone Trends Panel Report 1988 (WMO Ozone Project Report # 18, Vol. I) which he has contributed.
3.8 Conclusions to Section III
In summary, comparison of station data with some form of reference is a vital part of the re-evaluation process. Ideally, comparisons should be performed after a critical analysis of the station data, but it may also be the original means by which discrepancies in the data are identified. The four main comparisons that can be made are with mean annual and monthly data, with meteorological quantities, with another ground station, and with satellite data. once a comparison has been made, an interactive process is started, in which data are compared, corrected, and recompared. Setting up the necessary data sets and computing routines to do this is time consuming. However, once the initial data collection and programming are completed, iteration can be straightforward and relatively rapid.
It is hoped that in the coming few years all stations having long-term records will respond to the repeated calls by the WMO Executive Council, by the International Ozone Assessment Panels, and by the International Ozone Commission and will thoroughly re-evaluate their own data and thus contribute to the creation of a homogenized ozone data set so much necessary nowadays for detection of ozone trends as well as for verification of the implementation of the recommendations made by the amended Montreal Protocol.
References to Section III