A comparison of total column ozone data retrieved from the Global Ozone Monitoring Experiment (GOME), the Tiros Operational Vertical Sounder (TOVS), and the Total Ozone Mapping Spectrometer (TOMS) for the years 1996, 1997, 1998, and 1999 is presented. A statistical analysis and a spatial difference analysis were performed on a range of temporally and spatially averaged datasets. An analysis of globally averaged column ozone values showed a consistent offset with TOVS and TOMS values being consistently higher than GOME by ∼10–13 DU averaged over the 4-yr period. A 4%–5% drift was noted between the years 1996/97 and 1998/99 in the magnitude of the offset. The drift was identified as an increased offset of +25–30 DU between TOVS/TOMS and GOME occurring over latitudes above 70°N during 1998/99, and is a result of TOVS/TOMS ozone columns being higher during 1998/99 than in 1996/97.
Seasonal and latitudinal trends were noted in the global differences. In particular, TOMS and TOVS ozone values are consistently higher than GOME in the Southern Hemisphere from 30°–90°S. TOVS and GOME ozone columns show good agreement between 20°S and 20°N, with TOMS values approximately 10–15 DU higher than both TOVS and GOME in the same region. All three sensors show reasonable agreement between 20° and 60°N. However, there is no agreement above 60°N, where TOVS columns are higher than TOMS columns that in turn are higher than GOME columns. Results from a spatial difference analysis indicated further differences between GOME and TOVS ozone values that were not obvious from the global or latitudinal analysis owing to cancellation effects, including an area over Indonesia where GOME columns are higher than TOVS columns.
Satellite monitoring of the earth’s atmosphere is an important area of scientific research. Former, current, and planned instrumentation offers a formidable array of sensors, each one contributing substantial amounts of spatial and temporal information on physical parameters such as temperature, pressure, and trace gas concentrations. An important aspect of having such diversity is cohesion and continuity between datasets. There is an exacting requirement for consistent long-term datasets of geophysical parameters for use in applications such as climate change, numerical weather prediction, and pollution monitoring.
One particular area where satellite monitoring has proven to be useful is in tracking changes in global atmospheric ozone. Reviews by Krueger et al. (1980) and Miller (1989) and references therein describe a chronological overview of instrumentation flown from 1961 to 1988. Since 1988, notable additions include the Halogen Occultation Experiment (HALOE) and Microwave Limb Sounder (MLS) (Miller 1989) and the Global Ozone Monitoring Experiment (GOME) (Burrows et al. 1999). The finite lifetime of a space-based instrument dictates that series of instruments are required in order to obtain long time series, such as those currently available from the Total Ozone Mapping Spectrometer (TOMS) (Heath et al. 1975) and Tiros Operational Vertical Sounder (TOVS) (Smith et al. 1979) series. However, potential instrument failure and possible incomplete series overlap suggests that, within experimental error, all datasets of ozone concentrations ought to be comparable should a requirement to merge one dataset with another ever arise.
Several comparisons have been performed to test the feasibility of merging data products retrieved from different series of instruments and show significant variations in sensitivity to the retrieved ozone column arising from fundamental differences in sensor technology, weighting functions, and retrieval methods. As an example, several investigations have been performed on comparing ozone column amounts derived from the TOMS and TOVS sensors (Chesters and Neuendorffer 1991; Lefevrè et al. 1991; Neuendorffer 1996; Engelen and Stephens 1997; Susskind et al. 1997). The main difference between the TOMS and TOVS instruments is the spectral region used for ozone detection. The TOMS instrument measures UV at six wavelengths between 308 nm and either 360 nm (EP TOMS) or 380 nm (Nimbus-7 TOMS). The TOVS primarily measures infrared and microwave emission. In addition to providing a self-consistent dataset of column ozone values, the merging of TOMS and TOVS data would be beneficial from the standpoint that TOVS is able to monitor certain regions of the atmosphere that are inaccessible to TOMS. For example, at high latitudes during polar winters, where insufficient sunlight dictates that TOMS cannot reliably detect atmospheric ozone concentrations.
A relatively new addition to atmospheric remote sensing is GOME, launched on ERS-2 in April 1995. The GOME instrument is an ultraviolet–visible–near-infrared spectrometer similar in heritage to TOMS, but with superior spectral resolution. The greater spectral resolution has allowed retrieval of several products not previously observed using space-borne instrumentation (Burrows et al. 1999). Since deployment, GOME has proven suitable for the generation of long-term datasets of trace atmospheric constituents, a series soon to be augmented with the launches of SCIAMACHY on ENVISAT (Burrows et al. 1988) and GOME-2 on METOP-1. (SCIAMACHY is a GOME-type instrument with an increased spectral range into the IR spectral region.)
In this paper, results from a comparison of GOME, TOVS, and TOMS total column ozone values are presented as a follow-up to a preliminary investigation by Rathman et al. (1997). The work was performed as part of the GOME Data Interpretation, Validation and Application project (Goede et al. 2000), a European Community FP4 initiative aimed at improving the accuracy of the primary GOME products (column values of ozone and NO2), and the development of more advanced GOME products (e.g., ozone profiles).
The aims of the work are twofold. First, to test the validity of merging ozone products derived from multiple platforms and second, to validate GOME ozone columns on a global scale. Validation of global ozone columns against full-coverage fields derived from other satellites or numerical models is likely to expose systematic errors more efficiently than through a comparison with spatially irregular ground-based measurements. The time series of data analyzed (4 yr) and size of dataset (see later) is larger than that used in previous intercomparisons of satellite ozone values (e.g., Chesters and Neuendorffer 1991), allowing one to assess any differences over a wide range of spatial and temporal scales.
2. Instrument details and retrieval algorithms
A review of the three satellite instruments and ozone retrieval algorithms is beyond the scope of this paper. However, for GOME readers are referred to reviews by Burrows et al. (1999) for details of the instrument concept, and a discussion of the GOME data processing chain, including retrieval algorithms, is given in a series of technical reports (DLR 1994, 1995, 1996). Following its launch, an extensive validation campaign was carried out before the GOME data products were released to the general scientific community. Details of this campaign, and more importantly the accuracy of the GOME ozone columns, can be found in ESA (1996). Briefly, the accuracy of GOME columns at solar zenith angles less than 70° is approximately 5%–10%. Above 75°, a pronounced decrease in accuracy was noted (ESA 1996).
A general overview of the instruments that comprise the TOVS can be found in the work of Smith et al. (1979). Several algorithms can be utilised for retrieval of ozone concentrations from TOVS radiance measurements. Engelen (1996) has published an extensive assessment of five such retrieval methods and includes a more advanced method based on his conclusions. The algorithm of Neuendorffer (1996) was used to derive the TOVS ozone values used in this investigation (Long 1999, personal communication). Neuendorffer (1996) included a validation of TOVS data against both ground-based ozone columns and TOMS data and concluded that the TOVS data provided an estimate of column ozone to within ±25 DU, as TOVS is only sensitive to changes in lower stratospheric ozone. A discussion of the TOMS instrument and retrieval algorithms (version 7) can be found in publications by Heath et al. (1975) and McPeters et al. (1996). The TOMS dataset has been extensively validated against primarily Northern Hemisphere ground-based measurements, and the errors are reasonably well understood (McPeters and Labow 1996;Fioletov et al. 1999). The TOMS data is included in the analysis as a third party reference to both TOVS and GOME.
Datasets were prepared for the comparison from three sources.
TOVS total column ozone values obtained from either the NOAA Web site (http://www.cpc.ncep.noaa.gov/products/stratosphere/tovsto/archive/grib), or direct from NOAA where no data was available on the Web site.
GOME level-2 total column ozone values obtained on CD-ROM direct from DLR. This data is in two forms. First, the original dataset designated as GOME-R1, and second, the first reanalyses of ozone data, designated GOME-R2.
TOMS ozone data obtained from NASA Web site (http://toms.gsfc.nasa.gov/).
A 1° × 1° grid was chosen for the comparison since the majority of TOVS data was obtained in such a format. The remaining TOVS data was provided in a polar grid format, and was converted to the 1° lat × 1° long grid using existing code provided by NOAA. The GOME level-2 ozone data is distributed per ground pixel and was interpolated from its original coordinates onto the 1° × 1° grid. Although reference ellipsoids were not available for the TOVS and GOME column ozone data files, possible transformation errors were assumed to be minimal on the 1° × 1° scale used, and were therefore neglected. The TOMS data were obtained in the usual 1° lat × 1.25° long format. However, as the TOMS data was not included in the spatial difference analysis, it did not need to be regridded onto a 1° × 1° scale.
For a maximum grid size of 90°S–90°N × 180°W–180°E, there is a total possible number of points for analysis of 65 160 per dataset. However, a consequence of interpolating GOME data onto the chosen grid was that not all grid points were utilized, and so TOVS data points corresponding to null values of GOME ozone data were disregarded in the analysis. The inability of the TOMS instrument to retrieve ozone concentrations from areas of zero sunlight also applies to GOME, decreasing the latitudinal range of the instrument throughout polar winters. In the analysis, on average, 60 500 data points were available per dataset for each statistical computation. The initial analysis involved summing and averaging global datasets, followed by a simple correlation study. The formula for calculating the correlation coefficients was taken from Spiegel (1961).
4. Results and discussion
a. Global analysis
The first analysis carried out was a simple averaging of global weekly datasets. The results of this analysis, for the period January 1996 to December 1998, are shown in Fig. 1. The datasets used to produce Fig. 1 had to be restricted as the TOMS and GOME instruments cannot obtain true global coverage at any specific time of the year for reasons detailed earlier. To this end, the instrument with the most limited latitudinal coverage was used as the standard, with the latitudinal range of the other datasets (including TOVS) limited accordingly. The results presented in Fig. 1 clearly show that, on average, ozone values retrieved from both TOVS and TOMS are consistently higher than those retrieved from GOME (R1 and R2). This relatively simple analysis is sufficient to show marked differences between the retrieved ozone datasets. A simple linear correlation study was performed on all datasets, and an average offset for the 4-yr period was obtained, the results of which are shown in Table 1. When comparing GOME-R1 and TOVS, an average offset of ∼14 DU is apparent over the total time period, significantly lower than the value of ∼35 DU determined by Rathman et al. (1997) in their preliminary study. This significant change in offset is attributed to a much smaller spatial sampling rate (approximately one order of magnitude less) used by Rathman et al. (1997). However, we note that although the magnitude of the offset between TOVS and GOME was overpredicted by Rathman et al. (1997), their overall results are consistent with this study.
One unexpected feature of the correlation analysis is the small increase in correlation coefficient accompanied by a small increase in the magnitude of the offset when comparing GOME-R1 data and GOME-R2 with both TOMS and TOVS. One would expect that the outcome of any reanalysis would be an improvement in data quality and accuracy. The small increase in the correlation coefficient would suggest an improvement in data quality, but the increase in total offset would suggest that there are still deficiencies in the GOME retrieval algorithm when compared to both TOMS and TOVS. In addition, inspection of Fig. 1 shows a marked increase in the offset between the GOME data and both the TOVS and TOMS datasets of approximately 4%–5% when comparing the period from January 1996 to December 1997 and the period January 1998 to December 1999. The reason for this increase is not obvious from the global analysis and requires a change of spatial scale.
b. Latitudinal analysis
The results from the global-scale analysis show clear differences in TOVS and GOME ozone column amounts, including a drift of 4%–5% between January 1996–December 1997 and the period January 1998–December 1999. For the latitudinal analysis, the global datasets were subdivided into 10° latitude bands. Figure 2 shows the results obtained from averaging the 4-yr datasets over each latitude band. Three main trends are observed.
TOMS and TOVS ozone values are consistently higher than GOME-R1 and GOME-R2 in the Southern Hemisphere from 30° to 90°S.
TOVS and GOME ozone columns show good agreement between 20°S and 20°N. TOMS values are approximately 10–15 DU higher than both TOVS and GOME over this region.
All three sensors agree between 20° and 60°N. Above 60°N, there is no agreement, and TOVS values are higher than TOMS values that are in turn higher than GOME values.
To further clarify these observed trends, Table 2 shows a comparison of all latitudinal bands per instrument per year. Although minor differences are seen in Table 2, the numbers presented confirm the above trends and allow for a quantitative representation of the observed differences. It is worth noting that, for the latter analysis, only complete years of data were included to ensure that a full annual range of values was available, thus avoiding distortion of the mean value. The drift noted between the years 1996/97 and 1998/99 is apparent in Table 2, and appears to be an increased offset of TOVS/TOMS column ozone relative to GOME for latitude bands above 70°N. This increased offset is a result of higher ozone columns above 70°N being retrieved from TOVS/TOMS during 1997/98 than in 1996/97. This trend was not seen in the GOME data. In summary, there are clear latitudinal variations to the observed differences between TOVS, TOMS, and GOME ozone column values.
To test for seasonal dependency, the datasets were averaged into 3-month (seasonal) blocks covering January–March, April–June, July–September, and October–December. One example of this analysis is shown in Fig. 3, for the period January–March averaged over all 4 yr. It is apparent that two of the trends observed previously, low readings from GOME in the Southern Hemisphere and high TOMS readings over equatorial latitudes are prevalent during this particular season. However, the apparent disagreement between Northern Hemisphere values is not observed and all three datasets appear to agree very well.
The offset between Northern Hemisphere values is, however, apparent when results from all four seasonal analyses, presented in Tables 3–6, are analyzed. Inspection of Tables 3–6 suggests that all three trends detailed above are present from April to September each year, with the offset between all datasets in the Northern Hemisphere observed from March to December each year. The observed low ozone readings from GOME across the Southern Hemisphere are not present during October–December, during the formation and destruction of the Antarctic ozone hole. The seasonally dependant differences between the retrieved ozone columns analyzed here, relating to ozone columns retrieved from GOME, are likely to be caused by inaccuracies in the 3-monthly a priori climatologies used in the current GOME ozone retrieval algorithms (DLR 1994, 1996).
The conclusions drawn above were further confirmed by subtracting individual datasets prior to averaging. One such example, for TOVS minus GOME-R1 during 1996, is given in Fig. 4a. In Fig. 4a, the thick line is used to represent the average offset over the entire year, and the thinner lines are included to show the wide variation in offset seen at different times of the year. The results presented in Fig. 4a clearly show a trend where TOVS readings are higher than GOME-R1, indicated by the positive offset of the mean value. In addition, the two sensors are in agreement around the tropical latitudes and diverge toward both poles. A similar analysis has been carried for TOMS minus GOME-R1 and for TOVS minus TOMS, the results from which are presented in Figs. 4b and 4c.
The temporal variation of TOVS and GOME ozone column values for the period January 1996–December 1998 are shown in Figs. 5a and 5b, respectively. The data are prepared from weekly averaged ozone datasets, and are shown for a 1° latitude scale. To identify systematic and random differences, Fig. 5c contains a difference analysis of Figs. 5a and 5b obtained by simply subtracting GOME ozone values from TOVS ozone values. In Fig. 5c, positive values refer to points where TOVS ozone values were higher than GOME readings, with negative differences indicating points where GOME ozone values were higher than TOVS ozone values. Features apparent in Fig. 5c include:
increased ozone values from TOVS over Northern Hemisphere high latitudes (>60°N) from March to December each year and
increased ozone values from TOVS over Southern Hemisphere high latitudes (>60°S) from January to October each year.
These features confirm the apparent high ozone readings from TOVS when compared to GOME in each hemisphere, including the notable seasonal effect.
The current assessment of the GOME retrieval algorithm suggests that GOME total column ozone column values are thought to be accurate to 5% for solar zenith angles (SZAs) less than 70° and 10% for a SZA greater than 70° (ESA 1996). We would, therefore, expect to see differences between GOME and TOVS in high SZA regions, and it is likely that part of the observed differences in Fig. 5c could be attributed to a SZA dependence on the GOME ozone readings. However, this simple comparison analysis is not sufficient to quantify the effect of the SZA dependence as there is also a seasonal dependence related to GOME, and there are undoubtedly inaccuracies in the TOVS retrievals. Work is currently under way in our laboratory to quantify the observed differences through a systematic analysis of the retrieval algorithms. This analysis will be needed to fully understand the high variability observed between TOVS, TOMS, and GOME ozone columns.
c. Local-scale analysis
An alternative method to understand the differences between GOME and TOVS ozone values is to employ a spatial difference approach to identify differences on a local scale. Reasonable agreement was observed for certain conditions in the global and latitudinal analysis presented so far, for example, TOVS and TOMS over the Southern Hemisphere and TOVS and GOME around the equator. However, it is highly likely that cancellation of errors causes some of the observed agreement. As an example, a set of spatial difference maps is presented in Fig. 6, for July 1996, 1997, and 1998. Positive readings in Fig. 6 refer to points where TOVS readings are higher than GOME readings, and negative readings refer to points where GOME readings were higher than TOVS. Differences corresponding to ±20 DU have been removed, corresponding to an average error of ±10 DU per instrument. Consistent features observable in Fig. 6 include the following.
A band of high TOVS ozone over Northern Hemisphere high latitudes from 60° to 85°N. The feature is similar in magnitude in all 3 yr presented.
A band of high TOVS ozone from 35° to 60°S, consistent in shape over all 3 yr but with a higher maximum in 1997 and 1998 compared to 1996.
High TOVS ozone over the Sahara, increasing yearly from 1996 to 1998.
A region over part of the Indian subcontinent and parts of Indonesia where GOME readings are higher than TOVS readings by up to 40 DU. This feature is not consistent in shape, and has a high spatial variability.
The observation of high TOVS readings in high-latitude regions in both hemispheres confirms the earlier conclusions. However, to a first approximation, the high TOVS ozone over the Sahara region and the high GOME ozone over the Indian subcontinent will have canceled out during the global and latitudinal averaging due to the close proximity of the latitudes of the two areas.
The observation of high TOVS ozone values over the Sahara region was apparent in the analysis of nearly every month in this investigation, and has been observed before (Neuendorffer 1996; Susskind et al. 1997). It was suggested that this discrepancy in ozone values was due to an incorrect surface emissivity for desert region being used in the retrieval algorithm. An alternative theory for high TOVS readings over the Sahara is provided by Engelen (1996), who noted that all available TOVS retrieval algorithms did not properly account for stratospheric aerosols. A possible explanation for the high GOME readings is an increased tropospheric column of ozone resulting from biomass burning. For example, Burrows and coworkers (Burrows et al. 1998; Ladstätter-Weißenmayer et al. 1999) have reported the observation of excess tropospheric ozone over Indonesia. The TOVS retrieval algorithm of Neuendorffer (1996) adds a nominal tropospheric column to a retrieved stratospheric column. It is unlikely, therefore, that sudden increases in tropospheric ozone are correctly accounted for in the TOVS retrieval algorithm.
Total ozone column data from the GOME instrument have been compared to data from the TOVS and TOMS instruments. A simple global averaging analysis indicated marked differences between the datasets. An offset of ∼14 DU was noted between TOVS and GOME, and an offset of ∼10 DU between TOMS and GOME over the period from January 1996 to December 1999. A latitudinal analysis of all datasets showed consistent offsets allowing several conclusions to be drawn, including the following.
GOME ozone values are apparently lower than both TOMS and TOVS in the Southern Hemisphere. The difference varies from ∼5 DU at mid-to-low latitudes and varies almost linearly to ∼35 DU at the South Pole.
TOMS ozone values are ∼10–15 DU higher than both TOVS and GOME values in the equatorial latitudes.
All three sensors offer different readings in the Northern Hemisphere. Both GOME and TOMS ozone readings appear to be lower than TOVS values, and appear to vary almost linearly toward the North Pole, where TOVS values are higher than GOME by ∼35 DU and are higher than TOMS by ∼20 DU.
A spatial difference analysis confirmed some of the features observed in the global and latitudinal analysis, and highlighted further features that were originally hidden by cancellation effects. During the project, the first GOME reanalysis data was made available, and was subjected to the same analysis as the original dataset. This showed slightly improved correlation with both TOVS and TOMS data but showed a bigger offset, the reasons for which are not yet understood. However, it is apparent that the differences in ozone column between GOME-R1 and GOME-R2 data has slightly improved the stability of the dataset but has not improved the absolute accuracy when compared to TOVS and TOMS ozone data.
The authors wish to thank Craig Long of NOAA for his assistance in obtaining TOVS ozone data in polar gridded format, and Richard Engelen for the copy of his Ph.D. thesis. Also, the authors wish to thank other members of the GODIVA consortium for informative discussions about the GOME satellite project. This work is supported by the European Commission RTD Programme on Environment and Climate, Contract ENV4-CT 97-0418.
Corresponding author address: Dr. P. S. Monks, Department of Chemistry, University of Leicester, University Road, Leicester LE1 7RH, United Kingdom.