1. Introduction
The analyses produced operationally for the high-latitude areas of the Southern Hemisphere are generally assumed to be poorer than those covering the tropical and midlatitude areas, since there are few in situ observations from the region and the analyses are therefore based largely on satellite sounder data (Bengtsson 1989; Hart et al. 1993; Bourke 1994). However, over the Antarctic continent itself there are difficulties in using sounder data (Lutz and Smith 1988; Lutz et al. 1990) and many forecast centers only use such profiles at stratospheric levels. Although there was a considerable amount of research into the quality of Antarctic analyses during and immediately after the International Geophysical Year of 1957–58 (Taljaard 1967), there has been little activity on this subject since that time. However, the Antarctic First Regional Observing Study of the Troposphere (FROST) project1 (Turner et al. 1996) organized by the Scientific Committee on Antarctic Research has recently been a focus for studies into the meteorology of the Antarctic, and data collected during the FROST special observing periods (SOPs) allow an assessment to be made of the current status of analyses and forecasts distributed by the main weather services over the Global Telecommunications System (GTS). It is also possible to consider how non-GTS observations and novel forms of satellite data can improve the analyses.
In this paper we examine the operational analyses produced by the major numerical weather prediction (NWP) centers for the region south of 50°S [areas referred to are shown in Fig. 1; the locations of the research stations can be found in Colwell and Turner (1999, this issue)] during the winter SOP of July 1994 and compare these fields with each other and with improved, reanalyzed fields produced manually when late data and additional, high-resolution satellite imagery were available. The two quite distinct problems of analysis over the ocean areas and over the continent are considered separately since while over the ocean there are many conventional frontal depressions, over the interior of the Antarctic few lows are found and the satellite imagery and the analyzed fields are difficult to reconcile on many occasions.
The assessment of the analyses is undertaken using two approaches. First, the experiences of meteorologists who carried out the reanalysis of the fields in Hobart are summarized and their subjective assessment of the operational charts discussed. Second, more objective data on the analyses are presented via the difference fields between monthly mean fields produced using the analyses from the main centers—and between individual analyses. Finally, we make recommendations on the value of late non-GTS data and the new forms of satellite data that are becoming available.
2. The FROST reanalysis procedure
The goal here was to produce high quality surface and upper-air analyses taking the U.K. Meteorological Office (UKMO) hand-drawn charts (UKMO-HD) as the starting point and redrawing them in the light of late data and satellite imagery that was not available in real time. The UKMO-HD charts formed the basis of our work since they had scatterometer winds plotted. Different procedures were adopted for the reanalysis over the ocean areas and over the Antarctic continent itself.
Over the ocean, the pressure at mean sea level (PMSL) chart was the starting point for the reanalysis and these charts were redrawn for each day of the SOP at 0000 and 1200 UTC. Data used in the reanalysis consisted of surface synoptic observations (SYNOP), ship observations, drifting buoy data, and automatic weather station (AWS) observations. In order to make sure that as complete a set of observations as possible were compiled, two nodes of the GTS at Cambridge, United Kingdom, and Hobart, Australia, were used to collect data during the SOPs. As discussed in Turner et al. (1996), both locations received about the same number of SYNOP observations, but many of these were not received at the other site. When the data from both locations were merged, the new dataset contained about 30% more data than was received at either location. This merged dataset was then further supplemented by non-GTS observations received from contacts in the agencies who are responsible for operating the national meteorological observing programs in the Antarctic.
The reanalyzed surface charts had frontal positions indicated; these were based on satellite data (imagery and TIROS-N Operational Vertical Sounder thickness fields) as well as the available surface data. The satellite imagery available for the whole month included polar stereographic composite images covering the area south of 40°S, produced every 3 h by the University of Wisconsin—Madison (Stearns et al. 1995), together with imagery from the geostationary satellites GMS, Meteosat, and GOES-West, which was of value at lower latitudes. The charts produced from this initial reanalysis are referred to as FROST-1 analyses. It became clear during the preparation of these analyses that the satellite imagery was of great importance in the production of the surface fields so efforts were made to acquire high-resolution imagery to further refine the analyses over the ocean areas. Because of the large amount of effort required to process and utilize the high-resolution imagery, only one week of analyses, covering the period 22–28 July 1994, were redrawn for a second time and further refined over the Eastern Hemisphere, from Greenwich to 180°. This series of charts is here called FROST-2, and utilized imagery from the U.S. Defense Meteorological Satellite Program (DMSP), which had a horizontal resolution of 2.7 km, along with extra AWS data from Madison and 12-km resolution quick-look imagery from Casey Station (66.3°S, 110.5°E).
Upper-air height charts for the 500- and 250-hPa levels were prepared by converting the PMSL analyses to 1000-hPa heights and integrating them with 1000–500- and 500–250-hPa thickness fields produced from the temperature sounding (SATEM) thickness available over the GTS and additional radiosonde ascents received after the SOP. The means by which the SATEM profiles were produced during FROST is described by Burdsall Brown (1993).
Over East Antarctica, the main reanalysis effort was directed toward producing improved 500-hPa height fields, since this pressure surface is everywhere above the topography of the continent. For the reanalysis, spot heights were produced using the method of Phillpot (1991). His technique produces an estimate of the 500-hPa height using a model height as a first guess and adjusting it in the light of the surface pressure and temperature from an AWS. The surface data provide information on the temperature profile above a location and radiosonde data from the International Geophysical Year have been used to develop regression relationships providing adjustments to the 500-hPa height once the surface conditions are known. Over West Antarctica a modified version of the original technique was used to produce improved 700-hPa height fields.
Once the revised analyses had been prepared for the oceanic and continental areas the two sets of fields were merged to give 500- and 250-hPa fields for the whole area south of 50°S and PMSL fields northward of the Antarctic coastal region. The revised fields are currently being digitized and will be made available in gridded binary (GRIB) code format.
3. The NWP analyses
Numerical analyses for every 12 h throughout the SOPs from four NWP centers are available to the project, as follows.
The UKMO 19-level model, which has a horizontal grid length of about 90 km. In the following fields from the UKMO numerical analysis–forecast system are referred to as UKMO-NP.
The U.S. National Centers for Environmental Prediction (NCEP) 28-level, global spectral model with a triangular truncation at wavenumber 126, which is equivalent to a horizontal spatial resolution of about 100 km. Both the aviation analyses available via the GTS and the medium range forecast (MRF) analyses, which have a later cutoff time, are available to FROST.
The Australian Bureau of Meteorology Global Assimilation and Prediction (GASP) model, which has a rhomboidal truncation R53 and 19 levels.
The European Centre for Medium-Range Weather Forecasts (ECMWF) model, which has 31 levels in the vertical and a triangular truncation of T213.
All the models are dependent on GTS data in order to produce realistic analyses for the Antarctic. However, there were some significant differences in the way that the analysis systems operated and the data that they included. For example, the ECMWF and GASP systems have a data cutoff time of 7 h so that they usually contain most of the Antarctic observations on the GTS. Also, during SOP-1 in July 1994, the GASP model did not have the surface wind vectors that were generated from measurements from the scatterometer on the first European Remote Sensing satellite (ERS-1).
Although the models have different horizontal resolutions, the analyses and forecasts are all distributed over the GTS in the standard GRIB code format, which has a resolution of approximately 250 km.
4. Analyses over the ocean
a. Subjective assessment
During the preparation of the FROST-1 and FROST-2 revised charts the analysts gained a great deal of insight into the quality of the operational analyses and their strengths and weaknesses. In this section we summarize some general conclusions regarding the state of the products distributed over the GTS and present several examples to illustrate cases of note. As the analysts were taking the UKMO-HD surface charts and modifying these in the light of additional data, they were therefore most conscious of discrepancies in those charts. However, throughout the reanalysis exercise hardcopy versions of the GASP analyses were also consulted so that a picture was built up of the performance of that system too. Because of the lack of data over the ocean areas, even when all the late data had been collected, the satellite imagery was of paramount importance in determining the frontal locations and the positions of the vortex centers.
Much of the analyst’s time was spent considering the major synoptic-scale depressions over the Southern Ocean and their associated frontal systems. Over the whole of SOP-1 it was found that no large (diameter of several thousand kilometers) depressions were missed from the operational analyses. This was no doubt due to the availability of SATEM profiles across the ocean that provide the broadscale thermal structure of the atmosphere. In addition, the analysts at Bracknell can usually detect lows missed from the analyses on satellite imagery and then generate pseudo-observations to be inserted into a later run of the analysis system. Therefore, the main discrepancies in the analysis of depressions usually consisted of errors in the depth of the systems, location errors, and failures to resolve the complexity of the lows. Overall, the central pressures of the lows were generally handled better than the locations of the centers, but some major differences between the operational analyses did occur. For example, at 1200 UTC 6 July 1994 the NCEP analysis had a low in the northern Amundsen Sea with a central pressure of about 964 hPa (Fig. 2a). This is at variance with the UKMO-NP and GASP analyses, which had central pressures of 948 and 945 hPa, respectively (Figs. 2b and 2c). In this case there were also large differences in the location of the low in the difference analyses suggesting problems in resolving the situation using only SATEM observations.
It was found that during the preparation of the FROST-2 analyses some deepening or filling of the analyzed central pressures of the lows was necessary based on late AWS observations and data from coastal stations; however, these corrections were rarely greater than several hectopascals.
While the central pressures of the lows were found to be generally correct, the locations of the systems were often adjusted during the reanalysis process in the light of satellite imagery. An example of this occurred at 0000 UTC 23 July 1994 when the UKMO-HD analysis had a low near 63°S, 150°E in a region devoid of data (Fig. 3a). This analysis seemed to be supported by a swath of scatterometer winds to the north of the center. Inspection of the satellite imagery close to the analysis time showed clearly that the center of the vortex was in fact located at 60°S, 148°E and was analyzed at this position on the FROST-2 chart (Fig. 3b). This repositioning of the low center had a marked effect on the pressure gradient and implied much stronger winds to the northeast of the low center. Earlier research (Troup and Streten 1972; Carleton 1987) has shown that the relationship between the center of the cloud vortex and the level at which the pressure minimum occurs varies according to the stage of development of the cyclone. This must be borne in mind when reconciling the satellite imagery and the surface pressure field.
One change made repeatedly to the surface charts during the reanalysis process was to increase the complexity of the synoptic-scale lows over the Southern Ocean. Whereas on the initial operational charts large, single-centered lows were often analyzed, inspection of the satellite imagery revealed that these systems were often multicentered or had mesocyclones embedded within their circulations. This complexity of the flow over the Southern Ocean has been found in other investigations (Carleton et al. 1995). One case that illustrates the degree to which the operational analyses are often too simplistic occurred at 0000 UTC 25 July 1994. The GASP analysis for this time, which is shown in Fig. 4a, had a large area of low pressure centered close to 62°S, 100°E with a central pressure of 952 hPa. The analysis indicated a trough of low pressure stretching to the northwest in which the pressure gradient was very slack. The DMSP high-resolution imagery for this time (Fig. 4b) is dominated by a band of frontal cloud extending from the southern side of the low, extending around the eastern side of the system and then running northward. However, the complexity of the cloud in the center of the low is apparent and there are indications of a number of centers, as well as several frontal cloud bands. The final (FROST-2) analysis (Fig. 4c) had the main low pressure center slightly farther to the east, although the central pressure was unchanged at close to 952 hPa. However, the main change made to the analysis was the removal of the trough to the northwest of the main center, resulting in moderate southerly flow rather than the weak westerly wind suggested by the original analysis. In addition, the main low now has a double center and two minor centers have been added to the northeast of the main low. Also, the frontal structure was considerably revised when compared to the original UKMO-HD analysis in light of the high-resolution satellite imagery. As in a number of other cases, the fronts on the UKMO-HD analysis seem to have been moved forward by continuity, no doubt because of a lack of satellite imagery at the time the analysis was prepared.
A further situation where the operational analyses were seriously in error in resolving the details of a complex low occurred at 1200 UTC 22 July 1994 near 55°S, 115°E. The UKMO-HD analysis for this time is shown in Fig. 5a and indicates a complex area of low pressure with three centers near 60°S, 117°E; 52°S, 117°E; and 57°S, 132°E. When the Madison composite image was examined (Fig. 5b) it showed clearly that the two more northerly vortices were substantial frontal depressions but that there was no indication of the more southerly low. The final (FROST-2) analysis is shown in Fig. 5c and indicates the substantial changes made to the initial field with the southernmost low having been removed and replaced with an easterly flow.
While the satellite imagery was vital in improving most of the analyses, there were some occasions when the late surface data showed a vortex to be present but there was no cloud at all in the imagery. This was the case near Possession Island (72°S, 171°E) at 1200 UTC 26 July 1994, when the AWS at that location indicated a surface low to be present but there was no indication on the operational charts of this feature (Fig. 6a). When the Possession Island AWS PMSL value of 986.1 hPa was included, the surface chart changed dramatically, as can be seen from Fig. 6b. The imagery for 2300 UTC 26 July (Fig. 6c) shows very little cloud over the area, indicating that the atmosphere here was very dry.
A similar situation with a low on the charts and no well-defined cloud vortex was often found to occur with large occluded systems near the Antarctic coast. Such systems often have low central pressure and strong winds within their circulation. This was the case with a low near 67°S, 32°E at 0000 UTC 22 July 1994 (Fig. 7a). This system had a central pressure of less than 956 hPa but, as can be seen in Fig. 7b, only a disorganized mass of cloud. So the front appears to be incorrectly analyzed and according to the satellite imagery there is a region of cyclonic vorticity advection at 60°S, 35°E, where at least a well-defined surface trough should be analyzed.
In the course of the preparation of the FROST-2 analyses many changes were made to the fronts on the UKMO-HD charts as a result of the availability of the high-resolution DMSP imagery. Also, the fields of rain rate and cloud liquid water produced from the Special Sensor Microwave/Imager passive microwave imagery (Dalu et al. 1993; Goodberlet et al. 1989) were of particular value and also indicated which fronts were particularly active.
During the preparation of the upper-level charts a number of changes were made to the thickness fields that implied modifications to the thermal structure. Changes made included modifying the trough orientations and positions and, in a small number of cases, repositioning thermal ridges near to the Antarctic coast. Care was taken to reconcile the frontal locations as determined from the imagery and the thermal field.
Earlier studies (Turner and Thomas 1994; Carleton et al. 1995) have shown that few mesocyclones (depressions with a diameter of less than 1000 km) appear on the operational surface analyses. However, the satellite imagery available for FROST, and particularly the high-resolution DMSP imagery, showed many systems throughout the month, although, as to be expected, these were not resolved explicitly on the FROST analyses. One case of a mesocyclone with a diameter of less than 300 km is shown in Fig. 8a. The FROST-2 chart for the same time in Fig. 8b has the mesocyclone marked, but the pressure perturbation associated with this small low cannot be resolved with the data available.
As the zone of maximum frequency of anticyclones in the Southern Hemisphere is located north of 45°S throughout the year and migrates northward in winter (Taljaard 1972), traveling anticyclones are not regularly encountered in the FROST analysis region (i.e., south of 50°S). However, blocking anticyclones are not uncommon features in the higher latitudes and the frequency of occurrence of blocking in the hemisphere reaches a maximum in the Tasman Sea and New Zealand region during winter (Wright 1974; Lejenäs 1984; Trenberth and Mo 1985). According to the Australian Bureau of Meteorology Climate Monitoring Bulletin for July 1994, blocking activity as measured by a blocking index was below average in the Australasian region during SOP-1 but was above average near the Antarctic Peninsula and South America.
It is apparent from the FROST-1 analysis set that the blocking in the Peninsula sector developed from a ridge of high pressure extending southward from the western Pacific Ocean to link with the Antarctic ridge near the Ross Sea early in the month. The system continued to intensify as it migrated slowly eastward and by mid-July an anticyclone with central pressure above 1020 hPa had become established south of 60°S in the eastern Pacific. There was close agreement among the various analysis schemes on the position and central pressure of the high and the GASP mean sea level pressure analysis for 0000 UTC 15 July 1994 (Fig. 9) gives a typical treatment of the system. Scatterometer data available on some of the UKMO-HD analyses gave added confidence to the location of the system, as did the satellite imagery, which showed a region of little cloud. The anticyclone maintained its identity until 21 July, after which it gradually dissipated.
At the beginning of the FROST-2 analysis period (22 July) there was a weak high pressure center in the Atlantic sector and ridges near Australia and in the eastern Pacific. The Australian ridge moved slowly eastward during the reanalysis period and intensified as it became established in the western Pacific sector. A ridge from this anticyclone extended southwestward over George V Land toward the end of the FROST-2 period. Frontal bands approaching this ridge were observed to slow and weaken while several associated cyclonic centers subsequently migrated over the Antarctic continent (Pook and Cowled 1999, this issue).
The other region of significant anticyclonic activity in the FROST-2 period was in the vicinity of the Antarctic Peninsula. Here, an elongated lobe of the Antarctic anticyclone became apparent on 24 July and intensified to about 1010 hPa as it formed a high pressure cell approximately 24 h later. This anticyclone persisted in the region until 28 July when it rapidly dissipated. The PMSL situation for 0000 UTC 27 July 1994 is shown in Fig. 10.
Notwithstanding the number of alterations made to the positions of cyclonic systems between FROST-1 and FROST-2 on the PMSL analyses the positions of the anticyclones were largely unchanged. Partly, this can be explained by the absence of significant highs over the Indian Ocean and Australasian sector where the FROST-2 reanalysis was carried out and the tendency for slow-moving blocking highs to form at these latitudes. Nevertheless, the overall impression of the analysts was that the operational analyses had performed well in their treatment of anticyclones.
Many changes were made to the surface analyses between the preparation of the FROST-1 and FROST-2 analyses as a result of more data and improved imagery being available. The changes made to the synoptic-scale vortices located south of 45°S are summarized in Table 1. The additional data show the effect of the complexity of the lows toward the end of the “special week” with many new centers, both mesoscale and synoptic scale, making an appearance. Several centers were observed to move inland and were detected by AWS (e.g., dome C on 27 July when a drop in the surface pressure of approximately 3 hPa was recorded; for a full discussion of this case see Pook and Cowled 1999). Naturally, these centers were not analyzed in the first analysis set. The number of lows found to have been relocated by a large distance (>500 km) totaled 4 in an overall total of 161 on 14 charts, suggesting that the corrections to the positions of the centers was fairly small.
Many differences were noted between the analyses produced by the different centers on individual days, but there were some areas where particular problems were found. For example, the South Pacific, which is a region devoid of manned stations in the coastal region, was particularly poorly handed and this problem is discussed in the following section.
b. Objective assessment
The ability of the UKMO-NP, GASP, NCEP, and ECMWF models to represent correctly the mean conditions during July 1994 can be assessed from the mean PMSL fields for the month, which are shown in Fig. 11. It should be noted that a valid intercomparison can only be made over the ocean areas since the models use different methods for the estimation of PMSL under the Antarctic ice sheet from data at the lowest model levels. Across most of the Antarctic the surface, and therefore the lowest model level, is 2 km or more above mean sea level so that different methods for determining parameters at mean sea level, such as PMSL, can have a huge difference on the computed fields.
The mean fields in Fig. 11 show that all four models had the same broadscale structure for the PMSL field with a similar three-centered circumpolar trough in the Antarctic coastal zone. It should be noted that the July 1994 mean PMSL fields over the oceans resemble (in general form) the longer-term midwinter pattern. The mean low pressure centers close to 100° and 30°E had similar locations in all four models, although the central pressures were about 4 hPa lower in the UKMO-NP and NCEP fields. However, the low center in the vicinity of the Ross Sea showed greater differences between the models. The ECMWF, UKMO-NP, and NCEP models had this low in a similar position over the central Ross Sea and the central pressures were also similar at a little under 988 hPa. However, the GASP model had the low center displaced eastward to over the Amundsen Sea and the central pressure was well below 984 hPa. North of the circumpolar trough the analyses become much more similar, probably as a result of the greater volume of data available to the assimilation schemes.
The ability of the models to represent the position and depth of the circumpolar trough can be further examined via the zonal average of the PMSL for the month of July, which is shown in Fig. 12. All four models have very similar PMSL values for the deepest part of the trough and similar locations for the trough. North of the deepest part of the trough, the GASP model had slightly lower PMSL values than the other three models, although the difference is only a fraction of a hectopascal and north of 50°S all the models have very similar values. The mean PMSL values south of about 70°S are very different because of the different means used to reduce the data at the lowest model level to mean sea level.
The rms difference between the daily PMSL fields from the GASP and NCEP analyses (Fig. 13) gives some indication of how the analyses differ across the Southern Ocean. The differences are mostly less than 1 hPa, but with some small areas above this value close to 70°S, near 20° and 120°E. These areas are within the latitude where the circumpolar trough is deepest so that errors in the position and depth of depressions would be expected to give the largest differences. However, the largest rms difference was over the Amundsen and Bellingshausen Seas between the longitudes of 120° and 150°W. In this sector of the Antarctic there are no research stations on the coast, although there is an AWS at Mount Siple (73.2°S, 127.1°W), which provides some surface data over the interior of the Antarctic. The production of operational analyses over the ocean area is therefore dependent on SATEM data. Relatively data-rich areas, such as the Antarctic Peninsula, have only very small differences, indicating that the observations from stations in these areas are reaching the analysis centers in a timely fashion and are being used effectively. If we examine the mean difference (GASP − NCEP) between GASP and NCEP monthly mean PMSL fields (Fig. 14), we can again see very small differences along the Antarctic Peninsula, but some large differences over the ocean areas. The largest differences are again at the latitude of the circumpolar trough, with smaller values on the coast, where observations are available from the stations. In the Eastern Hemisphere, the differences are generally positive (GASP values higher), possibly suggesting differences in the way the SATEM observations are used.
The largest absolute differences are found in the Amundsen and Bellingshausen Seas area where the GASP field has values some 11 hPa below that of NCEP, with the greatest difference next to the coast. Examination of the daily difference fields for the month indicated that this large deviation was mainly the result of very large differences over the period 2–6 July. Throughout this time the GASP analyses had lower pressure in this region than the NCEP fields, with the maximum difference being 42 hPa at 0000 UTC 5 July (Fig. 15). Determining what is “truth” in such a data-sparse area is not easy, although the satellite imagery does allow the locations of the main low pressure centers to be determined with some confidence because of the associated cloud bands. Examination of the imagery for this period would suggest that close to the location of maximum difference (70°S, 135°W) pressure was high, as no cloud was present and the leads could be seen in the sea ice. This high pressure region was reflected in the UKMO-NP analyses, which had a ridge in the area, and the NCEP analysis, which had a closed high with central pressure of greater than 1010 hPa. The GASP analysis, on the other hand, had a trough of low pressure in this area extending southward from lower latitudes and pressure of below 970 hPa. It would therefore appear that the GASP analysis was in error on this occasion. An examination of the SATEM messages that went into the GASP assimilation scheme over the period showed that it had the same observations that went into the NCEP model, suggesting that the differences were a result of the use that was made of the data.
The mean 500-hPa height fields for July 1994 as produced by the four centers are shown in Fig. 16. These all show a similar three-wave pattern with troughs at approximately 100°E, 20°E, and 150° W. The field produced by the GASP model is again slightly different from the other three in having the Amundsen Sea trough slightly farther to the east, reflecting the analysis problems at the start of the month discussed earlier.
5. Analyses over the continent
As discussed by Hutchinson et al. (1999, this issue), the analyses over the continent were prepared using the technique of Phillpot (1991), with the main chart being prepared over West (East) Antarctica being at 700 (500) hPa. These levels were selected so that they were above the highest point of the orography in each area. Over the continent the preparation of the charts was reliant on the AWS observations. These indicated the rapid changes in conditions that are found over the plateau on a wide range of timescales. During SOP-1 the AWS surface pressure observations from East Antarctica sometimes indicated broadscale changes taking place, such as a major drop in pressure from 1 to 5 July. At other times much more localized changes were observed in the pressure data, suggesting that they were the result of small-scale weather systems or topographic effects. Since some of the AWS observations are put onto the GTS and assimilated into the NWP systems, the models should be able to reproduce some of the larger-scale synoptic events that take place on the plateau. However, not all the AWS observations are available on the GTS so that the reanalysis of the continental area using the full post-SOP dataset produced a much more detailed picture of the synoptic-scale activity on the plateau. An example of this is shown in Fig. 17 through the operational UKMO-NP 500-hPa height field for 0000 UTC 5 July 1994 and the revised analysis for this time based on using the Phillpot technique. The UKMO-NP field indicates that an elongated trough extended from the Pole along 90°E to the coast. A low was also present in the coastal area near 307°E while a major upper ridge was pushing into the interior between 150°E and 180°. The revised analysis in Fig. 17b has the same basic flow pattern but provides much more detail over the high plateau area. Rather than there being just a weak trough along 90°E, the revised analysis suggests a low with values of less than 468 dm at 83°S and two minor lows on either side of the 90°E meridian. During the reanalysis process great care was taken to ensure that the minor lows and highs included in the charts were not just spurious or transient features but could be followed between analyses. In fact some of the lows identified could be observed on the satellite imagery, indicating the value of using the Phillpot technique to construct the upper-air charts.
6. The value of various forms of data used in the reanalysis exercise
The reanalysis process carried out in Hobart allowed the value of the operational GTS data and the observations collected subsequently to be assessed for their contribution to the production of the high-quality analyses. Over the ocean areas perhaps the most valuable form of data was the high-resolution DMSP imagery, which allowed complex low pressure systems to be resolved and analyzed via their cloud signatures. On many occasions it was necessary to add additional centers to the large, rather featureless areas of low pressure on the operational analyses to make them consistent with the imagery. It was therefore found that in the comparison of the three sets of analyses consisting of the UKMO-HD operational fields, the analyses produced with the additional non-GTS data, and the analyses from the special week where DMSP imagery were available, that the fields for the special week were by far the most detailed and, in our opinion, closest to the truth.
The ERS-1 scatterometer surface wind vectors are an important form of data that have become available only recently. During the FROST SOPs these data were included in the UKMO-NP assimilation scheme and, as these fields formed the basis of our reanalysis exercise, their impact was felt throughout the production of the two subsequent sets of analyses. It is difficult to judge the impact of these wind vectors; earlier studies using winds from the Seasat scatterometer have pointed to them having a positive impact (Ingelby and Bromley 1991), while Hoffman (1993) suggests that ERS scatterometer data generally have a neutral impact on ECMWF forecasts but with greater impact over the Southern Ocean than elsewhere. However, the GASP assimilation system in use during the SOPs did not use the scatterometer winds and comparison of the UKMO-NP and GASP analyses did not generally suggest any major differences in the regions where the scatterometer data were used.
The assessment of the SYNOP and radiosonde data available on the GTS and available directly from the operators of the Antarctic stations (Colwell and Turner 1999) indicates that most of the observations are reaching the GTS in a timely fashion, so it is no surprise that the analyses prepared with the additional non-GTS observations have only small improvements over the operational products disseminated in real time. However, when the pressure and wind observations from the AWS data on the GTS were compared with the quality controlled AWS data obtained directly from the University of Wisconsin—Madison, it was found that a number of errors were present in the GTS data. This is not surprising considering that the AWS instruments are subject to very harsh conditions and the data have to be disseminated rapidly.
The value of high-quality AWS data on the GTS is illustrated by the case of the data from the Mount Siple AWS on 26 July. The pressure from this station was not on the GTS, and the GASP model had an error of about 5 hPa from the station value. In the FROST-1 reanalysis we did not have the Mount Siple AWS data and the redrawn chart was seriously in error as the depth of a low was underestimated. When the Mount Siple data became available it was found that the low in the coastal area should have been some 20 hPa deeper than drawn.
As discussed above, the AWS observations from over the continent were of great value in preparing the upper-air field using the Phillpot technique. But it was also found that the AWSs on islands around the Antarctic were of great value too, providing valuable data in very data-sparse regions.
7. Discussion and conclusions
During the FROST project a great deal of assessment of the operational analyses was undertaken both as part of the reanalysis exercise and during the examination of the operational NWP products distributed on the GTS. This allows some conclusions to be drawn regarding the current status of meteorological analysis south of 50°S.
Overall, the project found that the operational forecast systems were capable of resolving most of the major synoptic-scale weather systems around the Antarctic, but that significant differences could occur in data-sparse regions, such as the Amundsen and Bellingshausen Seas. Satellite imagery—with a horizontal resolution of 0.5 or 1 km—is of great value in showing the locations of depressions, although objective data are required in order that the depth of the systems can be determined. It is also important that such data are made available to the analysis centers in near-real time. The special week highlighted the value of the satellite imagery but suggested that an improved means of deriving objective data from the imagery is needed, such as the Guymer technique for determining surface pressure from the form of the cloud associated with a depression (Guymer 1978). More cloud drift winds for the high southern latitudes would also be of value, either derived from sequences of Advanced Very High Resolution Radiometer (AVHRR) images or from stretched imagery from the geostationary satellites.
The very large differences that were found between the GASP and NCEP analyses over the Amundsen and Bellingshausen Seas during the first week of July 1994 indicate problems that can occur in data-sparse regions and the ways that satellite sounder data are employed in the assimilation schemes.
Although the analyses for the Southern Ocean and coastal region of the Antarctic are now usually of a fairly high quality, there are still great difficulties in producing analyses over the interior of the continent. At the most basic level, the fact that all the major NWP centers use different methods to determine PMSL values from the data at the lowest model sigma levels results in very different PMSL fields being distributed on the GTS. This makes intercomparison of the model analyses and forecasts very difficult. The Phillpot technique for determining 500- and 700-hPa height data from AWS observations has proved to be very valuable and consideration should be given to producing these data on a routine basis for inclusion into the assimilation schemes.
The reanalysis over the continent indicated the rapidity with which many meteorological parameters can change as a result of meso- and synoptic-scale weather systems. Little research has taken place into these systems to date, but with increasing activity over the plateau concerned with projects such as ice core drilling there is a need for more case studies in this area to aid forecasting.
Acknowledgments
The work of staff from the Tasmanian Regional Forecasting Centre is greatly appreciated, in particular J. Beck, the leader of the technical officers, who was responsible for plotting all the TOVS and surface charts, and the meteorologists who carried out parts of the reanalysis, including M. Downing, R. Anderson, T. Gibson, S. McCulloch, and M. Jones. The work of H. Hutchinson, Regional Director (Tasmania) of the Bureau of Meteorology, in supporting the FROST project is also greatly appreciated. The authors would also like to thank the Scientific Committee on Antarctic Research for supporting FROST since 1992.
REFERENCES
Bengtsson, L., 1989: Numerical weather prediction at the Southern Hemisphere. Preprints, Third Int. Conf. on Southern Hemisphere Meteorology and Oceanography, Buenos Aires, Argentina, Amer. Meteor. Soc., 1–3.
Bourke, W. P., 1994: Review of the performance of the Bureau of Meteorology Global Assimilation and Prediction system (GASP). Research Report on the 1994 Modelling Workshop, BMRC Research Rep. 52, 47 pp. [Available from Bureau of Meteorology, GPO Box 1289 K, Melbourne, Victoria 3000, Australia.].
Burdsall Brown, E., 1993: TOVS and SSM/T operational processing at NOAA/NESDIS. Proc. Seventh Int. TOVS Study Conf., Reading, United Kingdom, ECMWF, 73–77.
Carleton, A. M., 1987: Satellite-derived attributes of cloud vortex systems and their application to climate studies. Remote Sens. Environ.,22, 271–296.
——, L. A. McMurdie, K. B. Katsaros, H. Zhao, N. M. Mognard, and C. Claud, 1995: Satellite-derived features and associated atmospheric environments of Southern Ocean mesocyclone events. Global Atmos. Ocean Sys.,3, 209–248.
Colwell, S., and J. Turner, 1999: Antarctic meteorological observations on the GTS during the FROST project. Wea. Forecasting,14, 811–816.
Dalu, G., C. Prabhakara, and J. Nuccarione, 1993: Optimization of an algorithm for the estimation of rainfall from the SSM/I data. J. Meteor. Soc. Japan,71, 419–424.
Goodberlet, M. A., C. T. Swift, and J. C. Wilkerson, 1989: Remote sensing of ocean surface winds with the Special Sensor Microwave/Imager. J. Geophys. Res.,94, 14 547–14 555.
Guymer, L. B., 1978: Operational applications of satellite imagery to synoptic analysis in the Southern Hemisphere: Melbourne. Australian Bureau of Meteorology Tech. Rep. 29, 83 pp. [Available from Bureau of Meteorology Research Centre, GPO Box 1289K, Melbourne, Victoria 3001, Australia.].
Hart, T., P. Steinle, P. Riley, R. Seaman, W. Bourke, and J. Le Marshall, 1993: Increasing the effectiveness of satellite temperature soundings in the Australian Bureau of Meteorology’s global analysis and prediction system. Preprints, Fourth Int. Conf. on Southern Hemisphere Meteorology and Oceanography, Hobart, Australia, Amer. Meteor. Soc., 30–31.
Hoffman, R. N., 1993: A preliminary study of the impact of the ERS-1 C-band scatterometer wind data on the European-Centre-for-Medium-Range-Weather-Forecasts Global Data Assimilation System. J. Geophys. Res. (Oceans),98, 10 233–10 244.
Hutchinson, H. A., and Coauthors, 1999: On the reanalysis of Southern Hemisphere charts for the FROST project. Wea. Forecasting,14, 909–919.
Ingleby, N. B., and R. A. Bromley, 1991: A diagnostic study of the impact of SEASAT scatterometer winds on numerical weather prediction. Mon. Wea. Rev.,119, 84–103.
Lejenäs, H., 1984: Characteristics of Southern Hemisphere blocking as determined from a time series of observational data. Quart. J. Roy. Meteor. Soc.,110, 967–979.
Lutz, H. J., and W. L. Smith, 1988: TOVS over polar regions. Proc. Fourth Int. TOVS Study Conf., Igls, Austria, Cooperative Institute for Meteorological Satellite Studies, University of Wisconsin—Madison, 168–181.
——, ——, and E. Raschke, 1990: A note on the improvement of TIROS operational vertical sounder temperature retrievals above the Antarctic snow and ice fields. J. Geophys. Res.,95, 11 747–11 754.
Phillpot, H. R., 1991: The derivation of 500 hPa height from automatic weather station surface observations in the Antarctic continental interior. Aust. Meteor. Mag.,39, 79–86.
Pook, M., and L. Cowled, 1999: On the detection of weather systems over the Antarctic interior in the FROST analyses. Wea. Forecasting,14, 920–929.
Stearns, C. R., J. T. Young, and B. Sinkula, 1995: Two years of IR images south of 40°S every three hours. Preprints, Fourth Conf. on Polar Meteorology and Oceanography, Dallas, TX, Amer. Meteor. Soc., 112.
Taljaard, J. J., 1967: Development, distribution and movement of cyclones and anticyclones in the Southern Hemisphere during the IGY. J. Appl. Meteor.,6, 973–987.
Trenberth, K. E., and K. C. Mo, 1985: Blocking in the Southern Hemisphere. Mon. Wea. Rev.,113, 3–21.
Troup, A. J., and N. A. Streten, 1972: Satellite-observed Southern Hemisphere cloud vortices in relation to conventional observations. J. Appl. Meteor.,11, 909–917.
Turner, J., and J. P. Thomas, 1994: Summer-season mesoscale cyclones in the Bellingshausen–Weddell region of the Antarctic and links with the synoptic-scale environment. Int. J. Climatol.,14, 871–894.
——, and Coauthors, 1996: The Antarctic First Regional Observing Study of the Troposphere (FROST) project. Bull. Amer. Meteor. Soc.,77, 2007–2032.
Wright, A. D. F., 1974: Blocking action in the Australian region. Melbourne, Bureau of Meteorology Tech. Rep. 10, 29 pp. [Available from Bureau of Meteorology Research Centre, GPO Box 1289K, Melbourne, Victoria 3001, Australia.].
A map of the Antarctic.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
(a) The NCEP PMSL analysis for 1200 UTC 6 Jul 1994. (b) The UKMO-HD PMSL analysis for 1200 UTC 6 Jul 1994. (c) The GASP PMSL analysis for 1200 UTC 6 Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
(a) The UKMO-HD PMSL analysis for 0000 UTC 23 Jul 1994. (b) The FROST-2 PMSL analysis for 0000 UTC 23 Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
(a) The GASP PMSL analysis for 0000 UTC 25 Jul 1994. (b) DMSP visible wavelength imagery for approximately 0145 UTC 25 Jul 1994, revealing the complexity of the depression to the north of Casey. Note that the horizontal black lines indicate data losses in this section of the orbit. (c) The FROST-2 PMSL analysis for 0000 UTC 25 Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
(a) The UKMO-HD PMSL analysis for 1200 UTC 22 Jul 1994. (b) The Madison IR composite satellite image for 1200 UTC 22 Jul 1994. (c) The FROST-2 PMSL analysis for 1200 UTC 22 Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
(a) The UKMO-HD PMSL analysis for 1200 UTC 26 Jul 1994. Possession Island is indicated by the small black square. (b) The FROST-2 PMSL analysis for 1200 UTC 26 Jul 1994. (c) The DMSP IR imagery for Victoria Land at approximately 2300 UTC 26 Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
(a) The UKMO-HD PMSL analysis for 0000 UTC 22 July 1994. (b) The DMSP IR image at approximately 0630 UTC 22 Jul 1994, showing the indistinct signature of an intense occluded low that was located on the 1200 UTC 22 Jul FROST PMSL analysis at 66°S, 24°E.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
(a) DMSP VIS image at approximately 2200 UTC 21 Jul 1994. The location of the mesocyclone is indicated by an arrow head. (b) The FROST-2 PMSL chart for 0000 UTC 22 Jul 1994. The mesocyclone at 52°S, 166°E is marked with a filled square.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
GASP analysis for 0000 UTC 15 Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
The FROST-2 PMSL analysis for 0000 UTC 27 Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
The mean PMSL fields for Jul 1994 as derived from the ECMWF, GASP, MRF, and UKMO-NP daily fields.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
Cross sections of zonal mean PMSL for Jul 1994 as produced from the UKMO-NP, MRF, GASP, and ECMWF analyses.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
The rms difference between the daily PMSL fields from the GASP and NCEP analyses for the month of Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
The mean difference (GASP–NCEP) between the daily PMSL fields from the GASP and NCEP analyses for the month of Jul 1994.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
The difference between the GASP and NCEP analyses at 0000 UTC 5 Jul 1994. Negative values indicate that GASP values are lower.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
The mean 500-hPa height fields for July 1994 as derived from the ECMWF, GASP, MRF, and UKMO-NP daily fields.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
(a) The UKMO-NP 500-hPa height field for 0000 UTC 5 Jul 1994. (b) The revised 500-hPa height field for 0000 UTC 5 Jul 1994 produced using the Phillpot technique.
Citation: Weather and Forecasting 14, 6; 10.1175/1520-0434(1999)014<0817:AAOOAA>2.0.CO;2
Differences in the locations of vortex centers between the FROST-1 and FROST-2 PMSL charts for the Eastern Hemisphere.
The data collected during the FROST project are available to the research community. Details of data collected and how they can be obtained are on the World Wide Web (http://www.nerc-bas.ac.uk/public/icd/FROST/).