Validation of operational numerical analyses in Antarctic ...

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 102, NO. D12, PAGES 13,761-13,784, JUNE 27, 1997

Validation of operational numerical analyses in Antarctic latitudes

Richard I. Cullather and David H. Bromwich

Polar Meteorology Group, Byrd Polar Research Center, Ohio State University, Columbus

Robert W. Grumbine

National Centers for Environmental Prediction, Camp Springs, Maryland

Abstract. Available rawinsonde, automatic weather station (AWS), ship, and synthesized long-term observations are used to evaluate the Antarctic numerical analyses of the European Centre for Medium-Range Weather Forecasts (ECMWF) and the U.S. National Centers for Environmental Prediction (NCEP) from 1985 to 1994. Twice-daily variations in the ECMWF surface pressure analyses compare closely with AWS units of the U.S. Antarctic Program and ship observations. The NCEP analyses over the same period show substantial improvement, particularly during the period 1985-1990. Surface air temperatures and winds do not agree so closely, which may result from analyses error, the localized nature of the fields, or a combination. Validation of the analyses standard pressure level fields using available rawinsonde data reveal a general long-term decrease in RMS errors with time for both analyses. RMS errors in NCEP 200 hPa geopotential heights of over 200 geopotential meters (gpm) for central plateau stations are evident only prior to May 1986. However, a significant upward trend from 1989 to 1993 in geopotential height RMS differences is apparent at several levels. The ECMWF analyses are generally found to be superior and offer a reasonable depiction of the broadscale atmospheric circulation; however, deficiencies in midtropospheric temperatures and lower tropospheric winds are evident. Comparisons of ship data from individual cruises of the S.A. Agulhas and the R/V Nathaniel B. Palmer to the numerical analyses reveal substantial agreement for pressure and temperature variables. Observations from the Nathaniel B. Palmer in the Amundsen and Bellingshausen Seas were not available to the weather forecasting centers. Results presented here indicate that a large amount of the available data is being incorporated and that large deficiencies identified in previous studies are being addressed, although areas of concern remain. Deficiencies in comparisons to specific stations are common to both analyses, implying continued communications problems. In particular, grid values corresponding to individual stations including the now-closed Leningradskaya base and Mirnyy are found to be conspicuously deficient at the 200 hPa level for both analyses.

1. Introduction

The Antarctic First Regional Observing Study of the Troposphere (FROST) is an international project of extensive data collection and analysis poleward of 50øS latitude covering three 1-month-long special observing periods (SOPs) in 1994 and 1995 [Bromwich and Smith, 1993; Turner et al., 1996]. The primary purpose of FROST is to verify current Antarctic analyses and model forecasts. In support of this program, an evaluation of the historical proficiency of the operational numerical analyses is conducted using Antarctic rawinsonde, ship, and automatic weather station (AWS) data. Numerical analyses have been viewed as a potentially valuable tool for atmospheric study [e.g., Trenberth and Olson, 1988a]. This is particularly true of high southern latitudes, where the scarcity of data

Copyright 1997 by the American Geophysical Union.

Paper number 96JD03328. 0148-0227/97/96JD-03328509.00

necessitates a complete assimilation of all meteorological observations available, including satellite data, to achieve an accurate depiction of the atmosphere.

In the recent past, numerical analyses have played an important role in a variety of atmospheric studies in high southern latitudes [e.g., Trenberth and Solomon, 1994; Quintanar and Mechoso, 1995; Heinemann, 1996]. Additionally, operational analyses have been utilized in computing the atmospheric moisture budget over Antarctica [Howarth, 1986; Masuda, 1990; Yamazaki, 1992, Budd et al., 1995]. Bromwich et al. [1995] evaluated the atmospheric moisture budget for three numerical analyses for the period 1985-1992 in high southern latitudes. The analyses produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) were found to be superior to those of the National Centers for Environmental Prediction (NCEP; formerly National Meteorological Center (NMC), and the Australian Bureau of Meteorology in comparisons with rawinsonde and glaciological data. Bromwich et al. [1995] limited the validation study to moisture transport variables. In

13,761

13,762 CULLATHER ET AL.: ANTARCTIC ANALYSES VALIDATION

this paper, a more comprehensive evaluation of the basic kinematic variables given by the tropospheric numerical analyses is conducted. In section 2, a description of the evaluated ECMWF and NCEP data sets is given. Section 3 provides results of surface and boundary layer comparisons using AWS stations maintained by the U.S. Antarctic Program, as well as observational climatologies. Results of comparisons with the international rawinsonde archive are discussed in section 4. An evaluation of the analyses using case-study ship observations along the periphery of the continent is given in section 5; and discussion and summary are given in section 6.

2. Analyses Description

The evaluated analyses are those of the ECMWF WCRP/TOGA archive 11 (TOGA is Tropical Oceans Global Atmosphere, a program under the World Climate Research Programme (WCRP)) and the NCEP analyses. Both data sets used in this study were obtained from the National Center for Atmospheric Research (NCAR). Both analyses are on a 2.5 ø latitude-longitude grid and are reported twice daily (0000 and 1200 UTC). The ECMWF archive consists of 14 standard pressure levels, the lowest six of which contain relative humidity data [Trenberth, 1992]. After 1991, a fifteenth level, at 925 hPa, was added to the archive. The NCEP archive consists of 12 standard pressure levels [Trenberth and Olson, 1988c], with standard levels in the troposphere identical to those of the ECMWF. Both archives contain

surface and boundary layer variables; however, not all computed fields are available. For the ECMWF, boundary layer values consist of 2 m temperature and dew point, and 10 m winds. The NCEP archive contains boundary layer potential temperature, relative humidity, and wind components. Prior to April 1985 the NCEP boundary layer variables correspond to the midpoint of a 50 hPa thick boundary layer, approximately the 200 m level. After April 1985 the boundary layer thickness was reduced to 10 hPa, with values approximately corresponding to the 40 m level.

Both the NCEP and the ECMWF analyses are produced from four-dimensional data assimilation incorporating accumulated observational data, including satellite data, with a 6-hour "initial guess" provided by a global spectral model. A difficulty in the use of numerical analyses is the effect of alterations in the data assimilation system on the climatic ensemble of analyses [Trenberth, 1992]. Changes in the analysis schemes result from resolution or parameterization adjustments in the spectral model. The analyses scheme may also be impacted by the discontinuation or introduction of data sources. The NCEP Global Data Assimilation System (GDAS) has evolved over the last 10 years [Kanamitsu, 1989; Kalnay et al., 1990; Ballish et al., 1992; Parrish and Derber, 1992; Kalnay et al., 1996]. A list of changes to the ECMWF data assimilation system has also been produced [Trenberth, 1992]. The analyses evaluated here are used by researchers, operational modelers, and operational forecasters. Reanalysis projects are under way to achieve the best possible uniform analysis [Schubert et al., 1993; Trenberth, 1995; Gibson etal., 1996; Kalnay et al., 1996], however operational analyses currently remain of interest due to the higher resolution of the operational analyses schemes.

Trenberth and Olson [1988a, b, c] and Trenberth [1992] comprehensively evaluated the ECMWF WMO (World

Meteorological Organization) archive and the NCEP analyses for the period 1979-1986 and found major problems in both analyses over and around Antarctica. While data scarcity is an acute problem for the southern hemisphere, these studies indicate that available observations were not incorporated and point to difficulties in communications as the primary source of error. In particular, the rawinsonde stations at South Pole (90.00øS) and McMurdo Station (166.67øE, 77.85øS) were found to be frequently truncated or otherwise unincorporated into the analyses. The NCEP analyses were found to contain large monthly RMS differences with observations, at times greater than 200 m for 200 hPa geopotential heights, until May 1986 [Trenberth and Olson, 1988b]. Trenberth and Olson [1988a] indicate that the ECMWF is the preferred analysis for the southern hemisphere, despite the substantial problems observed. A similar conclusion based on moisture budget validation was given by Bromwich et al. [ 1995] for the 1985-1992 time period, although validation of the moisture fluxes with rawinsonde and glaciological data generally presented a more favorable comparison for this later time period. Trenberth and Olson [1988a] note a marked improvement with time in the ECMWF Antarctic analyses, particularly in 1982 and 1983.

Several additional regional evaluations of the analyses have also been conducted. Hakkinen [1995] employed the ECMWF climatology from 1985 to 1989 in a coupled ice- ocean model. The inadequacy of the resulting sea-ice cover was attributed to apparent deficiencies in surface temperature and wind fields. Arpe and Cattle [1993] evaluated wind stresses derived from the analyses of the ECMWF and the United Kingdom Meteorological Office (UKMO) for 1988- 1990. Although large discrepancies arose from the acceptance of an individual observation in one analysis and the rejection in another, the two analyses were found to be generally similar over the Antarctic sea ice zone. Escoffier and Provost [ 1995] compared observed wind forcing over the southwestern Atlantic with the ECMWF analyses for 1986- 1990. The study found the major discrepancies to be due to land-sea effects not resolved by the model and that agreement with observed annual and semiannual signals was exceptional.

Brown and Zeng [1994] have derived cyclone surface pressures over data sparse oceanic regions using a planetary boundary layer model incorporating ERS 1 scatterometer data for 1992 and have made comparisons with the ECMWF and NCEP analyses. Most storms in the southern hemisphere were resolved by both analyses, although the scatterometer derivation generally yielded more detailed and stronger storms. The derived southern hemisphere minimum pressures were found to be on average 3 hPa lower than the ECMWF, with similar results for NCEP. Genthon and Braun [1995] evaluated the ECMWF model used for the analyses over Greenland and Antarctica for 1985-1991 and concluded that

the analysis and predictions "are in good agreement with available observations." Only two fields, surface temperature and precipitation, are presented however, and precipitation comparisons only utilize long-term averages at manned stations, while station observations are not employed in the surface temperature validation. In summary, the five studies noted above generally show an evolution in the analyses over time and a preference for the ECMWF analyses for atmospheric study of the southern hemisphere. They also indicate that the suitability of the analyses is highly dependent

CULLATHER ET AL.: ANTARCTIC ANALYSES VALIDATION 13,763

upon the application for which the analyses are utilized. In this context, three areas of interest are apparent: (1) whether the available data are being fully utilized by the analyses, (2) the relation of deficiencies to geographical location, or to particular variables, and (3) the significant data trends.

3. Surface and Boundary Layer Evaluation Over Antarctica

Validation Using Automatic Weather Stations

Near-surface and boundary layer conditions play a uniquely dominant role in the atmospheric circulation over Antarctica. Physical processes within the lowest few hundred meters of the atmosphere, primarily the near-surface katabatic wind regime, produce the most significant forcing of the atmospheric flows [e.g., Parish, 1988]. Consequently, the accurate reproduction of near-surface conditions is an essential component of the analyses.

About 30 automatic weather station units (AWS) have been deployed in Antarctica by the United States Antarctic Program [Stearns et al., 1993]. The AWS units and their purposes are described by Stearns and Wendler [1988] and Stearns et al. [ 1993]. The Scientific Committee for Antarctic Research (SCAR) Group of Specialists on Antarctic Research has recommended that the determination of moisture, heat,

and momentum budgets be a primary consideration for the use of the AWS units [Allison, 1983]. However, many of the AWS units have been deployed in support of mesoscale studies on the Ross Ice Shelf and barrier wind flow along the Antarctic Peninsula. Several stations, for example, are located near Ross, White, and other islands embedded within the Ross Ice Shelf near the Transantarctic Mountains. The

station, mountain range, and islands which influence surface conditions on the ice shelf may all fall into a single 2.5 ø analyses grid box. For purposes of validation, 15 stations have been selected based on site location and geographic

AEROVANE

WIND SPED AND WIND D•F.•ON

GROUND P•E

•.•:. ....

3.0 m ENCLOSU•

........ .- ...% .............

..... ::::..:..:• ..., •.: ......

'r.• r•"

Figure 2. Configuration of typical AWS unit (Bonaparte AWS, photographed by Phil Handley and obtained from Palmer Station Long-Term Ecological Research Program database).

Larsen Ice Shelf

Uranus Glacie•

Mount Sipk

South

Ocean

I

Butler Island ß ¾r". East • Sip!e West. ß .• PO le -. Antarctic a .......

Antarctica •i .... Byrd Howe ß

ß

Dome C

• -' Gill

J •n• .. D-80 Possession Island

_

Scott Island• • Young j Island

Figure 1. Location map of U.S. Antarctic Program automatic weather station (AWS) sites examined.

dispersion, shown in Figure 1. In Figure 2 a typical AWS station configuration is shown. Optimally, AWS data set values are available every three hours via the Argos data collection and location system onboard TIROS (Television and Infrared Observations Satellite) A-G and TIROS N series polar orbiting satellites. Not all AWS units report over the GTS (Global Telecommunications System). The Parascientific pressure transducer, located in the electronics enclosure, has a resolution of 0.05 hPa; the temperature resolution is 0.125øC. The relative humidity sensor is not included on many of the AWS stations, and those stations listed that do have the sensor have very few reportings after quality control by the University of Wisconsin. Missing data result from a variety of events that may occur, while the AWS is unattended for many months at a time. Stearns et al. [1993] indicate the wind measurement system (Bendix or Belfort aerovane) is the most susceptible component of the AWS unit to damage or the reporting of suspect data. In some locations, particularly along the Ad61ie Coast in East Antarctica (--140øE) and Terra Nova Bay [Bromwich, 1989], the force of the wind may at times be sufficient to render structural damage to the unit. The aerovane coupling is also vulnerable to rime or ice buildup in winter months. A


significant number of AWS units are deployed in the so- called "dog house" configuration and do not measure wind speed or direction, as oceanic island sites often get quickly buried in snow and ice. It is likely that AWS pressure and perhaps temperature reports only are used by the operational weather forecasting centers. This would appear to be reasonable given the potential error and the question of local versus synoptic scales.

Some significant issues arise in the computation of the analysis values for comparison with observations. While interpolation of four grid points is sufficient for locations of smooth topography such as the interior of the continent, it often results in large height differences between the station elevation and the analysis model. The problem is confused by the false topography utilized for the analysis models, referred to as "envelope" topography by ECMWF and "silhouette" topography used by NCEP prior to 1992, and is further compounded by changes made to model resolution, which in turn results in variations in the model elevation over time. Trenberth [1992] indicates that analysis values at high latitudes are more reflective of area rather than point values. Accordingly, the elevation of the four analysis grid points adjacent to each observation are computed from monthly averaged analyses, and values from the grid point closest in elevation are used for comparison. Even after the grid point elevation comparison, large differences in elevation between observation and analyses are still apparent for some stations. For results presented here however, no attempt was made to reduce the analyses' surface pressure values to the station datum. Genthon and Braun [1995, see their Figure 3] have examined the ECMWF analysis topography, which was found to be deficient not simply because of the envelope topography scheme but also due to the use of a comparatively old orographic database. An additional issue that arises is associated with the spectral nature of the model in close proximity to steep topography. This is clearly shown in

Figure 3, where ECMWF elevations as derived by NCAR for February 1987 are contoured for values below zero. Steep topography associated with the Transantarctic Mountains (left) and the Antarctic Peninsula (fight) result in large spectral topographic waves, or divots, of as much as 150 m or more below ground in the Ross and Larsen Ice Shelves, respectively. Similar topographic deficiencies exist for the NCEP analyses as seen from surface pressure comparisons with the AWS stations. Moreover, the NCEP boundary layer values correspond to approximately the 40 m level while ECMWF corresponds to the 2 m level. The AWS temperature, wind, and humidity measurements are made at the 3 m level. This likely will result in discrepancies, particularly for the interior plateau region under the intense Antarctic inversion. Consequently, the surface pressure evaluation is the least ambiguous comparison for temporal variations.

Figure 4 shows the monthly surface pressure correlation coefficients for each station. The overwhelming number of station observations are well correlated with the ECMWF

analyses. Most values are above the 0.80 correlation level, and 78% are above the 0.90 level. Overall interior stations

fare better than maritime or coastal stations, which generally do not have as complete a record as the continental stations. The record for Possession Island, for example (available 1992-1994), contains only half of possible 0000 and 1200 UTC reports for any given month while in operation. While this station has been included for completeness, all other data points denote at least 50 observations for the month. In general, there is a surprisingly flat trend over time in correlations with ECMWF. It is evident from Figure 4 that some months exist where multiple stations failed to correlate well with ECMWF. These months reflect changes to the analysis scheme that significantly impact correlations during the month. In contrast to the high consistency found for ECMWF, the NCEP analyses show a marked improvement

Figure 3. Below sea-level contours of European Centre for Medium-Range Weather Forecasts (ECMWF) elevation data for Ross Ice Shelf region (left), and Antarctic Peninsula (right), contoured every 20 m.


1

0.6

0.4

0.2

0

-0.2

-0.4 ' I ' I ' I ' I

1985 1987 1989 1991 1993

Date

1995

Byrd •D-80 &D-10 Dome C Larsen Gill Mt Siple Uranus Gl Pole Scott Is

1 Siple Stn Young Is Butler Is Possession

0.2

-0.2

X X

-0.4 I I I I I I I I

1985 1986 1987 1988 1989 1990 1991 1992 1993

Date

I

1994 1995

• Byrd ] •D-80 ] &D-10 I •.Dome C I O Larsen ß Gill + Mt Siple ß -Uranus Gl I ......... Pole I O Scott Is I ISiple Stn I A Young Is I X Butler Is • Possession

Figure 4. Monthly correlations of ECMWF (top) and NCEP (bottom) surface pressure analyses with AWS station data.

for the 1980s. For the early portion of the time series, bad station correlations are widespread and are not limited to the period prior to analysis scheme changes in May 1986. West Antarctic coastal stations as well as Gill on the Ross Ice Shelf

frequently correlate poorly during the 1980s. However, interior stations such as Dome C also have several monthly correlation coefficients below 0.50. The situation improves somewhat after 1990, with most points attaining monthly correlation coefficients above 0.70. The NCEP analyses compare more favorably to observations than the ECMWF for only a handful of monthly station comparisons for the 1 O-year period, however.

It is •uspected that the analyses tend to underestimate strong synoptic systems [Trenberth and Olson, 1988a; Brown and Zeng, 1994]. Although station comparisons indicate that monthly maximum and minimum analysis values tend to correspond to observation at least as accurately as other values in series, there is evidence that very intense lows are underestimated by one or both of the analyses when examining the strongest systems over the 10-year period. Figure 5 is an example of this, showing the lowest pressure observed at Scott Island. Here, the ECMWF analysis has overestimated the surface pressure by 15 hPa. The time series

indicates that station and analyses pressures are similar for other times. Interestingly, both analyses performed reasonably well in estimating the 932 hPa low that occurred

100o

990

980

970

960

950

940

930

920 I I [ I I I I I

10/1/93 10/2/93 10/3/93 10/4/93 10/5/93 10/6/93 10/7/93 10/8/93 10/9/93 10/10/93 00Z 00Z 00Z 00Z 00Z 00Z 00Z 00Z 00Z 00Z

Date

Figure 5. Nine-day surface pressure time series for Scott Island AWS (67.37øS, 179.97øW), and corresponding ECMWF and National Centers for Environmental Prediction

(NCEP) analysis values, in hectopascals, beginning October 1, 1993.


9

Uranus G!. 8

•,7

• 5

•.4

1

Month

9 [] ECMWF • rd

8 [] NCEP ,• 5

•3

1

Month

9

Month

Dome C

Month

9

8

•3

•.2

1

D-80

Month

9

Mt Siple 8

7

6

5

4

3

2

1

Month

Figure 6. Average monthly ECMWF and NCEP surface pressure standard error from regression with observations from six AWS units, in hectopascals.

3 « days prior to the lowest-pressure event. Figure 5 underscores the strong variability associated with the adjacent oceanic regions.

In Figure 6 the analysis standard error from regression with AWS surface pressure observations for selected stations, averaged for the years of available data, is plotted for the annual cycle. The amount of error for a particular time and location is contingent upon the amount of synoptic or mesoscale activity encountered. West Antarctic coastal stations such as Uranus Glacier and Mount Siple as well as Gill on the Ross Ice Shelf experience significant synoptic variability during the austral winter. Error for these stations is somewhat larger than interior stations and maximizes in winter months. This is generally true of the NCEP analyses, although large standard errors are also found for some summer months, such as at Mount Siple in January. At the

•AWS Uni! Uranus Glacier

930,

920 [ •' •

•870• i .: '" ' ' :•: ''i '"'l" 1985 1986 1987 1988 1989 1•0

Date

i I I

1991 1992 1993 1994 1995

Dome 710

,.700 t e 680 I;•. $

670 /:660 g 650

630

620

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 Date

interior station Dome C, there is little seasonal variability for Figure 7. Monthly averaged suface pressure for Uranus either analysis, which is not surprising as little synoptic Glacier(71.43øS, 68.93øW)andDomeC (74.50øS, 123.00øE) activity penetrates far inland beyond the coastal escarpment of AWS units in comparison with corresponding ECMWF and East Antarctica. Again the standard error for the NCEP NCEP analysis values, in hectopascals.

CULLATHER ET AL.' ANTARCTIC ANALYSES VALIDATION 13,767

9 9 ß ECMWF anus GI. rd

G8 8 I•t• •7 7

1 1

• •lonth < •

Gill 8

•7

•5

•4

•3

'• 1

• •lonth '•

9

t •)ome C 8

•lonth Z

D-80 8

,o. 7

5

'• 1

• • • </• • •lonth Figure 8. Average monthly ECMWF and NCEP near-surface temperature standard error from regression with observations from six AWS units, in degrees centigrade.

analysis surface pressure is larger than for the ECMWF for virtually all stations and time periods. In Figure 7, monthly mean values for Uranus Glacier and Dome C are plotted as a general reflection of analyses skill for seasonal to interannual timescales. Surface pressure over land is particularly susceptible to changes in the data assimilation scheme, due to the previously mentioned effects on topography. At both stations the substantial changes made to the NCEP analysis scheme in May 1986 are easily seen in both time series. While the corrections introduce a substantial bias, especially for Uranus Glacier, the NCEP time series, with some exceptions, appears to track reasonably well in comparison with the recorded observations. The ECMWF time series for

Dome C is extremely similar to observation for the 10 years recorded, which is greatly reassuring. However, for Uranus Glacier there is marked change in the ECMWF bias after the 1990-1992 data gap, and this is almost certainly related to a change in the NWP (numerical weather prediction) model resolution which occurred in September 1991.

No other surface or boundary layer field compares as closely with AWS values as the surface pressure field. It is

suggested that the surface pressure signal may be of a large enough spatial scale to be more easily reproduced by the analyses. Comparisons of near-surface temperature fields reveal a mixed bag, with skill highly variable and dependent upon each station. In Figure 8 the average monthly standard error for six AWS units is plotted for the annual cycle. Standard errors range from 0 ø to 9øC depending on station and month. Coastal and/or oceanic stations again fair poorly, although interior stations also show large standard errors. Unlike the pressure comparison the NCEP analysis is found to be superior for a significant number of observations, generally during summer months. As was mentioned previously, this seasonality is in part an artifact of the NCEP boundary layer temperature product, which is affected by the Antarctic inversion. Correlations with temperature observations average 0.5 for both analyses but may be considerably more or less. There are similar trends indicating analyses improvements to those found with surface pressure. In Figure 9 the time series of monthly mean temperatures are shown for Dome C. The station contains one of the more complete records for the total time period. The ECMWF analyses


-lO

-2o

-3o

-50

-60

-70

-80

'-'--'Dome C 3m Temp • ECMWF 2m

--•--NCEP Bndy Lyr

it i • . .". :. :• • - : . i '. '. . '

.. aa• . ,, ::

I

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995

Date

Figure 9. Average monthly near-surface temperatures for Dome CAWS unit (74.50øS, 123.00øE), in comparison with corresponding ECMWF and NCEP analysis values, in degrees centigrade.

closely follow observed values, while the NCEP analyses show a significant improvement during summer months over the 10-year period.

Comparisons with the near-surface wind observations do not fare so well as with temperature, with large errors occurring at many stations and all times of the year. Figures 10 and 11 show zonal and meridional wind standard errors, respectively, for four selected stations. The ability of the NCEP analyses to reproduce the twice-daily observed wind values varies greatly over time and from location to location. While the ECMWF generally shows more consistency, with standard errors generally in the range of between 2 to 4 m s -1, the errors encountered for individual times are of the same

order as the observations. The differences improve somewhat with temporal averaging. Figure 12 shows a comparison of the monthly averaged meridional wind for AWS unit D-80. Although in only very broad terms, this comparison shows some agreement with seasonal variation as well as significant improvement with time, particularly on the part of the

ECMWF analyses. The D-80 AWS is shown because of its location on relatively featureless topography in central Wilkes Land, although even at this location the downslope wind direction varies substantially with longitude. In summary, validation using the AWS wind data is a very messy activity due to resolution questions and the reliability of the observational data. A feature of important interest, however, is the ability of the analyses to reproduce the time-averaged katabatic wind regime for particular locations, and this is discussed below in comparison to modeling results.

Long-Term Climatologies

As a means of further evaluating the spatial distribution, the 10-year average temperature and wind fields are evaluated against long-term climatology and modeling studies. Validation of the long-term near-surface temperature field is conducted using the synthesis of observational data by Giovinetto et al. [1990], reproduced in Figure 13. In Figure 14 the corresponding fields for the ECMWF (left) and NCEP analyses (fight), averaged over 10 years, are shown. The high plateau minimum temperature in the ECMWF analyses is approximately 3øC warmer than the (-)60øC contour shown by Giovinetto et al., which may possibly result from decadal variability. The location of this minimum is also farther west in the ECMWF analyses, which is the result of topographic deficiencies. Again, the largest topographic errors found by Genthon and Braun [ 1995] are located near and to the west of the prime meridian, and there are some significant discrepancies in the temperature contours in this region. Apart from these discrepancies there is tremendous fidelity between the ECMWF and observed temperature fields for both distribution and magnitude. In particular, the (-)20øC contour running along the Amundsen and Bellingshausen Sea coastline in West Antarctica is exactly reproduced. There is also a clear representation of the Antarctic Peninsula in the temperature field. Overall the ECMWF temperature field compares quite well and adequately reproduces the gradients along the East Antarctic escarpment, which is consistent with the findings of Genthon and Braun [1995]. As previously

Uranus Glacier • i

: g:: : .'. • ECMWF • Byrd I ...... NCEP

.;

1985 1987 1989 1991 1993 1995 1985 1987 1989 1991 1993 1995 Date Date

9

8 t Dome C 7

6

3

2

1

0,

18 .... I .... 919 .... ! .... 1985 I 7 1989 I I 1993 1995 1995 Date

t Gill ....

1985 1 987 ! 989 1991 ! 993

Date

Figure 10. Average monthly ECMWF and NCEP near-surface zonal wind standard error from regression with observations from four AWS units, in meters per second.


8

1

Uranus Glacier

: . ?...,'..:i. '? .

:.: •.,..' :::• :.,.:;.: f\:,.: ..:"•.'

0 .... : .... : .... : .... : ....

1985 1987 1989 1991 1993 1995

Byrd -- ECMWF 1- ...... NCEP I

.... : .... : .... : .... :

1985 1987 1989 1991 1993

Date

9

s Dome

4

0 ..... . ß ,

1985 1987 1989 1991 1993 1995

Date

Gill

.

1985 1987 1989 1991 1993 1995

Date

Figure 11. Average monthly ECMWF and NCEP near-surface meridional wind standard error from regression with observations from four AWS units, in meters per second.

mentioned, the NCEP analysis values correspond to the boundary layer midpoint and are significantly warmer as a result. The NCEP contours are quite smooth and present a very simplified representation of the observational field. Contours are highly smoothed near the Antarctic Peninsula, and the gradient along the East Antarctic escarpment is significantly weaker than that shown by Giovinetto et al. [1990].

A realistic near-surface katabatic wind regime has been previously produced using a high resolution mesoscale model [Parish and Bromwich, 1991], results of which are reproduced in Figure 15. In Figure 16 the average July wind fields are shown from the ECMWF and NCEP analyses. The larger differences between the two analyses may be seen in the isotachs; however, there are some important differences in the vector fields as well. It may be seen from a comparison of vector fields in Figures 15 and 16 that the ECMWF adequately reproduces katabatic wind flow perpendicular to the coast near Byrd Glacier south of McMurdo Station and Terra Nova Bay to the north, while the NCEP analysis is more

10

1986 1986.5 1987 1987.5 1988 1988.5 1989

•D-80

---n-- ECMWF

I I

1989.5 1990 1990.5

Date

Figure 12. Monthly averaged meridonal wind for D-80 AWS unit (70.02øS, 134.72øE), in comparison with corresponding ECMWF and NCEP values, in meters per second.

parallel to the coast. Neither analysis accurately reproduces the southerly barrier winds along the eastern side of the Antarctic Peninsula. There has been some interest recently in the peninsula barrier winds, which may be a chronic problem for atmospheric models. Hines et al. [ 1995] found that errors in the representation of these winds result from differing parameterizations of the sensible heat flux. The lower portion of Figure 16 shows a large difference in isotach detail between the two analyses. The NCEP analysis again reflects a much smoother representation of the boundary layer winds. The ECMWF analysis correctly locates maxima in the interior of West Antarctica, which is surprising given the topographic errors found by Genthon and Braun [1995] in the region adjacent to the Weddell Sea. In contrast, the NCEP plot

The observed Ts 0

90W 90E

18O

Figure 13. Average annual near-surface temperature field synthesized from surface observations [Giovinetto et al., 1990], contoured every 5øC (in negative øC).


ECMWF NCEP

I : ,, :

61 ," ! i \ ',, ' : : ",

Figure 14. Average annt•al near-surface temperature fields for the ECMWF (left) and NCEP (right), contoured every 5øC (in negative øC).

shows a greatly smoothed representation of West Antarctic features. Both analyses locate three maxima associated with katabatic airflow convergence in Wilkes Land near 95øE, 120øE, and 145øE. The analyses maxima located near 120øE is not found in the model results as depicted in Figure 15 but is associated with enhanced model katabatic flow at Porpoise

Surface wind for winter - Parish and Bromwich 1991

'":<" '",, "', •' ..L•_'•,-•ii .!!i ..•,.-.ii: :.::. '* •- '•' .,. ',,;,,,,' ,'" ..... L o•'" ' •:•i'•i' :.•:;. •'.-';•"',,

-z•, "-. "."" ' ':•:•':':"-:"'q'•"•-';•"! ": " ,. • -.. .-'•. • ':i::•::• .:•:::•' • •, ), ,. •..,.•.:.•:. ,, , ,, ,,,,, ..

'. i..i ii!. I ß i • .... iii "-'• B .... ",:.i ......

.-• ', .... -:' • ,,•-.,,.X •o .•.•.• . .i • "•i • . - .• ' ::i• '•! .i•:-.: ,,."•,,,'--

...... -i• iii •'- .!iii:•' ,,-" i ij,.'• :.•,/.. I :i:: :'..:i•,

',, ..' ,/ ,, '--.. ; •5 :•:.,•..:•: • /' ß , ,, -,-,...~ • .... .f:,;•....' ,•, '-,. ,,

..-;<, ,," ," ,; "' i-li ...r..•.•:.•.x- ,,'-. ". .,' ,,' ,' ," :::', ' 'h•!!..':' •' /' ". ..- -....,., ,,, ,,' ,,, -:•. .:::,•:.:,•:........., .• ..•' ...

o•. " ..... _'" / L I•,i, "; "' L" -"' '" ß ... -.,'.. / ; .o•. ,., .• .•, ,, .0.,,!,.- ,,. O. 1•$I!I+01 • O.P.4$1ii+Og ,t o, •, , , ,

-- • CONTOUR FROM 5 TO 25 BY 5 Minimum Vector Maximum Vector

10-15 m/s > 15 m/s

Figure 15. Near-surface wind vector field and isotachs, in meters per second, for winter, after Parish and Bromwich [991.

Bay [Parish and Bromwich, 1991, see their Figure 3]. Both analyses show a considerably weaker field, with no values greater than 15 m s -•. Arpe and Cattle [1993] had similarly found ECMWF model surface wind stresses to be weak in

comparison to observation. The comparion is consistent with the validation of CCM2 in the work of Tzeng et al. [1994], who speculate that difference in magnitude results from the higher resolution and better topographic treatment employed by Parish and Bromwich [ 1991 ].

4. Rawinsonde Comparisons

Figure 17 shows the distribution of Antarctic rawinsonde stations available for comparison from operational weather centers for 1985-1993, compiled by NCAR (T.R. Parish, personal communication, 1996). The Antarctic rawinsonde record has been examined by Trenberth and Olson [ 1989] and Connolley and King [1993]. Stations have different schedules of launches, which may vary with season. Figure 18 shows the number of monthly reports from each station. For clarity the numbers of station reports are smoothed using a 4-month running mean, divided into three plots, and normalized to a maximum of 60 per month. The decreased number of observations in the early and later part of the series may be limitations only due to the archive employed here. Observations from the South African National Antarctic

Expedition Station (SANAE, 70.30øS, 2.35øW), for example, are shown to be missing in Figure 18 after 1992 but are available during July 1994 as part of the FROST SOP. However, several former Soviet stations have closed during the 1990s, including Leningradskaya, for which upper air reports became sporadic after 1990, while Vostok has apparently discontinued its rawinsonde program. Trenberth and Olson [1989] also note the loss of data from balloon


ECMWF NCEP

Figure 16. ECMWF (left) and NCEP (right) a_¾erage July surface wind field vectors (top) and scalar- averaged isotachs (bottom), contoured every 5 m s .

bursts, which is higher during the polar winter, as well as previously mentioned communication problems. Connolley and King [1993] present a review the rawinsonde equipment used, which varies from station to station. The most susceptible component of the rawinsonde measurement

Figure 17. Location map of Antarctic rawinsonde stations.

system appears to be wind measurements, which are derived via the OMEGA VLF navigational aid for a large number of non-Russian stations. OMEGA reception is poor at some locations, and the few OMEGA stations that are received have a small angular separation, resulting in errors in the derived wind components. Issues concerning humidity measurements in cold regions have also been previously reviewed by Bromwich and Robasky [1993] and Bromwich et al. [1995], who indicate that larger moisture transports are generally associated with warmer conditions, which may reduce measurement errors.

Standard level geopotential height, temperature, zonal and meridional wind, and derived specific humidity are used for comparison to the nearest analysis grid point without regard to topography. Quality control is performed by filtering for unrealistic values and removing values greater than three standard deviations from computed monthly means. Statistics are computed for months containing at least 20 reports where both observation and archived analyses values were simultaneously available. Conservatively, specific humidity values computed from dew points colder than -40øC are discarded. For wind comparisons at South Pole there is some ambiguity in the meaning of analyses values, referred to as the "pole problem" [Trenberth, 1992]. For wind validations only, rawinsonde data are compared with analysis values located one grid box from the station and located on the prime meridian, which results in consistency with the rawinsonde convention of wind directions corresponding to longitude.

13,772 CULLATHER ET AL.' ANTARCTIC ANALYSES VALIDATION

60 .

440 ! • g,30

•2o lO

0 • • I • I g---

1985.0 1987.0 1989.l•)at e 1991.0 1993.0

Molodezhnaya (MOLO)

..... Davis (DAVI)

Casey (CASE)

- - = SANAE (SANA)

............... Syowa (SYOW)

-'- Mawson (MWS•

• 40

•. 30

{ 20

10

0

1985.0 1987.0 1989.0 1991.0 1993.0 Date

Mimyy (MIRN)

..... Novolazarcvskaya (NOVO) Vostok (VOST)

- - - Amundsen-Scott

(POLE)

.................... Leningradskaya

-'- McMurdo (MCME

' Bellingshausen (BELL)

El Halley Bay (HAL•

40

30

0 I I 985.0 1987.0 1989.0

Date

I • t

1991.0 1993.0

Dumont d'Urville

(DUMT)

..... Neumayer (NEUM)

Vicecomodoro

Marambio (VMBO)

Figure 18. Monthly number of Antarctic rawinsondes available for validation, normalized to 60 per month maximum and plotted using 4-month running mean.

A summary of results is shown in Table 1, showing average monthly RMS (root mean square) errors by year, and Table 2, showing average statistical scores for selected stations and variables. Table 1 generally shows a decrease in geopotential height RMS errors with time for both analyses, with 1993 200 hPa errors approximately 11 geopotential meters (gpm) lower than in 1985 for the ECMWF analyses and approximately 22 gpm lower for the NCEP analyses. These numbers may be compared with the observed average monthly standard deviation, which is 114 gpm. The improvement in the ECMWF analyses is generally the result of incremental decreases in negative biases. Significantly, however, most of the reduction in error in the NCEP analyses resulted from the May 1986 changes, and a significant upward trend in error from 1989 to 1993 is found at several levels. Average NCEP monthly RMS for 200 hPa geopotential height increased by more than 6 gpm from 1989 to 1993. Analysis values at specific stations were found to frequently contain high RMS errors at the 200 hPa level for the period 1989 to 1993. Monthly RMS scores for both analyses at Leningradskaya and Mirnyy were found, on the average, to be greater than 40 gpm over the period. In

particular, there is substantial evidence of difficulties in incorporating Leningradskaya 200 hPa values into the ECMWF analyses. More than 40% percent of ECMWF RMS values for Leningradskaya are greater than 50 gpm, while 10% are greater than 100 gpm and are the largest errors observed for the period. Similar, although slightly smaller, errors are found for the NCEP analyses for this station and communications difficulties are suspected. Figure 19 shows the monthly 200 hPa geopotential height bias for the ECMWF and NCEP analyses. Approximately 87% of ECMWF 200 hPa values are biased by less than 30 gpm. The lower portion of Figure 19 shows biases for comparisons with the NCEP analyses, which are dominated by the early errors in the continental interior at South Pole and Vostok. For 1992

to 1993 the NCEP analyses become progressively negatively biased, primarily as the result of Casey and Mirnyy stations. The 1993 NCEP analyses bias for these two stations are typically-40 gpm. While some stations are noted as being notoriously faulty in both analyses at the 200 hPa level, several stations also show tremendous fidelity. Comparisons with Mawson in particular found maximum monthly RMS errors to be 37 gpm for the ECMWF and 42 gpm for NCEP,


Table 1. Average Monthly RMS for Comparison of ECMWF and NCEP Analyses with 17 Rawinsonde Stations (15 Available for 700 and 850 hPa Levels)

Variable Level, hPa Analysis Average Monthly RMS 1985 1986 1987 1988 1989 1990 1991 1992 1993

T, øC 850 ECMWF 3.1 3.2 3.2 3.2 3.3 3.3 3.3 3.9 2.7 NCEP 2.9 3.0 4.2 4.2 3.1 2.6 2.7 2.8 2.6

500 ECMWF 1.5 1.6 1.7 1.8 1.9 1.8 1.6 1.5 1.5 NCEP 2.5 2.1 1.6 1.6 1.5 1.6 1.6 1.4 1.3

200 ECMWF 1.6 1.5 1.5 1.5 1.4 1.4 1.4 1.3 1.3 NCEP 2.5 2.3 1.8 1.9 1.7 1.7 2.1 1.7 1.7

q, g kg -• 850 ECMWF 0.38 0.40 0.39 0.36 0.38 0.37 0.37 0.34 0.30 NCEP 0.33 0.38 0.39 0.38 0.32 0.31 0.34 0.32 0.28

700 ECMWF 0.29 0.25 0.22 0.21 0.22 0.22 0.21 0.22 0.20

U, m s -• 850 ECMWF 4.8 4.8 4.7 4.3 4.4 4.4 4.7 4.0 4.3 NCEP 5.5 5.2 4.8 4.9 4.8 4.5 5.3 4.4 4.5

500 ECMWF 4.0 3.7 3.5 3.6 3.3 3.4 3.5 3.8 3.8 NCEP 6.4 5.2 4.1 4.5 4.0 4.0 4.6 4.7 4.6

200 ECMWF 3.2 2.7 2.5 2.5 2.3 2.5 2.5 2.6 2.6 NCEP 6.0 4.6 3.7 3.8 3.3 3.3 4.0 3.9 3.8

V, m s -• 850 ECMWF 5.3 5.2 5.2 4.8 4.8 5.0 4.8 4.4 4.7 NCEP 4.5 4.3 4.5 4.3 4.6 4.3 4.6 4.3 4.4

500 ECMWF 3.5 3.4 2.9 3.0 2.8 3.0 3.0 3.2 3.1 NCEP 4.9 4.6 3.5 3.6 3.2 3.2 3.9 3.8 3.8

200 ECMWF 2.6 2.7 2.1 2.1 2.0 2.3 2.4 2.2 2.3 NCEP 4.3 4.0 3.1 3.1 2.7 2.7 3.4 3.2 3.2

Z, gpm 850 ECMWF 18.6 18.4 18.2 16.6 16.3 16.9 17.8 15.8 14.3 NCEP 27.7 27.2 24.8 27.5 19.4 20.7 26.6 22.4 22.3

500 ECMWF 25.9 22.8 20.2 19.2 21.2 20.2 19.8 18.6 18.2 NCEP 47.8 32.5 20.8 24.1 19.1 20.3 28.5 24.5 27.3

200 ECMWF 35.2 31.9 27.5 26.5 32.6 26.9 23.3 25.2 24.4 NCEP 56.8 42.6 28.7 29.5 28.6 29.0 33.2 33.4 34.7

ECMWF, European Centre for Medium-Range Weather Forecasts; NCEP, National Centers for Environmental Prediction.

with most values for both analyses less than 17 gpm. Analyses values for Molodezhnaya and Halley Bay also show typical RMS errors of 20 gpm. This is particularly encouraging for comparisons at Molodezhnaya, which contains the most complete twice-daily archive for the Antarctic, while the result at Halley Bay is typical of the excellent results found at stations located east of the Weddell

Sea.

While data truncation occurs for several stations at 200 hPa

the situation improves marginally at 500 hPa. RMS errors are still large for Leningradskaya and Mirnyy comparisons, although typical values are less than 40 gpm and average approximately 28 gpm for both stations. In Figure 20 the monthly averaged 500 hPa geopotential height time series for six stations are plotted with corresponding values from the numerical analyses. With the exception of NCEP analyses, comparisons for the continental interior stations of Vostok and McMurdo prior to May 1986, mean seasonal and interannual variability is closely reproduced by both analyses. For Bellingshausen there is a notable decrease in the negative biases of both analyses. Biases incrementally improve by an average of 21 gpm in the ECMWF analyses and 52 gpm in the NCEP analyses in 1993 as compared with 1985 differences. Other stations depicted show greater accuracy in reproducing monthly mean values.

Tables 1 and 2 show typical RMS errors for temperature to be 1 o to 3øC for all levels and find the NCEP analyses to be somewhat superior to the ECMWF for middle levels of the troposphere after 1987. Figure 21 shows the monthly mean time series at six stations for 500 hPa temperature. The largest discrepancies occur at Vostok. ECMWF biases at Vostok are characterized by differences typically less than 2øC until September 1986, large positive biases of between 4 ø and 7øC from September 1986 until September 1991, and positive values between 1 and 3øC from September 1991 until the end of the time series in December 1991. September 1986 corresponds to a revision in the use of observations in the analyses, which appears to have had an adverse impact here, while September 1991 again corresponds to a change in the forecast model resolution [Trenberth, 1992]. The large positive biases noted for Vostok in the ECMWF analyses are not found at South Pole or McMurdo, where on the average slightly negatively biases are found for both stations, although there is a small decrease in the bias at both stations after

1991. The largest temperature biases in the NCEP analyses occur at Leningradskaya but are generally less than a degree for all stations after May 1986. Largest RMS values at 500 hPa for both analyses occur at Vostok and South Pole, however, suggesting that both analyses and the ECMWF in particular have some difficulties in reproducing variability


Table 2. Average Monthly Statistical Scores for Comparison of ECMWF and NCEP Analyses With Four Rawinsonde Stations for 1989-1993

Variable Station Analysis RMS Bias Correlation Standard Error

700 hPa Casey ECMWF 25.0 -4.4 0.939 17.3 2', gpm NCEP 30.1 2.7 0.913 22.4

Halley Bay ECMWF 13.3 -4.2 0.985 11.3 NCEP 16.9 -6.5 0.976 14.1

Molodezhnaya ECMWF 9.4 -2.0 0.990 8.6 NCEP 14.5 -6.7 0.979 12.1

McMurdo ECMWF 33.6 -8.2 0.912 18.5

NCEP 34.2 -16.5 0.910 19.0

500 hPa Casey ECMWF 25.3 - 10.8 0.971 19.8 2', gpm NCEP 31.5 -9.1 0.951 25.8

Halley Bay ECMWF 13.1 -3.1 0.987 11.8 NCEP 18.2 -4.4 0.976 16.2


McMurdo ECMWF 28.9 -6.5 0.949 21.2

NCEP 31.1 -15.2 0.953 21.6

200 hPa Casey ECMWF 34.1 -25.5 0.981 19.6 Z, gpm NCEP 41.9 -25.5 0.965 26.5

Halley Bay ECMWF 14.5 -0.3 0.990 11.8 NCEP 21.8 3.0 0.977 18.4


McMurdo ECMWF 43.3 - 17.8 0.943 27.7

NCEP 48.0 -15.5 0.940 28.1

500 hPa Casey ECMWF 1.7 -0.4 0.899 1.4 T,øC NCEP 2.0 -1.1 0.893 1.5

Halley Bay ECMWF 1.1 0.3 0.949 0.9 NCEP 1.0 0.1 0.950 0.9

Molodezhnaya ECMWF 1.1 0.3 0.961 0.9 NCEP 1.1 -0.1 0.959 0.9

McMurdo ECMWF 1.9 -0.6 0.859 1.2

NCEP 1.9 -0.3 0.854 1.3

500 hPa Casey ECMWF 4.9 -2.0 0.909 3.7

U, rn s -• NCEP 5.8 - 1.4 0.851 4.4 Halley Bay ECMWF 2.8 0.4 0.912 2.7

NCEP 3.2 0.2 0.893 3.1

Molodezhnaya ECMWF 2.4 0.2 0.962 2.2 NCEP 3.3 0.2 0.930 2.9

McMurdo ECMWF 3.8 0.2 0.789 2.8

NCEP 4.2 -0.1 0.787 3.3

500 hPa Casey ECMWF 3.9 -0.8 0.859 3.1

V, rn s -• NCEP 5.0 -0.4 0.743 4.0 Halley Bay ECMWF 3.2 0.0 0.923 3.0

NCEP 3.4 -0.4 0.920 2.9

Molodezhnaya ECMWF 2.2 0.0 0.953 2.0 NCEP 2.7 0.3 0.930 2.4

McMurdo ECMWF 4.3 0.2 0.805 3.2

NCEP 4.0 -1.2 0.839 3.0


300

250

m 150

g

• 50 ß

• 0

g -50

• B ELL EC• --a--C^SE .... ß .... DAVI ....... • HALY • LENI •-+"-- •SN • MCMR .......................... MIRN ..... •;• ........ MOLO ----'t• --- NEUM

ß . • ..... '.• .../X ..... .•.•..•.•. NOVO •..• POLE ....... ß :•... ß SANA • SYOW • VICE

• • VOST

-150 I • t • • • • • • t

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Date

300 N•E• • BELL •CASE 250 ........ ß ........ DAVI

,.• -.,..,-•'.'-'t•,-- • DUMT • 200 • HALY ._ • LENI m• 150 ...... • ...... MWSN • • •O• MCMR • ....................... MIRN '• I & ........ •i: -•^" MOLO '• •i'.&&• _ "'"l• ':ø' NEUM I : ,0 • 0 •.• --•--•OL•- I

'"?"•"i --•!'-•- SANA ] • -5o '• ...:a½• --o--ViCE I -lOO +VOST ]

-150 I I I I I I I I

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994

Date

Figure 19. Monthly 200 hPa geopotential height bias for ECMWF (top) and NCEP (bottom) numerical analyses for 17 Antarctic rawinsonde stations, in geopotential meters.

above the semi-permanent Antarctic inversion over the plateau. For 1989 to 1993 the average analyses RMS is greater than 2øC for both stations. Other stations depicted in Figure 21 generally have smaller discrepancies in monthly mean values and in RMS. As with geopotential height, both analyses appear to have great success in reproducing values for stations located on the eastern coast of the Weddell Sea

(Halley Bay, Neumayer, and SANAE) for which typical RMS values are less than 1.5øC.

RMS errors in winds found in Table 1 show similar trends

in analyses improvements as was found for other variables. Zonal wind RMS is generally larger for both analyses, which is an indication of the differences in magnitudes between zonal and meridional winds that is generally found in high southern latitudes. At 500 hPa the wind bias averaged for all stations is essentially zero for zonal and meridional components after 1989. There is large variability between perimeter stations which is characterized by over or underestimation of zonal winds. Among stations with unique difficulties are South Pole and Vostok. At South Pole the

major problem is with the mean direction, where observations since 1987 indicate that average winds at 500 hPa are from the Weddell sector, while analyzed winds are more parallel

with the prime meridian. This record is highly biased by a lack of winter reports however. From the available reports at South Pole the magnitude of the ECMWF analyses are about 7% weaker while the NCEP analyses are in near parity with observation. At Vostok both analyses are very similar in producing 500 hPa winds that are more zonal than those observed, with average magnitudes in agreement with observations. Meridional RMS errors for Vostok average 4 m s -• for the ECMWF and 4.6 m s -• for NCEP. A feature of interest in the upper air wind analyses is the meridional wind field at 850 hPa, the first coastal standard level above the katabatic wind flow, and the only wind field for which the NCEP analyses are found to be consistently superior to the ECMWF. Figure 22 shows the mean value time series for six stations, again reflecting the tremendous variability in skill from station to station. Although values generally reflect interannual trends, biases are large, as is evident for Davis.

5. Maritime Surface Evaluation

As has been noted in the previous section, the majority of rawinsonde stations are located along the perimeter of the East Antarctic continent. These stations are periodically


5400 ............... Station •'5300 Bellingshausen Davis ECMWF

• i• ' .• .•.• •.• ' ...... NCEP •5200

4700 ' I

1985 1987 1989 1991 1993 198 Date

5400 • • •

]Halley Bay

1

ooo l

1985 1987 19•ate 1991 1993 1 3

5400

McMurdo

5300 5200 ,

4700 ' I ' I '

1985 1987 19•ate 1991 1993 1985 1987 19•ate 1991 1993

Figure 20. Monthly averaged 500 hPa geopotential height versus time for six rawinsonde stations, with corresponding ECMWF and NCEP analyses values, in geopotential meters.

visited by supply ships and research vessels. Observations taken by the supply ships may be available for incorporation into the analyses via the global telecommunications system (GTS). Although such observations by, themselves are inadequate for a numerical analysis of the Southern Ocean, which represents a major void of conventional data, they are necessary for the augmentation of the satellite data that are primarily used. In Figure 23 ship tracks of the S.A. Agulhas [Petty, 1995] are shown during the years 1991-1993. The S.A. Agulhas resupplies SANAE from Cape Town, South Africa. In Figure 24 the ship pressure and temperature reports are compared with numerical analysis values for the time period over which data are available. The pressure correlation is 0.95 for the ECMWF analyses and 0.86 for NCEP, and standard errors are 5.2 and 8.1 hPa for the ECMWF and

NCEP analyses respectively. Disturbingly, however, a low- pressure reading on October 28, 1992 (observation 60) of 946.1 hPa is overestimated by more than 30 hPa in the ECMWF analyses and 32 hPa in the NCEP analyses. The low-pressure value was attained near 60øS. In the comparison of temperature data, there again is general agreement with many data values. Correlation for the ECMWF analyses with

temperature observations is 0.94, while the NCEP correlation is 0.93. In early March 1993 (observations 126-132) the observations differ sharply with the analyses during a period of time when the ship is in close proximity to the Antarctic coast, possibly as a result of localized phenomena associated with the coastline. All other values are closely matched by the analyses. There is the general conclusion that available supply ship data are being incorporated into the analyses, however several caveats are apparent. On the basis of this example, the maximum intensity of mid-ocean cyclones may be underestimated by the analyses. Additionally, a comparison of the wind values found little correlation with the analyses.

As an analog to the S.A. Agulhas, results are presented for three ship cruises of the R/V Nathaniel B. Palmer [Jeffries, 1994] in the Bellingshausen and Amundsen Seas region for 1993-1994. Ship data for the three cruises were obtained from the University of Wisconsin and Antarctic Support Associates. Figure 25 shows ship tracks departing Puenta Arenas, Chile, for round trip shuttle to and from Palmer Station, McMurdo Station, and Auckland, New Zealand. Unlike Antarctic supply shipping, the Nathaniel B. Palmer


-20

,.•-25

?.

-30

• -40

:• -5o

Be!!ingshausen Davis '• '•'•'•'• S tation ECMWF

...... NCEP

-55 • I ' I ' I • I ' I ' I ' I ' l

1985 1987 1981•at e 1991 1993 1985 1987 1981•at e 1991 1993

alley Bay eningradskaya

F7-25 •-25 1 •-30 •-•0

_40 .40 -45 -50 :• -50

-55 -55 ' I '

1985 1987 1981•at e 1991 1993 1985 1987 1981•at e 1991 1993

-20

..,-25

•.-35

• -45

• -50

McMurdo -20

Vostok

-25

-30

-35

-45

~50

1985

-55 ' I ' I ' I ' I I ' I ' I ' I

1985 1987 1981%te 1991 1993 1987 1989 1991 1993 Date

Figure 21. Monthly averaged 500 hPa temperature versus time for six rawinsonde stations, with corresponding ECMWF and NCEP analyses values, in degrees centigrade.

was primarily engaged in physical and biological studies of the pack ice in the region. A large majority of the observations occurred near 70øS. No meteorological observations were transmitted via the GTS (W.M. Connolley, British Antarctic Survey, personal communication, 1996) and were not available to operational weather centers. Table 3 presents results of computed statistics for the comparison to the analyses. Figure 26 shows comparisons of pressure readings with the analyses (Cruise 1 comparison after Cullather et al. [1996]). The comparisons are exceptional given the scarcity of data in the region. The observed low pressure of 934.3 hPa is again overestimated in the analyses. The ECMWF analyses produce a value of 946 hPa, while NCEP shows 940 hPa. The other two cruises do not exhibit

the tremendous range found in Cruise 1, but it may be seen that the analyses closely reproduce the observed values. In comparison with Cruises 1 and 2 the performance of the analysis pressure values during Cruise 3 is not so close to observation. There is no directly attributable reason for this, although it may be seen from Figure 26 that the observational record for Cruise 3 is not so complete.

Figure 27 shows the temperature comparisons to the observed ship values. In Cruise 2, as depicted in Figure 25,

the ship is observed to follow sea ice leads to within a very close proximity to the continent. In fact, the ECMWF topography contains no grid points adjacent to the ship track that are near sea level, and this has an obvious detrimental effect on temperature comparisons. This problem is not shared by the NCEP analyses which compare more closely to observation. In Cruises 1 and 3 it may be seen that the ECMWF produces a greatly smoothed representation of the temperature variability. The NCEP analyses contain a temperature variability similar to the analyses but at some cost to accuracy. The standard error for the NCEP analyses is substantially larger than for the ECMWF for Cruises 1 and 3. Further study indicates that the observational temperature is approximately 60% correlated with wind direction. This is not surprising given the contrast between the cold southern continental air and relatively warm oceanic air to the north of the sea ice edge. The ECMWF analyses are not so closely correlated. This may be due to the lack of initialization in the ECMWF analyses. Averaged for all cruises, however, the analyses are within approximately 1.0øC of the observed values. In Figure 28, comparisons with the observed relative humidity are shown. Some of the observed trends are similar to variations in the ECMWF, however the values are


!0

Date 1985 1987 198•ate 1991 1993

6 mo•d•naj• --I •12 I 9 . , i • 0 : ' ,, • ' • 6

-• • I ' • • I ' ]

1985 1987 19•ate 1991 1993 1985 1987 19•ate 1991 1993

12 McMrd• SANAE•

3

-3 ' I ' I ' • '[ I

1985 1987 1981gat e 1991 1993 1985 1987 1981gat e 1991 1993

Figure 22. Monthly averaged 850 hPa meridional wind versus time for six rawinsonde stations, with corresponding ECMWF and NCEP analysis values, in meters per second.

approximately 10% too high. The ECMWF analyses report 2 m dew point, rather than relative humidity, and the conversion to relative humidity requires accuracy in both temperature and dew point fields. The overestimation of

0 o

• 60 øS

Figure 23. Ship track for S.A. Agulhas [after Petty, 1995, see his Figure 15].

moisture near the surface, however, is consistent with results from Bromwich et al. [1995] from comparisons against rawinsondes. NCEP relative humidity values show unusual variability throughout each cruise. While mean values for each cruise are similar to observation, the average standard error approaches 10%. Figures 29 and 30 present results for zonal and meridional wind comparisons respectively. The wind comparisons are very reasonable, with correlations of about 0.65 for the ECMWF and 0.57 for NCEP, which should be considered exceptional given the ship observations are translating through the analyses grid. The only significant departures from wind values occurred in Cruise 3 as the ship was under way for Auckland.

6. Summary

An evaluation of the Antarctic numerical analyses has been conducted using available rawinsonde, AWS, ship, and synthesized long-term observations. A comparison of ECMWF analyses with pressure readings from AWS units of the U.S. Antarctic Program reveals tremendous agreement using a variety of statistical measures for the 10-year period from 1985-1994. The NCEP analyses over the same period show substantial improvement, particularly during the period 1985 to 1990. Other AWS-observed variables do not agree so


-9O

-80 [ -70

-50

-40

-30

0 20 40 60 80 100 120 140 160 Obs. No.

1040

1030 1020 1010

lOOO

990 980

970 960

950

940 i i t

0 20 40 60

Agulhas A • ECMWF

80 i 00 120 140 160 Obs. No.

25

20 &

15

10

o• 5

-lO

-15

-2o

-25

0 20 40 60 80 100 120 140 160 Obs. No.

Figure 24. Time series of latitude for S.A. Agulhas (top). Comparison of mean sea level pressure observations from S.A. Agulhas with ECMWF and NCEP analyses (middle), in hectopascals. Comparison of temperature observations from S.A. Agulhas with ECMWF and NCEP analyses (bottom), in degrees centigrade. Observation numbers correspond to dates as follows: 1- 41, January 23, 1991 to February 18, 1991; 43- 84, October 18, 1992 to November 13, 1992; 86-101, December 4-12,1992; 103- 110, January 8,1993 to January 12, 1993; 112-154, February 28, 1993 to March 22, 1993; 156- 165, April 16-20, 1993; 167- 174: May 14-18, 1993.

closely, which may result from analyses error, the localized nature of the fields, or a combination. A validation of the

near-surface temperature and wind fields with synthesized long-duration studies also show the ECMWF to be superior to the NCEP analyses, which generally shows a smoothed representation of the observations. In particular, the ECMWF compared closely to the long-term temperature field synthesized from observational data. Neither analysis was able to adequately reproduce barrier winds near the Antarctic Peninsula, although several other features of the katabatic wind field are adequately reproduced by the ECMWF analyses.

Validation of the analyses standard pressure level fields using available rawinsonde data reveal a general long-term decrease in RMS errors with time for both analyses. Unacceptable errors in the NCEP analyses prior to May 1986

over the central plateau of the continent have largely been removed, however a significant upward trend from 1989 to 1993 in geopotential height RMS differences is apparent at several levels. Grid values at specific stations including the now-closed Leningradskaya base and Mimyy were found to be conspicuously deficient at the 200 hPa level for both analyses, while generally excellent results were obtained at Mawson and stations located east of the Weddell Sea. NCEP

midtropospheric temperature analyses were generally found to be superior to the ECMWF after 1987. Positive biases between 4 ø and 7øC were found in the ECMWF temperature analyses at Vostok, although both analyses show large RMS errors for interior stations. Statistical scores computed for wind component validation show large variability between stations at all levels.

Comparisons of analysis values with supply ship


Figure 25. Ship tracks of R/V Nathaniel B. Palmer.

observations were made using observations from the S.A. Agulhas during the years 1991-1993. Pressure, temperature, and relative humidity variables were in close agreement, although a low pressure of 946.1 hPa was greatly overestimated by both analyses. In general, however, there was substantial evidence that the majority of data was being incorporated into the analyses. Further ship comparisons were made using data from the R/V Nathaniel B. Palmer for three 1-month-long cruises in 1993-1994. Observations made by the Nathaniel B. Palmer were not transmitted via the GTS. Comparisons again show substantial agreement, despite the substantial conventional data void. Comparisons of pressure, temperature, relative humidity, and wind fields find the ECMWF to be in substantial agreement with the observational data, with the NCEP analyses also in adequate agreement.

Although the ECMWF analyses appear to be remarkably consistent over the time period examined, this study highlights the hazards of using operational analyses for climatic study. A preliminary investigation of the NCEP reanalysis product shows improvement over the corresponding operational fields including the near-surface temperature; however, there are several caveats. An error was incurred during the assimilation of PAOBs (paid observations) data (E. Kalnay, personal communication, 1996), which are southern hemisphere point estimates of sea level pressure obtained from manually analyzed charts [Seaman et al., 1993]. The effect of this error on Antarctic


[•)o] ajn:lejadma. l. • [Do] aJn:teJadma•l'.

t•

[Do] a.•n•eJadma,L •

[edq] •nss•d iaaa'I eaS uealAI [edq] a•nss•d iaaa'I eaS uealAI


[•_s m] pu!A• fl

N

[t_s m] pu!A• fl

N

[t_s m] pu!A • fl

[ %] •!plumH aa!l•!alt


10

-20 t -30

8/11/93 00 Z

Cruise 1 •N.B. Palmer

x ECMWF

. • --•--NEI•P t•

8/19/93 00 Z 8/27/93 00 Z 9/4/93 00 Z 9/12/93 00 Z 9/20/93 00 Z 9/28/93 00 Z Date

3O

Cruise 2 20

10

2/14/94 12 Z 2/22/94 12 Z 3/2/94 12 Z 3/10/94 12 Z 3/18/94 12 Z 3/26/94 12 Z 4/3/94 12 Z Date

Cruise 3 2O

-10

-20

-30

9/11/94 12 Z

I I I

9/19/94 12 Z

I

9/27/94 12 Z 10/5/94 12 Z 10/13/94 12 Z 10/21/94 12 Z Date

Figure 30. As in Figure 26 but for the meridional wind component, in meters per second.

fields is not known. Additionally, a spectral distortion in high-latitude moisture fields has been found [Cullather et al., 1996], and this requires further examination.

Several conclusions of this study are drawn with regard to the FROST program. There is the general determination that a large amount of the observational data is being incorporated, and the comparisons to ship data not available to the numerical weather centers imply that the satellite data are having a major beneficial impact on the analyses. Although significant improvements are found, procedures for transferring available observational data to the GTS require further refinement. The authors have been in contact with

station meteorological personnel at South Pole during the July 1994 FROST SOP. The transmission of South Pole

rawinsonde reports has been found to be unusually eventful. Reports are first transmitted to McMurdo for forwarding to New Zealand and eventual transmission onto the GTS. This

procedure of interstation data transfer through McMurdo is not always successful, however, and sometimes results in data exclusion from the analyses. In November 1991 an alternative route via Casey was established (S. Warren, personal communication, 1996). The Casey link has since

been abandoned due to an apparent increase in reliability at McMurdo; however, preliminary FROST results indicate losses still occur.

In addition to communication problems it is clear that the total amount of non-AWS direct observational data for

Antarctica, particularly the number of upper air reports, is presently in decline as a result of former Soviet base closures. It is hoped that this is only a temporary trend and will eventually reverse. Data scarcity and the loss and/or appearance of observational sites places additional priorities on the availability of satellite data and the NWP models for proper data assimilation. The evaluation of the global models used in the data assimilation will be addressed as part of the FROST exercise, and this may shed light on some of the deficiencies identified here.

Acknowledgments. The authors thank Grant Petty for providing data from the S.A. Agulhas, originally obtained from the South African Weather Bureau; Thomas Parish for unpacking and formatting the Antarctic rawinsonde data; William Connolley for examination of the GTS archive; and Linjuan Gong for assistance in data processing. Antarctic AWS station data were obtained from the University of Wisconsin, and numerical analyses and rawinsonde


data were obtained from NCAR. This research was sponsored by the National Science Foundation under grant ATM-9422104, and by the National Aeronautics and Space Administration under grant NAGW 3677. This is contribution 1025 of the Byrd Polar Research Center.

References

Allison, I., Proposals .for the Implementation of a Programme of Antarctic Research Contributing to the World Climate Research Programme, 65 pp., Scott Polar Res. Inst., Cambridge, England, 1983.

Arpe, K., and H. Cattle, A comparison of surface stress and precipitation fields in short-range forecasts over the Antarctic region, J. Geophys. Res., 98, 13,035-13,044, 1993.

Ballish, B., X. Cao, E. Kalnay, and M. Kanamitsu, Incremental nonlinear normal-mode initialization, Mon. Weather Rev., 120, 1723-1734, 1992.

Bromwich, D.H., An extraordinary katabatic wind regime at Terra Nova Bay, Antarctica, Mon. Weather Rev., 117, 688-695, 1989.

Bromwich, D.H., and F.M. Robasky, Recent precipitation trends over the polar ice sheets, Meteorol. Atmos. Phys., 51,259-274, 1993.

Bromwich, D.H., and S.R. Smith, Eds., U.S. FROST: Data and Science Plan, Report from the U.S. FROST Workshop, BPRC Rep. 7, 40 pp., Byrd Polar Res. Cent., Ohio State University, Columbus, 1993.

Bromwich, D.H., F.M. Robasky, R.I. Cullather, and M.L. Van Woert, The atmospheric hydrologic cycle over the Southern Ocean and Antarctica from operational numerical analyses, Mon. Weather Rev., 123, 3518-3538, 1995.

Brown R.A. and L. Zeng, Estimating central pressures of oceanic midlatitude cyclones, J. Appl. Meteorol., 33, 1088-1095, 1994.

Budd, W.F., P.A. Reid, and L.J. Minty, Antarctic moisture flux and net accumulation from global atmospheric analyses, Ann. Glaciol., 21, 149-156, 1995.

Connolley, W.M., and J.C. King, Atmospheric water-vapour transport to Antarctica inferred from radiosonde data, •.J.R. Meteorol. Soc., 119, 325-342, 1993.

Cullather, R.I., D.H. Bromwich, and M.L. Van Woert, Interannual variations in Antarctic precipitation related to E1 Nifio-Southern Oscillation, J. Geophys. Res., 101, 19,109-19,118, 1996.

Escoffier C., and C. Provost, Wind forcing over the southwest Atlantic: Comparisons between observations and ECMWF analyses, Mon. Weather Rev., 123, 1269-1287, 1995.

Genthon, C., and A. Braun, ECMWF analyses and predictions of the surface climate of Greenland and Antarctica, J. Clim., 8, 2324- 2332, 1995.

Gibson, J.K., A. Hernandez, P. K•tllberg, A. Nomura, E. Serrano, and S. Uppala, Current status of the ECMWF Reanalysis (ERA) project, in Proceedings of Seventh Symposium on Global Change Studies, pp. 112-115, Am. Meteorol. Soc., Boston, Mass., 1996.

Giovinetto, M.B., N.M. Waters, and C.R. Bentley, Dependence of Antarctic surface mass balance of temperature, elevation, and distance to open ocean, J. Geophys. Res., 95, 3517-3531, 1990.

H•ikkinen, S., Seasonal simulation of the Southern Ocean coupled ice-ocean system, J. Geophys. Res., 100, 22,733-22,748, 1995.

Heinemann, G., A wintertime polar low over the eastern Weddell Sea (Antarctica): A study with AVHRR, TOVS, SSM/I, and conventional data, Meteorol. Atmos. Phys., 58, 83-102, 1996.

Hines, K.M., D.H. Bromwich, and T.R. Parish, A mesoscale modeling study of the atmospheric circulation of high southern latitudes, Mon. Weather Rev., 123, 1146-1165, 1995.

Howarth, D.A., An analysis of the water vapor flux divergence field over the Southern Hemisphere. Ann. Assoc. Am. Geogr., 76, 190- 207, 1986.

Jeffries, M.O., R/V Nathaniel B. Palmer cruise NBP93-5: Sea-ice physics and biology in the Bellingshausen and Amundsen Seas, August and September 1993, Antarct. J. U.S., 29(1), 7-8, 1994.

Kalnay, E., M. Kanamitsu, and W.E. Baker, Global numerical weather prediction at the National Meteorological Center, Bull. Amer. Meteor. Soc., 71, 1410-1428, 1990.

Kalnay, E., et al., The NCEP/NCAR 40-year reanalysis project, Bull. Am. Meteorol. Soc., 77, 437-471, 1996.

Kanamitsu, M., Description of the NMC global data assimilation and forecast system, Weather Forecasting, 4, 335-342, 1989.

Masuda, K., Atmospheric heat and water budgets of polar regions: Analysis of FGGE data, Proc. NIPR Symp. Polar Meteorol. Glaciol., 3, 79-88, 1990.

Parish, T.R., Surface winds over the Antarctic continent: A review, Rev. Geophys., 26(1), 169-180, 1988.

Parish, T.R., and D.H. Bromwich, Continental-scale simulation of the Antarctic katabatic wind regime, J. Clim., 4, 135-146, 1991.

Parrish, D.F. and J.C. Derber, The National Meteorological Center's spectral statistical-interpolation analysis system, Mon. Weather Rev., 120, 1747-1763, 1992.

Petty, G.W., Frequencies and characteristics of global oceanic precipitation from shipboard present-weather reports, Bull. Am. Meteorol. Soc., 76, 1593-1616, 1995.

Quintanar, A.I. and C.R. Mechoso, Quasi-stationary waves in the Southern Hemisphere. Part I: Observational data, J. Clim., 8, 2659-2672, 1995.

Schubert, S.D., R.B. Rood, and J. Pfaendtner, An assimilated dataset for earth science applications, Bull. Am. Meteorol. Soc., 74, 2331- 2341, 1993.

Seaman, R., P. Steinle, and T. Hart, The impact of manually derived Southern Hemisphere sea level pressure data upon forecasts from a global model, Weather Forecasting, 8, 363-368, 1993.

Stearns, C.R., and G. Wendler, Research results from Antarctic automatic weather stations, Rev. Geophys., 26(1), 45-61, 1988.

Stearns, C.R., L.M. Keller, G.A. Weidner, and M. Sievers, Monthly mean climatic data for Antarctic automatic weather stations, in Antarctic Meteorology and Climatology: Studies Based on Automatic Weather Stations, Antarct. Res. Ser., vol. 61, edited by D.H. Bromwich and C.R. Stearns, pp. 1-21, AGU, Washington, D.C., 1993.

Trenberth, K.E., Global analyses from ECMWF and atlas of 1000 to 10 mb circulation statistics, NCAR Tech. Note, NCAR/FN- 373+STR, 191 pp., 24 fiche, Natl. Cent. for Atmos. Res., Boulder, Colo., 1992.

Trenberth, K.E., Atmospheric circulation climate changes, Clim. Change, 31,427-453, 1995.

Trenberth, K.E., and J.G. Olson, An evaluation and intercomparison of global analyses from NMC and ECMWF, Bull. Am. Meteorol. Soc., 69, 1047-1057, 1988a.

Trenberth, K.E., and J.G. Olson, Intercomparison of NMC and ECMWF global analyses: 1980-1986, NCAR Tech. Note, NCAR/TN-301+STR, 81 pp., Natl. Cent. for Atmos. Res., Boulder, Colo., 1988b.

Trenberth, K.E., and J.G. Olson, Evaluation of NMC global analyses, NCAR Tech. Note, NCAR/TN-299+STR, 82 pp., Natl. Cent. for Atmos. Res., Boulder, Colo., 1988c.

Trenberth, K.E., and J.G. Olson, Temperature trends at the South Pole and McMurdo Sound, J. Clim., 2, 1172-1182, 1989.

Trenberth, K.E., and A. Solomon, Implications of global atmospheric spatial spectra for processing and displaying data, J. Clim., 6, 531-545, 1994.

Turner, J., et al., The Antarctic First Regional Observing Study of the Troposphere (FROST) Project, Bull. Am. Meteorol. Soc., 77, 2007-2032, 1996.

Tzeng, R.-Y., D.H. Bromwich, T.R. Parish, and B. Chen, NCAR CCM2 simulation of the modern Antarctic climate, J. Geophys. Res., 99, 23,131-23,148, 1994.

Yamazaki, K., Moisture budget in the Antarctic atmosphere, Proc. NIPR Symp. Polar Meteorol. Glaciol., 6, 36-45, 1992.

D. H. Bromwich and R. I. Cullather, Polar Meteorology Group, Byrd Polar Research Center, The Ohio State University, Columbus, OH 43210-1002.

R. W. Grumbine, NOAA/NWS/NCEP, W/NMC21, Room 206, 5200 Auth Road, Camp Springs, MD 20746-4304.

(Received March 8, 1996; revised September 6, 1996; accepted October 8, 1996.)

Validation of operational numerical analyses in Antarctic ...

Documents