• Alpert, J. C., , M. Kanamitsu, , P. M. Caplan, , J. G. Sela, , G. H. White, , and E. Kalnay, 1988: Mountain induced gravity wave drag parameterization in the NMC medium-range model. Preprints, Eighth Conf. on Numerical Weather Prediction, Baltimore, MD, Amer. Meteor. Soc., 726–733.

  • Bengtsson, L., , S. Hagemann, , and K. I. Hodges, 2004a: Can climate trends be computed from reanalysis data? J. Geophys. Res., 109 .D11111, doi:10.1029/2004JD004536.

    • Search Google Scholar
    • Export Citation
  • Bengtsson, L., , K. I. Hodges, , and S. Hagemann, 2004b: Sensitivity of the ERA40 reanalysis to the observing system: Determination of the global atmospheric circulation from reduced observations. Tellus, 56A , 456471.

    • Search Google Scholar
    • Export Citation
  • Bouttier, F., , and G. Kelly, 2001: Observing-system experiments in the ECMWF 4D-Var data assimilation system. Quart. J. Roy. Meteor. Soc., 127 , 14691488.

    • Search Google Scholar
    • Export Citation
  • Bromwich, D. H., , and R. L. Fogt, 2004: Strong trends in the skill of the ERA-40 and NCEP–NCAR reanalyses in the high and midlatitudes of the Southern Hemisphere, 1958–2001. J. Climate, 17 , 46034619.

    • Search Google Scholar
    • Export Citation
  • Cadet, D., 1973: Étude du cisaillement du vent horizontal dans la stratosphèrePh.D. thesis, Université de Paris, 104 pp.

  • Cadet, D., , and J. Ovarlez, 1974: La mesure précise de la température dans la troposphère et la basse stratosphère: Équilibre thermodynamique du senseur. Tech. Rep. 62, Laboratoire de Météorologie Dynamique, CNRS, France, 13 pp.

  • Fels, S. B., , and M. D. Schwartzkopf, 1975: The simplified exchange approximation: A new method for radiative transfer calculations. J. Atmos. Sci., 32 , 14751488.

    • Search Google Scholar
    • Export Citation
  • Fourrier, J., , P. Morel, , and P. Sitbon, 1970: Ambient temperature measurements from constant-level balloons. J. Appl. Meteor., 9 , 154157.

    • Search Google Scholar
    • Export Citation
  • Hagan, M. E., , J. M. Forbes, , and F. Vial, 1993: A numerical investigation of the propagation of the quasi 2-day wave into the lower thermosphere. J. Geophys. Res., 98 , 2319323205.

    • Search Google Scholar
    • Export Citation
  • Hagan, M. E., , M. D. Burrage, , J. M. Forbes, , J. Hackney, , W. J. Randel, , and X. Zhang, 1999: GSWM-98: Results for migrating solar tides. J. Geophys. Res., 104 , 68136828.

    • Search Google Scholar
    • Export Citation
  • Hertzog, A., , C. Basdevant, , F. Vial, , and C. R. Mechoso, 2004: The accuracy of stratospheric analyses in the northern hemisphere inferred from long-duration balloon flights. Quart. J. Roy. Meteor. Soc., 130 , 607626.

    • Search Google Scholar
    • Export Citation
  • Hodges, K. I., , B. J. Hoskins, , J. Boyle, , and C. Thorncroft, 2003: A comparison of recent reanalysis datasets using objective feature tracking: Storm tracks and tropical easterly waves. Mon. Wea. Rev., 131 , 20122037.

    • Search Google Scholar
    • Export Citation
  • Jenne, R., 1999: Global observations for reanalysis, 1948-on. Proc. Second WCRP Int. Conf. on Reanalyses, WCRP Series Rep. 109, Waterfield Park, Reading, United Kingdom, World Meteorological Organization, 5–9.

  • Kållberg, P., , A. Simmons, , S. Uppala, , and M. Fuentes, 2004: The ERA-40 archive. ERA-40 Project Rep. Series 17, ECMWF, Shinfield Park, Reading, United Kingdom, 31 pp.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kanamitsu, M., 1989: Description of the NMC global data assimilation and forecast system. Wea. Forecasting, 4 , 334342.

  • Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82 , 247267.

    • Search Google Scholar
    • Export Citation
  • Lacis, A. A., , and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the Earth’s atmosphere. J. Atmos. Sci., 31 , 118133.

    • Search Google Scholar
    • Export Citation
  • Lott, F., , and M. J. Miller, 1997: A new subgrid-scale orographic drag parametrization: Its formulation and testing. Quart. J. Roy. Meteor. Soc., 123 , 101127.

    • Search Google Scholar
    • Export Citation
  • Marshall, G. J., 2002: Trends in Antarctic geopotential height and temperatures: A comparison between radiosonde and NCEP–NCAR reanalysis data. J. Climate, 15 , 659674.

    • Search Google Scholar
    • Export Citation
  • Mlawer, E. J., , S. J. Taubman, , P. D. Brown, , M. J. Iacono, , and A. S. Clough, 1997: Radiative transfer for inhomogeneous atmospheres: RRTM a validated correlated-k model for the longwave. J. Geophys. Res., 102 , 1666316682.

    • Search Google Scholar
    • Export Citation
  • Mo, K. C., , X. L. Wang, , R. Kistler, , M. Kanamitsu, , and E. Kalnay, 1995: Impact of satellite data on the CDAS-reanalysis system. Mon. Wea. Rev., 123 , 124139.

    • Search Google Scholar
    • Export Citation
  • Morel, P., , and W. Bandeen, 1973: The EOLE experiment: Early results and current objectives. Bull. Amer. Meteor. Soc., 54 , 298306.

  • Morel, P., , and M. Desbois, 1974: Mean 200-mb circulation in the southern hemisphere deduced from EOLE balloon flights. J. Atmos. Sci., 31 , 394407.

    • Search Google Scholar
    • Export Citation
  • Pan, H-L., , and W-S. Wu, 1995: Implementing a mass flux convection parameterization package for the NMC medium-range forecast model. NMC Office Note 409, 40 pp.

  • Simmons, A. J., , and J. K. Gibson, 2000: The ERA-40 project plan. ERA-40 Project Rep. Series 1, ECMWF, Shinfield Park, Reading, United Kingdom, 63 pp.

  • Simmons, A. J., , P. D. Jones, , V. da Costa Bechtold, , A. C. M. Beljaars, , P. W. Kållberg, , S. Saarinen, , S. M. Uppala, , and N. Wedi, 2004: Comparison of trends and low-frequency variability in CRU, ERA-40, and NCEP–NCAR analyses of surface air temperature. J. Geophys. Res., 109 .D24115, doi:10.1029/2004JD005306.

    • Search Google Scholar
    • Export Citation
  • Sitbon, P., 1975: Platform location and data collection by satellite systems. IEEE Trans. Geosci. Electr., 13 , 217.

  • Sterl, A., 2004: On the (in)homogeneity of reanalysis products. J. Climate, 17 , 38663873.

  • Tiedtke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117 , 17791800.

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., 1982: Seasonality in southern hemisphere eddy statistics at 500 mb. J. Atmos. Sci., 39 , 25072520.

  • Trenberth, K. E., 1991: Storm tracks in the southern hemisphere. J. Atmos. Sci., 48 , 21592178.

  • View in gallery

    Location of observations during the EOLE experiment.

  • View in gallery

    Number of temperature observations per day during the EOLE experiment. Only the observations that are considered to be of good quality and that have been used for the comparisons with the reanalyses are counted. A 5-day running mean has been applied. The tick marks indicate the beginning of the corresponding month.

  • View in gallery

    Heating of the temperature sensors vs solar zenith angle (solid line). The sunsets and sunrises occur at about 93.5° at the balloon altitude. The dashed curve shows the correction used to overcome the daytime warm bias.

  • View in gallery

    Temperature differences vs latitude: (left) NCEP–NCAR reanalysis minus EOLE observations; (middle) ECMWF reanalysis minus EOLE observations; (right) ECMWF reanalysis minus NCEP–NCAR reanalysis.

  • View in gallery

    Geographical structure of the reanalysis biases: (top row) NN50 minus EOLE, (middle row) ERA-40 minus EOLE, and (bottom row) ERA-40 minus NN50. Temperature, zonal-velocity, and meridional-velocity differences are shown in the left, middle, and right columns, respectively. Hatched areas correspond to boxes with less than 10 balloon observations.

  • View in gallery

    Mean meridional structure of (left) temperature and (right) zonal velocity during the EOLE experiment: observations (solid), NN50 (dashed), and ERA-40 (dashed–dotted). Note that the reanalysis meridional structure has been obtained by using only the values at the balloon locations. As a consequence, these plots are somewhat biased toward the summer season (cf. Fig. 2).

  • View in gallery

    Temperature differences for (left) winter and (right) summer seasons: (top) NN50 minus EOLE and (bottom) ERA-40 minus EOLE. Boxes where the reanalysis cold (warm) bias exceeds 3 K are colored in light (dark) gray. Hatched areas correspond to boxes with less than 10 balloon observations in each season. Winter (summer) months included in the comparisons are JJASO (DJF).

  • View in gallery

    Same as in Fig. 5, but for the reanalysis standard deviations. Note that the algorithm to remove the measurement noise has been applied to construct the figures in the first two rows (see section 4).

  • View in gallery

    Standard deviations of (left) temperature, (middle) zonal-velocity, and (right) meridional-velocity departures from monthly and zonally averaged fields: EOLE (solid), NN50 (dashed), and ERA-40 (dashed–dotted).

  • View in gallery

    Temperature standard deviations of reanalyses for (left) winter and (right) summer seasons: (top) NN50 and (bottom) ERA-40. Boxes where the reanalysis standard deviations exceeds 3 K (5 K) are colored in light (dark) gray. Hatched areas correspond to boxes with less than 10 balloon observations in each season. The seasonally averaged standard deviation is indicated in the lower-left corner of each panel. Winter (summer) monthes included in the comparisons are JJASO (DJF).

  • View in gallery

    Same as in Fig. 10, but for the meridional-velocity standard deviations. Boxes where the reanalysis standard deviations exceeds 12 m s−1 (15 m s−1) are colored in light (dark) gray.

  • View in gallery

    Mean geopotential height on the 200-hPa surface on 20 Feb 1972: NN50 reanalysis (solid) and ERA-40 reanalysis (dashed). The contours are plotted every 100 m. The boldface contours correspond to the 11 700-m geopotential height. The balloon winds for the same day are shown with arrows whose length is proportional to the wind speed. An arrow corresponding to a 30 m s−1 speed is plotted on the lower-left corner of each panel.

  • View in gallery

    Same as in Fig. 12, but on 14 Dec 1971. The boldface contours correspond to the 11 900-m geopotential height.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 7 7 2
PDF Downloads 2 2 1

An Assessment of ECMWF and NCEP–NCAR Reanalyses in the Southern Hemisphere at the End of the Presatellite Era: Results from the EOLE Experiment (1971–72)

View More View Less
  • 1 Laboratoire de Météorologie Dynamique, Université Pierre et Marie Curie, IPSL, CNRS, Palaiseau, France
© Get Permissions
Full access

Abstract

This article estimates the biases and standard deviations of the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) and the 50-yr National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) Reanalysis (NN50) in the upper troposphere and lower stratosphere in 1971–72. These estimates are obtained by comparing the reanalyzed temperatures and winds with EOLE observations, a dataset collected during 480 superpressure-ballon flights in the Southern Hemisphere (SH). Dedicated algorithms have been developped to control the quality of this dataset and a stringent selection has been performed on the observations. None of the atmospheric centers has assimilated the EOLE dataset, which is therefore fully independent from the reanalyses. It is furthermore argued that the statistics obtained in this study at the end of the presatellite era may be representative of the reanalysis accuracy since 1957. The results of these comparisons indicate that NN50 tends to be a few degrees colder than the observations in the SH subpolar latitudes, while ERA-40 is less hit by this cold-pole issue. Both reanalyses, on the other hand, are found to be warmer than the observations by about 1 K in the subtropics. In contrast, the wind comparisons only exhibit nonsignificant or small reanalysis biases, even though the reanalyzed subtropical jet is slightly displaced equatorward with respect to the observations. The ability of reanalyses to capture the atmospheric synoptic-scale variability in the upper troposphere is assessed by computing the standard deviations of the reanalysis minus observation differences. The ERA-40 and NN50 standard deviations show a maximum (i.e., a poorer reanalysis accuracy) in the SH storm track. However, ERA-40 standard deviations are found to be much larger than NN50 standard deviations. The standard deviations also exhibit a marked decrease above the continents, stressing the heterogeneity of the atmospheric observation network during the presatellite era. Finally, in contrast with previous studies, the reanalysis accuracy does not appear to be better during summer than during winter.

Corresponding author address: A. Hertzog, Laboratoire de Météorologie Dynamique, École Polytechnique, F-91128 Palaiseau CEDEX, France. Email: hertzog@lmd.polytechnique.fr

Abstract

This article estimates the biases and standard deviations of the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) and the 50-yr National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) Reanalysis (NN50) in the upper troposphere and lower stratosphere in 1971–72. These estimates are obtained by comparing the reanalyzed temperatures and winds with EOLE observations, a dataset collected during 480 superpressure-ballon flights in the Southern Hemisphere (SH). Dedicated algorithms have been developped to control the quality of this dataset and a stringent selection has been performed on the observations. None of the atmospheric centers has assimilated the EOLE dataset, which is therefore fully independent from the reanalyses. It is furthermore argued that the statistics obtained in this study at the end of the presatellite era may be representative of the reanalysis accuracy since 1957. The results of these comparisons indicate that NN50 tends to be a few degrees colder than the observations in the SH subpolar latitudes, while ERA-40 is less hit by this cold-pole issue. Both reanalyses, on the other hand, are found to be warmer than the observations by about 1 K in the subtropics. In contrast, the wind comparisons only exhibit nonsignificant or small reanalysis biases, even though the reanalyzed subtropical jet is slightly displaced equatorward with respect to the observations. The ability of reanalyses to capture the atmospheric synoptic-scale variability in the upper troposphere is assessed by computing the standard deviations of the reanalysis minus observation differences. The ERA-40 and NN50 standard deviations show a maximum (i.e., a poorer reanalysis accuracy) in the SH storm track. However, ERA-40 standard deviations are found to be much larger than NN50 standard deviations. The standard deviations also exhibit a marked decrease above the continents, stressing the heterogeneity of the atmospheric observation network during the presatellite era. Finally, in contrast with previous studies, the reanalysis accuracy does not appear to be better during summer than during winter.

Corresponding author address: A. Hertzog, Laboratoire de Météorologie Dynamique, École Polytechnique, F-91128 Palaiseau CEDEX, France. Email: hertzog@lmd.polytechnique.fr

1. Introduction

The European Centre for Medium-Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) have produced comprehensive atmospheric reanalyses known respectively as the 40-yr ECMWF Re-Analysis (ERA-40) and the 50-yr NCEP–NCAR Reanalysis (NN50). NN50, which begins in 1948, is continuously updated and thus extends to the present (Kalnay et al. 1996; Kistler et al. 2001). On the other hand, ERA-40 begins with the International Geophysical Year (IGY) in September 1957 and stops in 2002 (Simmons and Gibson 2000).

The basic idea of reanalyses is to use a general circulation model coupled with an assimilation system that is kept constant for the whole analyzed period. The output atmospheric fields are consequentely free of secular changes that may result from changes in the physics of the model or in the way observations are treated. Reanalyses are particularly well suited for the detection of long-term changes or oscillations in the atmosphere and have become widely used among the community for such studies. However, the global meteorological observing system has significantly changed during the second half of the twentieth century. For instance, upper-air observations mostly relied on radiosoundings (as well as pilot balloons) after 1957 up to the second half of the 1970s, after which satellite observations have progressively become more and more important. This change in the observing system can result in spurious trends in the reanalyzed fields (Marshall 2002; Bengtsson et al. 2004a; Bromwich and Fogt 2004). This issue is particularly relevant in the Southern Hemisphere (SH) where the radiosounding station network was (and still is) far more sparse than in the Northern Hemisphere. The reanalysis skills were therefore much more improved by the use of satellite observations in the SH than in the Northern Hemisphere (Mo et al. 1995; Jenne 1999; Kistler et al. 2001; Simmons et al. 2004).

A possible way to assess this issue is to compute trends in dedicated experiments in which the observation network is kept constant for a long period (Bengtsson et al. 2004a). Alternatively, one can also try to estimate the accuracy of presatellite reanalyses by comparing (postsatellite) reanalyses where the satellite data have and have not been assimilated (Mo et al. 1995).

Another worthwhile approach is to compare presatellite reanalyses with observations gathered at the same time (Marshall 2002; Simmons et al. 2004). This method provides direct estimates of the reanalysis bias and skill in capturing atmospheric variability, which can then be used as inputs that characterize the reanalysis accuracy in trend computations. However, one of the drawbacks of this approach is that the observation–reanalysis comparisons are performed at discrete locations on the earth, and consequently the results are not necessarily representative of the reanalysis global behavior. Furthermore, as pointed out by Simmons et al. (2004), most of these presatellite observations have been assimilated by the reanalyses so that the comparisons may tend to overestimate the reanalysis actual performance.

In this study, we will also directly estimate the accuracies and biases of ERA-40 and NN50 reanalyses but we will use an observationnal dataset that has neither been assimilated by ECMWF nor by NCEP–NCAR (Jenne 1999; Kållberg et al. 2004). It is therefore fully independent from the reanalyses. This dataset has been gathered in 1971 and 1972 during flights of superpressure balloons in the SH upper troposphere and lower stratosphere. This balloon experiment, named EOLE after the French name for the Greek god of the winds, was designed by the Laboratoire de Météorologie Dynamique and by the French Space Agency (CNES) in collaboration with the National Aeronautics and Space Administration (NASA). The dataset size (roughly 80 000 temperature and wind observations), its spatial coverage of the whole SH, and temporal extension of approximately 1 yr actually allow one to obtain reliable and relevant statistics on the accuracy of reanalyzed fields.

In contrast with Marshall (2002) and Bromwich and Fogt (2004) who compared long time series of observed and reanalyzed monthly mean fields, almost instantaneous fields are compared in this study. Hence, while the former articles are primarily aimed at quantifying the impact of a changing observationnal network in the detection of atmospheric trends in reanalyses, the results obtained in this study will also characterize how well the shorter synoptic-scale variability is captured in the reanalyzed fields. Hodges et al. (2003) also adressed this issue in their study that intercompared various reanalyses.

The outline of the article is as follows. Section 2 presents the EOLE experiment, the observations collected during the campaign and the accuracy of the meteorological measurements. The last part of that section deals with the quality control that has been performed on the observations to identify spurious measurements. Section 3 briefly describes the main characteristics of the reanalyses that have been used in this article. The characteristics of the observation network in the SH in 1971–72 are also shortly discussed. Section 4 presents the methodology that has been applied to compare the observations with the reanalysis. Sections 5 shows the results of this comparison (i.e., the geographical structure of the reanalysis biases and standard deviations). The last section summarizes and briefly discusses the implication of the results obtained in this study.

2. The EOLE experiment

a. Description of the experiment

The EOLE experiment took place between August 1971 and December 1972. In total, 480 superpressure, 3.70-m-diameter balloons were launched from three sites in Argentina (Mendoza, 33°S; Neuquen, 39°S; Ushuaïa, 55.5°S) during the first four months of the experiment (Morel and Bandeen 1973). The mean balloon lifetime was 103 days, but some balloons kept on sending data for more than 1 yr. At the maximum (October–December 1971), more than 200 EOLE balloons were drifting in the SH atmosphere. After July 1972, only about 30 balloons were still drifting and no data after that date were considered in this study.

Superpressure balloons have the property to keep their volume constant in flight and therefore to drift on constant-density (isopycnic) surfaces. The nominal flight level of the EOLE balloons was 0.328 kg m−3. On the other hand, the atmospheric pressure varies slightly during the flights: the mean pressure recorded during EOLE is 208 hPa and more than 99% of pressure records are between 190 and 230 hPa. The EOLE balloons thus drifted in the SH upper troposphere and lower stratosphere (UTLS). Figure 1 shows the balloon positions for the whole dataset considered in this study. One can notice the good coverage of the SH between 20° and 70°S achieved during the experiment. In particular, the Antarctic Ocean, which is perhaps the largest data-void area on the earth at that time, is very well sampled. On the other hand, only a few balloon performed observations south of 75°S or north of 15°S.

The whole EOLE dataset amounts to about 120 000 observations, from which 80 000 were considered appropriate for the comparisons with the reanalyses. The observing system, as well as the quality check that was performed to select the observations, are described in the following section. Figure 2 shows the number of selected temperature observations per day during the EOLE experiment. During the campaign peak, there were more than 600 daily temperature observations. Similar figures are obtained for the winds (Morel and Desbois 1974). This number is to be compared with the daily number of radiosondes and aircraft data assimilated by NN50 south of 20°S at the same period, which reaches about 400 and 100, respectively.

b. Meteorological observing system

1) Pressure

The EOLE pressure sensors, which are used to determine the balloon vertical positions, had an accuracy of 2 hPa, corresponding approximately to an altitude accuracy of 60 m (Morel and Bandeen 1973). A small number of pressure sensors were leaking and were erroneously indicating low pressure levels. This problem has been corrected as indicated at the end of this section.

2) Horizontal position

The horizontal position of the EOLE balloons was determined by the ancestor of the modern ARGOS space localization system: a satellite devoted to the balloon localization was launched just before the begining of the campaign in August 1971. The satellite orbit had a 50° inclination over the equator, which did not enable the localization of balloons that were south of 75°S.

The horizontal-localization process was based on the measurement by the satellite of the delay and Doppler shift of a signal transmitted by the balloon. The algorithm used to locate the balloons results in the determination of two possible balloon positions, almost symmetrical with respect to the satellite ground track (Sitbon 1975). Both locations were recorded in the dataset and the continuity of the balloon trajectories was the main criteria used to flag the correct one. The final accuracy of the horizontal positioning system (which includes timing issues, frequency shifts, and ionospheric noise) was measured at the time of the EOLE experiment and estimated at 2 km (1σ; Morel and Desbois 1974; Sitbon 1975).

The wind horizontal components are deduced from these locations by centered finite differences. They were computed only when the time delay between two successive balloon locations was less than 300 min, which corresponds to three orbital periods. Owing to the satellite-balloon communication protocol, there are typically three such closely spaced positions per balloon and per day. The balloon velocities are thus 1.5–5-h averages of the wind velocities and are therefore well suited for the comparisons with the 6-hourly reanalyses. The corresponding 1σ accuracy of the horizontal-velocity estimates is less than 0.5 m s−1.

However, since pressure records are used to interpolate the balloon positions in the reanalyses, the noise in pressure measurements induces a corresponding noise in the interpolated velocities. Taking into account the vertical shear of the horizontal wind at 200 hPa in the SH, which can reach 1.5 10−2 s−1, the additional noise in the interpolated velocities amounts to 1 m s−1 (1σ; Cadet 1973). The final standard deviation on interpolated velocities is therefore estimated at 1.1 m s−1.

3) Temperature

Special attention was paid during the campaign to the use of very small thermistors and to mount them on almost transparent glass plates in order to avoid as much as possible the solar heating of the sensors and their environment. The temperature accuracy, which was estimated at the time of the campaign, is 0.5 K (Fourrier et al. 1970).

However, despite these precautions, we have found that a warm bias affects daytime temperatures, as shown in Fig. 3. This bias results from the sensor heating by solar radiations. We empirically estimated it and corrected the raw observations as follows.

  • First, a mean nighttime temperature Tn(λ) is computed for 1° wide latitude bands between 50° and 8°S. Nighttime temperatures are actually expected to be almost unbiased (Fourrier et al. 1970; Cadet and Ovarlez 1974). The latitudinal southern limit was used to avoid severe seasonal effect (i.e., comparing winter, nighttime temperatures with summer, daytime temperatures at polar and subpolar latitudes).
  • The solar zenith angle (SZA) is computed for each balloon position, and individual “heatings” (ΔT) are estimated according to ΔT(SZA) = T(SZA, λ) − Tn(λ).
  • The individual heatings are binned into 1° SZA boxes and the mean heating is computed for each box (solid curve in Fig. 3).
  • The daytime thermistor heating ΔT is estimated by fitting an exponential law to these observed heatings at SZA < 93.5° (which corresponds to the day–night transition at the balloon altitude).
  • The resulting correction (dashed curve in Fig. 3), which reaches 1.3 K at SZA < 70°, is finally substracted from the daytime temperature observations.

In this algorithm, the observed daytime warming is entirely attributed to the thermistor heating by the solar radiation. In particular, any natural warming of the UTLS is neglected. Diurnal variations of temperature at 200 hPa are indeed expected to be very small: relevant models, for example, the Global Scale Wave Model (Hagan et al. 1993, 1999), predict that typical amplitudes of diurnal and semidiurnal tide components at the balloon altitude do not exceed one-tenth of a degree. Furthermore, the observed diurnal cycle in the measured temperatures does not look like a combination of a small number of sine and cosine functions, which should be the case for tidal disturbances.

The diurnal thermistor heating determined as above is found to be very similar to the one obtained on recent balloon flights that use the same measurement technique but with a better sampling rate (e.g., Hertzog et al. 2004). This higher temporal resolution allows an easier characterization of the sensor behavior. The fact that both results are almost identical gives further confidence in the restitution of the thermistor heating during EOLE.

It should also be noted that a significant number of thermistors (∼20%) encountered problems during the flights. When the balloon passed into clouds or into almost-saturated areas, some thermistors actually broke or experienced ice deposit on their surface, which modify their behavior for the rest of the flight. The algorithm used to discard such erroneous records is described in the following section.

As for the velocities, we used the observed pressure to define the balloon vertical coordinate when we interpolate the reanalysis at the balloon position. Thus, the uncertainty on pressure observations similarly adds a contribution to the uncertainty on interpolated temperatures. Statistics performed in ERA-40 temperature fields show that the maximum absolute temperature gradient in the UTLS is about 10 K km−1, which leads to a further uncertainty of 0.6 K in the interpolated temperature. The total standard deviation on temperature is therefore estimated at 0.8 K.

c. Data quality control

This article aims at comparing gridded reanalyses with balloon-borne observations, which is achieved through the interpolation of the modeled fields onto the balloon position. The data quality control must therefore address two different issues. First, the atmospheric variables that are compared (temperature or wind) have to be checked to remove measurement errors. Second, the 3D balloon locations have to be carefully controlled to avoid erroneous interpolation.

The horizontal locations have been validated by visually checking the trajectory continuity for each balloon. As recalled previously, much of the work needed to flag the correct balloon positions was done at the time of the campaign. We only swap a very small amount of those flags, which were associated with obvious location errors.

The way of detecting erroneous vertical positions and physical measurements differs whether one deals with wind comparisons or with temperature comparisons, in which case it is simpler.

1) Temperature comparisons

If the pressure and temperature records are correct, one can compute the balloon density. Alternatively, if one of the records is incorrect, so is the density. Now, superpressure balloons have the property to stay on isopycnic surfaces. Besides, the nominal EOLE-balloon density is known to about 3% because of small variations in the balloon volume or craft mass. Thus, for each flight, the following procedure has been adopted to flag the correct temperature/pressure observations: a mean flight density has been computed by considering only the records that agree to within 3% with the expected flight level (0.328 kg m−3). Next, within each flight, the records that show an absolute departure of more than 5% from the mean flight density have been discarded from the temperature comparisons. (These small density variations during the flight arise from two sources: the natural oscillation of the device about its equilibrium level and changes in balloon volume due to changes in radiative fluxes impinging on the balloon). The EOLE-balloon material was actually not able to bear density fluctuations (and the corresponding overpressure variations) greater than 5% (P. Cocquerez 2005, personal communication).

In total, 88 158 records have been finally selected for the temperature comparisons. An inspection of discarded points reveals that the vast majority of them are associated with a failure of the temperature sensor.

2) Wind comparisons

Since the erroneous horizontal positions of the balloons have already been discarded by the trajectory inspection, the aim here is to detect erroneous vertical positions. The previous algorithm cannot however be used as such, for it would lead to the rejection of many points that are actually associated with a temperature-measurement problem (and not a pressure problem).

The selection of records used for wind comparisons nevertheless starts with the points for which the density (and thus pressure) is correct. These records, which will be used for the wind comparisons, also serve to compute a flight-mean pressure (Pd) and temperature (Td). Then, among the records with an incorrect density, those that verify both following conditions are also included in the comparisons.

  • The relative variation of temperature is 4 times greater than the relative variation of pressure, that is,
    i1520-0493-134-11-3367-e1
    where i is the record index. This ensures that the density issue most likely comes from the temperature measurement.
  • The relative variation of pressure is less than 10%, that is,
    i1520-0493-134-11-3367-e2
    which excludes doubtful pressure measurements.
Finally, among all these selected records, couples of points less than 300 min apart are searched for the computation of the wind velocity. In total, 79 131 couples of record have been found to meet all these criteria.

3. Reanalyses

a. ERA-40

Aspects of the ERA-40 reanalysis are given in Simmons and Gibson (2000) and Kållberg et al. (2004). (The full description of the model and assimilation system is available online at http://www.ecmwf.int/research/ifsdocs/CY23r4.)

Briefly, the model used to produce the ERA-40 reanalysis is a T159 spectral model, with 60 hybrid levels in the vertical from the surface to 0.1 hPa. The primitive equations are integrated with the help of a semi-Lagrangian scheme. The physical parameterizations include the Tiedtke convective scheme (Tiedtke 1989), a rapid radiative code (Mlawer et al. 1997), and the subgrid-scale orographic-drag scheme of Lott and Miller (1997). A three-dimensional variational data assimilation (3DVAR) system is used to incorporate the observations in the model.

The observation network in the SH during the EOLE experiment (mid-1971 to mid-1972) was particularly sparse. Upper-air observations assimilated at that time by ERA-40 come almost entirely from radiosoundings and pilot-balloon soundings over the continents. Over the oceans, a few radiosoundings launched from ships were assimilated, as well as a few in situ aircraft data. In particular, no satellite data have been used by ERA-40 at the time of the campaign. The radiance assimilation of the eight-channel Vertical Temperature Profile Radiometer (VTPR) on board the National Oceanic and Atmospheric Administration (NOAA) satellite NOAA-2 begins in November 1972. Similarly, the pressure pseudoobservations (PAOBs) issued by the Australian Bureau of Meteorology, which are aimed at filling the observation gap over the SH oceanic areas, are assimilated since November 1972. Both satellite and PAOB datasets are known to significantly contribute to the analysis improvement in the SH (Bouttier and Kelly 2001; Bengtsson et al. 2004b).

b. NN50

The NN50 reanalysis is described in Kalnay et al. (1996) and Kistler et al. (2001). Briefly, the NN50 reanalysis uses a T62 spectral model with 28 sigma levels in the vertical. The model top is located at 3 hPa. A semi-implicit scheme is used for the temporal integration of the primitive equations (Kanamitsu 1989). A comprehensive set of physical and subgrid-scale parameterizations is used, including a simplified Arakawa–Schubert convective scheme (Pan and Wu 1995), a radiative heating parameterization (Lacis and Hansen 1974; Fels and Schwartzkopf 1975), and a gravity wave drag based on the mean orography (Alpert et al. 1988). The assimilation of observational data is performed with a 3DVAR assimilation system.

Similar to ERA-40, most of the upper-air data assimilated by NN50 during the EOLE experiment in the SH comes from continental radiosounding stations. Some aircraft data and radiosondes launched from ships are assimilated over the ocean. Similar to ERA-40, neither PAOB nor satellite radiance are assimilated in NN50 in 1971–72 (NN50 begins the assimilation of VTPR data in 1975).

Still, in contrast with ERA-40, NN50 assimilates the first satellite-derived cloud winds since 1967 (Jenne 1999). The Applications Technology Satellite-1 (ATS-1) was actually launched in December 1966, followed 1 yr later by ATS-3. Both satellites were able to make photographs of the earth cloud cover with a sufficient quality and sampling rate to estimate atmospheric winds from cloud motions. The satellites were respectively located at 151°W (central Pacific) and 69°W (Central America) and contributed to fill in the data gap over the Pacific Ocean. Yet, much of the cloud winds are obtained in the intertropical belt (30°S–30°N), with the exception of some few observations down to 45°S over the eastern Pacific and western Atlantic (typically less than two observations per month and per 2.5° × 2.5° box during EOLE). It is therefore believed that the impact of this dataset in the SH midlatitudes at the time of the EOLE experiment is rather small.

c. Final remarks

The IGY in 1957–58 was a major step in the development of the SH observation network. For instance, Antarctic stations were installed at that time and some of them have been permanently occupied since then. Up to the end of the late 1970s, however, the network did not change significantly, and radiosondes were almost the only source of upper-air observations. After 1975 (1972 in ERA-40), the progressive introduction of space observations (i.e., VTPR radiances) helped the reanalyses to perform better in the SH (Kistler et al. 2001). But the major breakthrough occurred in 1979 with the assimilation of the various atmospheric sounders on board the NOAA Television Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS) satellite. That year also coincides with the intensive atmospheric observations performed during the First Global Atmospheric Research Program (GARP) Global Experiment (FGGE), as well as with a marked increase in aircraft observations (Jenne 1999).

The EOLE experiment therefore took place at the end of the presatellite era. Hence, even though the balloon campaign lasted only 1 yr, the results presented in this study are likely representative of the whole 1957–79 period as the upper-air SH observation network was almost frozen during that period. Kistler et al. (2001) and Simmons et al. (2004) have for instance shown that the NN50 and ERA-40 performance in southern latitudes keeps almost constant since 1957 to the middle of the l970s.

4. Methodology

The interpolation of the 6-hourly reanalyzed fields onto the balloon positions has been performed using cubic splines in time and space. The 8 × 8 horizontal grid points that surround the balloon positions, as well as the six temporal states about the balloon observation time have been used to compute the spline. In the vertical, the interpolation was made with the help of nine standard levels: 50, 70, 100, 150, 200, 250, 300, 400, and 500 hPa. The logarithm of pressure was used as the vertical coordinate onto which the spline is applied.

For each selected balloon observation, the difference between the reanalyzed and observed temperatures (as well as horizontal velocities) has been computed. We also computed for the same points the difference between both reanalyses. For instance, the whole set of temperature differences are represented in Fig. 4 as a function of latitude.

These differences have then been binned into latitude–longitude boxes of 4° × 10°, respectively, in order to assess the geograpical structure of the observation/reanalysis discrepancies. Within each box with at least 10 correct observations during the whole EOLE experiment, we finally have computed the mean and standard deviation (i.e., the square root of the variance) of the reanalysis minus observation differences.

To clearly illustrate the meanings of these statistics, let us for instance consider the set of temperature differences within one box:
i1520-0493-134-11-3367-e3
where TiEOLE is the ith EOLE observation in the box, TiREA is the reanalyzed (either NN50 or ERA-40) temperature interpolated at the observation location. This difference can be decomposed into
i1520-0493-134-11-3367-e4
where TiATM is the real atmospheric temperature at the same time and place. Thus, within the box, the mean of the differences reads as
i1520-0493-134-11-3367-e5
We will now assume that the EOLE observations are statistically unbiased, so that the second term in the rhs of the previous equation vanishes. The mean of the reanalysis/observation differences is therefore an estimate of the reanalysis bias in the box. The standard deviation of the differences on the other hand reads as
i1520-0493-134-11-3367-e6
where we have made the assumption that both sets of differences (TiREATiATM and TiEOLETiATM) are independent (which is guaranteed by the fact that the EOLE dataset was not assimilated). The second term in the square root in Eq. (6) is associated with the measurement noise, which has been estimated in section 2 (for both temperature and wind). Equation (6) can therefore be inversed to estimate σ (TiREATiATM). This term, which will be called the reanalysis standard deviation in the rest of the article, therefore quantifies the ability of reanalyses to capture the atmospheric variability in the box.

5. Results

a. Reanalysis bias

The reanalysis biases are shown in Fig. 5. We will first comment the results on temperature (i.e., the left column in Fig. 5).

1) Temperature

The bias of NN50 with respect to EOLE observations looks almost zonally symmetric. NN50 tends to be warmer than EOLE at tropical and subtropical latitudes, whereas the converse is true at subpolar latitudes. The mean meridional temperature gradient during the EOLE experiment is shown in Fig. 6. EOLE observations are significantly weighted by summer observations (cf. Fig. 2), and consequentely polar temperatures are higher than tropical temperatures. Note that, in order to keep coherent with the observations, the meridional structure in the reanalyses shown in this figure has been obtained by using only the interpolated values at the balloon locations. NN50 obviously tends to underestimate the observed meridional temperature gradient. Quantitative statistics are reported in Table 1. The NN50 warm bias in the subtropics is about 1 K, while the NN50 cold bias reaches 3 K at 70°S. The amplitude of this high-latitude cold bias in NN50 is in agreement with the results of Marshall (2002), who compared NN50 with radiosoundings performed on the edge of the Antarctic continent. The results of a Student’s t statistical test show that all the temperature biases reported in Table 1 are significantly different than 0 at the 95% confidence level.

ERA-40 on the other hand performs better in capturing the temperature meridional gradient, and the cold-pole problem is much smaller than in NN50 (∼0.5 K). Simmons et al. (2004), who compared ERA-40 with monthly mean station data, reported comparable bias values over the Antarctic. Nevertheless, ERA-40 tends also to overestimate upper-tropospheric temperatures in the Tropics by approximately the same amount as NN50.

Direct comparisons between both reanalyses clearly show that ERA-40 is warmer than NN50 at polar and subpolar latitudes, while they tend to agree at lower latitudes. To analyze a step further these discrepancies, we reperformed the previous analysis and separated the summer and winter seasons. [Note that December–February (DJF) months were used to compute the reanalysis summer bias, whereas June–October (JJASO) months were included in the winter bias, due to the smaller number of EOLE observations in this latter season.] Figure 7 shows for both seasons the areas where the absolute differences between the reanalyses and the observations are greater than 3 K. Clearly, the largest discrepancies are found at polar and subpolar latitudes. The extent of the strong cold biases is greater in the summer season (when the polar UTLS is the warmest) in both reanalyses. In the winter on the contrary, the polar UTLS is the coldest and ERA-40 is warmer than the observations, whereas NN50 stays colder than the observations. The extent of the strong NN50 cold biases is however smaller than in summer.

Overall, the picture is thus that both reanalyses have difficulties in capturing the correct amplitude of the annual cycle in temperature in the polar UTLS. NN50 is generally colder than the observations in this region, while ERA-40 is warmer in winter and colder in summer, which leads to a small bias when averaged over the entire year.

2) Wind

The geographical structure of the reanalysis bias in zonal velocity in less clear than in temperature (Fig. 5). Figure 6 shows that both reanalyses essentially succeed in capturing the meridional variations in zonal velocity, although the modeled tropospheric jet is somewhat displaced toward the north with respect to the observed jet. Consequentely, NN50 and ERA-40 overestimate the zonal velocity in the subtropics. ERA-40 has also difficulty in reproducing the zonal-velocity peak at 40°S: the ERA-40 zonal component exhibits a dip at this latitude, leading to a double-jet structure. It can also be noted in Fig. 5 that the midlatitude bias in zonal velocity is generally smaller on the continents than on the oceans.

As for the meridional velocity, the reanalysis biases are generally small and not significant (cf. Tables 1 and 2). The number of EOLE observations at low latitudes is likely too small to correctly capture the upper branch of the Hadley cell and therefore the observed mean magnitude of the meridional velocity is homogeneously small. With no surprise, both reanalyses reproduce this mean structure.

b. Reanalysis standard deviation

The structure of the reanalysis standard deviation is shown in Fig. 8. As outlined in section 4, this statistic quantifies the ability of the reanalyses to reproduce the atmospheric variability observed during the EOLE experiment.

1) General structure

Figure 8 shows that the general structure of the reanalysis standard deviations is similar in NN50 and in ERA-40, although the values are different. In a first approach, we will thus only discuss the ERA-40 results (middle panel in Fig. 8), but most of the comments could also apply to the NN50 standard deviations.

The most striking feature is the annular pattern of the standard deviation maximum: values of standard deviation are generally small in the Tropics and close to the Antarctic, while they exhibit a clear maximum in the SH midlatitudes. This pattern is very similar to the structure of storm tracks in the SH (e.g., Trenberth 1991). To further illustrate this similarity, we have estimated in the balloon dataset the observed short-period atmospheric variability during EOLE. This has been done by computing, for each calendar month, the standard deviations of departures from zonal means. These standard deviations, which quantify the intensity of atmospheric disturbances with periods shorter than 1 month, are shown in Fig. 9 as a function of latitude. Similarly, we estimated the atmospheric variability captured by the reanalyses, by using as previously only the values interpolated at the balloon locations. Figure 8 (reanalysis accuracy) together with Fig. 9 (atmospheric variability) show that the latitudes where the short-term atmospheric variability is the greatest (i.e., the storm stracks) corresponds to the latitudes where the reanalysis standard deviations are the largest (i.e., where the reanalysis accuracy is the poorest). Note in particular that, as in Trenberth (1991), the largest atmospheric variability in zonal velocity is located equatorward of the maximum variability in temperature and meridional velocity (respectively at about 35° and 45°S), and that this latitudinal shift is also exhibited by the reanalysis standard deviation (Fig. 8).

Figure 9 also reveals that both reanalyses succeed in capturing the meridional shape of the atmospheric variability at 200 hPa and that they agree particularly well with each other. Still, there is less energy in the reanalyzed disturbances equatorward of 50°S, and in particular the reanalyses underestimate the observed maximum of synoptic activity. This issue likely results from the almost complete lack of observations over large areas in the SH. Bengtsson et al. (2004a), who compared reanalyses in which satellite observations are/are not assimilated, also noticed that the largest difference in total atmospheric kinetic energy occurred in the SH extratropics. Furthermore, since synoptic disturbances are responsible for the poleward transport of heat in the midlatitudes, this underestimation of synoptic activity may be at least partly responsible for the cold bias of reanalyses in the SH high latitudes.

2) Differences between ERA-40 and NN50

As stated previously, the differences between both reanalyses are not found in the standard deviation geographical structure, but rather in the magnitude of these standard deviations (see Tables 1 and 2 for exact values). NN50 standard deviations are actually significantly smaller than ERA-40, which means that NN50 performs better in reproducing the upper-tropospheric disturbances that are observed by the balloons. Between 40° and 50°S for instance, NN50 standard deviations in temperature, zonal, and meridional velocity amount respectively to 3.8 K, 9.5 m s−1, and 10.1 m s−1, while ERA-40 standard deviations reach 5.3 K, 13.3 m s−1, and 15.2 m s−1. Figure 8 even shows that the reanalysis standard deviations can reach larger values above the oceans (which is further discussed below).

The magnitudes of these reanalysis standard deviations can be compared with those induced by synoptic disturbances, which are shown in Fig. 9. In the midlatitudes, the ERA-40 1σ uncertainties and the root-mean-square magnitudes of atmospheric disturbances are almost equal, which states the poor correlation between the real and ERA-40 SH disturbances. Note also that, although the NN50 standard deviations are less than those of ERA-40, they are still of the same order than those induced by the synoptic activity. Both reanalyses have therefore difficulties in reproducing the real synoptic-scale activity in the SH in 1971–72. Examples will be shown at the end of this section.

3) Influence of continental observations

The influence of continental observations is obvious in the reanalysis standard deviations shown in Fig. 8: in the midlatitudes where the reanalyses experience difficulties in reproducing the atmospheric variability, standard deviations are much smaller above the continents than above the oceans. The most spectacular example occurs above Australia and New Zealand: a sharp decrease in the reanalysis standard deviations (particularly for the velocities) is observed on the western coast of Australia, and these low values extend toward the east up to New Zealand and even a few tenths of degrees farther east. The influence of New Zealand is also particularly clear in the temperature standard deviations at 50°S.

The same decrease of standard deviations is also observed above South America and Africa. This zonal modulation stresses the lack of global observations in the SH in the begining of the 1970s. The accuracy of the reanalyses is thus much higher above the continents (from where most of the radiosondes are launched) than above the oceans. A general feature is also that the positive effect of continents is transported at some distances to the east: following the motion of synoptic disturbances, the reanalysis accuracy is higher on the eastern side of the continents than it is on the western side. The information obtained over the continents is then progressively lost over the oceans, and consequently the reanalysis standard deviations increase from west to east above the Pacific, Atlantic, and Indian Oceans.

A final corollary related to this continental effect is worth stressing: the fact that the EOLE dataset shows so clearly the reasonable improvement of the reanalysis accuracy above continents is an a posteriori check of the consistency of the EOLE observations. In particular, this supports that the balloons were well located and that the EOLE temperature observations were accurate.

4) Seasonal modulations

Trenberth (1982) noted that the latitudinal extent of SH storm tracks is broader in winter than in summer, and that winter disturbances are more energetic than summer ones. An interesting issue is therefore to look at whether such seasonality is also found in the reanalysis standard deviations. We thus divided as previously the EOLE dataset into summer (DJF) and winter (JJASO) months. The corresponding statistics for temperature and meridional velocity are shown in Figs. 10 and 11, respectively. The equivalent plot for zonal-velocity standard deviations is not shown, as it looks very similar to the meridional-velocity figure.

These figures first show that whatever the season the ERA-40 standard deviations are larger than those of NN50. Quite surprinsingly however, the reanalysis standard deviations have either comparable magnitudes in summer and winter (temperature), or are even larger in summer (meridional velocity). These results are in contrast with those of Bromwich and Fogt (2004), who mainly compared surface observations from high-latitude continental stations to the reanalyses. They indeed found that NN50 and ERA-40 are more accurate during summer than during winter. Our results, on the other hand, are primarily estimates of the reanalysis accuracy over the southern oceans at the tropopause level. Two reasons may explain this discrepancy: first, since the SH storm track is somewhat displaced to the north in the winter season and is of broader latitudinal extent, the winter disturbances may likely be partly constrained by the observations performed at continental stations in South America, Africa, Australia, and New Zealand. On the other hand, the summer disturbances mainly circulate above the data-free ocean and may be largely missed by the presatellite observation network. Second, Trenberth (1982) also showed that summertime synoptic-scale disturbances have relatively shorter lifetimes than wintertime synoptic-scale disturbances. Hence, the atmospheric decorrelation time is shorter in summer and the information assimilated by the models over the continents is not valid as far away over the oceans as in winter. This characteristic may explain why the meridional-velocity standard deviations increases more rapidly toward the east on the Pacific or Indian Ocean in summer than in winter.

c. Case studies

To further illustrate some characteristics of the differences between the balloon observations and the NN50 and ERA-40 reanalyses, as well as between both reanalyses, we present in this last part two specific case studies.

The first one took place on 20 February 1972, above the Indian Ocean and is typical of the reanalysis behavior over the southern oceans. It corresponds to a situation where the agreement between NN50 and the balloon observations is particularly good, while on the contrary ERA-40 disagrees with these two datasets. The mean geopotential heights on the 200-hPa surface for this day are shown in Fig. 12, together with the balloon winds collected the same day. Above the Indian Ocean, the NN50 geopotential field reveals the development of an upper-level trough located at about 50°S, 80°E. Some balloons were embedded in this disturbance, and the associated winds enable one to capture the jet entrance on the western side of the trough, the jet exit on the other side of the trough, and even the low winds in its core. On the other hand, this disturbance is totally missed by ERA-40, which rather predicts an anticyclonic ridge in the center of the trough. Consequently, large discrepancies occurred that day between the balloon and the ERA-40 winds: on the western side of the trough for instance, the flow in ERA-40 is almost zonal, whereas the balloons are advected by the jet toward the north. More generally, the two reanalyses produce on that day geopotential-height oscillations above the Indian Ocean that are almost anticorrelated. One can notice on the other hand that both fields agree much better with each other as well as with the balloon winds on the eastern side of Australia and above New Zealand. This emphasizes once again the significant impact of continental observations in these presatellite reanalyses, and the difficulty of models to reproduce the atmospheric state above the data-free southern oceans.

The second case took place on 14 December 1971 above the Pacific Ocean. This time the better agreement is obtained between the observations and the ERA-40 reanalysis, while the NN50 reanalysis tends to disagree with the two previous fields. Such cases are less frequent than the previous one, in agreement with our results on the reanalysis standard deviations. The meteorological situation is shown in Fig. 13 and exhibits a trough roughly located at 40°S, 150°W. Several balloons were located close to this cyclone and the balloon winds capture particularly well the jet structure on the eastern side of the trough. Even though a corresponding low is found in NN50 reanalysis, its intensity is significantly underestimated. On the contrary, ERA-40 succeeds in capturing the large meridional displacements of the balloons induced by this disturbance. Note also that on that day, both reanalyses (and the balloon winds) agree once again on the eastern side of Australia and New Zealand.

6. Conclusions

This study compared NN50 and ERA-40 reanalyses with observations collected during a large balloon campaign that took place in the SH in 1971–72. One of the assets of these balloon observations is that they are not restricted above continental areas. They therefore offer a global perspective on the reanalysis accuracy in the SH before the satellite era. In particular, the EOLE observations achieve very good coverage of the southern oceans, which were otherwise poorly observed at the time of the campaign. The results presented in this study are therefore estimates of the reanalysis bias and standard deviation in the upper troposphere–lower stratosphere during the beginning of the 1970s. Numerous studies already reported that the most significant improvement in the reanalysis skill in the SH occurred in the second half of the 1970s with the progressive assimilation of satellite observations (e.g., Kistler et al. 2001; Marshall 2002; Bromwich and Fogt 2004; Simmons et al. 2004; Sterl 2004). Before that period, the SH upper-air observation network was primarily based on radiosoundings and did not change significantly after the International Geophysical Year (1957). Consequently, the results presented in this study are, at least qualitatively, likely representative of the reanalysis accuracy during the 1957–75 presatellite era.

The overall picture that emerges is that NN50 exhibits a significant cold bias at subpolar latitudes (∼3 K), which is much smaller in ERA-40 (∼0.5 K). In the Tropics, both reanalyses are slightly warm biased (∼1 K). Hence, as stated in Bromwich and Fogt (2004), the smaller ERA-40 bias over very poorly sampled regions tends to support the better climatology of the ECMWF model as compared to the NCEP–NCAR model. These biases are a serious issue, which limits the relevance of long-term trends computed with reanalyses over areas where the observation density changed significantly with time (e.g., Bengtsson et al. 2004a; Bromwich and Fogt 2004).

The reanalysis standard deviations obtained in this study have revealed the difficulty of the reanalyses to adequately reproduce the atmospheric short-term variability over the Southern Oceans. Indeed, the modelization of the atmosphere in the SH subpolar latitudes is likely one of the greatest issue for reanalyses during the presatellite era: this region is actually at the same time almost void of observations and the place where the tropospheric variability is the largest. Nevertheless, it has been found that this issue is more severe in ERA-40 than in NN50. This result agrees with those of Bromwich and Fogt (2004), who found a much better correlation of upper-air observations in the SH high latitudes with NN50 than with ERA-40. The ERA-40 standard deviation in meridional velocity for instance reaches 15.2 m s−1 between 40° and 50°S, which is typically the rms intensity of atmospheric variability at these latitudes. Two case studies have illustrated the large discrepancies that are found in this area between the balloon observations and the reanalyses, as well as between both reanalyses. A much better agreement between both reanalyses is observed above the emerged lands (and especially over Australia and New Zealand), where the reanalysis standard deviations are found to be significantly smaller than above the oceans. The comparisons with the balloon observations have also showed that the positive impact of continental observations propagates somewhat eastward with the disturbances. Finally, over the oceans, the reanalysis standard deviations is found to be pretty much independent of the season.

The differences between NN50 and ERA-40 reported in this study likely do not result from differences in the observations that are assimilated, which are actually almost the same. They rather point toward differences in the model physics or in the relative observation weights in both assimilation systems.

The general issue of observation heterogeneity before and after the satellite era will surely not be fully resolved by continuoulsy adding new, nonassimilated datasets. Nevertheless, the assimilation of the EOLE dataset in future reanalysis projects and the assessment of its impact is certainly worth further studying. It may also provide some information on the contribution of a global balloon-borne observation network to modern assimilation systems.

Acknowledgments

We would like first to express our gratitude to NCAR, who provided us with a copy of the EOLE dataset. We also thank Philippe Cocquerez (CNES) who gave us detailed information on the EOLE balloons. NCEP–NCAR reanalysis data were provided by the NOAA–CIRES Climate Diagnostics Center, Boulder, Colorado, from the mirror Web site (http://climsev.ipsl.polytechnique.fr/). The ECMWF is also gratefully acknowledged for giving the ERA-40 data used in this study. This study has been partly supported by the European Commission Hibiscus Project (Contract EVK2-2001-000111).

REFERENCES

  • Alpert, J. C., , M. Kanamitsu, , P. M. Caplan, , J. G. Sela, , G. H. White, , and E. Kalnay, 1988: Mountain induced gravity wave drag parameterization in the NMC medium-range model. Preprints, Eighth Conf. on Numerical Weather Prediction, Baltimore, MD, Amer. Meteor. Soc., 726–733.

  • Bengtsson, L., , S. Hagemann, , and K. I. Hodges, 2004a: Can climate trends be computed from reanalysis data? J. Geophys. Res., 109 .D11111, doi:10.1029/2004JD004536.

    • Search Google Scholar
    • Export Citation
  • Bengtsson, L., , K. I. Hodges, , and S. Hagemann, 2004b: Sensitivity of the ERA40 reanalysis to the observing system: Determination of the global atmospheric circulation from reduced observations. Tellus, 56A , 456471.

    • Search Google Scholar
    • Export Citation
  • Bouttier, F., , and G. Kelly, 2001: Observing-system experiments in the ECMWF 4D-Var data assimilation system. Quart. J. Roy. Meteor. Soc., 127 , 14691488.

    • Search Google Scholar
    • Export Citation
  • Bromwich, D. H., , and R. L. Fogt, 2004: Strong trends in the skill of the ERA-40 and NCEP–NCAR reanalyses in the high and midlatitudes of the Southern Hemisphere, 1958–2001. J. Climate, 17 , 46034619.

    • Search Google Scholar
    • Export Citation
  • Cadet, D., 1973: Étude du cisaillement du vent horizontal dans la stratosphèrePh.D. thesis, Université de Paris, 104 pp.

  • Cadet, D., , and J. Ovarlez, 1974: La mesure précise de la température dans la troposphère et la basse stratosphère: Équilibre thermodynamique du senseur. Tech. Rep. 62, Laboratoire de Météorologie Dynamique, CNRS, France, 13 pp.

  • Fels, S. B., , and M. D. Schwartzkopf, 1975: The simplified exchange approximation: A new method for radiative transfer calculations. J. Atmos. Sci., 32 , 14751488.

    • Search Google Scholar
    • Export Citation
  • Fourrier, J., , P. Morel, , and P. Sitbon, 1970: Ambient temperature measurements from constant-level balloons. J. Appl. Meteor., 9 , 154157.

    • Search Google Scholar
    • Export Citation
  • Hagan, M. E., , J. M. Forbes, , and F. Vial, 1993: A numerical investigation of the propagation of the quasi 2-day wave into the lower thermosphere. J. Geophys. Res., 98 , 2319323205.

    • Search Google Scholar
    • Export Citation
  • Hagan, M. E., , M. D. Burrage, , J. M. Forbes, , J. Hackney, , W. J. Randel, , and X. Zhang, 1999: GSWM-98: Results for migrating solar tides. J. Geophys. Res., 104 , 68136828.

    • Search Google Scholar
    • Export Citation
  • Hertzog, A., , C. Basdevant, , F. Vial, , and C. R. Mechoso, 2004: The accuracy of stratospheric analyses in the northern hemisphere inferred from long-duration balloon flights. Quart. J. Roy. Meteor. Soc., 130 , 607626.

    • Search Google Scholar
    • Export Citation
  • Hodges, K. I., , B. J. Hoskins, , J. Boyle, , and C. Thorncroft, 2003: A comparison of recent reanalysis datasets using objective feature tracking: Storm tracks and tropical easterly waves. Mon. Wea. Rev., 131 , 20122037.

    • Search Google Scholar
    • Export Citation
  • Jenne, R., 1999: Global observations for reanalysis, 1948-on. Proc. Second WCRP Int. Conf. on Reanalyses, WCRP Series Rep. 109, Waterfield Park, Reading, United Kingdom, World Meteorological Organization, 5–9.

  • Kållberg, P., , A. Simmons, , S. Uppala, , and M. Fuentes, 2004: The ERA-40 archive. ERA-40 Project Rep. Series 17, ECMWF, Shinfield Park, Reading, United Kingdom, 31 pp.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kanamitsu, M., 1989: Description of the NMC global data assimilation and forecast system. Wea. Forecasting, 4 , 334342.

  • Kistler, R., and Coauthors, 2001: The NCEP–NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82 , 247267.

    • Search Google Scholar
    • Export Citation
  • Lacis, A. A., , and J. E. Hansen, 1974: A parameterization for the absorption of solar radiation in the Earth’s atmosphere. J. Atmos. Sci., 31 , 118133.

    • Search Google Scholar
    • Export Citation
  • Lott, F., , and M. J. Miller, 1997: A new subgrid-scale orographic drag parametrization: Its formulation and testing. Quart. J. Roy. Meteor. Soc., 123 , 101127.

    • Search Google Scholar
    • Export Citation
  • Marshall, G. J., 2002: Trends in Antarctic geopotential height and temperatures: A comparison between radiosonde and NCEP–NCAR reanalysis data. J. Climate, 15 , 659674.

    • Search Google Scholar
    • Export Citation
  • Mlawer, E. J., , S. J. Taubman, , P. D. Brown, , M. J. Iacono, , and A. S. Clough, 1997: Radiative transfer for inhomogeneous atmospheres: RRTM a validated correlated-k model for the longwave. J. Geophys. Res., 102 , 1666316682.

    • Search Google Scholar
    • Export Citation
  • Mo, K. C., , X. L. Wang, , R. Kistler, , M. Kanamitsu, , and E. Kalnay, 1995: Impact of satellite data on the CDAS-reanalysis system. Mon. Wea. Rev., 123 , 124139.

    • Search Google Scholar
    • Export Citation
  • Morel, P., , and W. Bandeen, 1973: The EOLE experiment: Early results and current objectives. Bull. Amer. Meteor. Soc., 54 , 298306.

  • Morel, P., , and M. Desbois, 1974: Mean 200-mb circulation in the southern hemisphere deduced from EOLE balloon flights. J. Atmos. Sci., 31 , 394407.

    • Search Google Scholar
    • Export Citation
  • Pan, H-L., , and W-S. Wu, 1995: Implementing a mass flux convection parameterization package for the NMC medium-range forecast model. NMC Office Note 409, 40 pp.

  • Simmons, A. J., , and J. K. Gibson, 2000: The ERA-40 project plan. ERA-40 Project Rep. Series 1, ECMWF, Shinfield Park, Reading, United Kingdom, 63 pp.

  • Simmons, A. J., , P. D. Jones, , V. da Costa Bechtold, , A. C. M. Beljaars, , P. W. Kållberg, , S. Saarinen, , S. M. Uppala, , and N. Wedi, 2004: Comparison of trends and low-frequency variability in CRU, ERA-40, and NCEP–NCAR analyses of surface air temperature. J. Geophys. Res., 109 .D24115, doi:10.1029/2004JD005306.

    • Search Google Scholar
    • Export Citation
  • Sitbon, P., 1975: Platform location and data collection by satellite systems. IEEE Trans. Geosci. Electr., 13 , 217.

  • Sterl, A., 2004: On the (in)homogeneity of reanalysis products. J. Climate, 17 , 38663873.

  • Tiedtke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117 , 17791800.

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., 1982: Seasonality in southern hemisphere eddy statistics at 500 mb. J. Atmos. Sci., 39 , 25072520.

  • Trenberth, K. E., 1991: Storm tracks in the southern hemisphere. J. Atmos. Sci., 48 , 21592178.

Fig. 1.
Fig. 1.

Location of observations during the EOLE experiment.

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 2.
Fig. 2.

Number of temperature observations per day during the EOLE experiment. Only the observations that are considered to be of good quality and that have been used for the comparisons with the reanalyses are counted. A 5-day running mean has been applied. The tick marks indicate the beginning of the corresponding month.

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 3.
Fig. 3.

Heating of the temperature sensors vs solar zenith angle (solid line). The sunsets and sunrises occur at about 93.5° at the balloon altitude. The dashed curve shows the correction used to overcome the daytime warm bias.

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 4.
Fig. 4.

Temperature differences vs latitude: (left) NCEP–NCAR reanalysis minus EOLE observations; (middle) ECMWF reanalysis minus EOLE observations; (right) ECMWF reanalysis minus NCEP–NCAR reanalysis.

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 5.
Fig. 5.

Geographical structure of the reanalysis biases: (top row) NN50 minus EOLE, (middle row) ERA-40 minus EOLE, and (bottom row) ERA-40 minus NN50. Temperature, zonal-velocity, and meridional-velocity differences are shown in the left, middle, and right columns, respectively. Hatched areas correspond to boxes with less than 10 balloon observations.

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 6.
Fig. 6.

Mean meridional structure of (left) temperature and (right) zonal velocity during the EOLE experiment: observations (solid), NN50 (dashed), and ERA-40 (dashed–dotted). Note that the reanalysis meridional structure has been obtained by using only the values at the balloon locations. As a consequence, these plots are somewhat biased toward the summer season (cf. Fig. 2).

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 7.
Fig. 7.

Temperature differences for (left) winter and (right) summer seasons: (top) NN50 minus EOLE and (bottom) ERA-40 minus EOLE. Boxes where the reanalysis cold (warm) bias exceeds 3 K are colored in light (dark) gray. Hatched areas correspond to boxes with less than 10 balloon observations in each season. Winter (summer) months included in the comparisons are JJASO (DJF).

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 8.
Fig. 8.

Same as in Fig. 5, but for the reanalysis standard deviations. Note that the algorithm to remove the measurement noise has been applied to construct the figures in the first two rows (see section 4).

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 9.
Fig. 9.

Standard deviations of (left) temperature, (middle) zonal-velocity, and (right) meridional-velocity departures from monthly and zonally averaged fields: EOLE (solid), NN50 (dashed), and ERA-40 (dashed–dotted).

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 10.
Fig. 10.

Temperature standard deviations of reanalyses for (left) winter and (right) summer seasons: (top) NN50 and (bottom) ERA-40. Boxes where the reanalysis standard deviations exceeds 3 K (5 K) are colored in light (dark) gray. Hatched areas correspond to boxes with less than 10 balloon observations in each season. The seasonally averaged standard deviation is indicated in the lower-left corner of each panel. Winter (summer) monthes included in the comparisons are JJASO (DJF).

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 11.
Fig. 11.

Same as in Fig. 10, but for the meridional-velocity standard deviations. Boxes where the reanalysis standard deviations exceeds 12 m s−1 (15 m s−1) are colored in light (dark) gray.

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 12.
Fig. 12.

Mean geopotential height on the 200-hPa surface on 20 Feb 1972: NN50 reanalysis (solid) and ERA-40 reanalysis (dashed). The contours are plotted every 100 m. The boldface contours correspond to the 11 700-m geopotential height. The balloon winds for the same day are shown with arrows whose length is proportional to the wind speed. An arrow corresponding to a 30 m s−1 speed is plotted on the lower-left corner of each panel.

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Fig. 13.
Fig. 13.

Same as in Fig. 12, but on 14 Dec 1971. The boldface contours correspond to the 11 900-m geopotential height.

Citation: Monthly Weather Review 134, 11; 10.1175/MWR3256.1

Table 1.

NN50 bias and standard deviation in 1971–72 at 200 hPa.

Table 1.
Table 2.

ERA-40 bias and standard deviation in 1971–72 at 200 hPa.

Table 2.
Save