Glenn H. White

Global Modeling Branch/Environmental Modeling Center

National Centers for Environmental Prediction/National Weather Service

National Oceanic and Atmospheric Administration/U.S. Dept. of Commerce

5200 Auth Rd., Washington, DC 20233, USA


Operational data assimilation for numerical weather prediction (NWP) analyses large quantities of many different types of data after carefully checking the data and interpolates the data in space and time using highly accurate short-range forecasts from a complex numerical model of the atmosphere.  (In the near future coupled land, ocean and atmospheric models will be used for operational forecasts.)  Complex parameterization of atmospheric physics produce air-sea fluxes estimates on a regular grid at 3 to 6 hour intervals a few hours after observation time.  Large numbers of diverse users, including marine forecasters, routinely scrutinize operational analyses and forecasts and provide an invaluable check on them.  Forecasting the behavior of a system is a rigorous test, but not the only test, of how well we understand that system.  The forecast skill of NWP systems has increased greatly in recent decades.


Many data assimilation systems now assimilate satellite radiances directly, rather than fields derived by other systems from satellite data.  Over 80% of the observations now used at NCEP come from satellites, but surface-based observations, especially rawinsondes, are still crucial.  The amount of satellite data is expected to increase by a factor of 100,000 in the next 10 years, posing an enormous challenge to data assimilation.  


Operational analyses and forecasts are under constant development; major changes in operational analysis/forecast systems occur at least once a year, resulting in improved products, but also causing changes and disrupting the ongoing climate record contained in such analyses.  Such changes are rigorously tested and examined for their effects on synoptic forecast fields, but not necessarily for their effect on air-sea fluxes.  On May 15, 2001, a new parameterization of cloudiness using prognostic cloud liquid water and a parameterization of cumulus momentum mixing were introduced in the NCEP operational global analysis/forecast system to replace an older diagnostic cloud parameterization.  As Fig. 1 shows, the new physics significantly increases evaporation and surface net short wave in the tropics and decreases surface net short wave in mid-latitudes.  Preliminary comparisons with climatological estimates of surface short wave from the NASA Langley Surface Radiation Budget project suggest that the new system has improved the surface net short wave. Further evaluation and improvement of the new cloud parameterization is planned.


Updating and improving one component of the physics in NWP systems may not automatically improve the performance of the physics as a whole nor improve the forecast skill of the system.  Because of the large uncertainty that exists in the magnitudes and the correct parameterization of physical fields, the wrong component of the physics may have been erroneously adjusted to correct past problems in the model physics and offsetting mistakes introduced.  For poorly observed fields, such as divergent flow in the tropics or moisture, the effect of observations can be quickly lost in forecasts if the observations are inconsistent with the model physics.  Figure 2 shows global mean precipitation and evaporation as a function of forecast length in the NCEP global system during October 1998.  In the first 24 hours of the forecast substantial adjustment, known as "spin-up", of the precipitation occurs, as the model fields adjust to the impact of observations.  Later a more gradual adjustment, known as "climate drift", occurs as the model forecast moves towards the model climate away from the climate of the observations.  The most balanced fluxes may not be found in the shortest length forecasts, but in longer forecasts after the "spin-up" phase.  For this reason, ECMWF recommended the use of fluxes from the 12-24 h forecasts in ERA15, not the 0-6 h forecasts.


To provide a more consistent ongoing record, reanalysis projects have used frozen analysis/forecast systems to analyze decades of observations and produce a consistent climate record without discontinuities due to changes in data assimilation, although the observation network itself can change significantly from decade to decade. The NCEP/NCAR reanalysis project (NCEP1) has produced an ongoing record from 1948 to the present and continues as a Climate Data Assimilation System (CDAS) in which the frozen system is run on current data, extended the consistent climate record. The NCEP/DOE reanalysis (NCEP2) corrected mistakes in the NCEP/NCAR reanalysis from 1979 to the present and introduced some improvements in the physics.  ECMWF produced a 15-year reanalysis from 1979 to 1993 and is beginning a 40-year reanalysis.  Even in the reanalyses discontinuities in the climate record can be seen due to changes in the data.  ERA15 used satellite radiances and had problems correcting biases in the radiances, introducing a major discontinuity in 1986 (Trenberth et al., 2001).  In 1979 the modern satellite era began in NWP; before 1979 the observation network was not truly global.  As a result, a major discontinuity can be seen in the NCEP1 climate record at the beginning of 1979.  The observation network has continued to change in recent years, with more and more space-based observations and fewer traditional ground-based observations such as ship reports and radiosondes.

Fig.1: Time-mean change in evaporation (left) and surface net short wave radiation (right) due to new parameterizations of cloudiness and cumulus momentum mixing in the NCEP global analysis/forecast system.  Fields are from 0-6 h forecasts of the NCEP global system averaged over April 14-May 8, 2001.


The reanalyses are the most evaluated of all the air-sea flux data sets and provide consistent, complete global fields of near-surface meteorology and surface fluxes over decades.  The NCEP-1 reanalysis has been examined by a large number of scientists from different fields and compared to many independent estimates of fluxes from many different periods, often over many years.  Run in near real-time as CDAS, it provides a useful benchmark for the operational global analysis/forecast system; however, it is based on the 1995 operational model and therefore is only a distant ancestor of the present operational model.

Fig. 2 Global mean precipitation (P) and evaporation (E) in mm/day as a function of forecast length for October 1998 in the NCEP operational global analysis/forecast system.  Forecast lengths shown are (from left to right) first time step (10 minutes, indicated as 0), 0-6 hr, 0-12 hours, 12 –24 h, … 348-360 h.

Horizontal resolution of the system changed from T126 to T62 after 168 hrs.


Section 11.4 of the Working Group on Air-Sea Fluxes report (Taylor, 2001) presents a review of studies of air-sea fluxes from the reanalyses and a detailed comparison of monthly mean air-sea fluxes from 4 reanalyses (NASA’s GEOS reanalysis, NCEP-1 and 2, ERA-15) with independent estimates of air-sea fluxes from COADS (da Silva et al., 1994) for 1981-92 and surface radiative fluxes from NASA Langley’s Surface Radiation Budget (SRB) for July 1983-June 1991 (Darnell et al., 1992; Gupta et al., 1992). Table 1 displays time-mean values of air-sea fluxes globally averaged over the ocean from the different estimates.  Highlighted values indicate extreme values.









Sensible heat







Latent heat







Net short wave







Net long wave







Net heat flux







Table 1 Global mean surface energy balance in W/m2 estimated from COADS data by da Silva et al. (1994) (without tuning), from the four reanalyses and from satellite-based estimates of surface radiation (SRB)


            Considerable differences can be seen even in the global means in Table 1.  The COADS estimate has a large global imbalance, as does the GEOS1 reanalysis.  Even though the other three reanalyses have smaller energy imbalances, they still do not give realistic oceanic heat transports (Trenberth and Caron, 2001). GEOS1 also has larger net radiative fluxes than other estimates.  ERA15 and NCEP2 have more evaporation than other estimates, while NCEP2 has smaller sensible heat flux.  The net long wave estimate from the SRB is considerably lower than other estimates (Kiehl and Trenberth, 1997); the resulting SRB net radiation is rather large and implies a considerably higher evaporation than even ERA15 or NCEP2.  Most comparisons with in-situ observations suggest that reanalysis evaporation estimates are already too high. 


Even when reanalysis fields have time-mean biases, their variability can still be quite realistic.

Monthly mean sensible and latent heat fluxes and wind stresses correlate well with COADS estimates where there is ample ship data and the different reanalyses show similar levels of agreement with COADS, although the magnitude of latent heat flux varies significantly and may be too large in the NCEP and ECMWF reanalyses.  The reanalyses all appear to have substantial errors in surface radiation fields, reflecting problems in clouds and clear sky radiation.  The NCEP-1 reanalysis (and no doubt other flux estimates as well) benefits from offsetting mistakes.  It has too much downward surface short wave radiation and too high an ocean surface albedo as well as too little cloud and too high cloud albedo.  ECMWF, GEOS and NCEP-2 had too few low-level stratus clouds.  Precipitation patterns from the first ECMWF reanalysis showed the best agreement with independent estimates of precipitation, especially in the northern mid-latitudes, but tropical precipitation appeared too large and to vary too much from month to month.  The reanalyses= estimates of oceanic precipitation in middle and high latitudes may be a useful supplement to satellite-based estimates of precipitation, since satellite-based estimates of precipitation tend to be tuned to the tropics.


Data assimilation can provide realistic estimates of air-sea fluxes, although satellite estimates of surface net short wave radiation are probably more accurate.  Clouds are a major problem.  The Abest set@ of air-sea fluxes may include both fields from data assimilation and from satellites, adjusted to give a reasonable net heat transport.



da Silva, A., C.C. Young,  and S. Levitus, 1994:  Atlas of Surface Marine Data 1994.  Vol. 1: Algorithms and Procedures.  NOAA Atlas NESDIS 6, U.S. Department of Commerce, Washington, DC, 83 pp.

Darnell, W.L., W.F. Staylor, S.K. Gupta, N.A. Ritchey, and A.C. Wilber, 1992: Seasonal variation of surface radiation budget derived from International Satellite Cloud Climatology Project C1 Data.  J. Geophys. Res., 97, 15 741-15 760.

Gupta, S., W. Darnell, and A. Wilber, 1992:  A parameterization of long wave surface radiation from satellite data:  Recent improvements.  J. Appl. Meteor., 31, 1361-1367.

Kiehl, J.T., and K.E. Trenberth, 1997:  Earth’s annual global mean energy budget.  Bull. Amer. Meteor. Soc., 78, 197-208.

Taylor, P.K., Ed., 2001:  Intercomparison and validation of ocean-atmosphere energy flux fields.  Joint WCRP/SCOR Working Group on Air-Sea Fluxes Final Rep., WCRP-112, WMO/TD-No. 1036, 306 pp. 

Trenberth, K.E.,and J.M. Caron, 2001:  Meridional atmosphere and ocean heat transports.  This volume.

-----, J.W. Stepaniak, and M. Fiorino, 2001:  Quality of reanalyses in the Tropics.  J. Climate, 14, 1499-1510.




(These sites tend to be under continual development; appropriate caution should be taken.)

National Centers for Environmental Prediction:

Environmental Modeling Center:

Global Modeling Branch: or

Global analysis documentation:

Global model documentation:

Global model changes since 1991:

Comparisons to observations: and

Surface fluxes and systematic errors: and

Tests of proposed changes to global system:

Systematic errors in tests of proposed changes:

NCEP1 and 2 reanalysis page: