A set of global observing-system experiments is run by excluding classes of
observations from the European Centre for Medium-Range Weather Forecasts (
ECMWF) four-dimensional variational (4D-Var) data assimilation and forecast
system. This indicates how efficiently the observations are used in this p
articular system. The observing systems tested were mainly TIROS (Televisio
n Infra-Red Observation Satellite) Operational Vertical Sounder (TOVS) radi
ances. SATellite OBservation (SATOB) atmospheric-motion winds, radiosondes,
aircraft, drifting buoys and Australian pseudo observations (PAOBs). The i
mpact is assessed in both summer and winter 1999, over a total period of si
x weeks, in terms of the average objective quality of the deterministic for
ecast of tropospheric wind and geopotential height at medium (5-7 day) and
short (1-3 day) ranges. It is shown that the choice of verification method
matters in data-poor areas such as the tropics and the southern hemisphere.
Some interesting conclusions can be drawn about the ECMWF 4D-Var system. Ea
ch observing system tested has a notable positive impact on the medium-rang
e forecast performance. This is not always true at shorter ranges. The TOVS
radiances have a large impact on all areas and ranges. Drifting buoys and
PAOBs both have some impact in the southern hemisphere. In the northern hem
isphere, the relative impact of TOVS, aircraft and radiosondes depends on t
he area considered. For Europe, the statistical significance of the results
is limited, but the most important observing system among those tested app
ears to be the radiosondes.