Observation Influence Diagnostic of a Data Assimilation System - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 98

GOES-Ra O3

MTSAT-Rad

MET-Rad

AMSU-B

MERIS
MHS

TMI-1

SSMIS

AMSR-E

GPS-RO

IASI

AIRS

AMSU-A

HIRS

SCAT

MODIS-AMV

MTSAT-AMV
Meteosat-AMV

GOES-AMV

PROFILER

PILOT

DROP

TEMP

DRIBU

Aircraft

SYNOP

0 0.2 0.4 0.6 0.8 1 1.2 1.4

OI

Fig. 4.2 Observation Influence ( OI ) of all assimilated observations in the ECMWF 4DVar system
in October 2011. Observation types are described in Table 4.1

of 'All-Sky' observations (TMI and SSMIS) have increased their influence in the
analysis ( Cardinali and Prates 2011 ; Geer and Bauer 2011 ).
In Sect. 4.2 it has been shown that tr( S) can be interpreted as a measure of the
amount of information extracted from the observations. In fact, in non-parametric
statistics, tr( S ) measures the 'equivalent number of parameters' or degrees of
freedom for signal (DFS) . Having obtained values of all the diagonal elements of
S (using 4.16 ) we can now obtain reliable estimates of the information content in
any subset of the observational data. However, it must be noted that this theoretical
measure of information content does not necessarily translate on value of forecast
impact. Figure 4.3 shows the information content for all main observation types. It
can be seen that AMSU-A radiances are the most informative data type, providing
23 % of the total observational information, IASI follows with 17 % and AIRS with
16 %. The information content of Aircraft (10 %) is the largest among conventional
observations, followed by TEMP and SYNOP ( 4

%). Noticeable is the 7 % of
GPS-RO (4th in the satellite DFS ranking) that well combines with the 0.4 value
for the average observation influence. In general, the importance of the observations
as defined by e.g. the DFS well correlates with the recent data impact studies by
Radnoti et al. ( 2010 ).
Similar information content of different observation types may be due to different
reasons. For example, DRIBU and OZONE information content is similarly small
but whilst OZONE observations have a very small average influence (Fig. 4.2 )and
dense data coverage, DRIBU observations have large mean influence but much
lower data counts (Fig. 4.2 ). Anyhow, the OZONE data are important for the ozone
assimilation in spite of their low information content per analysis cycle. In fact,
OZONE is generally a long-lived species, which allows observational information
to be advected by the model over periods of several days.