Global vs mesoscale ATOVS
assimilation at the Met Office
Global
Mesoscale

Large obs error (4 K)

Smaller obs error (2 K)

NESDIS 1B radiances

Direct broadcast radiances

NOAA-15 & 16

NOAA-15 & 16

HIRS and AMSU

AMSU only

thinned to 154 km (2°)

thinned to 0.4°

±3hr assim window

± 1½ hr assim window
AMSU-B
Approximate
MetO mesoscale
model domain
Example of full resolution AMSU-B 89 GHz imagery
Typical data coverage plots for 15Z and 18Z runs
15Z
18Z
Issues and summary

Accuracy of the navigation of direct broadcast NOAA-15.

Latent heat nudging scheme is applied after the 3D-var analysis.

HIRS gridded AMSU-B data used (36 observations averaged).

Trials: 20 April - 1 May & 14 May - 1 June 2001 + case studies.

28 days for each main forecast. 4 large impacts (2 improved, 2
degraded) others close to neutral verified subjectively by a
trained (and unbiased!) forecaster.

(-) ve impact: surface visibility, 6 hourly precipitation, cloud base
height, surface temperature (v synop* data).

(+) ve impact: total cloud amount, surface wind (v synop data).

Using an objective weighted mean of key meteorological fields
the skill of the forecast v synop data was reduced by 0.7%.
* Synop data: these are routine hourly observations from meteorological stations: cloud observations are made
by trained observers and other variables are from standard meteorological station equipment (rain gauge,
screen thermometers etc.). No radar or satellite data is used.
Fit of radiances to background (six
hour forecast)
Control = No ATOVS
assimilated
1.2
Trial = ATOVS
assimilated
1
Background
0.8
Control analysis
0.6
Trial analysis
0.4
The mesoscale
model analysis pulls
strongly to the
ATOVS data in the
first assimilation
cycle
0.2
AMSU-20
AMSU-19
AMSU-18
AMSU-9
AMSU-8
AMSU-7
AMSU-6
AMSU-5
0
AMSU-4
RMS Obs - Forward modeled TB / K
1.4
AMSU-B positive impact
Operational
Radar
Reduction in rain rate
Trial
AMSU-B negative impact
Operational
No ATOVS
assimilation
Note that AMSU-B only has a large impact
in the real time trial, where AMSU-B has
been assimilated for 8 days before this case.
The case study had only 24 hours of
ATOVS assimilation and impact was
minimal.
Note satellite data confirmed that the
operational stratus analysis was very
accurate.
Stratus
probabilities
Real time trial
8 days of ATOVS
assimilation
Case study
24 hours of ATOVS
assimilation
Conclusions

Moisture analysis and background fit to AMSU is improved.

AMSU-B had 2 positive and 2 negative impacts - rest neutral.

Overall statistical objective verification (v synops) showed slight
degradation in forecast skill  not made operational.
Next steps

Previous results are being compared with satellite cloud products.

More trials are being run to increase sampling and for which fullresolution (unmapped) AMSU-B can be used.

Rain radar and cloud observation processing will be included in 3Dvar in the future rather than as a separate nudging scheme.

Are 3D-var error correlations, cloud detection, radiative transfer
model, quality control, model physics etc. good enough?