Geostationary Satellite Winds in HIRLAM
•Characteristics of AMVs used in this Study
• Assimilation of AMVs in HIRLAM
• Assimilation Experiment Description and
Discussion of Results
• Conclusions and Future Work Outlook.
SATELLITE VECTORS USED IN THIS STUDY
• Vectors from GOES, M7 and M5 Satellites.
• Two types of Vectors: Vectors derived from VIS (0.5
μ) and from WV(about 6 μ) (only cloud tracking
vectors included) consecutive images.
• Increased Sampling in Space and Time. Reliable
Quality Information included in the BUFR messages.
ASSIMILATION OF AMVs in HIRLAM
• Screening: Background Cross-Check Settings.
• Calibration of Screening Settings with AMVs
• Recent developments: observation operators
for AMVs and bias correction schemes.
Settings of First Guess Check for Wind Data:
Compare 0.5 * [ ( uo - ub )2 / σ2 + ( vo - vb )2 / σ2 ] .vs. x(i) * factor
σ2 = σ o 2 + σ b 2
σo ( m/s) x(1), x(2), x(3) factor Rejection
Aircraft 4. 8, 18, 20 0.33 ~ 2.4 σ
Ship 3. 8, 10, 20 1.0 ~ 3.0 σ
AMVs 2. 8, 18, 20 Sb – So < 4 m/s < 1.9 σ
(low level) Sb – So > 4 m/s
AMVs 5. 8, 18, 20 Sb – So < 4 m/s < 1.3 σ
(high level) Sb – So > 4 m/s
[ 0., 0.075]
Aircraft/AMVs FG Check Results Compared
Ship/AMVs FG Check Results Compared
FG check results compared with QI values
Characterising and Searching out a Bias Correction Method for Satellite AMVs
Assimilation Experiment Brief Description
• Assimilation Algorithm: Hirvda_6.0.0
• HIRLAM Forecast System Version: 6.1.0
• Geographical Area: 15.5N-65N; 66.5W-30E
• Spatial Resolution: 0.5 x 0.5 degrees, 31 levels
• Time period: June 2003, 6 hours assimilation cycle
• Observation usage:
CNT; default + wind data from SHIPS.
EXP; as CNTL + AMVs presented in this talk.
a) Assimilation time window for AMVs shrunk to
+/- 15 minutes about nominal analysis time.
b) QI above 0.6
c) Thinning at 0.5 degrees and 50 hPa by selection of
vector with best QI.
Spatial Structure of Observation Errors for AMVs
(N.Bormann, S.Saarinen, J.N Thépaut, G.Kelly)
Impact on Fit to Analysis of Aircraft/Ship Data
Variational Quality Control
Contribution to the cost function of a given observation split
up in gross-error and gaussian error
JoiQC = - ln poi = - ln ( a + b exp ( - zoi / 2. ) )
ω = grad JoiQC / grad Joi; if smaller than a given threshold the
observation is rejected
“a priori” probability of gross Range of uniform distribution
AIRCRAFF 0.015 5 σo
Rest of observations 0.003 5 σo
FG VARQC FG VARQC
AIRCRAFT 2.3 % 1.4 % 2.2 % 1.4 %
SHIPS 0% 0.3 % 0% 0.3 %
AMVs 8.8 % 0% 8.3 % 0%
AMVs 27.8 % 0% 22.2 % 0%
Wind Speed at Model Level 28
Wind Speed at Model Level 16
Surface Pressure (Mean Difference EXP-CNT )
Surface Pressure (t-test)
Geopotential at Model Level 25
Geopotential at Model Level 16
CONCLUSIONS AND PROSPECTS
• New AMVs have proved to have a moderate positive
impact for low European latitudes.
• There is still some tuning work to do.
• Future work should consider 3D-Var assimilation
algorithm enhancements like FGAT and new
• New vectors from recently de-commissioned MSG
satellite should be included as well in future studies.