4dvar presentation

1
Precipitation Assimilation at ECMWF P. Bauer P.
Lopez A. Benedetti M. Bonazzola
2
1D4D-Variational Analysis using Rainfall
Observations
4D-Var
3
Forecasted Rainrate from 26-12-2002 at 12 UTC
ZOE
PATER/RR
1D/RR
CONTROL
1D/TB
6 H
12 H
24 H
48 H
4
RMS-Error of Tropical Geopotential Height
Forecasts, 09/2003
500 hPa
200 hPa
1000 hPa
850 hPa
5
Candidate Problems with 1D4D-Var Analysis
Too small observation error for TB-exp tested
with 2?o Nearest grid-point TB vs.
grid-box averaged 3h vs 30 time
window
3h 30
6
TCWV FG vs. AN Departure Statistics
1D-Var RR 1D-Var TB
RAIN CLEAR
FG AN RAIN CLEAR FG
AN
FG AN
7
Observation-Model Bias, SSM/I 19v,h 22v Channels
8
TCWV FG vs. AN Departure Statistics (w/
Bias-correction)
before after
FG AN RAIN CLEAR FG
AN
FG AN
9
Impact of TB Bias-correction on 4D-Var Analysis
Hurricane Fabian 2003-09-01 1200 UTC
10
RMS-Error of Tropical Geopotential Height
Forecasts, 09/2003
500 hPa
200 hPa
1000 hPa
850 hPa
11
Isabel Track Forecast from 1D4D-Var Analyses
(cycling)
Analysis 2003091212
Analysis 2003091512
Analysis 2003091612
control rain
assimilation observed
12
Plans for 2004

• Technical Issues
• Perform 1D-Var with physics package for access
  to all physical variables in observation
  operator at full spatial/temporal resolution (40
  km/30 ? 25 km/30).
• Test 1D4D-Var (w/ bias-correction) using global
  SSM/ISSMIS data (22 GHz).
• Investigate performance (observation errors, new
  Jb-formulation) using dynamics/precipitation
  scores, track forecasts, PR data, model
  spin-down.
• Activate with December 2004 model cycle.
• Science Issues
• Improve on bias-correction (physics, RT) for
  better trade-off between information and bias.
• Investigate direct 4D-Var TB-assimilation for
  stronger impact on dynamics.
• Add more complementary channels (e.g. SSMIS
  22183 GHz channels, possibly T-sounding
  channels) for more consistent cloud-precipitation
  analysis.

Global data assimilation systems provide
excellent tool for GPM data monitoring Individua
l data quality/consistency is monitored by model
fit to all other observations related to
atmospheric dynamics, moisture, temperature and
surface state.