Upload
belinda-lambert
View
217
Download
0
Embed Size (px)
Citation preview
Data assimilation and observing systems strategies
Pierre GauthierData Assimilation and Satellite Meteorology DivisionMeteorological Service of CanadaDorval, Québec CANADA
Co-chair of the THORPEX working group on DAOS
Data assimilation and observing Data assimilation and observing strategiesstrategies
• Optimal use of observations– Adaptive observations (targeted observations)
• Deploy observations over regions where small changes lead to substantial changes in the forecasts
– Better use of existing observations, particularly satellite data
• Satellite data• Data assimilation methodology• A few scientific objectives
Satellite dataSatellite data
• Relatively low proportion of received data makes its way to the assimilation (<20%)
• Observation error– Biases: assimilation is bias blind and
innovations cannot distinguish between model and observation bias
– Observation error correlation• Characterization of surface emissivity to
assimilate many satellite data types
Distribution of ATOVS satellite data Distribution of ATOVS satellite data received over a 6-h windowreceived over a 6-h window
Distribution of ATOVS satellite data Distribution of ATOVS satellite data assimilated assimilated over a 6-h windowover a 6-h window
Channel selection of IASI radiances in meteorologically Channel selection of IASI radiances in meteorologically sensitive areas (Fourrié and Rabier, 2003)sensitive areas (Fourrié and Rabier, 2003)
Current and Planned Satellites (1/2)Current and Planned Satellites (1/2)
Source: JCSDA (Joint Center for Satellite Data Assimilation) 13th AMS Conf. 2004
Current and Planned Satellites (2/2)Current and Planned Satellites (2/2)
Source: JCSDA
Data assimilation methodsData assimilation methods
• Several NWP centres have now implemented 4D-Var– Significant impact on the forecasts– Better usage of satellite and asynoptic data– Issues on specific aspects of the implementation,
particularly when it comes to humidity analysis• Assimilation with a numerical model
– Leads to model improvements and assimilation methodology
– Attention needs to be paid to the details of the implementation
3D and 4D data screening3D and 4D data screening
4D-Var
0-h-3h +3h
3D-Var
0-h-3h +3h
Type 4D-Var 3D-Var Difference
Aircrafts 75707 26147 +189%
Radiosonde 66605 66603 ~0%
Satwind 82160 41604 +97%
ATOVS 71517 46832 +53%
GOES 3612 1979 +83%
Profilers 13040 2196 +494%
Data assimilated Data assimilated 4D-Var 4D-Var vsvs 3D-Var3D-Var (12Z 16 (12Z 16
February 2005)February 2005)
Anomaly correlation: winter periodAnomaly correlation: winter period4D-Var3D-Var
Impact of the various components of 4D-Var
Type Outer loops Simplified
physics
Observation
thinning
3D-Var 1 - 3D
3D-Var(FGAT)
1 - 3D
3D-Var(FGAT)
1 - 4D
4D-Var 1 (simpler) 4D
4D-Var 2 (simpler,simpler) 4D
4D-Var 2 (simpler, better) 3D
4D-Var 2 (simpler, better) 4D
August 2004
RMS error
GZ 500 hPa
Southern Hemisphere
Impact of the various components of 4D-Var
4D-Var
3D-Var
4D-Var (simpler)
4D-Var (simpler,1 loop)
4D-Var (thinning 3D)
7% (better simplified physics)
3% (Updated trajectory)
35% (thinning 4D)
(TL/AD dynamics) 55%
FGAT (thinning 3D)
FGAT (thinning 4D)
16%
18%
Impact of the various components of 4D-Var
Information Content (%)
0
5
10
15
20
25
Total influence (%) of satellite and in-situ observations when assimilated by ECMWF 4DVar System. From Cardinali et al. 2004.
Further developments in data Further developments in data assimilation methodsassimilation methods
• Background term– Up to now: little (but positive) impact– Requirements for the assimilation of fine scale
structures, particularly in the humidity field– Hybrid methods (EnKF +4D-Var?)
• Nonlinearities– Observation and physical parameterizations
• Weak-constraint 4D-Var– Extending the assimilation window
(Fisher, 2004)– Dealing with model error
• Surface analyses, high-resolution analysis for mesoscale models
Verification of 48-h forecastsagainst radiosondes observations over North America
Regional forecast issued directly from the 4D-Var global analysis
12-h regional assimilation cycle initiated from the 4D-Var global analysis
Impact of 4D-Var analysis on regional (15 km) forecasts (24 winter case)
Impact of 4D-Var global analysis on regional 3D-Var cycle1 case : 48 hr forecast valid on November 16th 2004, at 12z
0%
10%
20%
30%
40%
50%
60%
70%
80%
Gz 500 hPa MSL pres QPF
EquivalentReg 3DReg 4D
Subjective Evaluation (Winter 2004-2005)Subjective Evaluation (Winter 2004-2005)% in favor of % in favor of 3D-Var3D-Var or or 4D-Var4D-Var
A few scientific objectives (1)A few scientific objectives (1)
• THORPEX regional campaigns– Storm Winter Reconnaissance Program (US)
over the North Pacific since 1998– Fall of 2003 in the North Atlantic (A-TReC 2003)– Pacific campaign: 2007-2008
• Seattle meeting 6-7 June 2005• What needs to be observed to improve the
large scale forecasts– Design of TReCs by learning from previous
ones– Recommendations for future campaigns
A few scientific objectives (2)A few scientific objectives (2)
• Improving the assimilation of existing satellite data– What is not currently well observed (e.g., winds)– Estimation of observation error characteristics– Targeting methods
• Impact of large-scale improvements on local short-term forecasts (downscaling)– Relevant weather elements for socio-economic studies
often need the magnifying glass of a higher resolution model
• Ensemble prediction– Impact of changes in the observation network on the
estimated variability in ensemble prediction systems
Error variance estimated with a Kalman filterError variance estimated with a Kalman filter(Radiosonde coverage only) (Radiosonde coverage only) (Gauthier (Gauthier et al.et al., 1993), 1993)
Error variance estimated with a Kalman filterError variance estimated with a Kalman filter(Radiosonde and satellite coverage)(Radiosonde and satellite coverage)
The EndThe End
ATReC 026
48-h Singular vector SV1 at initial time (Zadra and Buehner)
Valid time: 5 Dec. 2003 12 UTC
MSC-GEMSimplified physics• Vertical diffusion• orographic blocking and GWD• stratiform condensation• convection
Computed with dry physics
Computed with moist physics