11 th jcsda science workshop on satellite data assimilation
DESCRIPTION
11 th JCSDA Science Workshop on Satellite Data Assimilation. Latest Developments on the Assimilation of Cloud-Affected Satellite Observations Tom Auligné National Center for Atmospheric Research - PowerPoint PPT PresentationTRANSCRIPT
11th JCSDA Science Workshop on Satellite Data Assimilation
Latest Developments on the Assimilation of Cloud-Affected Satellite Observations
Tom AulignéNational Center for Atmospheric Research
Acknowledgments: Gael Descombes, Greg Thompson, Francois Vandenberghe, Dongmei Xu (NCAR)Thomas Nehrkorn, Brian Woods (AER)Funding provided by the Air Force Weather Agency
1
Introduction
Approaches for the use of cloud-affected radiances
• Nowcasting• COP + advection• 3D cloud fraction + WRF dynamical transport
• Oklahoma U. / GSI-RR cloud analysis
• Expansion of analysis control variable• Total water + linearized physics• Microphysical variables
2
Introduction
Approaches for the use of cloud-affected radiances in NWP
• Nowcasting• COP + advection• 3D cloud fraction + WRF dynamical transport
• Oklahoma U. / GSI-RR cloud analysis
• Expansion of analysis control variable• Total water + linearized physics• Microphysical variables (WRF Data Assimilation)
3
Outline
• Control Variable Transform• Multivariate Background Error
Covariances• Hybrid Ensemble/Variational• Displacement Analysis
• Processing of Cloudy Satellite Observations
• Experimental Demonstration
4
Outline
• Control Variable Transform• Multivariate Background Error
Covariances• Hybrid Ensemble/Variational• Displacement Analysis
• Processing of Cloudy Satellite Observations
• Experimental Demonstration
5
Pseudo Single Observation Test (PSOT)
• Multivariate covariances for qc, qr, qi, qsn• Binning option: dynamical cloud mask• Vertical and Horizontal autocorrelations via Recursive Filters• 3D Variance 6
Outline
• Control Variable Transform• Multivariate Background Error
Covariances• Hybrid Ensemble/Variational• Displacement Analysis
• Processing of Cloudy Satellite Observations
• Experimental Demonstration
7
NEW ALGORITHM: update of the ensemble perturbations inside the Variational analysis
through Lanczos minimization(without external EnKF)
Ensemble covariance included in 3D/4DVar through state augmentation(Lorenc 2003, Wang et al. 2008, Fairbairn et al., 2012)
Ensemble/Variational Integrated Localized (EVIL)
with
8
Outline
• Control Variable Transform• Multivariate Background Error
Covariances• Hybrid Ensemble/Variational• Displacement Analysis
• Processing of Cloudy Satellite Observations
• Experimental Demonstration
9
ForecastCalibration &Alignment
Hurricane Katrina OSSE
Synthetic observations(TCPW)
Balanced displacement
Nehrkorn et al., MWR 2013 (submitted) 10
Outline
• Control Variable Transform• Multivariate Background Error
Covariances• Hybrid Ensemble/Variational• Displacement Analysis
• Processing of Cloudy Satellite Observations
• Experimental Demonstration
11
• Unified processing for Radiances and COP Retrievals
• VarBC: Variational Bias Correction (unchanged predictors)
• Revisited QC: extended Gross and First-Guess check
(to conserve cloudy data)• Huber Norm: robust definition of
observation error• Land Surface: Tskin introduced as a sink variable• Field of View: advanced interpolation scheme• CRTM Jacobians: rescaled base state
(floor and ceiling values for cloud parameters)
• Middle Loop: Multiple re-linearizations of obs. operator
Unified Processing of Cloud-affected Satellite data
12
Observation
Background
Radiance(GOES imgrchannel 4)
• AIRS, IASI, AMSU-A, AMSU-B, MHS, HIRS, SSMI/S BUFR files from NCEP
• MODIS HDF files (Terra & Aqua) from NASA
• GOES Sndr ASCII files (E & W) from Uwisc.
• GOES Imgr NetCDF files from NCAR/RAL
Cloud-Affected Radiances
13
Observation
Background
Log10(COD)(GOES imgr)
• GOES Imgr NetCDF files from Uwisc.• MODIS NetCDF files from AFWA
with (Benedetti and Janiskova, MWR 2008)
€
Log10(COD)
€
rlog = log10(1+ε r2)
Cloud Optical Properties
14
• Unified processing for Radiances and COP Retrievals
• VarBC: Variational Bias Correction (unchanged predictors)
• Revisited QC: extended Gross and First-Guess check
(to conserve cloudy data)• Huber Norm: robust definition of
observation error• Land Surface: Tskin introduced as a sink variable• Field of View: advanced interpolation scheme• CRTM Jacobians: rescaled base state
(floor and ceiling values for cloud parameters)
• Middle Loop: Multiple re-linearizations of obs. operator
Unified Processing of Cloud-affected Satellite data
15
• Unified processing for Radiances and COP Retrievals
• VarBC: Variational Bias Correction (unchanged predictors)
• Revisited QC: extended Gross and First-Guess check
(to conserve cloudy data)• Huber Norm: robust definition of
observation error• Land Surface: Tskin introduced as a sink variable• Field of View: advanced interpolation scheme• CRTM Jacobians: rescaled base state
(floor and ceiling values for cloud parameters)
• Middle Loop: Multiple re-linearizations of obs. operator
Unified Processing of Cloud-affected Satellite data
16
• Unified processing for Radiances and COP Retrievals
• VarBC: Variational Bias Correction (unchanged predictors)
• Revisited QC: extended Gross and First-Guess check
(to conserve cloudy data)• Huber Norm: robust definition of
observation error• Land Surface: Tskin introduced as a sink variable• Field of View: advanced interpolation scheme• CRTM Jacobians: rescaled base state
(floor and ceiling values for cloud parameters)
• Middle Loop: Multiple re-linearizations of obs. operator
Unified Processing of Cloud-affected Satellite data
17
Huber Norm: robust observation errors
Figure from Fisher (2008)
Innovations
Normalized Obs Weight
Distribution of innovations18
• Unified processing for Radiances and COP Retrievals
• VarBC: Variational Bias Correction (unchanged predictors)
• Revisited QC: extended Gross and First-Guess check
(to conserve cloudy data)• Huber Norm: robust definition of
observation error• Land Surface: Tskin introduced as a sink variable• Field of View: advanced interpolation scheme• CRTM Jacobians: rescaled base state
(floor and ceiling values for cloud parameters)
• Middle Loop: Multiple re-linearizations of obs. operator
Unified Processing of Cloud-affected Satellite data
19
• Unified processing for Radiances and COP Retrievals
• VarBC: Variational Bias Correction (unchanged predictors)
• Revisited QC: extended Gross and First-Guess check
(to conserve cloudy data)• Huber Norm: robust definition of
observation error• Land Surface: Tskin introduced as a sink variable• Field of View: advanced interpolation scheme• CRTM Jacobians: rescaled base state
(floor and ceiling values for cloud parameters)
• Middle Loop: Multiple re-linearizations of obs. operator
Unified Processing of Cloud-affected Satellite data
20
Satellite Field of View (FoV) interpolation
• Calculate polygons (ellipses)
• List model grid points inside ellipse
• Use average input for CRTM• Currently testing for AIRS and IASI
21
• Unified processing for Radiances and COP Retrievals
• VarBC: Variational Bias Correction (unchanged predictors)
• Revisited QC: extended Gross and First-Guess check
(to conserve cloudy data)• Huber Norm: robust definition of
observation error• Land Surface: Tskin introduced as a sink variable• Field of View: advanced interpolation scheme• CRTM Jacobians: rescaled base state
(floor and ceiling values for cloud parameters)
• Middle Loop: Multiple re-linearizations of obs. operator
Unified Processing of Cloud-affected Satellite data
22
• Unified processing for Radiances and COP Retrievals
• VarBC: Variational Bias Correction (unchanged predictors)
• Revisited QC: extended Gross and First-Guess check
(to conserve cloudy data)• Huber Norm: robust definition of
observation error• Land Surface: Tskin introduced as a sink variable• Field of View: advanced interpolation scheme• CRTM Jacobians: rescaled base state
(floor and ceiling values for cloud parameters)
• Middle Loop: Multiple re-linearizations of obs. operator
Unified Processing of Cloud-affected Satellite data
23
Cost Function
First Guess Second Guess Third Guess
First Analysis Second Analysis
AIRS Window Channel #787
3DVar Middle loop
24
First Guess Second Guess Third Guess
Observation
Update of qcloud, qice in WRF
ObservationsAIRS Window Channel #78725
Outline
• Control Variable Transform• Multivariate Background Error
Covariances• Hybrid Ensemble/Variational• Displacement Analysis
• Processing of Cloudy Satellite Observations
• Experimental Demonstration
26
WRF-ARW model, CONUS 15km, Thompson microphysicsFirst Guess = Mean of 50-memb ens. from EnKF expt
(Romine)
Single analysis at 20012/06/03 (12UTC), 5 middle-loops
GOES-Imager: Radiances (Thompson) & COD (Uwisc.)
• CTRL no DA • COD Cloud Optical Depth (3DVar)• FCA Cloud Optical Depth
(Displacement)• 3DVAR Radiances (3DVar)• EVIL Radiances (Hybrid)
Experimental Framework
27
Incremental fit to obs
Cloud Optical Depth
3DVar
EVILRadiances
Obs
28
qcloud
Level 10
Analysisincrements
COD FCA
3DVAR EVIL
29
qcloud
Level 10
Analysisincrements
COD FCA
3DVAR EVIL
30
0.12°0.25°0.5°
1° 2° 3°
Multi-scaleverification
31
Multi-scale verification
Analysis Forecast
EVIL3DVARCTRL
Impact of all-sky radiances on forecast up to 3-4h Best forecast with EVIL system
EVIL3DVARCTRL
32
Conclusion
• Expansion of analysis vector to clouds• Multivariate, flow-dependent background errors
• Displacement pre-processing• Updated processing of cloud-affected satellite data (bias correction, QC, interpolation, RTM, middle-loop)
• Sustained impact in short-term forecast
• More work required…
33