assimilation of direct satellite radiance data at ncmrwf
TRANSCRIPT
Assimilation of Direct Satellite Radiance Data at
NCMRWF
V.S. PrasadNCMRWF
Building of data assimilation system
– Observing system – Data handling system – Forecast model– Computational resources– understanding the available knowledge about
observations and statistics– Human resources – Verification and monitoring system
Forecasting System at NCMRWF
A new Global Data Assimilation &Forecast System at T254L64 has been implemented recently
Data handling• NCMRWF got access to dds.nesdis.noaa.gov
through an exclusive user account.• Radiance data from NPOES satellites in level 1B
data format• Derived products in WMO BUFR format-120 km pre-processed NPOES-ATOVS retrievals- quickscat scatterometer winds- SBUV/2 ozone retrievals- DMSP SSM/I retrievals
Current Status (NCMRWF):Global Modeling & Data Assimilation System
Global Model & Data Assimilation at T254L64 resolutions Data Assimilation is based on
3-dimensional Variational Scheme
Global Model & Data Assimilation at T170/L28 & T80/L18 Resolutions also
Ensemble Prediction at T80/L18 Resolution8 Member Ensemble (Breed Vector)
Mesoscale Model & Data Assimilation System
Mesoscale Data Assimilation and Forecast over India – WRF
Other Mesoscale Models routinely used at NCMRWF for mesoscale forecast over India
MM5ETARSM
These Models are run using Initial and Boundary Conditions from NCMRWF Global Models
• The analysis is that choice of model variables which most closely fits both the observations (via forward model) and prior estimate , as measured by the following objective function:
J= ½ [ (X –Xb)T B-1(X –Xb ) + (yo –HX)TR-1 (yo –HX)]•• Where• X – resultant analysis vector• Xb - the background or first guess vector• B - background error covariance matrix• yo - the observation vector• H - forward model• R - Observational error covariance matrix•• The objective function J is a weighted least square fit ; the
model state which minimizes the J provide an optimal estimate if the error covariances B and R known and are defined by normal probability distribution.
•
Solution Algorithm• Solve series of simpler problems with some
nonlinear components eliminated• Outer iteration, inner iteration structure• Outer iteration– QC– More complete forward model• Inner iteration– Preconditioned conjugate gradient– Often simpler forward model– Variational QC– Solution used to start next outer iteration– Possibly lower resolution
Atmospheric analysis problem (Practical)Outer (K) and Inner (L) iteration operators
Variable K operator L operatorTemperature – surfaceobs. at 2m
3-D sigma interpolationadjustment to differentorography
3-D sigma interpolationBelow bottom sigmaassumed at bottom sigma
Wind – surface obs. at10m over land, 20m overocean, except scatt.
3-D sigma interpolationreduction below bottomlevel using model factor
3-D sigma interpolationreduction below bottomlevel using model factor
Ozone – used as layers Integrated layers fromforecast model
Integrated layers fromforecast model
Surface pressure 2-D interpolation plus orography correction
2-D interpolation
Precipitation Full model physics Linearized model physics
Radiances Radiances Full Radiances Full
Assimilation of wind and humidity The observation operator : The variational formulation makes it possible in principle to assimilate any type of observation as long as a forward model can be defined which
will converts the model variables to the observation type.
• Observations are not made at model grid points and are often not of model variables
• So need to derive model estimate of the observation for comparison.
• This is done with the ‘observation operator’ H
• H may be a simple interpolation from model grid to observation point
• or may be complicated radiative computation for satellite data
IRSP4-Oceansat-I• IRS-P4 / MSMR Launched 1999
• Swath 1360 Km 6.6 , 10.6 , 18 and 21 G Hz (V, H)
• Over data sparse oceanic region ( 150 km grid data used )
Surface wind speedTotal precipitable water content
The operational assimilation code was modified for including the ocean surface wind speeds and TPWC from the microwave channel retrievals in GDAS. The changes were extended to DMSP-SSM/I
Satellite data Satellite data differ from many conventional
data inthat the observations are often indirect
observationsof meteorological parameters– If x is the vector of meteorological parameters
we are interested in and– y is the observation,– then y = K(x,z),• where z represents other parameters on
which the observationsis dependent• K is the physical relationship between x, z
Satellite radianceExample –– y are radiance observations,– x are profiles of temperature, moisture and
ozone.– K is the radiative transfer equation and– z are unknown parameters such as the
surface emissivity(dependent on soil type, soil moisture, etc.), CO2 profile,methane profile, etc.
• Radiative transfer (CRTM) developed and maintained by JCSDA
Level 1B format• NOAA digital data is supplied to users in a predefined format called
Level-1B format defined by NOAA. Other organizations in the world, who supply the data, follow the same format. It is a packed format and all the band data exists in a 10 bit format. The data product, in addition to video data, contains ancillary information like Earth Location Points (ELPs), solar zenith angle and calibration. All the commerciallyavailable Image Analysis Packages have the facility to read data in Level-1B format.
• The current Level-1B format has been modified from NOAA – 15 data onwards. The new format is called NOAA – KLM format. The data in this format can be read on any system. All variables are declared as integers. The real values are also converted into integers with scaling factor. The earlier format had some problems regarding real numbers ie., ELPs, calibration values etc. They followed either IEEE or VAX floating point format. Only the latest versions of image analysis packages have the facility to read the data in this format. Older versions will not have this feature.
The raw counts in the level 1b files are transformed using the calibration coefficients in the data file to antenna temperatures and then to brightness temperatures (for AMSU-A data) using the algorithm of Mo (1999). Then, the following data is extracted from orbital data is then binned in 6 hour periods (+/-3hrs) of the analysis time for use in the assimilation system
year Sensor id
month Scan element
day Land/sea flag
hour Satellite Zenith angle
Minute Solar Zenith angle
second Surface elevation
Latitude Satellite height
Longiude Brightness Temp. of each channel
Satellite id
• The radiance assimilation scheme has several essential components, viz.
1 A fast radiative transfer model to relate analysis variables to observations2 A quality control (QC) procedure to remove observations that are
unreliable or beyond the simulation capacity of RT model and NWPmodel. Also for the selection of channels and assignment of their observation errors according to the observation conditions and the removal of observational biases.
3 A thinning procedure to remove redundancy and horizontal correlation of observationsThese procedures are very sensitive to a NWP system and because they depend on the analysis scheme, forecast model accuracy, computational resources and data usage policy. Further the RT codes are requires satellite sensor specific information to include them in it. Therefore, it is necessary to develop specific procedure for each NWP system and for specific sensor
AMSU/A
• Total number of channels 15• Ch’s 11-14 are not used because of vertical
resolution of the forecast model.• Because of difficulties in estimating the surface
emissivities and surface skin temperature over snow/ice/land, ch’s 1-6 are used only on open water.
• Ch 1-6 and 15 are assumed contaminated by precip./ large cloud droplets,and not used if ch’s 1-3 are simulated properly.
HIRS/3• Total Channels : 20• Ch #1 and 16-20 not used• Over land only channel 2 and 3 used and over
ocean channel 2 to 15 are used • Over ocean Ch # 4 –15 used, only if the surface
sensing channel 8 simulated with proper accuracy.• Ch 2 and 3 in most cases are not affected by the
clouds. For this reason too many channel 2 and 3 observations for the assimilation system to handle. Therefore for these channels , every 5th HIRS observation is selected.
GOES Image• NCEP and ECMWF assimilate Clear Sky Brightness
Temperatures(CSBT) from GOES images.• CSBT data are averaged over boxes of ~ 50 km. Each box
consists of 187 (11rows by 17 columns field of view (fov’s)• For each box average BT for each ir band and albedo for visible
band are calculated with the average clear and cloudy BT’s.• Additional parameters are the number of clear and cloudy fov’s,
center lat and long of box, center local zenith and solar zenithangles of box, land sea flag, standard deviation of average clear and cloudy BT’s and two qc flags.
• The quality indicator flags provide information on the likelihood of a particular observation being effected by sunglint and relative quality of SST.
• CIMSS extract the above information from GOES image data and packs it in BUFR format and make it available.
CRTM
• CRTM is updated RTM based on Optran transmittance model described bye Kleespies et al. (2004) A summary of these development works are given by Weng et al., (2005).
• The CRTM computing algorithms based heavily on a set of pre-prepared data sets. These data are stored in several binary files, and all loaded into the CRTM data variables during the CRTM initialisation phase.
• These files may be dived into two groups: - one with data specific to a collection of sensorsa) Spectral coefficient (SpcCoeff) file.
b) Optical depth (TauCoeff) coefficient file- the other valid for all sensors or sensors in whole
spectral region such as Microwave and Infrareda) Cloud Coefficient (cloudCoeff) file : The file contains cloud
optical parameters and look up tables such as mass extinction coefficients, single scattering albedo, asymmetry factors and Legendre expansion coefficients.
b) Surface Emissivity coefficient (EmisCoeff) file : The file currently contains coefficients data for computing infrared ocean surface emissivity.
c) Aerosol coefficient (AerosolCoeff) file : Currently it is a dummy file.
Spectral coefficient file
• Spectral coefficient (SpcCoeff) file contains the following information for each channel.
• NCEPSID,WMOSID,WMOCID,SENCHN, Freq., Wave number, Planck-C1,Planck-C2, Band-C1, Band-C2, ISMic, Polarisation, Cosmic Background(CB) Temp,CBRadi, ISC, Solar irradiance, Blackbody irradiance
• NCMRWF is having 4.2 version spcCoeff file and it supports the following 535 satellite channels. It is compatible to 5.3 version TauCoeff file
• It is upgraded to 842 channels so it supports airs and Metop satellites also. This is carried out last week and experiments are underway
Fixed TAU information for all channels
Release 5version 3N-orders (IO) 10 (Maximum polynomial order used to
reconstruct the regression coefficients )N_predict (IU) 6 (Number of predictors used in the gas
absorption regression)N_absorb(J) 3 ( Number of gaseous absorbers)N-Chn (L) 535Absorber_ID 12, 113, 114Alpha 13., 10., 4 (Array containing the alpha values
used to generate the absorber space levels)Alpha C1 First constant (slope) used in defining the Alpha to
absorber space equationAlpha C2 Second constant (slope) used in defining the
Alpha to absorber space equation
Tau precomputed information for each channel
Order_index Array containing the polynomial orders to use in reconstructing the gas absorption modelTYPE: INTEGER( Long ) DIMENSION:
0:n_absorber x J x L
Predictor_index Array containing the predictor indices used to identify which predictors to use in the gas absorption model.TYPE: INTEGER( Long ) DIMENSION: 0:n_Absorber x J x L
Coefficients Array containing the gas absorption model coefficients. TYPE: REAL( Double ) DIMENSION: 0:n_predictors x 0:n_absorbers x J x L
INSAT SYSTEM• INSAT system is commissioned with launch of INSAT-1B. It is a joint venture of
Dept’s of Space, Telecommunications,Indian Meteorological Dept., All India Radio and Doordarshan.
• The meteorological payload consist of Very High Resolution Radiometer(VHRR) with 2km resolution in visible and 8 km in infrared band.
• The INSAT space segment consists of INSAT-1D(74oE), the last of INSAT series launched in 1990’s and three ISRO built satellites INSAT-2A (74oE)launched in jul,1992, INSAT-2B ( 93.5oE) launched in Jul, 1993 and INSAT- 2C launched on Dec,1995( 93.5oE).
• INSAT-2E launched in April,1999. It is located at 83oE and after INSAT-2B this satellite again carries meteorological payloads. It carries , a new Charge Coupled Device (CCD) camera operating in visible, near infrared, and shortwave infrared band with 1km resolution along with normal VHRR. The CCD camera is still operating while VHRR is shut off.
• INSAT-3B(83oE) and INSAT-3C(74oE) are exclusively communication satellitesand they have been advanced to precede that of 3A to cater the immediate requirement of the extended C band capacity that was depleted due to INSAT-1D failure.
• INSAT- 3A (93.5E) launched on April 2003, and Kalpana-I (74 E), launched on September 2002 , are having operational meteorological payloads at present.
• INSAT-3D which is scheduled to be launched soon will first time carry a sounder along with VHRR
Launching of METSAT on 12th Sep 2002
PSLV-C4 / METSAT Mission
•METSAT
• VHRR on board of METSAT has capability to take images in 3 spectral bands – visible, water vapor and thermal infrared. The pictures provide a spatial resolution of 2kmX2km in visible and 8 kmx8 km in water vapour and thermal infrared bands
• In addition to VHRR METSAT carries a Data Relay Transponder (DRT) configured to collect local meteorological information from unattended Data collection platforms and relay them to Meteorological Data utilisation center at New Delhi.
• The 3 axis stabilised spacecraft will be stationed at 74oE.
INSAT• The daily rainfall analysis procedure was developed to
provide grided rainfall values by merging INSAT QPE and raingauge values.
• Experiments were conducted to include INSAT OLR data in the Global Data Analysis (GDAS) system.
• The satellite imagery can provide data as much better resolution, but its direct use in operational models is not yet common. This information can be used in models by generating proxy data sets such as satellite image based moisture profiles and tropical cyclone bogus.
• NCMRWF already tested various moisture bogusing schemes and these schemes may be used to nudge moisture in mesoscale models.
• AMSUA Brightness Temperature for a typical day 15 Jul 2007. a) Satellite observed (NOAA-16 and NOAA18). B) Difference between Model simulated and Observed BT’s. C) Difference between simulated BT with and no Bias correction
Impact study
• A low resolution impact (T62L64) study experiment to study the impact of radiance data was conducted during monsoon-2007 period.
• Level 1B radiance data sets from NOAA-16 and 18 viz AMSU-A , AMSU-B and HIRS/3 data sets were used in the experiment.
• Experiment was conducted by running GDAF cycle with and without including radiance data sets.
• Analysis increment at the first cycle of exp. 00z01Jun2007
• A) no radiation• B) with radiation
850 hPa Analysed Zonal mean Zonal Wind(m/sec) Jun 07
250 hPa Analysed Zonal mean Temperature(0k) for Jun 07
250 hPa Analyses Zonal mean u wind(m/sec) for Jun 07
850 hPa Analyses Zonal mean Temperature(0k) for Jun07
RMSE observed-background fields for June 2007 over N.H
No rad
With rad
Mean error observed-background fields for June 2007 over N.H
No rad
With rad
RMSE observed-background fields for June 2007 over S.H
No rad
With rad
Mean Error observed-background fields for June 2007 over S.H
No rad
With rad
RMSE observed-background fields for June 2007 over Tropics
No rad
With rad
Mean error observed-background fields for June 2007 over Tropics
No rad
With rad
RMSE observed-background fields for June 2007 over Indian region
No rad
With rad
Mean error observed-background fields for June 2007 over Indian region
No rad
With rad
0
1
2
3
4
5
6
7
1 2 3 4 5
Day
RM
SE
No-RadWith-RadT254
850 HPa Wind over Indian region July 07
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
1 2 3 4 5
Day
Cor
rela
tion
850 HPa Geo Potential Height
0
10
20
30
40
50
60
70
1 2 3 4 5
Day
Skill
Sco
re
No-RadWith-RadT254
250 HPa Geo Potential Height -NH
The future observation type• Targeted drop-sondes, as successfully trialed by the
Americans • Drift sondes, as proposed for testing during THORPEX • Unmanned aerial vehicles (UAVs), • IASI, AIRS and CrIS satellite sounders.• GIFTS geostationary satellite• Satellite-based Doppler wind lidar (ADM) • Satellite-based radar (Global Precipitation Mission)• GPS radio occultation.• Meteosat Second Generation with 12 channels, • Humidities from AMDAR.• Land-use information from AVHRR and MODIS