Title: Impact study of AMSR-E radiances in NCEP Global Data Assimilation System
1Impact study of AMSR-E radiances in NCEP Global
Data Assimilation System
- Masahiro Kazumori(1)
- Q. Liu(2), R. Treadon(1), J. C. Derber(1) , F.
Weng(2), S. J. Lord(1) - (1) NOAA/NCEP/EMC
- (2)NOAA/NESDIS
2Contents
- Purpose of this study
- Development of Microwave Ocean Emissivity Model
- Data Assimilation Experiment
- Results
- Conclusions
3Purpose of this study
Investigate the impact of AMSR-E radiance on NCEP
global model AMSR-E (Advanced Microwave Scanning
Radiometer for EOS) observes the radiance from
the Earth with 6 microwave dual-polarized
channels.
Frequency GHz Polarization Physical Observable
6.925 V,H SST
10.65 V,H SSW
18.7 V,H WV
23.8 V,H WV
36.5 V,H SSW
89.0 V,H Rain
AMSR-E Sensor Unit
Aqua
These low frequency channels are sensitive to SST
and SSW and less sensitive to hydrometeor in the
atmosphere. They can be assimilated in the all
weather condition.
4Development of Microwave Ocean Emissivity Model
for AMSR-E
- Community Radiative transfer model (CRTM) has two
options for Microwave Ocean Emissivity Model - FASTEM (Developed by UKMO)
- NESDISEM (Developed by NESDIS)
operational use
Comparison of TBcal - Tbobs
FASTEM
NESDISEM
00z 16 August 2005
Both models have large bias(about 3K) in 10.65GHz
(H).
Necessary to develop a new microwave ocean
emissivity model
5Design of New Microwave Ocean emissivity model
- Wind speed dependent model
- Fresnel Reflectivity in a calm sea
- Two-Scale Ocean roughness model
- Small Scale correction (Liu1998, Bjerkaas1979)
- Large Scale correction (Modified Storyn1972)
- Foam emissivity and foam fraction (Modified
Storyn1972,Rose2004) - Coefficients were derived from the fitting to
satellite measurements (AMSR-E, SSMI and AMSU-A). - TL and AD models with respect to SSW and SST
6Comparison of (TBcal - Tbobs) KAMSR-E 10.65
GHz (H)
00z 16 August 2005
NESDISEM
FASTEM(operational)
New model
Biases are substantially reduced.
7Comparison of (Tbcal-Tbobs) vs Wind Speed AMSR-E
10.65 GHz (H)
FASTEM
NEWMDL
Bias is depend on surface wind speed. New Model
has smaller bias than operational (FASTEM).
8Comparison of FASTEM NEWMDLin AMSR-E channels
Horizontal-polarization
Vertical-polarization
Statistic period1-5 December 2005
Bar BIAS LineSTD
FASTEM
New Model
New model is better in the low frequency (lt
20GHz).
9Data Assimilation ExperimentConfiguration
- Analysis NCEP GSI 3D-Var assimilation system
- Forecast NCEP global model (as of May 2006)
- 00z Initial 180 hour forecast
- Resolution T382L64 (same as operational, about
50km in horizontal) - Cntl Same as operational
- Test1 Cntl AMSR-E
- with FASTEM ( all
microwave frequency range) - Test2 Cntl AMSR-E
- with NEWMDL (lt20GHz only)
and FASTEM (gt20GHz) - Period 12 Aug.-11 Sep. 2005
AMSR-E 6.925GHz channels(V,H) are not used
because their FOV size are too large
(43.2x75.4km)
10Data Assimilation ExperimentQuality Control of
AMSR-E radiance data
- Select ocean data and thin with 160km distance
- Remove rain and cloud affected data(Criteria are
based on CLW) - Remove land or ice contaminated data (FOV size is
29.4x51.4km at 10.65GHz) - Remove sun glint affected data in the ascending
orbit - Gross error check
- (Tbobs- Tbcal lt Threshold )
Tbcal-Tbobs K 10.65 GHz (V) 00z 16 Aug. 2005
TB bias correction term FOV dependent
air-mass dependent
0.1 of all data are used for the assimilation.
A few thousand / analysis
11ResultsImpact on Analysis
TK
Test1
Mean difference Test-Cntl T Q at 850hPa
PeriodAug.12-Sep.11 2005
No systematic bias in temperature and moisture
Qg/kg
12ResultsImpact on Analysis
TK
Test2
Mean difference Test-Cntl T Q at 850hPa
PeriodAug.12-Sep.11 2005
Increase of Temperature (about 0.2K) in the high
latitude. Decrease of moisture (about 0.1g/kg)
over ocean.
Qg/kg
13ResultsImpact on Forecast (A.C. at 1000hPa
Height)
N.H. Almost Neutral S.H. Positive
(Test1Test2)
AMSR-E radiance assimilation is positive for the
S.H.
Period00z 12 Aug.-00z 11Sep. 2005
14ResultsImpact on Forecast (A.C. at 500hPa Height)
N.H. Almost Neutral S.H. Positive
(Test1Test2) Test2 is slightly better than Test1
Period00z 12 Aug.-00z 11Sep. 2005
15ResultsImpact on Forecast (Fits to RAOB wind)
RMSE of 24 and 48 hour Vector Wind forecast are
reduced in the S.H.
Test1
dotted Test solid Cntl Black24hr
forecast Red 48hr forecast
Test2
16ResultsImpact on Forecast
Zonal mean of 5-day Temperature Forecast RMSE
against initial
RMSE Difference (Test Cntl)
Blue color means improvements
Test1
Test2
17Case study Hurricane Track Prediction (Katrina
2005)
- 5 samples in the experiment period
- (00z 25 August 00z 29 August, 00Z initial
forecast)
Test2 is better than Test1.
18Conclusions(1/2)
- A MW Ocean emissivity model was developed for
AMSR-E - The model is an empirical two scale roughness
model, the coefficients were derived from the
fitting to the satellite measurements. - The model has a better performance for low
frequency channels than FASTEM. - Impact study of AMSR-E radiances in NCEP global
data assimilation system - The new MW ocean emissivity model was used in
CRTM for the experiment. - Three cycle experiments were conducted.
- Cntl same as operational
- Test1 Cntl AMSR-E
- (with FASTEM)
- Test2 Cntl AMSR-E
- (with New model lt 20GHz,
with FASTEM gt20GHz)
19Conclusions(2/2)
- Impacts on analysis
- Increase of Temperature in high latitudes,
decease of moisture over ocean at 850hPa. - Impacts on forecast
- Positive for the S.H. (A.C., RMSE, Fits to
RAOB) - Neutral for the Tropic and the N.H.
- New emissivity model showed better results.
- The new emissivity model can extract the
information on the ocean surface (SSW, SST)
effectively from AMSR-E radiances in the data
assimilation system.
20Thank you
21backup
22Microwave Ocean emissivity
( p h or v )
Total Reflectivity
Frequency
Zenith angle
- In a calm sea, the ocean surface is specular.
- Reflectivity can be calculated by Fresnel law.
sea surface
23Microwave Ocean emissivity
- When wind starts blowing, it makes small ripples
on the ocean surface. - The height variance is
Ocean roughness spectrum function (Bjerkaas1979)
Small-scale height variance is
cutoff wave number
( p h or v )
Small Scale roughness correction
24Microwave Ocean emissivity
- Large scale roughness correction
- A function of wind speed, incidence angle and
frequency
Large Scale roughness correction
Coefficients were obtained from the fitting to
the satellite measurements (AMSR-E,SSMI and
AMSU-A)
25Microwave Ocean emissivity
- Foam emissivity
- Foam fraction
- Total reflectivity
Modified Stogryn1972 function based on
Rose2004 FASTEM uses a constant (1.0) for both
polarization.
Stogryn1972
10m wind speed
FASTEM use Monahan(1986)
26ResultsImpact on Forecast (Fits to RAOB wind)
For the N.H. and the Tropics, impacts are almost
neutral for Test1 and Test2.
27Zonal mean of RMSE of 500 hPa height forecast
against initial. Difference ( Test Cntl )
1-day forecast
1-day forecast
Test1 (AMSRE with FASTEM) Cntl (W/O AMSR-E)
5-day forecast
3-day forecast
5-day forecast
3-day forecast
Negative value indicate improvement
28Zonal mean of RMSE of 500 hPa height forecast
against initial. Difference ( Test Cntl )
1-day forecast
1-day forecast
Test2 (AMSRE with NEWMDL) Cntl (W/O AMSR-E)
5-day forecast
3-day forecast
5-day forecast
3-day forecast
Negative value indicate improvement
29Conclusions
- Impact on analysis
- In Test1, no systematic bias in mean analysis
field (850hPa temperature, humidity). - In Test2, increase 850hPa temperature (0.2K) in
the high latitude. - decrease 850hPa humidity (0.1g/kg) over ocean.
- decrease guess TPW bias
- no significant difference mean 6-hour rain (not
shown). -
30Conclusions
- Impact on forecast
- Positive
- A.C. of 500hPa for S.H., A.C. of 1000hPa N.H.
and S.H. - Fits to RAOB of 24, 48 hour vector wind
forecast in the S.H. - RMSE of 500hPa height for 3day and 5day
forecast - RMSE of temperature from 1000 to 100hPa for 3,5
day forecast - (Test2 has
larger improvement than Test1) - RMSE of 200hPa vector wind (negative for FASTEM
case) not shown - Neutral
- A.C.500hPa of N.H. (Slightly positive for Test1
case) - Fits to RAOB of 24 and 48 hour vector wind for
the Tropics, N.H. - Negative
- RMSE of 850hPa vector wind in the Tropics (not
shown) - A Case Study of Hurricane Track prediction
(Katrina) - Test1(FASTEM) degrade a hurricane track
prediction. - Test2(New model) keeps the accuracy
31ResultsImpact on Analysis (Total Precipitable
water kg/m2)
Test1
Test2
Zonal mean
Bias in guess
Bias of total precipitable water in guess field
are reduced slightly.
32ResultsImpact on Forecast
Zonal mean of 3-day Temperature Forecast RMSE
against initial
RMSE Difference (Test Cntl)
Blue color means improvements
Test1
Test2