Assimilation of ATOVS level1C radiance and MODIS polar winds at JMA

July 19, 2005 EMC Sack Lunch Seminar

Assimilation of ATOVS level1C radiance and MODIS polar winds at JMA

Masahiro Kazumori

Japan Meteorological Agency / EMC Visiting scientist


Introduction of JMA NWP modelsRSMGSM MSM

Global Model (GSM) Regional Model (RSM) Meso-scale Model (MSM)

3-7day forecastBoundary for RSM

1-2day forecastBoundary for MSM

Disaster prevention information

55km, 40levels (0.4hPa) 90 hours (00UTC)216 hours(12UTC)

20km, 40levels (10hPa) 51 hours (00,12UTC)

10km, 40levels (10hPa) 18 hours (00,06,12,18UTC)

4D-Var (Since 2003) 4D-Var (Since 2002) 4D-Var (Since Feb. 2005)

Purpose:

Resolution, Forecast time:

Assimilation system:


Satellite data in JMA global data assimilation system

• Radiance AMSU-A(NOAA15/16, Aqua) AMSU-B(NOAA15/16/17)

• Wind GOES(9/10/12),METEOSAT(5/7) MODIS(Terra/Aqua) QuikSCAT

• Radiance AIRS (Aqua) SSMI,TRMM,AMSR-E GOES-9,METEOSAT,MTSAT-1R

• Wind METEOSAT(8),MTSAT-1R

• SSMIS(DMSP-16)

• HIRS,AMSU-A,HSB(NOAA-18)

• IASI (Metop)

• ASCAT(Metop)

Now using operationally Now testing

Planning


Progress on global data assimilation in 2004-2005

• Dec. 2004 : – Direct assimilation of ATOVS level-1C data replaced that of lev

el-1D data– The radiative transfer model for the radiance assimilation was u

pgraded from RTTOV-6 to RTTOV-7• Feb. 2005 :

– Introduction of 4D-Var in operational system• March. 2005:

– Start of Aqua/AMSU-A radiance data assimilation

ATOVS level 1C radiance

MODIS polar winds• May 2004 : – Assimilation of polar winds from Terra/MODIS and

Aqua/MODIS in the north polar region started. • Sep. 2004 :

– Assimilation of polar winds from Terra/MODIS and Aqua/MODIS in the south polar region started.

4D-Var


Assimilation of ATOVS level 1C radiance

Masahiro KazumoriHiromi OwadaKazuyo Fukuda

Numerical Prediction DivisionJapan Meteorological Agency


ATOVS level 1C radiance data

Dec. 2004Dec. 2004

ATOVS data was changed from ATOVS data was changed from level-1D datalevel-1D data(NESDIS pre-processe(NESDIS pre-processed data) to d data) to level-1C datalevel-1C data(unmapped instrument data) (unmapped instrument data)

At the same time, RTM upgrade was performed.At the same time, RTM upgrade was performed.•JMA uses RTTOV (Saunders et al. 1998) as RTM

•The update was from RTTOV-6 to RTTOV-7

•RTTOV-7 treats each sensor separately

•RTTOV-7 has an ability to calculate AIRS radiance

Since May 2003, ATOVS radiance data have been assimilated Since May 2003, ATOVS radiance data have been assimilated directly into the global model with JMA 3D-Var system.directly into the global model with JMA 3D-Var system.

But…, it had been using ATOVS level 1D radiance data


AAPP & MSC direct read out data

AAPPAAPP （（ ATOVS and AVHRR Processing PackageATOVS and AVHRR Processing Package ））Direct broadcast data are received at JMA Meteorological Satellite Center.Direct broadcast data are received at JMA Meteorological Satellite Center.

Global level 1B data from NESDISGlobal level 1B data from NESDIS

Direct read out data (HRPT) at MSCDirect read out data (HRPT) at MSC AAPPLevel 1C radianceLevel 1C radiance

decodedecode

The limit of data receiving.The limit of data receiving.

Data coverage of ATOVS level 1c for early analysis.Data coverage of ATOVS level 1c for early analysis.Blue points are the global data from NESDIS and red points are Blue points are the global data from NESDIS and red points are direct read out data at MSC at JMA. 12UTC 31 July 2004.direct read out data at MSC at JMA. 12UTC 31 July 2004.

Level 1D radianceLevel 1D radiance

JMA


Difference of ATOVS data between level 1C and level 1D

Level 1D dataLevel 1D dataNo high latitude data.No high latitude data.Thinned and removed by NESDISThinned and removed by NESDISNESDIS provide some quality flag for NESDIS provide some quality flag for observationobservation

Level 1C dataLevel 1C dataAll observation data are available.All observation data are available.thinning and quality check are needed.thinning and quality check are needed.

Level 1D dataLevel 1D data

Level 1C dataLevel 1C dataBlueBlue ：： NOAA15NOAA15 　　 RedRed ：：NOAA16NOAA16


Difference of ATOVS data between level 1C and level 1D

Level 1D dataLevel 1D dataNo high latitude data.No high latitude data.Thinned and removed by NESDISThinned and removed by NESDIS

Level 1C dataLevel 1C dataAll observation data are available.All observation data are available.Original data thinning and quality Original data thinning and quality check are needed.check are needed.

Level 1D dataLevel 1D data

Level 1C Level 1C datadata


Mapping (computing sounder data to another sounder grid)

HIRS AMSU-A

In level 1D data,In level 1D data,

Mapping of AMSU-A FOV to Mapping of AMSU-A FOV to HIRS FOV are conducted.HIRS FOV are conducted.

In level 1C data,In level 1C data,

We can assimilative those We can assimilative those data in their original data in their original observation points.observation points.

FOVFOV


Assimilation scheme of ATOVS 1C radiance

• Data thinning

– 240km for AMSU-A, 180km for AMSU-B

– Land sea decision: 0.25*0.25degree map and considering the maximum size of FOV for each sensor.

• Quality Control

– Rain detection ( based on scattering index)

– Cloud detection( based on amount of cloud liquid water)

• Observation error

– Remove the adjustment for each FOV

• Bias correction

– Scan bias correction ( fixed for each channel)

– Air mass correction (fixed for all season)

• Predictor: Calculated brightness temperature of AMSU-A Ch5,7,10, Surface temperature.

• Coefficient was calculated from collocated data set with RAOB

discontinued


Land Sea Mask based on FOV

AMSU-A CH4 Departure(monthly mean)

[K]

Land sea grid

Land

Sea0.25degree

0.25degree

A Observation point

Field of View

Land sea mask are modified based on FOV size for each sensor.

Contamination by land


Results of cloud, rain detection

Level 1C dataLevel 1C dataThere are many rain observationThere are many rain observation（（ Level 1C have original all observation dataLevel 1C have original all observation data ））Many cloud around high latitude in winter.Many cloud around high latitude in winter.

For level 1C,For level 1C,

HIRS can not be used for HIRS can not be used for cloud detection (no mapping)cloud detection (no mapping)

Clear Clear →→Thin cloudThin cloud

Level 1DLevel 1D

Level 1CLevel 1C

Used Retrieval Algorithm for QC

CLW from AMSU-A (English et al. 1997) > 100g/m^2 rejectedRAIN (Scattering index method) AMSU-A (over ocean) SI = ETB15 – TB15 >10 rejected AMSU-B ( over ocean) SI = TB1-TB2>3 and Median filter QC for CH1


Mean departure for each scan position

Level1C (AMSU-A)Level1C (AMSU-A) Level1D (AMSU-A)Level1D (AMSU-A)

　　 Level 1D data has an effect of mapping to HIRS field of Level 1D data has an effect of mapping to HIRS field of viewview

RedRed ：： W Bias correctionW Bias correctionBlueBlue ：： W/O Bias W/O Bias correctioncorrection

Stop the use of Stop the use of Ch14Ch14


Difference between level 1C and level 1D• Example: AMSU-A Channel 7 　 (2004/10/07/12UTC)

Level 1DLevel 1D

Level 1CLevel 1C

Level 1C dataLevel 1C dataNo high latitude data.No high latitude data.Thinned and removed by NESDISThinned and removed by NESDIS(NESDIS pre-processed data)(NESDIS pre-processed data)

Level 1C dataLevel 1C dataAll observation data are available.All observation data are available.Data thinning and quality check Data thinning and quality check should be done by ourselves.should be done by ourselves.


Monthly mean O-B distribution (AMSU-A)

level1Dlevel1D

level1Clevel1C

The bias of level1C seems smaller than 1D The bias of level1C seems smaller than 1D It means the QC of level 1C works better.It means the QC of level 1C works better.


Assimilation experiment of ATOVS 1C• Period:

– Summer ： From July 13, 2004 To September 9, 2004

• Verification Period ： From July 26, 2004 to August 31, 2004

– Winter: From December 27 2003 to February 2004

• Verification Period: From January 1,2004 to January 31 2004

• Setting of experiments

Analysis ： Global 3D-Var, Forecast ： T213L40 　 GSM0407 　 NAPEX 　 R111

　 CNTL: same with operational （ For summer, with MODIS polar winds in the S.H.)

　 TEST:CNTL + replace of ATOVS data ( from 1D to 1C)

　　　　　 Update of RTM from RTTOV-6 to RTTOV-7

• Used level1C

　 NOAA15,16 AMSU-A,B, NOAA17AMSU-B

• Discontinued level 1D data

　 NOAA15,16AMSU-A,B, NOAA16 HIRS


TEST GUESSTEST GUESS

TEST ANALTEST ANAL CNTL ANALCNTL ANAL

CNTL GUESSCNTL GUESS

TESTTEST--CNTLCNTL

TESTTEST--CNTLCNTL

ANALANAL--GUESSGUESS ANALANAL--GUESSGUESS

Summer Z500Summer Z500


Impact on analysis

• TEST:

– ATOVS level 1C data

– RTTOV-7

• CNTL:

– ATOVS level 1D data

– RTTOV-6

Zonal mean difference of Temperature for August 2004

Change at the high latitudes in the troposphere Change at the high latitudes in the troposphere And large change in the stratosphere.And large change in the stratosphere.

TEST-CNTL


level 1Clevel 1C

level 1Dlevel 1D

2004 Aug. N.H.2004 Aug. N.H. 2004 Aug.2004 Aug. 　　 S.H.S.H.

2004 Jan.2004 Jan. 　　 N.H.N.H. 2004 Jan.2004 Jan. 　　 S.H.S.H.

Improvement of RMSE of 500hPa GeImprovement of RMSE of 500hPa Geopotential heightopotential height

Z500RMSE

Impacts on forecast

Operational use of level 1C data started on 2 Dec. 2004


Where can I find the improvements?

1day1day

3day3day

5day5day

summersummer winterwinterDifference of RMSE between TEST and CNTL (blue color means improvement)Difference of RMSE between TEST and CNTL (blue color means improvement)

S.H. and the polar S.H. and the polar region were improved.region were improved.


• Target

　　　 The 11 typhoon in the summer test period ( maximum sample number 62 )

red ： TEST, green ： CNTL

Comparison of typhoon track prediction

T0409T0409 T0410T0410 T0411T0411 T0412T0412 T0413T0413

T0414T0414 T0415T0415 T0416T0416 T0417T0417 T0418T0418

T0419T0419

There are some difference between There are some difference between TESTTEST and and CNTLCNTL..But, statistically, the impact was almost neutral.But, statistically, the impact was almost neutral.


Summary

JMA use ATOVS level-1C radiance since Dec. 2004 operationally.

The use of ATOVS level-1D radiance was discontinued.

– Land sea mask are modified for each sensor.

– Cloud, rain detection scheme are introduced.

– Use of direct read out data received at MSC/JMA

– Change of the RTM from RTTOV-6 to RTTOV-7.

– OSE showed the improvement of forecast score on 500hPa height.


Plan for 2005 on ATOVS at JMA

KMA and JMA are going to exchange the direct received ATOVS data.

For Early analysis,These data make the data coverage expand.

Data will be available within 50 min. after the observation

Red: JMA(Tokyo)Blue:KMA(Seoul)

At present, data impact study is being conducted.


Plan for 2005 on ATOVS at JMA

Japan has a ground station(ice station) in the Antarctic. SHOWA-kichi

JMA is getting the direct received ATOVS data in the Antarctic through National Institute of Polar Research (NIPR).

The data is available within 50 min. after observation.

The data is coming through International Mobile Satellite Organization.

Data acquisition is going well.


Exchange of ATOVS Direct Broadcast Data in Eastern Asia

• JMA will contribute to establish a RARS together with the associated inter-regional data exchange mechanisms;

• - JMA is willing to perform its part of the re-transmission functions through the GTS and/or the Internet, in co-operation with CMA, ABoM and other centres;

• - The APSDEU forum should be used to co-ordinate the Asian RARS activities;

• - Responsibility for implementation of functions and operations to be shared between centres.

Reference:CGMS/WMO REGIONAL ATOVS RE-TRANSMISSION SYSTEM(RARS) WORKSHOP REPORT


Data coverageEARS(EUMETSAT ATOVS Retransmission Service ) RARS(Eastern Asia)

If we use both, we will get much data in Early

analysis.

http://www.eumetsat.int/en/dps/atovs/images/coverage.gif


Next…


Assimilation of MODIS polar winds at JMA

Masahiro KazumoriYoshiyuki Nakamura

Numerical Prediction divisionJapan Meteorological Agency


MODIS polar winds• AMVs from Geostationary Satellites have been used at JMA

• But, the polar regions have been remained as data poor regions because no winds from geostationary satellite for these regions. RAOB and Aircraft-network are also sparse. And ATOVS 1D data had no high latitude data.

• Since July 2002, CIMSS(Cooperative Institute for Meteorological Satellite) at Univ. of Wisconsin have been produced AMVs from MODIS on Terra and Aqua for the polar regions.

Orange TerraOrange Terra

Green AquaGreen Aqua

N.H.N.H. S.H.S.H.

Data distribution of AMVs from Satellite

MODIS polar winds fill the gap of observation and will improve the accuracy of MODIS polar winds fill the gap of observation and will improve the accuracy of analysis in the polar regions, and bring a better forecast in the mid-latitudes.analysis in the polar regions, and bring a better forecast in the mid-latitudes.


Difference between CIMSS and NESDIS• Terra/MODIS WV 400hPa N.H. Wind Speed O-B and number

Period:Period:

From 27 December 2003From 27 December 2003

To 9 February 2004To 9 February 2004

Difference in QualityDifference in Quality

And CoverageAnd Coverage

O-BO-B

NumberNumber

CIMSSCIMSS NESDISNESDIS

JMA use CIMSS MODIS winds.


Quality of MODIS Polar Winds• BIAS and RMSE of Wind Speed against first guess

ALL dataALL dataQC PassQC Pass


MODIS polar winds data assimilation Experiments• Period of the experiments

– From 27 June 2003 to 9 August 2003– From 27 December 2003 to 9 February 2004

• ConfigurationsJMA Global Spectral Model (GSM)+ 3D-Var T213L40 CNTL the same as JMA operational run. MODIS polar winds were passively monitored. TEST CNTL+Terra/MODIS+Aqua/MODIS

• Used MODIS polar winds (Only in the Arctic)　 Over ocean 　 IR: above 700hPa WV: above 550hPa　 Over land 　 IR,WV: above 400hPa.

• Data thinning 150km(horizontal) 100hPa(vertical)• The data in the Antarctic have large bias. That degraded the

forecast scores in another experiment.


Impacts on Analysis

• Mean field

– Temperature ： Rise in the lower troposphere and fall in the upper troposphere ( about 0.5 degree in zonal mean)

– 500hPa height:Increase over ocean and decrease over land, especially over Siberia area.

• Comparison with Radiosonde observation(RAOB)

– Analysis and first guess in the TEST became close to RAOB, especially over Siberia

– Improvements on background bring better forecasts.


Zonal mean difference of Temperature

• Rise in the lower troposphere and fall in the upper troposphere

( about 0.5 degrees in zonal mean)

Monthly mean for July 2003Monthly mean for July 2003 Monthly mean for January 2004Monthly mean for January 2004hPahPa


Impacts on Analysis

• Mean field

– Temperature ： Rise in the lower troposphere and fall in the upper troposphere ( about 0.5 degree in zonal mean)

– 500hPa height:Increase over ocean and decrease over land, especially over Siberia area.

• Comparison with Radiosonde observation(RAOB)

– Analysis and first guess in the TEST became close to RAOB, especially over Siberia.

– Improvements on background bring better forecasts.


Change of 500hPa Z • Monthly mean error for July 2003

Analysis (TEST-CNTL)Analysis (TEST-CNTL)

First guess (TEST-CNTL)First guess (TEST-CNTL)

TEST first guess against RAOBTEST first guess against RAOB

CNTL first guess against RAOBCNTL first guess against RAOB

Analysis and first guess in the TEST became Analysis and first guess in the TEST became close to RAOB, especially over Siberia.close to RAOB, especially over Siberia.

(m(m))

(m(m))

(m(m))

(m(m))


Impacts on Forecast

• Anomaly Correlation and RMSE at 500hPa– Large improvement for both seasons in the N.H.– Neutral for the Tropics and the S.H. (no MODIS assimilation in these regions in this test.)– Change in the Arctic spread to the lower latitudes.

• RMS forecast error of wind vectors were reduced.– Especially, improvements at 500hPa was remarkable. ( 500hPa was the level with maximum data number)

• Improvements on typhoon track prediction

– Small, but positive impacts were found at the later stage in the forecasts.


Anomaly 　 Correlation of 500hPa height

TEST:With MODISTEST:With MODISCNTL:Without MODISCNTL:Without MODIS

Large improvements were found for the forecast scoreLarge improvements were found for the forecast score..

MODIS polar winds assimilation in the Arctic


Impacts on Forecast by MODIS polar winds• RMSE at 500hPa

– Large improvement for both seasons in the polar region

– Change in the polar regions spread to the lower latitudes.

RMS forecast error difference for 500hPa Z TEST minus CNTL

• Monthly mean difference for July 2003

1day forecast1day forecast 3day forecast3day forecast 5day forecast5day forecastPositive Impacts of MODIS (negative difference) spread to mid-latitudes with Positive Impacts of MODIS (negative difference) spread to mid-latitudes with procession of forecast.procession of forecast.


Impacts on Forecast

• Anomaly Correlation and RMSE at 500hPa– Large improvement for both seasons by 9-day forecasts– Neutral for the tropics and the S.H. (no MODIS assimilation)– Change in the Arctic spread to the lower latitudes.

• RMS forecast error of wind vectors were reduced.– Especially, improvements at 500hPa were remarkable. ( 500hPa was the level with maximum data number)




Zonal mean of RMSE of Wind Speed at 500hPa（ 5-day forecasts,January2004 ）

ImprovementImprovement

ImprovementImprovement


Impacts on Forecast

• Anomaly Correlation and RMSE at 500hPa– Large improvement for both seasons by 9-day forecasts– Neutral for the tropics and the S.H. (no MODIS assimilation)– Change in the Arctic spread to the lower latitudes.

• RMS forecast error of wind vectors were reduced.– Especially, improvements at 500hPa was remarkable. ( 500hPa was the level with maximum data number)




Mean positional error of typhoon track predictions

• 22 events in July 2003

• Neutral or slightly positive at the later stage in the forecast time.Neutral or slightly positive at the later stage in the forecast time.


Summary• The MODIS polar winds assimilation experiments were performed at JMA

– Period:July 2003, January 2004

– Used data:Aqua/MODIS,Terra/MODIS in the Arctic

– QC

• Over land, above 400hPa for IR and WV

• Over ocean, above 700hPa for IR and 550hPa for WV

– Data thinning 150km(horizontal) 100hPa(vertical)

• Results

– Improvements on the analysis and first guess in the Arctic.

– Large positive impacts on forecasts for both seasons. ( height, temperature, wind fields)

– Improvements on the typhoon track prediction.

Since 27 May 2004, operational use in the Arctic

Since 16 Sep 2004, operational use in the Antarctic


The change of Quality of MODIS polar winds

After June 2004, both data became similar.Same algorithm for retrieval, and same first guess for height assignment.

Wind speed comparison between CIMSS and NESDIS

CIMSS changed the first guess for height assignment from NAVY model to GFS in June 2004.


Next…


Improvement of operational forecast score

12-month averageForecast score of JMA global model is rapidly improving.


RMSE of 500hPa height against initial

1day forecast 5day forecastN.H. N.H.

5day forecast S.H.1day forecast S.H.


Conclusion

• ATOVS level 1C radiance data are used in JMA global data assimilation system operationally.

• MODIS polar winds data are used in JMA global data assimilation system operationally.

• In virtue of these data and 4D-Var system, operational forecast score of JMA global model is improving rapidly.

Filling data poor region(space) with satellite data make better analysis and better forecast.

Next step:

Effective data usage for 4D-Var. Intelligent data thinning and Quality control for 4D-Var time slot in the assimilation window.

Use of new satellite data (AIRS, SSMI, AMSR-E, SSMIS, etc…)


Thank you.

Documents

Assimilation of ATOVS level1C radiance and MODIS polar winds at JMA