21
Recent Developments in assimilation of ATOVS at JMA 1.Introduction 2.1DVar preprocessor 3.Simple test for 3DVar radianc e assimilation 4.Cycle experiments 5.Conclusion and plan Kozo Okamoto , Yoshiaki Takeuchi, Yukihiro Kaido, Masahi ro Kazumori NWP Division, Forecast Dept, Japan Meteor ological Agency

Recent Developments in assimilation of ATOVS at JMA

  • Upload
    micheal

  • View
    25

  • Download
    0

Embed Size (px)

DESCRIPTION

Recent Developments in assimilation of ATOVS at JMA. Kozo Okamoto , Yoshiaki Takeuchi, Yukihiro Kaido, Masahiro Kazumori NWP Division, Forecast Dept, Japan Meteorological Agency. 1.Introduction 2.1DVar preprocessor 3.Simple test for 3DVar radiance assimilation 4.Cycle experiments - PowerPoint PPT Presentation

Citation preview

Page 1: Recent Developments in assimilation of ATOVS  at JMA

Recent Developments in assimilation of ATOVS at JMA

1.Introduction

2.1DVar preprocessor

3.Simple test for 3DVar radiance assimilation

4.Cycle experiments

5.Conclusion and plan

Kozo Okamoto,

Yoshiaki Takeuchi, Yukihiro Kaido, Masahiro Kazumori

NWP Division, Forecast Dept, Japan Meteorological Agency

Page 2: Recent Developments in assimilation of ATOVS  at JMA

Recent Change in the JMA NWP system

• Mar. 2001 : Replace the supercomputer (768GFlops, 640GByte, 80node)

GSM T213L30 => T213L40 (model top : 10=>0.4 hPa)• Sep. 2001 : Global 3DVar system started in operational data assimilation

system

• Mar. 2002 : Meso 4DVar system is going to start in operational data assimilation system (H.Res.: 10km, Assimilation window: 3h)

Page 3: Recent Developments in assimilation of ATOVS  at JMA

Use of ATOVS in the JMA assimilation system

NESDIS/MSCT,Q retrievals

・ conversion・ QC・ select region

Present Status Retrieval Use

3DVar

NESDIS 120km BUFR TBB

・ QC ・ Channel Selection ・ Obs Error Assignment ・ Bias Correction

PlanTBB Use

3DVar

1DVar as preprocessor

dZ( -1000hPa)Bias Corrected TBBTskin

Page 4: Recent Developments in assimilation of ATOVS  at JMA

ATOVS 1DVar as Pre-processor (1)Quality Control (QC)

• Geographical check : reject data over the coast, lake and river ..

• Edge scan check: reject data with outer edge swath

• Gross check : reject data for TBB >400K or <100K

• Rogue check-1: reject data including some channels with |dTBB|>a*Ostd

• Minimize check: reject data not converged within 12 iterations

• Jend check: reject data with Jend>8*used channel number

• Rogue check-2: tighter Rogue check-1

0200400600800

100012001400160018002000

all geo-graphic

edgescan

gross rogue1 minimize J end rogue2

NOAA15NOAA16

pass data number for each QC : 00Z 20Dec2001

Page 5: Recent Developments in assimilation of ATOVS  at JMA

ATOVS 1DVar as Pre-processor (2)Bias Correction

• The TBB bias for each channel j can be described by

– y: background TBB (TBbg) of AMSU-5,7,10

– TPW: background total column precipitable water

– : satellite scan angle, Ts:skin temperature

– overbar represents spatial and temporal mean

• The regression coefficients aji are updated every day using previous 2 weeks d

ata and calculated for NH/Trop/SH and each analysis time.

• The bias-correction is not applied to HIRS11,12,AMSU13,14 because of large systematic errors in the JMA forecast model

cos

1)()(

)(

654

3

10

jSSjj

iii

jijj

aTTaTPWTPWa

yyaaBIAS

Page 6: Recent Developments in assimilation of ATOVS  at JMA

AM

SU

-A

ATOVS 1DVar as Pre-processor (3)

Channel Selection and Observation Errors

The channels to be used and observation errors for each observation condition : Clear/Cloudy and Sea/Ice/Land

– Clear Sea : HIRS1-8, HIRS10-16, AMSU5-14

– Land : only HIRS1-3 and AMSU 8-14 are used.

• Observation errors used in 3DVar are multiplied by 1.5.

• At the moment,

– Cloud detection is based on NESDIS flag

– Ice detection based on SST<1K and the classification is corrected as sea when TBob - TBbg <-50 for AMSU1

chClearSea

CloudySea

ClearSea Ice

CloudySea Ice

ClearLand

CloudyLand

1 HIRS1 1.40 1.40 1.40 1.40 1.40 1.402 HIRS2 0.35 0.35 0.35 0.35 0.35 0.353 HIRS3 0.30 0.30 0.30 0.30 0.30 0.304 HIRS4 0.20 0.20 0.205 HIRS5 0.30 0.306 HIRS6 0.40 0.807 HIRS7 0.60 1.208 HIRS8 1.10 2.209 HIRS910 HIRS10 0.80 0.8011 HIRS11 1.10 1.1012 HIRS12 1.50 1.50 1.5013 HIRS13 0.50 0.5014 HIRS14 0.35 0.3515 HIRS15 0.30 0.3016 HIRS16 0.80 0.8017 HIRS1718 HIRS1819 HIRS1920 HIRS2021 MSU122 MSU2 0.30 0.30 0.60 0.6023 MSU3 0.22 0.22 0.22 0.22 0.22 0.2224 MSU4 0.25 0.25 0.25 0.25 0.25 0.2525 SSU1 0.60 0.60 0.60 0.60 0.60 0.6026 SSU227 SSU328 AMSU129 AMSU230 AMSU331 AMSU432 AMSU5 0.40 0.40 0.80 0.8033 AMSU6 0.40 0.40 0.80 0.8034 AMSU7 0.40 0.40 0.40 0.4035 AMSU8 0.40 0.40 0.40 0.40 0.40 0.4036 AMSU9 0.40 0.40 0.40 0.40 0.40 0.4037 AMSU10 0.40 0.40 0.40 0.40 0.40 0.4038 AMSU11 0.40 0.40 0.40 0.40 0.40 0.4039 AMSU12 0.50 0.50 0.50 0.50 0.50 0.5040 AMSU13 1.60 1.60 1.60 1.60 1.60 1.6041 AMSU14 2.50 2.50 2.50 2.50 2.50 2.5042 AMSU15

HIR

S

Page 7: Recent Developments in assimilation of ATOVS  at JMA

Surface type and TBob-TBbg

• Due to mis-classimication of surface type, TBbg is quite different from TBob.– The mis-classification of the coast accounts for 95% of data with TBob-TBbg >50K– The mis-classification of the sea ice accounts for 98% of data with TBob-TBbg <-50K

Distribution of data with large TBob-TBbg for AMSU A1 (10 Oct - 11 Nov 2001)

Page 8: Recent Developments in assimilation of ATOVS  at JMA

JMA 3DVar

• Incremental method

– Outer loop : T213L40

– Inner loop : T106L40

• Background error covariance is calculated by using the NMC method

– Horizontal homogeneous

• Observation operator for radiance data

– RTTOV6 ADJ and TL model

Page 9: Recent Developments in assimilation of ATOVS  at JMA

Evolution of Cost function J and Gradient of J with iteration

• The minimization is continued for 100 iterations

• Case of 12Z on 18th Dec. 2001

Radiance Assimilation Retrievals Assimilation

Cos

t J

|gra

dJ|

All

Radiance

Others Others

All

Z

Page 10: Recent Developments in assimilation of ATOVS  at JMA

Cross Section along observation longitude(137E)

Q[g/kg] U[m/s]

0.410

100

300500700

0.410

100

300500700

0.410

100

300500700

0.410

100

300500700

Analysis Increment for 1ch-1point observation• Only one HIRS4 observation with TBB departure of +10*Observation error STD is assimilated at the point of

35N,137E• Analysis Increments are large in the stratosphere because of the large background error covariance and wide

spread RT sensitivity.

T[K] Z[m]

Page 11: Recent Developments in assimilation of ATOVS  at JMA

Analysis Increment for 1ch-1point observation

At the 35th level of JMA eta level (around 10hPa)

Q[g/kg]

T[K] Z[m]

U[m/s]

Page 12: Recent Developments in assimilation of ATOVS  at JMA

ATOVS Radiance Assimilation Impacts on NWP -Parallel Assimilation Experiments (Jul 2001)-

• TEST : 1DVar preprocessor + 3DVar Radiance Assimilation • CNTL: 3DVar Retrieval Assimilation• Data Configurations

– TEST : ATOVS TBB from 120km BUFR• note: All HIRS and AMSU-14 radiances from NOAA15 are not

used due to instrumental problems– CNTL: ATOVS NESDIS retrievals (BUFR + SATEM)

• System – 6hourly intermittent data assimilation– forecast model : T106L40 (model top 0.4hPa) global spectral model,

216h forecasts for 12Z initial– analysis model : 3DVar Incremental method

• 1 month run

Page 13: Recent Developments in assimilation of ATOVS  at JMA

RMSE and Bias of Analysis/Guess verified

against radiosonde

• Temperature on the standard pressure levels from 1000 to 10 hPa

• Case of 30th Jul 2001

Test AnalCntl Anal

Test GuesCntl Gues

Bias RMSE

N.H.

Trp.

S.H.

Page 14: Recent Developments in assimilation of ATOVS  at JMA

RMSE and Bias of Analysis/Guess verified

against radiosonde

• Wind Speed on the standard pressure levels from 1000 to 10 hPa

• Case of 30th Jul 2001

Test AnalCntl Anal

Test GuesCntl Gues

Bias RMSE

N.H.

Trp.

S.H.

Page 15: Recent Developments in assimilation of ATOVS  at JMA

Forecast Errors verified against radiosonde for 500hPa Z

• Improvements especially in the S.H.

• But in the N.H. and Tropics, the improvements diminish beyond day 5 of the forecast.

Test

Cntl

BiasRMSE

N.H.

Trp.

S.H.

Page 16: Recent Developments in assimilation of ATOVS  at JMA

Forecast Errors verified against radiosonde for 250hPa Wind Speed

• Nearly Neutral Impact on forecast

Test

Cntl

BiasRMSE

N.H.

Trp.

S.H.

Page 17: Recent Developments in assimilation of ATOVS  at JMA

Averaged Zonal Mean for Forecast Error at day 5 and Analysis difference

• Average during 13th - 29th Jul 2001

• Large systematic forecast errors around 10 hPa and above 3hPa, especially in the S.H. are obvious.The value is positive around 10hPa while negative above 3hPa.

• Averaged analysis difference is also obvious. Unfortunately Test fits radiosonde worse than Cntl for the 10hPa temperature.

10hPa

Averaged Zonal Mean Forecast error (Fcst - Init ) at day 5 for temperature from 850 to 1 hPa

Averaged Zonal Mean Analysis difference between Test and Cntl for temperature from 850 to 0.4 hPa

10

1hPa

100

-10

10

90N90S

1hPa

100

10

-3

3

90N90S

Page 18: Recent Developments in assimilation of ATOVS  at JMA

Conclusion and Plan

• JMA global 3DVar started operationally since Sep. 2001. At the moment NESDIS and MSC thickness retrievals are assimilated.

• The direct radiance assimilation system is being developed. QC, channel selection and bias correction are performed in the 1DVar pre-processing system.

• Parallel assimilation experiments have been run. Some improvements for analyses and forecasts are given but are not found beyond day 5 of the forecast.

• The problem can be attributed to QC, observation error assignment and data selection ( thinning ). Besides forecast systematic error in the stratosphere probably have something to do with it.

• We have other plans to– assimilate AMSU-B radiance– improve QC– use level 1B data

Page 19: Recent Developments in assimilation of ATOVS  at JMA

AMSU-B Assimilation : initial results• Accuracy of AMSU-B 1DVar products verified against radiosonde

observations for specific humidity below 100 hPa • Studying the impact of AMSU-B radiance on analysis and forecast

AMSU-B retrieval

First Guess

Bias RMSE

N.H.

Trp.

S.H.

Page 20: Recent Developments in assimilation of ATOVS  at JMA

Improve QC (1)• Detect clear/thin cloud/thick cloud/rain using only observation information (not guess)• The system is based on AAPP.

• Cloud detection J = ( y-m )T C-1 ( y-m )– y: TBob of HIRS1-4, 13-15, AMSU4-5 for thin cloud detection

AMSU1-3 for thick cloud detection– m:average clear TBB , C: clear TBB covariance

• designate as cloudy when J>J0

0

2

4

6

8

10

1 3 5 7 9 11 13 15 17 19

HIRS CH

NESDIS STDTEST STD

STD of clear TBob-TBbg over land Histogram of TBob-TBbg for HIRS8 over sea

Clear

Thin cloudy

Thick Cloudy

TBob-TBbg

Page 21: Recent Developments in assimilation of ATOVS  at JMA

Improve QC (2)• Rain detection : Scattering Index SI = TBcal(A15) - TBob(A15)

– TBcal(A15) is calculated based on a statistical regression approach with predictors of AMSU1-3

• designate as rainy when SI > 10.– The threshold 10 is determined based on collocated TRMM TMI and PR rain

NOAA1610Oct2001 - 31J an2002, over Sea

0

5

10

15

20

25

30

HIR

S-1

HIR

S-3

HIR

S-5

HIR

S-7

HIR

S-9

HIR

S-1

1

HIR

S-1

3

HIR

S-1

5

HIR

S-1

7

HIR

S-1

9

AM

SU

-2

AM

SU

-4

AM

SU

-6

AM

SU

-8

AM

SU

-10

AM

SU

-12

AM

SU

-14

STD

of TB

ob-T

Bbg clear

thin cloudythick clou d yrain

TBob-TBbg STD of each HIRS and AMSU channel for clear/cloudy/rain over sea