A23
Recent Development of ATOVS usage in Korea Meteorological Administration
Sang-Won Joo, Eun-Ju Lee, and Seung-On Hwang
Korea Meteorological Administration, Seoul, Korea
Introduction
Introduction
Direct Assimilation Algorithm
History of (A)TOVS assimilation at KMA
What we
want talk
about today!
2000: 1dVar for TOVS + 3dOI
2002: 1dVar for ATOVS + 3dOI
Details of the H operator for radiance
ATOVS Radiance
Scan angle and air mass bias correction is done before
3DVAR(Joo and Okamoto, 2000)
1DVAR
1DVAR
2004: Direct assimilation of ATOVS in 3dVar
Method
Method
Radiance departure
- Bias correction depends on the cloud condition and latitude
- No bias correction for the stratospheric channels
QC, bias correction &
Thinning
H operator is linearized in the background states
and no update within the inner loop
Observation error(Joo
Joo &
& Lee
Lee 2002)
2002)
3dVar Formulation
J (η ) = η Tη + (Hη − D) R −1 (Hη − D)
η = L−1 (x − xb ), B = LLT
T
Inner loop
(T63)
D = ( y − H ( xb ))
Level 1d data of HIRS 1-8,10-15,AMSU-A 5-15
channels are used at the same time
J (η ) = η Tη + (D − Hη ) R −1 (D − Hη )
T
Outer loop
(T213)
Diagnostics
Diagnostics
R = (H (xb ) − yo )(H (xb ) − yo ) − HBHT
T
η = L−1 [δζ , δDu , δ (T , Ps )u , δ ln q ]σs
• Control variables:
xb = [ζ , D, T , Ps , q ]s
σ
• Model variables :
Adjoint check and Normalized Cost Function
• Background error
in radiance space
• Square of innovation
• First estimates of
Derber and Wu (1999)
• Observation error
• RTM error and
instrument error
Satellite
sounding
(Hdx)T*(Hdx)
dxTHTHdx
1
0.025767163
0.025767163
2
0.514098014
0.514098014
3
0.744231604
0.744231604 J
4
0.864059397
0.864059397
5
0.758332559
0.758332559
6
0.557038524
0.557038524
7
0.146815814
0.146815814
y = [φ , u, v, T , q] , Ps , Ts , us , vs , qs , Rad
p
g
• Observed variables :
One point observation experiment
Effect of fixed TL
in basic state
Mass-wind relationship
Put an ATOVS observation of which innovation is 3
deg in all channels at the point of 30N / 125E
Temperature increment
results in a reasonable
Variable
operator is not updated
usually.
Is it true?
1dVar retrievals with and
Temperature increments are positive except tropopause
and humidity is decreased. It is what we expected!
without H update are
compared.
Background temperature
Schematic diagram explains
error correlation
the negative correlation
30
Mass and wind balance is maintained by the linear
balance equation and it makes the wind increment by
the ATOVS assimilation in 3dVar.
25
15
10
xa
a
Tb
Fixed basic state of H is
Ta
The negative temperature analysis increment near the
tropopause is caused by the negative background error
correlation of temperature.
b’
a’
5
5
10
15
20
25
acceptable
Bias
10
11
21
31
41
51
61
71
81
91 101 111
0.998
925 T
1
0.569
0.998
850 T
1
0.315
0.999
700 T
1
0.003
1
600 T
1
-0.31
1.001
500 T
1
-0.837
1.003
400 T
1
-1.426
1.005
300 T
0.998
-1.384
1.005
250 T
0.999
-0.274
1.001
200 T
1
0.395
0.998
150 T
1
-0.445
1.002
100 T
1
0.086
1
70 T
1
0.141
0.999
50 T
1
0.103
1
30 T
1
0.033
1
20 T
1
0.011
1
10 T
1
-0.005
1
Reg Coef
Variable
Corr
Coef
Intercept
Slope
SFC q
1
0.022
0.998
1000 q
1
0.027
0.997
925 q
1
0.035
0.993
850 q
0.999
0.033
0.987
700 q
0.999
0.049
0.983
600 q
0.999
0.06
0.985
500 q
0.998
0.069
0.985
400 q
0.992
0.06
0.982
300 q
0.993
0.017
0.987
MSLP
1
0.683
0.999
SFC U
1
-0.032
1.001
SFC V
1
-0.007
1.001
7
•In order to correct the warm bias, the ATOVS
stratospheric channel data is assimilated without bias
correction.
8
•However the negative innovation of stratospheric
channels are increased.
11
Zonal averaged T diff (DG3V-G3OI)
•The warm bias is corrected in the stratosphere.
Pressure
50
3doi
3dvar
d ire c t
700
8
5
0
90
20
5
10
-2
0
2
4
6
0
8
1
2
3
Fit well in the
stratosphere
20
30
50
50
70
100
3doi
3dvar
d ir e c t
-1 0
-8
-6
6
70
100
150
200
250
300
400
500
G3OI
G3VR
DG3V
3doi
3dvar
d ir e c t
700
8
5
0
90
20
5
10
-4
-2
0
ATOVS assimilation makes
the north pacific high
stronger and then corrects
the north-east bias of
typhoon track forecast.
7
STDV is small
for G3OI at
Troposphere
10
30
150
200
250
300
400
500
5
Tropics
20
700
8
5
0
90
20
5
10
4
T e m p e ra tu r e
2
4
6
8
T e m p e ra tu re
0
1
2
3
4
5
6
7
T e m p e r a tu re
3500
SH
10
20
30
30
50
50
70
Pressure
20
70
100
150
200
250
300
400
500
3doi
3dvar
d ir e c t
700
8
5
0
90
20
5
10
-1 0
-8
100
150
200
250
300
400
500
3doi
3dvar
d ir e c t
700
8
5
0
90
20
5
10
-6
-4
-2
0
T e m p e r a tu re
2
4
6
0
8
1
2
3
4
5
T e m p e r a tu r e
Dec. 24,2003 –Jan. 7 2004
6
7
0
168
192
216
240
0
24
48
72
Forecast hour
TY0407
600
400
200
0
-200
200
192
216
DG3V
Summary
Summary&&Plans
Plans
BIAS
15
Summary
10
-800
120
108
96
84
72
60
48
36
24
12
5
4
3
CNTL-EXPR 0408 Track Error
61712
2
1
62112
62012
61912
61612
61512
61812
Initial Time
400
DG3V
G3VR
TY0408
200
0
4
5
6
7
8
9
10
NH 500Z 2004 12
BIAS
140
CNTL-EXPR 0410 Track Error
62512
0
70312
70112
62712
62912
40
20
Initial Time
0
1
2
3
800
TY0410
600
120
200
100
4
5
6
SH 500Z 200 412
DG3V
140
400
7
8
9
10
BIAS
- Careful investigation is needed to correct model bias by
using the ATOVS data because of the vertical error
correlation of model error
Plans
80
60
-200
- Direct assimilation of ATOVS improves forecast
performance and it becomes operation at KMA in
August 2004.
- Typhoon forecast errors are reduced by the ATOVS
direct assimilation and it is assumed to be caused
by the better analysis of the north pacific high.
60
0
-200
3
DG3V
160
0
062612
062712
062812
062912
063012
070112
070212
070312
2
80
62312
1000
600
1
100
-600
6
0
120
-400
0
-200
40
-400
96
84
72
60
48
36
24
12
AC
72
400
0
70112
70312 70212
63012
Initial Time
0
1
2
3
4
5
6
7
8
9
10
-400
84
72
60
48
36
24
12
6271262612
62912 62812
- FGAT for ATOVS direct assimilation
20
r
ou
48
168
0
tH
24
144
062312
062412
062512
062612
062712
062812
062912
063012
070112
070212
070312
s
ca
re
Fo
0
061512
061612
061712
061812
061912
062012
TY0406
600
ur
Ho
st
ca
DG3V
G3VR
0.2
120
800
800
re
0.4
96
Forecast Hour
Reduction of error is about 200km at 72 hour forecast.
Fo
0.6
72
5
SH
0.8
48
Forecast Hour
Distance[km]
G3VR
24
1000
AC
DG3V
0.2
0
216
r
0.4
192
u
Ho
0.6
168
20
1200
Tropics
0.8
144
st
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
96
120
144
168
192
216
240
0
Forecast hour
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
96 120 144 168 192 216 240 0
120
ca
96
120
144
Forec as t hour
96
re
72
72
Fo
48
48
ay
D
st
ca
re
Fo
24
24
One month averaged RMSE of 500hPa
25
1000
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
AC
DG3V
G3VR
0.2
0
Distance[km]
NH
0.4
1000
April 2, 2004 –July 31, 2004
1
0.6
1500
0
G3VR
– DG3V
- TY0406,TY0407,TY0408,TY0410-
500
450
400
350
300
250
200
150
100
50
0
DG3V
2500
2000
500
Forecast verification against its own analysis
0.8
G3VR
•The bias correction should not apply selectively.
New experiments BIAS is performed to solve the problems in the stratosphere
BIAS : DG3V + Bias correction is applied in the stratospheric channels depending on the latitude
Mean Diff of TY Track Error(CNTL- EXPR)
Typhoon Track Forecast Error
3000
Distance(km)
10
•It is caused by the negative vertical temperature error
correlation in 3dVar.
Distance(km )
-4
Distance[km]
-6
•However, the warm bias is strongly increased from 300
to 100hPa.
Z500 DIF(DG3V-G3VR) 200407
3doi
3dvar
d ir e c t
Distance[km]
-8
70
150
200
250
300
400
500
Pressure
Pressure
0.998
0.624
100
T e m p e ra tu re
Pressure
0.481
1
30
70
-1 0
A n o m a l y c o r r e l a tio n
Slope
1
20
50
150
200
250
300
400
500
700
8
5
0
90
20
5
10
A n o ma ly c o rre la t io n .
Reg Coef
Intercept
SFC T
10
G3VR has strong northeast bias in typhoon
track forecast
Best track
10
30
Pressure
Standard deviation
NH
20
100
0
1
•KMA global model has strong warm bias in the
stratosphere.
6
DG3V
G3VR
G3VR
Forecast
tracks
RAOB fit of analysis(2003121400-122818)
1
Corr
Coef
1000 T
AMSUA-5
Typhoon Track Forecast Error / TY0406,07,08,10
•DG3V : Direct assimilation + 3dVar
•G3VR : 1dVar + 3dVar
•G3OI : 1dVar + 3dOI
0
0.0
Problems in stratosphere
Experiments
1 0
0.2
Time series of global averaged Innovation
RESULTS
RESULTS
10
0.4
30
Level
Anomaly correlation.
Level
b
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
Direct assimilation
0.6
In order to get the 3dVar
time, basic state of the H
20
3dVar
0.8
Iteration
Humidity increment(DPD)
xb
1.0
0
73112
73012
72912
72812
Initial Time
725
72712 72612
Typhoon track forecast errors are small at the
initial stage of typhoon development !
0
0
1
2
3
4
5
6
7
8
9
10
Bias correction is important to
improve forecast skill !
- ATOVS level 1C data assimilation in the Unified 3dVar
which was developed to run the global and regional
application with the same code.