munehiko yamaguchi 1 masayuki kyoda 1 ryota sakai 1 1: numerical prediction division, japan...
TRANSCRIPT
Munehiko YAMAGUCHI1
Masayuki Kyoda1
Ryota Sakai1
1: Numerical Prediction Division, Japan Meteorological Agency
Typhoon Ensemble Prediction System at
Japan Meteorological Agency
科研費打ち合わせ@気象研究所 19 Mar. 2007
Probabilistic Track Forecast for CHABA (1)C
urre
nt f
orec
ast
Fut
ure
plan
Black: best trackBlue: unperturbed run (control run)Red: perturbed run(s)Mark is plotted every 24 hoursTEPS
original
tracks
The ensemble track forecasts grasp the observed CHANBA’s track.
Initial time: 12UTC 24Aug 2004
Current forecast
original
tracks
The spread of typhoon track
forecasts enable to optimize the error
circle’s size, which is
determined based on statistics in the current system.
Initial time: 12UTC 28Aug 2004
Black: best trackBlue: unperturbed run (control run)Red: perturbed run(s)Mark is plotted every 24 hoursTEPS
Probabilistic Track Forecast for CHABA (2)
決定論的進路予報の精度1
回目
の実
験
1回
目の
実験
2回
目の
実験
2004 年 夏実験 2005 年 夏実験
2005 年 WGNE
決定論的進路予報の精度台風アンサンブル予報の開発計画
試験運用での台風アンサンブル予報の仕様台風モデルによる台風予報と比較した、
台風アンサンブル予報の仕様
予報初期時刻と予報時間に関して変更はなく、台風アンサンブル予報では 1 日 4 回 (00, 06, 12, 18UTC) 、 132 時間予報を行う。
予報モデルは高解像度全球モデルの低解像度版 (TL319L60) である。予報対象とする台風の個数は、従来の 2 つから 1 つ増えて 3つとなる。
132 時間
CkTDSV
PRESV
SV_RSMC SV_J3SV_J2SV_J1
PTB_RSMC PTB_J3PTB_J2PTB_J1
初期摂動算出部分のジョブネット図Yhtcファイルを参照して台風の有無をチェックする。 Jフラグが立っていない場合は後続の JOBは実行されない。 (1分 )
特異ベクトル計算用に高解像度モデル解析値から T63L40 の解析値を作成する。 (1 分 )
特異ベクトル計算 (7 分 )
特異ベクトルからアンサンブル初期摂動を作成する。 (1 分 )
JG : Ti
予報部分のジョブネット図
Step1. 特異ベクトル計算用に高解像度モデル解析値から TL319L60 の解析値を作成する。 Step2. Ti で求めた初期摂動を解析値に足す ( 又は解析値から引く ) 。 Step3. 84 時間予報を行う。 Step4. 台風の tracking を行う。 ( 計 18 分 )
後処理。 (1 分 )
JG : Te
Tf00
Tf01m Tf02m Tf03m Tf04m Tf05m
Tf01m Tf02m Tf03m Tf04m Tf05m
Pstn
Case Study: T0607 (MARIA)
Black line: best trackT
YM
and
GS
MC
urre
nt f
orec
ast
TY
M f
orec
asts
GS
M f
orec
asts
In the early stage of MARIA, TYM and GSM failed to express the recurvature.
All track forecasts by TYM and GSM for MARIA.
Initail time is from 2006.08.05.00 to 2006.08.10.18 UTC.
Initial time: 00UTC 5Aug 2006
Typhoon Ensemble Forecast for MARIADeterministic forecast at
2006.08.05.00 UTC.
Black: best track
Red: GSM Green: TYM
Typhoon track forecasts by TEPS at 2006.08.05.00 UTC init..
TEPS capture the probability of MARIA’s recurvature.
Why is RSM good ?Deterministic forecast at
2006.08.05.00 UTC.
Black: best track Red: GSM Green: TYM Blue: RSM
Initial field of wind at 700hPa by RSM (black), GSM (red) and TYM (green).
RSM is good!!
RSM’s southerly wind around east area against TC center position is strengthen, compared with GSM and TYM.
Geographical distribution of 2nd SVInitial field comparison of Wind at 700hPa .
Black is control Red is 02m
TC center position
Wind component of 2nd SV at 700hPa.
(the amplitude of the initial perturbation of u or v is limited to 4 m/s, which almost correspond with the observ
ation error of satellite wind SATOB)
2nd SV is characterized by southerly wind around east area
against TC center position.
Singular Vector Structure
02m member, which initial condition is made using the 2nd singular vector, expresses the MARIA’s recurvature.
The SV is explained by the wind and wv (specific humidity) energy under about 500hPa.
Enegy profile of 2nd SV Enegy spectra of 2nd SV
Track Forecast for MARIAInitial time: 00UTC 5Aug 2006
Typhoon ensemble forecast
TY
M a
nd G
SM
Typhoon EPS captured the probability that MARIA close to Tokai area even in the first forecasting for MARIA.
Cur
rent
for
ecas
t
Case Study: T0408 (CONSON)
Best track of CONSON
Win
d &
Z a
t 250
hPa
Win
d &
Z a
t 500
hPa
Weather map
Initial time: 2004.06.08.1200UTC
SV Calculation for DOTSTAR cases (2)
http://box.mmm.ucar.edu/uswrp/thorpex/symposium_Dec2004/wednesday/Wed_1130_Wu.pdf
DOTSTAR for Typhoon CONSON(T0404)The all points where dropson
des are dropped.
JMA operational data assimilation system obtained these observed data from GTS
OSE result
What the additional observation change?W
ind
& Z
at 2
50hP
aW
ind
& Z
at 5
00hP
a
Weather map
Green(Z): With Sonde
Black(Z): No Sonde
Red(Wind): With Sonde – No SondeEnergy Profile of the analysis increment
TE field of 1st SV at initial tme
TE field of 1st SV at final tme
SV structure for CONSON
Initial field perturbed by SV
SV-Perturbed Run
OSE resultPerturbed Run
Singular vector method at JMAA singular vector method is developed at JMA to make initial conditions in both ensemble prediction systems, the medium-range EPS and the typhoon EPS, which is planned to be newly operational in 2007.
•wq is set to 0 above about 500hPa
•vertical integration is limited under about 150hPa
•wt and wq are decided as follows
Tr = 300K, Pr = 800hPa, Γ=(2/3)Γd
Cp: specific heat of dry air at constant pressure
Lc: latent heat of condensation
Rd: gas constant for dry air
The norm for a SV calculation is based on a total energy norm (Barkmeijer 2001).
Wind component are dominant in a calculated singular vector.
The linearized model and its adjoint version are derived from the global 4D-Var analysis system.
They consist of full dynamics based on Eulerian integrations and full physical processes containing representations of vertical diffusion, gravity wave drag, large-scale condensation, long-wave radiation and deep cumulus convection.
The resolution of the linearized model is T63L40.
wq
Initial norm
1
Final norm
1
I3f3 (Γ= 0) i3f3_t3 (wq = 0) i2f2_t3
Energy Profile by different normN
orm
aliz
ed e
nerg
y
Nor
mal
ized
ene
rgy
Nor
mal
ized
ene
rgy
初期摂動の振幅
台風周辺域の摂動の振幅は、下記の値を超えないように、算出された特異ベクトルを定数倍して求める。
RSMC 領域の乾燥特異ベクトルの振幅は、バリアンスミニマム法を用い、 850hPa 高度の気温の気候学的変動の値を用いて規格化する
Track Forecast
I3f3 (Γ= 0) i3f3_t3 (wq = 0) i2f2_t3
Munehiko YAMAGUCHI1
Tetsuo NAKAZAWA2
1: Numerical Prediction Division, Japan Meteorological Agency
2: Typhoon Research Department, Meteorological Research Institute
Intercomparison of Sensitivity Analysis Guidance for Tropical Cyclones in the Western North Pacific
科研費打ち合わせ@気象研究所 19 Mar. 2007
THORPEX contributes to the developing of an interactive forecast system
observationdata
assimilationforecast user
current system
interactive forecast system
targeting observation
sensitive analysis
observationdata
assimilationforecast user
A sensitive analysis technique is needed to maximize the effect on a numerical prediction and to minimize the cost of the observation.
sensitive area
targeting observation©Vaisala
©JAXA
©NASA
Strategies for targeting observation
ensemble forecasting
singular vector (SV) method
ETKF
A Comparison of Adaptive Observing Guidance for Atlantic Tropical CyclonesS. J. Majumdar, S. D. Aberson, C. H. Bishop, R. Buizza, M. S. Peng, and C. A. ReynoldsMon. Wea. Rev., 134, 2354–2372
Global trend …
The Intercomparison of Adaptive Observing Guidance for the Western Pacific Tropical Cyclones
Sensitivity analysis methods to be compared
Motivation
IWTC-VI recommends cooperative studies on THORPEX-related activities such as T-PARC.
Following this recommendation, the intercomparison of adaptive observing guidance for TCs in the Western Pacific is designed by participants of IWTC-VI.
The results of this intercomparison could be expected to contribute to T-PARC which aims at performing TC-targeted observation.
1. The NCEP ensemble DLM wind variance
2. ETKF based on ensembles from NCEP
3. NOGAPS Singular Vector (SV) method
4. JMA SV method
5. ECMWF SV method
6. MM5 SV method
7. MM5 ADDSV method
Configuration of the experiment
Targeted TC --- All typhoons in 2006
(totally 92 cases in the initial date)
Geographical distributions of a sensitivity region of each method are compared.
Contribution toward T-PARCJMA has resolved to provide the sensitive analysis products by the singular vector method on real-time base. The data will be opened to the THORPEX Community in Asia on an internet homepage which needs user authentication.
top page
click
The first column is the singular vector number.
The second column shows the typhoon track forecasts using the singular vector as the initial perturbation (red line is control run, green is positively perturbed run and blue is negatively perturbed run).
The third column shows the sensitive analysis result which is expected to indicate where to observe.
The fourth and fifth column show the vertical and spectrum structure of the singular vector.
First guess comparison, Ra and GaFirst guess wind field at 700hPa for 2006.08.05.12UTC.
Black results from regional analysis for RSM, and red does from global analysis for GSM.
事例検証: T0413, 2004.08.08.12UTC 初期値
1 回目の実験 2 回目の実験
1 回目の実験 2 回目の実験
従来の予報 台風アンサンブル予報
T0416 Typhoon CHABA Best track of CHABA Accumulated Precipitation
Maximum momentary wind speed
Setouchi area suffered enourmous damage by tidal wave. At Takamatus Port and Uno Port, the sea level attained to the highest level in their history.
Let’s see the ensemble track forecasts at CHABA’s early stage and the pre-recurvature stage
early satage
pre-recurvature stage
Track Forecast for CHABA in its early stageC
urre
nt f
orec
ast
Fut
ure
plan
Black: best trackBlue: unperturbed run (control run)Red: perturbed run(s)Mark is plotted every 24 hoursTEPS
original
tracks
The ensemble track forecasts grasp the observed CHANBA’s track.
Initial time: 12UTC 24Aug 2004
Current forecast
original
tracks
The spread of typhoon track
forecasts enable to optimize the error
circle’s size, which is
determined based on statistics in the current system.
Initial time: 12UTC 28Aug 2004
Track Forecast for CHABA in its pre-recurvature stage
Black: best trackBlue: unperturbed run (control run)Red: perturbed run(s)Mark is plotted every 24 hoursTEPS
T0607 Typhoon MARIA Best track of MARIA
TY
M f
orec
asts
GS
M f
orec
asts
In the early stage of MARIA, TYM and GSM failed to express the recurvature.
The track forecast for MARIA starts from 00UTC on 5th August. Track forecast information available at that time is TYM and GSM only. One-Week EPS is yet to be executed.
2006.08.05.00UTC
recurvature point
Track Forecast for MARIAInitial time: 00UTC 5Aug 2006
Typhoon ensemble forecast
TY
M a
nd G
SM
Typhoon EPS captured the probability that MARIA close to Tokai area even in the first forecasting for MARIA.
Cur
rent
for
ecas
t
Typhoon ensemble forecast Medium-range ensemble
Typhoon ensemble forecast Medium-range ensemble
Initial time Control run Perturbed run
06.00UTC
3 days before
07.00UTC
2 days before
07.00UTC
1.5 day before
Radar Amedas
09.00UTC
Accumulated rain from the 6 hours before
Observed Precipitation (mm/hr) at Irozaki
Control run Perturbed run (03p) Perturbed run (04m)
Init
ial t
ime:
07.
00U
TC
RR
24 a
t T+
48
T+24 T+48
Rain Forecast by TEPSObserved Precipitation (mm/24hrs)
Observed Precipitation (mm/hr) at Irozaki
Control run Perturbed run (03p)
Init
ial t
ime:
070
0UT
C
08.0800UTC
Rain Forecast by TEPSObserved Precipitation (mm)from 0800UTC to 1000UTC
08.1000UTC
High resolution GSM (20km)
Initial time 0612UTC
Observed Precipitation (mm/hr) at Irozaki
Control run Perturbed run (03p)
08.0715UTC
Rain Forecast by TEPSObserved Precipitation (mm)from 0715UTC to 1000UTC
08.1000UTC
High resolution GSM (20km)
Initial time 0612UTC Initial time 0700UTC
2006.08.0708UTC
i3f3 i3f3_t3 i2f2_t3
Contents
1. JMA singular vector method
2. OSE for 2004 DOTSTAR cases
3. OSE for 2004 DOTSTAR cases using JMA SV method
addition. Typhoon track forecasts for 2004 DOTSTAR cases using JMA SV method
1. JMA singular vector method
JMA currently operates a medium-range ensemble prediction system using a breeding method for generating initial perturbations. JMA plans to introduce a new EPS for typhoon track forecasting in 2007. That EPS will be operated four times a day when typhoons exist in the northwestern Pacific. For this purpose, a singular-vector (SV) method seems more appropriate than a breeding method, because the former method allows us to generate initial perturbations when necessary and to generate perturbations that is optimally chosen for typhoon track forecasts.
Motivation to develop a SV method
A SV method also enables us to calculate sensitive regions in initial conditions for targeting observations.
•The model for calculating a forecast trajectory is a low-resolution version of JMA-GSM (T63L40).
•Its tangent linear and adjoint models including dry or moist physical processes is prepared by JMA-GSM4DVAR and integrated with a horizontal-vertical resolution of T63L40.
•A norm to measure perturbation growth is based on the total energy (TE).
JMA Singular Vector Method
Total energy norm
Ealuation of JMA moist SVs
Similarity Index(left) and growth rate (right) between non-linear growing moist SVs and evolved moist SVs. Initial norm is TE w/o q term and final norm is TE below 17th model level w/o q term. Target area is a Tropical area(S20-N20). Evaluation time is 24 hours. Initial time is 4th August 2003.
02468
101214161820
1 2 3 4 5 6 7 8 9 10Number of SVs
Gro
wth
Rat
e
linear growth
non lineargrowth
Evaluation of JMA moist SVs
Non-linear growing moist SVs and evolved moist SVs have a high similarity!!
“moist SVs” means SVs which are calculated by a tangent linear and its adjoint models with moist physical processes.
Targeting Technique of a SV methodTarget area
9th
SV
1st
SV
Initial time is 4th August 2003.Evaluation time is 24 hours.
Initial norm TE
Final norm TEdec17w/oq
high similality!!
If the target area is decided around this 9th SV…..
Computer Resources for a SV calculation
HITACHI SR11000(JMA updated its super computer at 2006.03)
Iteration times: 50
evaluation time : 24 hours
SR11000 8 nodes
dry SVs moist SVs
10 minutes 13 minutes
2. OSE for 2004 DOTSTAR cases
SV Calculation for DOTSTAR cases (1) DOTSTAR: Dropsonde Observation for Typhoon Surveillance near the Taiwan Region. http://typhoon.as.ntu.edu.tw/DOTSTAR/English/home2_chinese.htm
http://box.mmm.ucar.edu/uswrp/thorpex/symposium_Dec2004/wednesday/Wed_1130_Wu.pdf
What is DOTSTAR ?
SV Calculation for DOTSTAR cases (2)
http://box.mmm.ucar.edu/uswrp/thorpex/symposium_Dec2004/wednesday/Wed_1130_Wu.pdf
OSE for Typhoon CONSON(T0404)Best track of CONSON The all points where dropson
des are dropped.
JMA operational data assimilation system obtained these observed data from GTS
SV Calculation for DOTSTAR cases (3)
Initial time is 2004.06.08.12UTC.
If observed data hadn’t been used in JMA operational data assimilation… (1)
We re-tried a data assimilation with no dropsonde data and compared the initial field (NDS) with a operational one (RTN).
Vertically accumulated total energy field of the difference between
NDS and RTN initial field
Center position of CONSON
The difference between NDS and RTN initial
field is relatively larger at northeastward area against center position
of CONSON
SV Calculation for DOTSTAR cases (3)
Initial time is 2004.06.08.12UTC.
If observed data hadn’t been used in JMA operational data assimilation… (2)
We re-tried a data assimilation with no dropsonde data and compared the initial field (NDS) with a operational one (RTN).
Vertical distribution of total energy field of the difference between
NDS and RTN initial field The difference between NDS and RTN initial
field is mainly explained by the effect of Rotation
(green) above 500hPa and specific humidity (blue) below 500hPa.
500hPa250hPa
850hPa
SV Calculation for DOTSTAR cases (3)
Initial time is 2004.06.08.12UTC.
If observed data hadn’t been used in JMA operational data assimilation… (3)
We re-tried a data assimilation with no dropsonde data and compared the initial field (NDS) with a operational one (RTN).
Analysis Increment at 500hPa wind field
Analysis increment of southerly wind at northeastward area against center position of CONSON
Typhoon track forecasts by RTN and NDS
Initial time is 2004.06.08.12UTC.
NDS cannot perform CONSON’s northeast
ward movement.
RTN track forecast has slow bias but the direction is quite similar with best track!!
3. OSE for 2004 DOTSTAR cases using JMA SV method
SV Calculation for DOTSTAR cases (6)
Initial time is 2004.06.08.12UTC.
Target Area : 20N-25N, 120E-130E.
Norm : dry total energy norm(the effect of upper model levels is neglected at evaluation time)
optimization time : 24 hour
SVs : moist SVs
SV Calculation for Typhoon CONSON
Vertically accumulated total energy field of forecast error by JMA medium-range
EPS CTL forecaste(FT+24h).
Target area is decided to include the area where forecast error is large.
Calculated 1st SVInitial time is 2004.06.08.12UTC.
Vertically accumulated total energy field of 1st SV
Center position of CONSON
Vertical and spectra distribution of total energy norm of 1st SV
500hPa
1st SV is calculated around northeastward area against TC center
position and has a peak of perturbation around 500hPa explained by Rotation
and Temperature.
Green line is by the effect of Rotation and yellow line is by effect of Temperature
OSE using JMA SV method (1)
Vertically accumulated total energy field of 1st SV
The all points where dropsondes are dr
opped.
If dropsond data only where 1st SV is calculated are used in data assimilation, how the typhoon track forecast changes…
Initial time is 2004.06.08.12UTC.
OSE using JMA SV method (2)Initial time is 2004.06.08.12UTC.
We performed 4 experiments.1. All dropsonde data are used (RTN; already done)2. All dorpsonde data are not used (NDS; already done)3. Dropsonde data only where 1st SV is calculated are used (SDS)4. Dropsonde data only where 1st SV is not calculated are used (NSA)
SDS: 3, 4, 5, 7, 9, 11, 12, 14 are used
NSA: 6, 8, 10, 13, 15, 16 are used
Typhoon track forecasts by 4 EXPsInitial time is 2004.06.08.12UTC.
NSA cannot perform CONSON’s northeastward movement. (almost same with NDS)
SDS track forecast has slow bias but the direction is quite similar with best track!! (almost same with R
TN)
Typhoon track forecasts by 4th EXPInitial time is 2004.06.08.12UTC.
forecast filed of each experiments (PSEA, FT+90h)
SDS RTN NSA
Typhoon exists
NDS is almost same with NSA
Suggestion
TRMM image for T0404(CONSON)
2004.06.08.16 UTCThe result of this experiments suggests that JMA SV method may analyze a sensitive region in initial condition and be useful to find a optimized observation area.
Sensitive region in initial condition
addition. Typhoon track forecasts for 2004 DOTSTAR cases using JMA SV method
Left Figure: TC track forecasts for typhoon Conson by JMA medium-range EPS CTL (green) and ensemble member (red) using 1st SV.
Typhoon track forecast using JMA SVs
Right Figure: TC tracks by 3 ensemble members which are made by each component of the 1st SV, Temperarute, Wind and Specific Humidity .
Win
d co
mpo
nent
con
trib
ute
the
mos
t to
the
impr
ovem
ent
of T
C tr
ack
fore
cast
.
TC
trac
k fo
reca
st is
impr
oved
by
usi
ng a
SV
.