multi-perturbation methods for ensemble prediction of the mjo multi-perturbation methods for...
TRANSCRIPT
Multi-Perturbation Methods forEnsemble Prediction of the MJO
Seoul National University
A paper appearing in Climate Dynam-ics (2013)
In-Sik Kang and Pyong-Hwa Jang
Post Processing
Predictability Research
Prediction
Initialization
Atmosphere LandOcean
CGCM
OceanModel
AtmosphericModel
Intraseasonal Prediction System
2
3
Boreal Winter Boreal Summer
ABOM EC GFDL NCEP CMCC ECMWF JMA
Ensemble number 10 10 10 5 5 15 6
Ref. CLIVAR/ISVHEIntraseasonal Variability Hindcast Experiment
Corr
ela
tion
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
Corr
ela
tion
5 10 15 20 25 30 35
Leadtime (DAY)
RMM index
4
SNU CGCM
MJO simulation
Initialization/Perturbation methods
MJO Ensemble Prediction Results
CONTENTS
5
Model Description References
AGCM SNU AGCM
T42, 21 levels (2.8125X2.8125)SAS cumulus convection2-stream k-distribution radiationBonan (1996) land sur-face
Kim (1999)Kang et al. (2002)Kang et al. (2004)Kim et al. (2003) Lee et al. (2003)
OGCM SNU OGCM
MOM2.2 + Mixed Layer Model1/3o lat. x 1o lon. over tropics(10S-10N), Vertical 32 levels
Noh and Kim (1999)Noh et al. (2003a) Noh et al (2003b)Kim et al. (2004) Noh et al. (2004) Noh et al. (2005)
CGCM SNU CGCM SNU AGCM + MOM2.2
-1-day interval exchange- Ocean : SST- Atmosphere : Heat, Salt, Mo-mentum Flux- No Flux Correction is applied
M M
H H
K S ql
K S ql
Vertical Eddy Viscosity:
Vertical Eddy Diffusivity:
,M HS S: empirical Constantwhere
2 / 2q : TKE
l : the length scale of turbulence
Mixed Layer Model Coupling Strategy
SNU CGCM
6
CNTL CGCM
Bias
Annual mean SST
• Convective momentum trans-port
• Diurnal coupling• Tokioka constraint
(alpha=0.1)• Auto conversion time scale
(3200s)
Improved CGCM
Bias
Annual mean SST
Improvement, Ham et al. (2012) Climate Dyna-mics
7
Performance of SNU CGCM ver.2
Power Spectrum (averaged 15S-15N),summer (May-Oct)
OLR, OBS OLR, SNUCGCM
U850, OBS U850, SNUCGCM
9
SNU CGCM V.2
SNU AGCM + MOM2.2T42, 20 levels resolution
Coupled GCM
GODAS- SST & salinity Nudging
(relaxation time: 5 day)
ERA interim- U, V, T, Q, Ps Nudging
(relaxation time: 6hr)
Ocean
Atmosphere
Initialization
+
Data for Nudg-ing
Perturbation method
Lagged average fore-cast
(LAF)
Breeding Method
Empirical Singular Vector
Intra-seasonal Prediction System
10
ATM: ERA interim (T,U,V,Q,Ps)
OCN: GODAS (temperature, salinity)
Nudging
Initialization Prediction
LAF
Breeding
4 ensemble members
(6hr lag)
breeding interval : 1 day, 3 days, 5 days
BD (+) perturbation
BD (-) perturbation
3(breeding intervals) × 2(mirror im-ages)
= 6 ensemble members
Rescaling factor (percentage) : 10%
L()= : VP200, U200, U850: VP200, U200, U850
ESV
ESV (+) perturbation
ESV (-) perturbation
2 ensemble members
Performance of SNU CGCM
11
EV of the combined EOF of 20-100 day filteredsummer (May-Oct)
OLRU200U850
STD of VP200 perturbation
(b) Bred Perturbation 3%
(a) Bred Perturbation 0.3% (c) Bred Perturbation 10%
- Characteristics of Bred perturbations
Initial Perturbation - Breeding
13
14
Bred Vector 1day Bred Vector 5day
Unit : ×105 (m²/s)
Bred Vector 3day1st mode
2nd mode
1st mode
2nd mode
1st mode
2nd mode
Initial Perturbation - Breeding
- Characteristics of Bred perturbations
EOF of perturbations
15
Bred Vector mode1
Bred Vector mode2
• Breeding rescaling factor : 10%
• Breeding interval : 5 day
- EOF modes of VP200 perturbation- Winter
(X 1e+6)
Characteristics of Bred perturbation To be up-date
16
Final VP200 anomaly is on the east of initial VP200 anomaly
Eastward propagating mode
Observa-tion
Model
Initial Perturbation - ESV
- Singular mode
17
Initial Perturbation - ESV
- Singular mode of ESV
0
1
2
3
Singular value
1st mode
2nd mode
3rd mode
4th mode
5th mode
5lag 10lag 15lag 20lag
Total 360 summer cases Include all MJO phase
Summer : 12 < LAF(4) + BD1day(2) + BD3day(2) + BD5-
day(2) + ESV(2)> cases × 18 cases/year × 20 years
= 4320 predictions
19
- Outline
1 May11 May
21 Oct
45 Days Integration
whole
SU
MM
ER
Ensemble Prediction
- Correlation skill of Real-time Multivariate MJO (RMM) Index
20
LAF [4]
BD1day+ BD3day [4]BD1day [2]
BD1day+ BD3day+BD5day [6]BD5day [2]BD3day [2]
Ensemble Prediction* The parenthesis refers ensemble members
Corr
ela
tion
Coeffi
cie
nt
Lead Days
- Correlation skill of Real-time Multivariate MJO (RMM) Index
21
Ensemble Prediction* The parenthesis refers ensemble members
ESV [2]LAF [4]
22
- Correlation skill of Real-time Multivariate MJO Index
LAF (4)
ESV (2) ALL (12)
BD1day+BD3day+ BD5day(6)
: 4 ensemble mem-bers with 6 hours lag intervals
Multi-Perturbation Ensemble* The parenthesis refers ensemble members
23
- Correlation skill of RMM Index for each MJO phase
Multi-Perturbation Ensemble* The parenthesis refers ensemble members
* The parenthesis refers ensemble members
(a) 5 day
(b) 10 day
(c) 15 day
24
- Correlation Skill of U850 for lead times
MP (12)= LAF (4) + BRED (6) + ESV (2)
Multi-Perturbation Ensemble
25
Summary
1. Various perturbation methods (LAF, Breed-ing, ESV) produce different unstable modes and different perturbations.
2. MJO predictability is not sensitive to the perturbation methods.
3. Multi-perturbation ensemble prediction sys-tem is slightly better than that of any pertur-bation method.
27
Multi-Perturbation Ensemble
- The ensemble spread of 200-hPa zonal wind averaged 15°S-15°N
LAFBDESV
28
Initial Perturbation
- Variance of initial perturbation of VP200 Unit : ×105 (m²/s)
(a) LAF (b) BD
(c) ESV
29
- Correlation skill of RMM Index for each MJO amplitude
Multi-Perturbation Ensemble* The parenthesis refers ensemble members
MP predic-tions
LAF predic-tions