Global Ensemble Prediction System at KMA

1
Global Ensemble Prediction System at KMA
The 3rd Ensemble user workshop

• Sunok Moon
• NWPD / KMA
• 1. Nov. 2006

Numerical Weather Prediction Division, Korea
Meteorological Administration
2
Outline

• Operational global ensemble model

• History of the EPS

• Current issues ? New experiments

Factor Rotation
Stochastic Perturbation

• EPS Products

Probability of Precipitation
Probability of Maximum Wind
Ensemble Plumes
Ensemble Mean and Spread
Spaghetti diagram

• Future plans of EPS

3
Operational Global Ensemble Model
4
History of the EPS at KMA
GBEPS- Global Bred-vector Ensemble Prediction
System
5
Improvement Increasing resolution
RMSE T106L30 vs T213L40
T106L30
T213L40
6
Improvement Increasing member
RMSE Spread 16 vs 32
7
Improvement Increasing member
Brier Skill Score 16 vs 32
8
Current Issues of the EPS at KMA

• Operational EPS at KMA usually does not
• produce spread enough.

The bred vectors seem to have similarity. ?
Introduced factor rotation
The growth rate of perturbation is small. ?
tested stochastic perturbation
9
Factor Rotation
Theoretical base for factor rotation Factor
analysis ? a statistical technique
What is the Factor analysis? Use
relationship among variables ? factors
How to produce rotated BV with factors?
Rotate the factors obtained from factor analysis
with keeping quasi orthogonal ? New
perturbations ( RBV )
? introduced in operational system, last year
10
Factor Rotation
Breeding
D day
D day 12hr
D1 day
D1 day 12hr
11
Improvement Factor rotation
RMSE Spread OBV vs RBV
12
Positive Impact of factor rotation - probability
of precipitation
ini.
2004 July 13 12UTC
PoP July 16 12UTC After factor
rotation scheme, probability of
precipitation successfully reproduces the
heavy rainfall as observed.
Observation daily rainfall
Operation
Factor rotation
13
Stochastic Perturbation
Estimate error of the model physics from
background error covariance used in 3dVar.
? random (stochastic) forcing ltlt model error
Apply the stochastic perturbation to tendency
every 1 hour.
It is computationally expensive for this
method. ? Stochastic
perturbation should be optimized.
14
Improvement Stochastic perturbation
RMSE Spread Oper vs Stoc
RMSE oper
RMSE stoc
Spread oper
Spread stoc
15
Improvement Stochastic perturbation
RMSE Spread Oper vs Stoc
RMSE oper
RMSE stoc
Spread stoc
Spread oper
16
(No Transcript)
17
Improvements of the EPS this year
Increasing resolution and ensemble size
T106L30M16 ? T213L40M32 once a day (12Z)
? twice a day (00Z, 12Z)
18
EPS products
Probability of Precipitation Probability
of Wind Gust Ensemble Plumes Ensemble
mean and spread Spaghetti diagram Stamp
map
http//web.kma.go.kr/eng/wis/gws_14.jsp
19
KMA EPS Homepage
20
Future Plans
21
Thanks for your attention !
Numerical Weather Prediction Division, Korea
Meteorological Administration
22
Improvements of the EPS this year
Increasing resolution and ensemble size
T106L30M16 ? T213L40M32 once a day (12Z)
? twice a day (00Z, 12Z)
23
Preparing for TIGGE
TIGGE Project

Preparing GRIB 2 Format under development
Global Ensemble Data Exchange
?TIGGE THORPEX Interactive Grand Global Ensemble
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
Experiment Results time series at Seoul for
500Z and 850 T

ini. 2004 Aug. 15 12UTC
32
?? ??
33
Time series of RMSE - monthly average
34
Regional Rescaling

• difference of vorticity between unperturbed
  and perturbed field
• norm global enstrophy on wave space at time t (
  24 hours )
• zeronorm e Initial Norm
• Rescaling Factor

( Filtering with latitude )