KMA
Numerical weather forecast and Digital forecast at KMA
Young-Youn ParkNumerical Prediction Center/
Korea Meteorological Administration, currently visiting ECMWF
With contributions from
SangWon Joo, SunOk Mun, Young-Kyoung Seo
11th MOS workshop 12-16 Nov 2007
KMA
Outline
█ Overview of KMA’s NWP Systems █ Medium-range forecast
Recent Changes in KMA’s EPSTIGGE : Multi-model test
█ Digital Forecast System
11th MOS workshop 12-16 Nov 2007
KMA
Temp, POP(statistical
model)
GWAM(wave model,
1.25º, 10days)
RTSM(1/12º,48hrs)
GDLM( statistical
model,10days)
Data Acquisition & Q.C
EPS
(T213L40, 16members,10days,00/12)
KWAM(1/12º, 66hrs)
KLAPS(15/5km, 18hrs)
(Downscale,5km/33L, 1day)
RDAPS
RWAM(0.25º,66hrs)
On-Going Project
Storm-Scale Model
1~5Km grid modelRadar assimilationIntensify model physics
GDAPS(3dVar, FGAT
T426/L40, 10days)
RDAPS(Downscale,
FDDA, 30km/33L,66hrs)
RDAPS(3dVar,1day10km/33L)
Typhoon Model(30km, 72hrs)
Digital ForecastMIN/MAX T
PoP3Hr TRH
Wave HeightWind S/DPrep. Typ.Sky Cond.Rain/Snow
KWRF(3dVar
10km/30L,60hrs)
NWP systemTIGGE
Digital ForecastTMNTMXPoPPTYSKY
11th MOS workshop 12-16 Nov 2007
KMA GBEPS 1.1.1~ GBEPS 1.2.1
GBEPS 2.1.1~ GBEPS 2.3.1 GBEPS 3.3.1
Operation period 2001.3.1 ~ 2003.11.1 ~ From 2006.7.Data assimilation 2dOI → 3dOI 3dOI → 3dVar 3dVar
Model GDAPS T106L21 GDAPS T106L30 GDAPS T213L40Vertical
resolution 21 levels 30 levels 40 levels
Perturbation method Breeding Breeding +
Factor RotationBreeding +
Factor Rotation
Target area (BV) Global Northern Hemisphere
Northern Hemisphere
Run per days 1 (12UTC) 1 (12UTC) 2 (00, 12UTC)Lead time 10 days 8 days 10 daysEnsemble members 16 (16 members + 1 control) (16+1)*2
members
EPS (Ensemble Prediction System)
11th MOS workshop 12-16 Nov 2007
KMA
Recent Changes█ Increasing resolution and ensemble size
T106L30M16 → T213L40M32 once a day (12Z) → twice a day (00Z, 12Z)
█ Statistical Post-Processing : Bias Correction Decaying averaging bias estimate ( from Bo cui )
█ EPS does not produce enough spreadBred vectors seem to have similarity
Introduce factor rotationIntroduce factor rotationGrowth rate of perturbation is small
Test stochastic perturbationTest stochastic perturbation
11th MOS workshop 12-16 Nov 2007
KMA
Factor Rotation
█ What is the Factor analysis?Use relationship among variables Find the structure of the relationship and Tie to similar variables Factors
█ How to produce perturbation with factors?RRotateotate the factors obtained from factor analysis With keeping orthogonal feature among factors
⇒⇒ New perturbationsNew perturbations
11th MOS workshop 12-16 Nov 2007
KMA
Breeding+Factor Rotation
AnalysisAnalysisDD
AnalysisD+PerturbationPert. run
Control run
normalization
AnalysisAnalysisD+1D+1
Breeding
D day D day + 12hr D+1 day D+1 day +12hr
AnalysisAnalysisDD
AnalysisD+Perturbation
Pert. ru
n
Control run Rotation
normalization
AnalysisAnalysisD+1D+1 Rotation
Breeding + Rotation
11th MOS workshop 12-16 Nov 2007
KMA
Bias Corr. & Stochastic Pert.Bias Correction█ Motivation: strong +bias over east Asia █ Method: Decaying averaging bias estimation (Bo Cui)
Stochastic Perturbation█ Estimate error of model physics from background
error covariance used in 3dVar.random (stochastic) forcing << model error
█ Apply stochastic perturbation to tendency every hour
11th MOS workshop 12-16 Nov 2007
KMA
Impact of Factor rotation
RMSE OBVRMSE OBVRMSE RBVRMSE RBV
Spread RBVSpread RBVSpread OBVSpread OBV
IC: 2004 July 13 12UTC
PoP : July 16 12UTC (+3d)After factor rotation,probability of precipitation successfully reproduces the heavy rainfall as observed.
Operation(+3d) Factor rotation(+3d)
OBS : 24h rain
11th MOS workshop 12-16 Nov 2007
KMA
Impact of Resolution & Membership
T106L30T106L30
T213L40T213L40
RMSERMSE M16M16RMSE M32RMSE M32
Spread M32Spread M32SpreadSpread M16M16
11th MOS workshop 12-16 Nov 2007
KMA
Impact of Bias Correction
RMSE of Z500 SUMMER 2006
10
20
30
40
50
60
70
1 2 3 4 5 6 7
Lead time (Day)
RM
SE (m
)
before(NH)after(NH)before(EA)after(EA)
RMSE of Z500 WINTER 2006
10
20
30
40
50
60
70
80
90
1 2 3 4 5 6 7
Lead Time (Day)
RM
SE (m
)
편자보정 전(NH)
편자보정 후(NH)
편자보정 전(EA)
편자보정 후(EA)
11th MOS workshop 12-16 Nov 2007
KMA
TIGGE: Multi-model testMany Questions/Choices█ Which analysis to use█ How to combine ?
Use all the centers/members or only good centers/members?Simple or weighted combine How to determine the weight ?Using calibration? How to calibrate?
Multi-model ensemble test█ Combine test period: 8 Jun – 1 Sep (84 cases)█ Centres: ECMWF, UKMO, JMA, CMA█ Method:
Simply combine all the members With/without bias-correction (use previous 30 days)Calculate bias for each centres, cf/pf separately
11th MOS workshop 12-16 Nov 2007
KMA
15JUL
2007
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 1 2 3 4 5 6 7 8 9 10 11 12 13 14AUG
2007
5760
5764
5768
5772
5776
5780
()
ecmwf ukmo ncep jma bmrc cmaz500hPa mean of an area n.hem int 12h
Which an ? (Area mean of an)
15JUL
2007
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 1 2 3 4 5 6 7 8 9 10 11 12 13 14AUG
2007
289.2
289.4
289.6
289.8
290
290.2
290.4
290.6
290.8
()
ecmwf ukmo ncep jma bmrc cmat850hPa mean of an area tropics int 12h
15JUL
2007
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 1 2 3 4 5 6 7 8 9 10 11 12 13 14AUG
2007
5444
5448
5452
5456
5460
5464
5468(
)
ecmwf ukmo ncep jma bmrc cmaz500hPa mean of an area s.hem int 12h
11th MOS workshop 12-16 Nov 2007
KMA
Raw/Cal data ? (RMSE/RPSS : Z500 NH)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
10
20
30
40
50
60
RM
S (m
)
ecmwf ukmo cma jma eu eu_bccases 20070608-20070901_N84, area n.hemz at 500hPa 12 UTC (ecmwf_as_an) RMSE of Ens. Mean
EC/UK(BC)
EC/UK
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Ran
ked
Prob
abilit
y Sk
ill Sc
ore
ecmwf ukmo cma jma eu eu_bc10 categories, cases 20070608-20070901_N84, area n.hemz at 500hPa (ecmwf_as_an)
EC/UK(BC)
EC/UK
11th MOS workshop 12-16 Nov 2007
KMA
Raw/Cal data ? (RMSE/RPSS : Z500 TR)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
2
4
6
8
10
12
14
16
RM
S (m
)
ecmwf ukmo cma jma eu eu_bccases 20070608-20070901_N84, area tropicsz at 500hPa 12 UTC (ecmwf_as_an) RMSE of Ens. Mean
EC/UK(BC)
EC/UK
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Ran
ked
Prob
abilit
y Sk
ill Sc
ore
ecmwf ukmo cma jma eu eu_bc10 categories, cases 20070608-20070901_N84, area tropicsz at 500hPa (ecmwf_as_an)
EC/UK(BC)
EC/UK
11th MOS workshop 12-16 Nov 2007
KMA
Which centers ? (RPSS – Z500 NH/TR)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Ran
ked
Prob
abilit
y Sk
ill Sc
ore
ecmwf eujc eujc_b10 categories, caz at 500hPa (ecm
c euj euj_bc eu eu_bc e_bcses 20070608-20070901_N84, area n.hemwf_as_an)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Ran
ked
Prob
abilit
y Sk
ill Sc
ore
ecmwf eujc eujc_bc euj euj_bc eu eu_bc e_bc10 categories, cases 20070608-20070901_N84, area tropicsz at 500hPa (ecmwf_as_an)
EC/UK/JMA/CMA(BC)
EC(BC)
EC/UK(BC)
EC/UK/JMA(BC)
11th MOS workshop 12-16 Nov 2007
KMA
Sensitivity to parameters ?
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Ran
ked
Prob
abilit
y Sk
ill Sc
ore
ecmwf eujc eujc_bc euj euj_bc eu eu_bc e_bc10 categories, cases 20070608-20070901_N84, area n.hemz at 500hPa (ecmwf_as_an)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Ran
ked
Prob
abilit
y Sk
ill Sc
ore
ecmwf eujc eujc_bc euj euj_bc eu eu_bc e_bc10 categories, cases 20070608-20070901_N84, area n.hemt at 850hPa (ecmwf_as_an)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Ran
ked
Prob
abilit
y Sk
ill Sc
ore
ecmwf eujc eujc_bc euj euj_bc eu eu_bc e_bc10 categories, cases 20070608-20070901_N84, area tropicsz at 500hPa (ecmwf_as_an)
EC/UK/JMA/CMA(BC)
EC(BC)
EC/UK(BC)
EC/UK/JMA(BC)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15fc-step (d)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1R
anke
d Pr
obab
ility
Skill
Scor
e
ecmwf eujc eujc_bc euj euj_bc eu eu_bc e_bc10 categories, cases 20070608-20070901_N84, area tropicst at 850hPa (ecmwf_as_an)
EC/UK/JMA/CMA(BC)EC(BC)
EC/UK(BC)
EC/UK/JMA(BC)
11th MOS workshop 12-16 Nov 2007
KMA
9JUN
12151821242730 3JUL
6 9 12151821242730 2AUG
5 8 11141720232629 1
1.2
1.35
1.5
1.65
1.8
1.95
2.1
2.25
2.4
2.55
rmse
_em
(K)
ecmwf ukmo eu eu_bct850hPa 12UTC step: 72h rmse_em area: n.hem
Daily RMSE – t850 +3d/+9d NH
9JUN
12151821242730 3JUL
6 9 12151821242730 2AUG
5 8 11141720232629 1
2.4
2.6
2.8
3
3.2
3.4
3.6rm
se_e
m(K
)
ecmwf ukmo eu eu_bct850hPa 12UTC step: 216h rmse_em area: n.hem
11th MOS workshop 12-16 Nov 2007
KMA
Summary of Preliminary Result █ More gains on probabilistic scores than deterministic scores █ AS, NA, NH: small gains, distinctive gains over tropics (where
ecmwf shows under-dispersion)█ Difference btn parameters and areas█ Bc better in early period, but nbc better in the later period of
forecast█ Larger benefit of multi-model over strong bias area█ Decision should be made on the bases of areas, parameters
of interest█ Lower level parameter (T2m, precipitation) ?█ Advanced calibration ? █ Optimum weight for each centers?
11th MOS workshop 12-16 Nov 2007
KMA
Digital Forecast : Motivation
Quantitative
Detailed
Applicable
VariousContents
█ Strong demand for detailed, quantitative meteorological services
█ Public does not want weather at station, but at their house
█ The use of meteorological data is too difficultPrivate sector requires easy to handle data to reproduce new application information
Nwp output is used by only very limited area
one of important barrier to develop application information
11th MOS workshop 12-16 Nov 2007
KMA
Change
Current Forecast
Goal of Digital ForecastGoal of Digital Forecast
When, Where, How Much, Computer Friendly form
QuantitativeDetailed
Applicable
VariousContents
Digital Forecast
11th MOS workshop 12-16 Nov 2007
KMA
Conceptual Structure
NWP
GDAPS/RDAPS
GWAM/ReWAM
Observation
Remote Sens.
Synoptic Obs.
AWS
Interpretationof NWP
MOS
PPM
Obs analysis
5km grid obs
Forecaster’s Editing (GEM)
Temporal Editing
Spatial Editing
Digital Digital ForecastForecast
DBDB
Verification/Intranet/InternetService
Spatial analysis
Spatial analysis
Graphical Editing
Sat: Sky Cover
Radar QPE
WEM
11th MOS workshop 12-16 Nov 2007
KMA
Status of Digital ForecastDigital Forecast range Short range Medium range
NWP Model RDAPS GDAPS
Forecast length/interval 48hours/3hours 10.5day/12hours
Resolution 5km 30km
Frequency (per day) 8 2
Forecast parameter 12 5
Grid/element transform module 11 5
Statistical analysis module (MOS) MOS 8(/9) developing
Spatial analysis module (2d-OI) apply, improvement not apply
Graphical editing module (GEM) semi-operation, improvement
Web service module (WEM) semi-operation, improvement
Verification system operation, improvement
11th MOS workshop 12-16 Nov 2007
KMA
Interpretation of Model Output : shortDFS
• Combine NWP Model and statistical methods• Generate 5 km-grid forecast, Predict every 3 hour up to 48 hours
MOS modelsat 103 stations
MOS
PPM
T, WIND, HUMIDITY PRESSURE, RAIN,
RDAPS:Forecast 66hourResolution : 30km/3hour
ReWAMForecast 66hourResolution : 0.25deg/3hourly
Sea surface wind, Wave height,
3-hour/Max/Min Temp., Probability of Precip., RH,
Wind, single-station eq.
Sky cover, Precip. TypeSnow-depth
Generalized-operator eq.
Precip.
Effective wave height
11th MOS workshop 12-16 Nov 2007
KMA
█ Need tool that mange and produce MOS eq. efficiently according to the changes of NWP models
█ Using this tool, make the indicator for MOS modelo Speed up for developing and improving MOS eq.o Standardization of MOS modelo Total system of development/verification/management of MOS
efficiency, accuracy, convenience
MOS Tool
11th MOS workshop 12-16 Nov 2007
KMA
1. Sampling Module
2. Statistics Module
3. Display/GUI Module
MOS Tool
11th MOS workshop 12-16 Nov 2007
KMA
Forecaster’s Editing (GEM)Weather Station
Regional Met. Office
KMA Headquarter
Time Series Editing Spatial Editing Issue Forecast
Objective Analysis
11th MOS workshop 12-16 Nov 2007
KMA
Lua Editor
Lua Script filesSmart Tool manager
Lua Wrapper
SmartToolInterface Library
GEM3.0
Lua
GEM Editing Window Apply Smart Tool
GEM
11th MOS workshop 12-16 Nov 2007
KMA
Spatial analysis module (2d-OI)
Observation Grid point
First Guess (Y0f)
ffj
oj
m
j
j YXXhY 01
0 )( +−=∑=
guessFirstincrementweight +×=∑ )(
X
X
X
X
increment
0YfjX
ojX
Weight determine the reliability of analysis (Y0)
2 dimensional-Optimal Interpolation
( X ) ( Y0 )
determine weight that minimize (Y0-X0 )
11th MOS workshop 12-16 Nov 2007
KMA
h1h2...hm
σ01σ02...σ0m
σ11 , σ12 , .... , σ1mσ21 , σ22 , .... , .. . .. . .. . . σm1 , ......…. ,σmm
-1
=
Weight co-variance of obs to obs co-variance of obs to grid (fixed) (alter)
Weight (h) : 2dOI P
igjjiij
P
ijj
H μλλμμ =+∑=
)( 07
1 b
oσσ
λ =
i,j : station g: digital grid μij
P : BE covariance μijo: OE covariance
μigP : BE covariance between grid & station
assume OE covariance ~ 0
(Forecaster editing ~ true)
P
igj
P
ij
N
jH μμ =∑
=)(
1
Weight (h): DFS 2dOI
Weight in 2d-OI
11th MOS workshop 12-16 Nov 2007
KMA
21 braP
ij +=μ
r : distance a : error ratio (fixed, 1 )
b : error cov wrt r (fixed, 1/1600 )
Lorentz’s type function fitting
0.0
0.2
0.4
0.6
0.8
1.0
0 50 100 150 200 250 300
a
R(km)
Rd=100Rd=50
Rd=112 Rd=200
a=0.9a=0.8
a=0.5
a=1.0
Rd=112
)bR(1af(R) 2+
=
2dR
1 b =
OLD NEW
Use BE cov between stations
jjii
ijij
σσ
σρ =
(75х75)
ijσ : covariance
• Construct full matrix
- isotropic ⇒ anisotropic- homogeneous⇒ inhomogeneous
ρ11 , ρ12 , .... , ρ1mρ21 , ρ22 , .... , .. . .. . .. . .
ρm1 , ......…. , ρmm
=Pijμ
Correlation between obs stations
11th MOS workshop 12-16 Nov 2007
KMA
21 braP
ig +=μ
r : distance a : error ratio (fixed, 1 )
b : error cov wrt r (fixed, 1/1600 )
OLD NEW
1.Use BEC between stations2.Use BEC between station and AWS
3. Use grid GIS (PRISM :weight wrt distance, altitude, slope, ,)
assume: correlation of spatial error ~ correlation of geographic
information
PRISM0.6
0.05
0.2
0.15
0.1
0.4
0.1
0.4
∑
∑=
=
=K
kgk
K
kPkigkP
ig
1
1
ω
ρωμ
Pikρ
Pgkω
: BEC of neighboring station : weight of neighboring stations
Correlation between grid and station
11th MOS workshop 12-16 Nov 2007
KMA OLD - OBS NEW - OBS
RMSE: 2.82 RMSE: 2.24
Temperature ( 06UTC 5 OCT 2007)
Impact: ERROR
11th MOS workshop 12-16 Nov 2007
KMA
WEB SERVICE MODULE (WEM)
IMAGE-GRAPHIC TIME SERIES
WORDED/ VOICED GRID-POINT
11th MOS workshop 12-16 Nov 2007
KMA
Accuracy : Short range DFS
Verification items FCST MOS NWP IMPROVEMENT % (NWP vs.FCST) Remarks
RMSE TMP(C) 2.1 2.0 2.72.93.011.70.872.90.440.7418.0
RMSE TMAX(C) 2.4 2.422.217.220.011.11.53.42.274.05
RMSE TMIN(C) 2.4 2.4RMSE PPTN(mm) 10.5 -
ACC PPTN IDF 0.86 -RMSE WS (m/s) 2.8 -
RF WD (deg) 0.45 -ACC2 SKY 0.77 - Comparison with PPM
RMSE RH (%) 14.0 12.4 22.2 MOS since March 2006
DEC 2005 - SEP 2006
11th MOS workshop 12-16 Nov 2007
KMA
Distribution of RMSE
WINTER SPRING SUMMER
Mountainous area large error
Summer better than winter
11th MOS workshop 12-16 Nov 2007
KMA
Accuracy: Medium range DFS
Birer Skill Score(Warm Season)
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
+12h +36h +60h +84h +108h +132h +156h +180h +204h +228h +252hForecast time
BSS
103STN_MOS 76STN_MOS 41STN_MOS103STN_GDP 76STN_GDP 41STN_GDP
Birer Skill Score(Cold Season)
-0.2
-0.1
0
0.1
0.2
0.3
0.4
+12h +36h +60h +84h +108h +132h +156h +180h +204h +228h +252hForecast time
BSS
103STN_MOS 76STN_MOS 41STN_MOS103STN_GDP 76STN_GDP 41STN_GDP
BSS (Warm Season) BSS (Cold Season)
GDAPS
Medium MOSMedium MOS
GDAPS
11th MOS workshop 12-16 Nov 2007
KMA
Skill
Short term (’07-’09) Medium term(’10-’12) Long term (’13-’15)-Smart tool-MOS improvement-Editing module imp.
-5km KWRF -linkage with specialweather system-short, medium rangeunified system
-unified NWP system-dynamic time-spaceediting-seamless WID
Time
High Quality Digital Forecast
70%
80%
90%
100%
Rain or notTmaxTmin
95 %
69 %
62 %
97 %
84 %82 %
96 %
77 %
73 %
2009 2012
Extended public service-quantitative and detail forecast
-support to disaster prevension
Service for life benefit-life index, fire index
-Service of 3hr forecast
Service of industry-management of water source
-management of energy and harvest
deve
lopm
ent
Appl
icatio
n pl
anPlan : Digital Forecast
KMA
Thank You~