GRAPES Model Research Progresses at CMA Chen D.H., Wang J.J., Shen X.S.

Download Report

Transcript GRAPES Model Research Progresses at CMA Chen D.H., Wang J.J., Shen X.S.

GRAPES Model Research Progresses
at CMA
Chen D.H., Wang J.J., Shen X.S. et al.
Numerical Weather Prediction Center
China Meteorological Administration
with thanks to our colleagues who contribute to the presentation
(The 4th THORPEX-ASIA Science Workshop and ARC-8 Meeting
30 Oct.~3 Nov., 2012, Kunming, China )
Outline
•
•
•
•
•
•
1 Current Operational NWP Systems
2 Efforts for improving GRAPES_GFS
3 Progresses in GRAPES_VAR
4 Implementation of GRAPES_TYM
5 High resolution modeling activities
6 Future Plan
Numerical Weather Prediction Center of CMA
Director: Dr. WANG Jianjie
Chief Engineer: Dr. CHEN Dehui
Deputy-directors:
Dr. GONG Jiandong and Dr. SHEN Xueshun
System & Operation Division
R&D Division
Dynamic
process group
Data assimilation
group
General Office
Model verification
group
Observation
data quality
control group
Ensemble
prediction
Group
Physical
process group
Post process and
products
development
group
Parallel computing
group
Regional
model group
Typhoon
prediction
group
System preoperational test
group
Model version manage
and information
technology group
The restructured organization of Numerical Prediction Center
1 Current Operational NWP
Systems at CMA
Current NWP Operational System in NMC
Models Global Spectral Meso Scale Model Global Ensemble
(GRAPES_Meso)
(T213L31)
Model
specified
(TL639L60)
Forecast
Range
Forecast
domain
Global Medium- Regional shortrange forecast range foreecast
Global
Horizontal TL639(0.28125o)
resolution
Vert. levels /
60
Top
0.1hPa
Forecast
240hours
hours (initial (00, 12UTC)
time)
33
10hPa
Initialization
Global GSI
(NCEP)
Typhoon
Ensemble
forecast
Typhoon
forecast
10 day forecast
China/East Asia
(8340km5480km)
Global
15km
T213 (0.5625 o)
33
10hPa
72 hours
(00, 12UTC)
GRAPES_3VAR
31
10hPa
240hours
240hours+BGS
(00, 12UTC)
(00, 12UTC)
15members
15members
Initial Perturb. by
BGM
BGM+NCEP
SSI + vortex
relocation,
intensity
adjustments
In general, there were no big changes in the operational NWP systems
GRAPES_TCM at Shanghai Typhoon
Institute for East C.S.
Fig: Topography of the domain of
GRAPES_TCM
• Configuration
– Domain: E90º~E170º,N0º~N50º
– Hor. Res.: 0.25ºx0.25º
– Grids: 321x201
– V. res.: 31(ztop: 35000m)
• Physics
– Cumulus:KF-eta
– PBL: YSU
– Micro: NCEP cloud3
– LSM: SLAB scheme
– Radia.: RRTM scheme
(From Wang et al., 2010)
Assessment of TC forecast methods
0.35
GRAPES_TCM
0.3
TRaP
TAPT
0.25
ETS
• TRaP: extrapolating
method based
satellite-estimated
precipitation
• TAPT: tropical cyclone
precipitation analogue
method
• GRAPES_TCM:
numerical forecast
0.2
0.15
0.1
0.05
0
0mm
0.1mm
10mm
25mm
50mm
100mm
(From Wang et al., 2010)
Evolution of yearly mean track errors
hrs
hrs
Bogus initialization + cumulus schemes
(From Wang et al., 2012)
GRAPES_TMM at Guangzhou Tropical
Meteor. Institute for S. C. S.
Domains of GRAPES_TMM
0.12o
0.36o
0.03o
GRAPES_TMM(Tropical Meteorological Model), which is a three-nested
model system
(From Wan et al., 2010)
Since 2003, GZ began to operationally implement
GRAPES_3DVAR, and then GRAPES_Meso for establishment
of GRAPES_TMM, which is a three-nested model system:
Global model
GRAPES_TMM
(0.36o)
GRAPES_TMM
(0.12o)
GRAPES_TMM
(0.03o)
MOM-sea
flow model
Storm surge
Sea waves
CHAF-1h-cyc
SWIFT-nwcst
+ 5d Tro. weather forecast
+ T. Cyclone forecast
+ SST, sea flow forecast
+ 36 hrs Meso-scale forecast
+ S.C. fine w. forecast
+ sea waves, surge forecast
+ hourly rapid cycling anal.
+ 1~3 hrs nowcast
+ 3~12 hrs sort-term forecast
Radar-extrap.
(From Wan et al., 2010)
路径误差(km)
Evolution of Yearly Mean Track Errors
450
400
350
300
250
200
150
100
50
0
24小时预报
24 hrs F.
TL
2003
2004
2005
48小时预报
48 hrs F.
Grapes
2006
2007
2008
2009
2010
2011
逐年变化
DAS/optimal use of data+
cumulus/PBL schemes
Mean Track errors2012
24h
48h
72h
96.5 km 176.7km 235.6km
(From Chen et al., 2012)
Obs.
Initial time: 00Z21June2012
F. length: 48hrs
(From Chen et al., 2010)
GRAPES_TMM
Obs.
Initial time: 00Z22June2012
F. length: 48hrs
(From Chen et al., 2010)
GRAPES_TMM
Complicated Track of Prapiroon-2012
(From Chen et al., 2012)
Inter-comparison to ECMWF, JMA, T639 and
GRAPES_TMM (Initial Time: 12UTC, 00UTC)
(From Chen et al., 2012)
Implemented
GRAPES_Meso forecast system
1.
2.
3.
4.
5.
GRAPES_Meso: operation in NMC
GRAPES_RUC: quasi-operation in NMC
GRAPES_TCM: operation in Shanghai I.
GRAPES_TMM: operation in Guangzhou I.
GRAPES_SDM: operation in CAMS
Extended to GRAPES_HMM: Basin flooding height and volume Prediction
GRAPES
Model
GRAPES_VAR
Prediction of 6hrs precipi. accumulated
Obs.
Obs.
GRAPES prediction
Estimated on hydro. stat
(Flooding height and volume; Initial time at 00UTC, 29th August, 2009.
from Wang and Chen, 2010)
(From Wang et Chen, 2012)
2 Efforts in improvements of
GRAPES_GFS
Flow chart of GRAPES_GFS
Sat. data
First Quess
Quality Control
GRAPES_3DVAR
Initial F.
Conventnl. data
Cycling
Assimilation
and Forecast
Pre-Processing
Digital Filter
Pre-Processing
GRAPES_GFS
Quality Control
10 d forecast
Efforts in improving the forecast skill of
GRAPES_GFS
-toward operation-
• More satellite data
– ATOVS(NOAA-19,METOP,FY3)
– AIRS
– IASI
• Assimilation from pressure level to model grid space
• Improve model performance
– The dynamic core refinement: conservation issue
– Hybrid vertical coordinate: from terrain-following to terrainfollowing & Z
– Increase the vertical resolution and model top lift-up
– Tuning of physical processes
•
•
•
•
•
•
Land surface: CoLM
GWD
SSO
Microphysics + fractional cloud treatment
Cumulus scheme tuning
cloud-radiation interaction issue
(From shen et al., 2012)
GRAPES Global Forecast System(preoperational)
S. Hemis.
N. Hemis.
2007
2008
2009
2011
N.H
5.5
5.8
6
6.5
S.H
4.0
4.7
5.3
6.9
forecast verification 200906-200908 12UTC
geopotential 500hPa
Correlation coefficent of forecast anomaly
NH Extratropics Lat 20.0 to 90.0 Lon -180.0 to 180.0
1.0
forecast verification 200906-200908 12UTC
geopotential 500hPa
Correlation coefficent of forecast anomaly
SH Extratropics Lat -20.0 to -90.0 Lon -180.0 to 180.0
1.0
0.9
0.9
0.8
0.8
0.7
0.7
0.6
0.6
ACC
ACC
ACC>0.6
0.5
GRAPES-GFS 2011
0.4
0.5
GRAPES-GFS 2011
0.4
0.3
0.3
T639
T639
0.2
0.1
S. Hemis.
0.2
N. Hemis.
GRAPES
GRAPES
0.1
0.0
0.0
1
2
3
4
5
6
7
8
1
2
3
Forecast days
4
5
6
7
8
Forecast days
reforecasts for 200906~200908
(From shen et al., 2012)
3 Progresses in GRAPES_VAR
Milestone of GRAPES variational data assimilation system
2001
Serial regional P3DVAR using pressure coordinate
2005
2005
Serial global P3DVAR
Serial regional M3DVAR using height-based terrain
following coordinate
2008
2009
Parallel global P3DVAR
2009
Quai-operation running
Black: developed
Blue: in progress
Orange: Operation
Red: in the future
Serial regional 3DVAR
operation running
2005
2010
Serial regional 4DVAR
Serial global M3DVAR
2010
2010
Paral. Reg. M3DVAR/4DVAR
Serial global 4DVAR
2013
Parallel global 4DVAR
(From Gong et al., 2012)
GRAPES model level analysis (GRAPES_M3DVAR)
and pressure level analysis (GRAPES_P3DVAR)
GRAPES_M3DVAR
GRAPES_P3DVAR
Vertical
coordinate
Charney-phillips , Z terrain
following, vertical stagger grid
Pressure level analysis, no stagger
grid
Horizontal grid
Arakawa C, horizontal stagger grid
Arakawa A grid
Analysis variable
Model state variable: π, θ, u, v, q Partial model state variable:Φ, u,
v, RH(q)
Control variable
Ψ, χ, Πu, RH/q/RH*
Observation
operator
Physical variable transform, horizontal bi-linear interpolation, vertical
linear interpolation or/and 3rd spline interpolation
Control variable
transform order
Vertical EOF transform firstly, then
horizontal spectral transform
Ψ, χ, Φu, RH/q
Horizontal spectral transform
firstly, then vertical EOF transform
(From Gong et al., 2012)
OBS
OBS
OBS
forward integration using non-linear model at the higher resolution
outer
loop
observation
increments
observation
increments
observation
increments
cost function J
forward integration using tangent-linear model at the lower resolution
inner
loop
forcing term
forcing term
forcing term
J
backward integration using adjoint model at the lower resolution
minimization
process
analysis
incrementδx
GRAPES 4DVAR
analysis
results
(From Zhang et al., 2012)
Since Aug.2010
1-month running
GRAPES_MESO V3.0 vs 4DVAR
Model: GRAPES_MESO V3.0
Resolution: 15 km (502x330), 31 levels
Time Step: 300 seconds
Analysis System: GRAPES-4DVAR
Outer loop resolution: The same resolution
as the model
Inner loop resolution: 45 km (167x111), 31
levels
Physics process: LSP; MRF PBL; CUDU
convection
Outer loop: 1 iteration
Obs: TEMP, SYNOP, AIREP, SHIPS
Assimilation Window: [-3, 0]
Analysis Time: 00UTC and 12UTC
Background Fields: TL639L60 12-hours
forecast
Forecast Range: 48 hours
(From Zhang et al., 2012)
1-month averaged Ts score of 24 hour precipitation
forecast over whole China
0.7
3DVAR
4DVAR
0.6
0.5
0.4
0.3
0.2
0.1
0
Light
1
Moderate
2
Heavy
3
Torrential
4
(From Zhang et al., 2012)
Flow Chart of Cloud Analysis Scheme
(From Zhu et al., 2012)
Result of Cloud Cover
background
used Surface data
used radar reflectivity
used satellite cta
used satellite tbb
increment
(From Zhu et al., 2012)
3h forecast
With cloud
analysis
Without cloud
analysis
observation
6h forecast
12h forecast
(From Zhu et al., 2012)
The first hour precipitation 12:00-13:00
Hourly accumulated precipitation
Hourly accumulated precipitation
With cloud analysis
Without cloud analysis
Hourly accumulated precipitation
OBS
(From Zhu et al., 2012)
6h forecast composite reflectivity
Without cloud analysis
With cloud analysis
Radar OBS
(From Zhu et al., 2012)
4 Implementation of
GRAPES_TYM
4.1 Quasi-operational
implementation in NMC
(From Ma et al., 2012)
GRAPES_TYM
Model
Domain
Grid points
Initial time
Initialization
F. lenth
Interval-out
Physical schemes
GRAPES_MESO3.0
90º~171ºE,0º~51ºN
541341
00UTC、12UTC
Bogusrelocated+intensityadjustment
72hrs
3hrs
Micro:WSM6
Cumul:SAS
PBL:YSU
LSM:SLAB
(From Ma et al., 2012)
Development of GRAPES_TYM for
Typhoon intensity forecast
350
20
Min SLP Error (hPa)
200
Track error
150
100
50
0
0/218
24/179
48/144
72/110
400
GRAPES_TYM
NCEP_GFS
T213
15
10
5
0
FstHour/Samples
Track Error (km)
Track Error (km)
250
20
GRAPES_TYM
T213
Minimum SLP error
0/218
24/179
48/144
72/110
FstHour/Samples
Max V10m Error (m/s)
300
25
GRAPES_TYM
T213
15
10
5
Maximum V10m error
0
0/218
24/179
48/144
72/110
FstHour/Samples
GRAPES_TYM
GRAPES_TCM
GRAPES_TMM
Mean track errors of
GRAPES_TYM to
GRAPES_TMM、
GRAPES_TCM
300
200
100
0
0/173
24/148
48/107
FstHour/Samples
72/77
(From Ma et al., 2012)
Mean Track Error (km)
Case of 2012-13 KAITAK
900
800
700
600
500
400
300
200
100
0
GRAPES_TYM
GRAPES_PHY
GRAPES_NEW
0/11
24/9
48/7
72/5
FstHour/Samples
(From Ma et al., 2012)
Case of 2012-13 KAITAK
16
GRAPES_TYM
GRAPES_PHY
GRAPES_NEW
12
8
4
0
0/11
24/9
1000
995
990
985
980
975
970
965
960
955
950
13/00 14/00 15/00 16/00 17/00 18/00
Min PSL (hPa)
MAE of PSL (hPa)
20
48/7
72/5
Valid Time (DD/HH)
FstHour/Samples
Valid Time (DD/HH)
1000
995
990
985
980
975
970
965
960
955
950
13/00 14/00 15/00 16/00 17/00 18/00
Min PSL (hPa)
Min PSL (hPa)
1000
995
990
985
980
975
970
965
960
955
950
13/00 14/00 15/00 16/00 17/00 18/00
Valid Time (DD/HH)
(From Ma et al., 2012)
Case of 2012-13 KAITAK
50
GRAPES_TYM
GRAPES_PHY
GRAPES_NEW
16
45
Max SPD (m/s)
MAE of SPD (m/s)
20
12
8
4
40
35
30
25
20
15
10
0
0/11
24/9
48/7
72/5
13/00 14/00 15/00 16/00 17/00 18/00
Valid Time(DD/HH)
FstHour/Samples
50
50
45
40
35
Max SPD (m/s)
Max SPD (m/s)
45
30
25
20
15
10
13/00 14/00 15/00 16/00 17/00 18/00
Valid Time(DD/HH)
40
35
30
25
20
15
10
13/00 14/00 15/00 16/00 17/00 18/00
Valid Time(DD/HH)
(From Ma et al., 2012)
Case of 2012-11 HAIKUI
Mean Track Error (km)
300
200
GRAPES_TYM
GRAPES_PHY
GRAPES_NEW
100
0
0/14
24/12
48/10
72/8
FstHour/Samples
(From Ma et al., 2012)
Case of 2012-11 HAIKUI
16
GRAPES_TYM
GRAPES_PHY
GRAPES_NEW
1010
1000
12
990
Min PSL (hPa)
MAE of PSL (hPa)
20
980
8
970
4
0
960
950
0/14
24/12
48/10
72/8
940
FstHour/Samples
03/00 04/00 05/00 06/00 07/00 08/00 09/00
Valid Time (DD/HH)
1010
1010
1000
1000
Min PSL (hPa)
990
Min PSL (hPa)
990
980
980
970
970
960
960
950
950
940
940
03/00 04/00 05/00 06/00 07/00 08/00 09/00
Valid Time (DD/HH)
03/00 04/00 05/00 06/00 07/00 08/00 09/00
Valid Time (DD/HH)
(From Ma et al., 2012)
Case of 2012-11 HAIKUI
50
GRAPES_TYM
GRAPES_PHY
GRAPES_NEW
8
45
Max SPD (m/s)
MAE of SPD (m/s)
10
6
4
2
0
0/14
24/12
48/10
72/8
FstHour/Samples
40
35
30
25
20
15
10
03/00 04/00 05/00 06/00 07/00 08/00 09/00
Valid Time(DD/HH)
50
50
45
40
Max SPD (m/s)
Max SPD (m/s)
45
35
30
25
20
15
40
35
30
25
20
15
10
10
03/00 04/00 05/00 06/00 07/00 08/00 09/00
Valid Time(DD/HH)
03/00 04/00 05/00 06/00 07/00 08/00 09/00
Valid Time(DD/HH)
(From Ma et al., 2012)
4.2 The Coupled Typhoon-Ocean Model
Regional air-sea
Coupled model
Atmosphere
Ocean
Wind stress
Heat flux
Water flux
GRAPES_TYM
(0.15*0.15)
Coupler
(Oasis 3.0)
Regional
ECOM-si
(0.25*0.25)
SST
Initial conditions/
Lateral boundary
condition
GFS
Initial conditions/
boundary condition
Global HYCOM
(From Sun et al., 2012)
Model domain
ECOM:
Horizontal resolution: 0.25°x 0.25°
Domain:104°E~145°E, 8°N~43°N
Provided:SST
GRAPES:
Horizontal resolution : 0.15°x 0.15°
Domain:100°E~150°E, 5°N~45°N
Provided: wind stress, solar flux, heat flux,
water flux;
Fluxes are exchanged every 360s.
(From Sun et al., 2012)
OASIS3 coupler
OASIS: Ocean Atmosphere Sea Ice Soil
---------Developed since 1991 in CERFACS
•performs:
synchronisation of the component models
coupling fields exchange and interpolation
I/O actions
A
O
O
A
•External library and module used:
NetCDF/parallel NetCDF
libXML, mpp_io, SCRIP
MPI1 and/or MPI2
Oasis3
O
A
O
(From Sun et al., 2012)
SST forecasted by the coupled model
---Typhoon Muifa
NCEP AVHRR + AMSR-E SST analysis
at 08/08/11, 00UTC
72 hour forecasted SST by the coupled model
Initialized at 00UTC 05 AUGUST, 2011
 The coupled model reproduces the sea surface cooling that
is closed well to the analysis.
(From Sun et al., 2012)
Typhoon Muifa – impact of coupling
Tropical Cyclone Muifa (2011)
INITIAL TIME 00:00 UTC, 5 August 2011
Black –observation
Red-Uncoupled model
Green-Coupled model
Too strong in GRAPES_tym
Coupling weaken the intensity


(From Sun et al., 2012)
Forecast verification for MUIFA
Number of cases (21, 21,19,17)
(From Sun et al., 2012)
Typhoon MUIFA intensity forecast
Minimum sea level pressure forecast
GRAPES_tym
Minimum sea level pressure forecast
Coupled model
Maximum wind forecast
GRAPES_tym
Maximum wind forecast
Coupled model
(From Sun et al., 2012)
Typhoon SINLAKU – impact of coupling
NCEP AVHRR + AMSR-E SST analysis
at 15/09/08, 00UTC

Too strong in GRAPES_TYM model
 Coupling weaken the intensity
(From Sun et al., 2012)
Tropical Cyclone SINLAKU (2008)
INITIAL TIME 12:00 UTC, 12 September 2008
Forecast verification for Typhoon SINLAKU
Number of cases (21, 21,19,17)
Maximum wind error (m/s)
16
14
12
10
8
6
4
Uncoupled
2
Coupled
0
(From Sun et al., 2012)
Forecast hour
Forecast verification of Nine TC in 2011
Number of cases (72,72,56,56,49,44,44)
Mean minimum sea level
pressure error (hpa)
25
Uncoupled
20
Coupled
15
10
5
0
0
Mean maximum wind error
(m/s)
16
14
12
10
8
6
4
2
0
(From Sun et al., 2012)
12
24
36
48
Forecast hour
60
72
Uncoupled
Coupled
0
12
24
36
48
Forecast hour
60
72
Intensity forecast of Nine TC in 2011
Number of cases (72,72,56,44)
Minimum sea level pressure forecast
Coupled model
Minimum sea level pressure forecast
GRAPES_tym
Maximum wind forecast
GRAPES_tym
Maximum wind forecast
Coupled model
60
Forecast
50
40
30
T+0h
20
T+24h
T+48h
10
T+72h
0
0
(From Sun et al., 2012)
10
20
30
Observation
40
50
60
5 High resolution modeling
activities
5.1 High Resolution Modeling Activities at CMA
Based on GRAPES_Meso
Recent activities
• Vertical coordinate from terrain-following Z to hybrid coordinate (Schar,
2002)
• Inclusion of thermal expansion effect in continuity equation
• Improve the interpolation accuracy in physics-dynamics interface
• Refinement of 2-moment microphysics scheme
• Some bug fix in land surface scheme
• Refinement of back ground error covariance in 3DVAR
Modification of TF coordinate
• In order to design a new TF coordinate, we
rewrite the formulation of Gal-Chen and
Sommerville (1975) in a common formulation:
z  Z s ( x, y)
zˆ  Z T
Z T  Z s ( x, y)
z  zˆ  b  Z s ( x, y)
with b  (1 
zˆ
)
ZT
It is a decaying coefficient of the coordinate surface with
height. It is possible to use different “b” to accelerate the decaying.
(From Li et Chen, 2012)
New TF coordinates
• The different decaying coefficients “b” can
be defined as:
G.C.S.
b  (1 
zˆ
)
ZT
(Gal-Chen and Sommerville, 1974)
sinh[(Z T  zˆ) / h* ]
SLEVE1 bh 
sinh[Z T / h* ]
(Schar, 2002)
h*: scale of ref-topography; h*1 and h*2:
sinh[(Z T  zˆ) / hi* ]
SLEVE2 bH  
sinh[Z T / hi* ]
i 1
2
COS

zˆ
1 
bC   Z T

0
    zˆ 

  cos
   2 zˆc 
large and small-scale of ref-topogr.
n
zˆ  zˆ c
zˆ  zˆc
(similar to Klemp, 2011)
“n>2”: an empirical number; zc :
a reference height from which
the coordinate surface becomes
horizontal.
(From Li et Chen, 2012)
1D test design
•Test Objective :to compare the errors of PGF calculation of four
coordinates in rest atmosphere over an artificial terrain.
•Test design:
•Reference rest atmosphere:

 gz 
  exp  

C
T

p 0 

g  9.81, T0  287.0

 gz 

  T0 exp  C T  ,
 p 0

•Classical algorithm used for PGF calculation
C p z   C p zˆ  C p J b 
with J b 

 b  zˆ ( Z S ( x, y ))
zˆ
zˆ
z
(From Li et Chen, 2012)
Errors of PGF calculation induced by using TF coordinates
top
pressure gradient force error
0.005
0.0025
0
-0.0025
-0.005
-50000
0.02
0.01
0
-0.01
-0.02
-50000
0.05
0.025
0
-0.025
-0.05
-50000
0.1
0.05
0
-0.05
-0.1
-50000
0.1
0.05
0
-0.05
-0.1
-50000
bottom
Gal.C.S coordinate
L40
0
L30
0
L20
0
L10
0
L2
0
SLEVE1
SLEVE1 coordinate
15000 0.005
10000 0.0025
0
5000 -0.0025
0
-0.005
50000
-50000
15000 0.02
10000 0.01
0
5000 -0.01
0
-0.02
50000
-50000
15000 0.05
10000 0.025
0
5000 -0.025
0
-0.05
50000
-50000
15000
0.1
0.05
10000
0
5000 -0.05
0
-0.1
50000
-50000
15000
0.1
10000 0.05
0
5000 -0.05
0
-0.1
50000
-50000
L40
0
L30
0
L20
0
L10
0
L2
0
SLEVE2
COS
SLEVE2 coordinate
15000 0.005
10000 0.0025
0
5000 -0.0025
0
-0.005
50000
-50000
15000 0.02
10000 0.01
0
5000 -0.01
0
-0.02
50000
-50000
15000 0.05
10000 0.025
0
5000 -0.025
0
-0.05
50000
-50000
15000
0.1
0.05
10000
0
5000 -0.05
0
-0.1
50000
-50000
15000
0.1
10000 0.05
0
5000 -0.05
0
-0.1
50000
-50000
COS coordinate
15000
L40
10000
5000
0
L30
0
50000
15000
10000
5000
0
L20
0
50000
15000
10000
5000
0
L10
0
50000
15000
10000
5000
0
L2
0
50000
15000
10000
5000
0
0
50000
0.005
0.0025
0
-0.0025
-0.005
-50000
0.02
0.01
0
-0.01
-0.02
-50000
0.05
0.025
0
-0.025
-0.05
-50000
0.1
0.05
0
-0.05
-0.1
-50000
0.1
0.05
0
-0.05
-0.1
-50000
15000
L40
10000
5000
0
L30
0
50000
15000
10000
5000
0
L20
0
50000
15000
10000
5000
0
L10
0
50000
15000
10000
5000
0
L2
0
50000
15000
10000
5000
0
0
50000
On different vertical levels: L2, L10, L20, L30 and L40 from bottom to top
(From Li et Chen, 2012)
height
G.C.S
Relatively Reduced Errors: SLEVE1(SLEVE2, COS) against GCS
R.R.E. is defined as:
EGal  ESLEVE1/ SLEVE 2/ COS

100%
EGal
Vertical levels
SLEVE1
SLEVE2
COS
L40
67%
99%
100%
L30
62%
99%
100%
L20
51%
99%
99%
L10
31%
95%
75%
L2
4%
30%
2%
(From Li et Chen, 2012)
2D test design (cont.)
Initial wind:
Analysis density
distribution :
1

  z  z1 

u ( z )  10  sin 2  

 2 z2  z1 

0

4km  z  z
z  4km
 2  r
 x  x0   z  z0 
) r 1
cos (
 ( x, z )  0  
,r 
2


R
R
x
z





r 1
0
before mount
flow from L to R
5km  z
over
after mount
density distribution
(From Li et Chen, 2012)
Advection test : air mass moves over a topographic obstacle
15000
15000
b
Gal.C.S coordinate
12500
10000
10000
GCS
height
height
a
12500
7500
5000
2500
-50000
-25000
0
x
25000
50000
12500
12500
10000
10000
7500
0
x
25000
50000
75000
50000
75000
7500
5000
2500
2500
0
-75000
-50000
-25000
0
x
25000
50000
0
-75000
75000
15000
-50000
-25000
0
x
25000
15000
e
f
SLEVE2 coordinate
12500
12500
10000
10000
SLEVE2 coordinate
SLEVE2
height
height
-25000
SLEVE1 coordinate
SLEVE1
height
height
d
SLEVE1 coordinate
5000
7500
5000
7500
5000
2500
2500
0
-75000
-50000
-25000
0
x
25000
50000
0
-75000
75000
15000
-50000
-25000
0
x
25000
50000
75000
15000
g
h
COS coordinate(Zc=15km)
12500
COS coordinate(Zc=15km)
12500
10000
10000
COS-zc=15km
height
height
-50000
15000
c
7500
5000
7500
5000
2500
2500
0
-75000
-50000
-25000
0
x
25000
50000
0
-75000
75000
15000
-50000
-25000
0
x
25000
50000
75000
15000
i
j
COS coordinate(Zc=10km)
12500
12500
10000
10000
COS coordinate(Zc=10km)
COS-zc=10km
height
height
5000
0
-75000
75000
15000
7500
5000
7500
5000
2500
2500
0
-75000
-50000
-25000
0
x
25000
50000
0
-75000
75000
15000
-50000
-25000
0
x
25000
50000
75000
0
x
25000
50000
75000
15000
k
regular grid
l
12500
regular grid
12500
without topography
10000
height
10000
height
7500
2500
0
-75000
7500
5000
7500
5000
2500
0
-75000
Gal.C.S coordinate
2500
-50000
-25000
0
x
25000
50000
75000
left:density distribution at 0s,5000s,10000s
0
-75000
-50000
-25000
right:the errors at 10000s after mountain
(From Li et Chen, 2012)
The errors of the simulations
、
Defining two parameters as following, according to Williamson (et.al,1992)
2

1
 m n
2 2




( i ,k ) num  ( i ,k ) ana  
 
k 1 i 1





1
2 2

m
n





 ( i , k ) ana  

k

1
i

1






 MAX ( i ,k )num  ( i ,k )ana

MAX ( i ,k )ana


( i ,k )num








SLEVE1
SLEVE 2 COS




SLEVE1
SLEVE 2 COS
1
 m n
2 2




( i ,k ) num  ( i ,k )ana  
 
k 1 i 1




 
1
2 2

m
n





 ( i ,k ) ana  

k

1
i

1






 MAX ( i ,k )num  ( i ,k )ana

MAX ( i ,k )ana


(i,k )ana
is numerical solution
0.6




wi t hout
t er r ai n
t er r ai n
is analytical solution
0.6
Gal.C.S coordinate
SLEVE1 coordinate
SLEVE2 coordinate
A coordinate(Zc=15km)
A coordinate(Zc=10km)
0.5
Gal.C.S
0.4
2
Gal.C.S coordinate
SLEVE1 coordinate
SLEVE2 coordinate
A coordinate(Zc=15km)
A coordinate(Zc=10km)
0.5
计
算
误
差
SLEVE1
0.3
SLEVE2
error
0.2
Gal.C.S
SLEVE1
0.4

0.3
SLEVE2
0.2
COS(15KM)
0.1
0
1000
2000
3000
4000
5000
6000
7000
COS(15KM)
0.1
COS(10KM)
0
-0.1








wi t hout
8000
0
9000
integral time
left : temporal evolution of
10000
-0.1
COS(10KM)
积分时间
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
integral time
2
right : temporal evolution of

(From Li et Chen, 2012)
The preliminary results with new TF
coordinates in GRAPES_Meso
• The preliminary results with regional GRAPES (15km) are quite
encouraging:
Monthly mean of 24h
forecast of geopotential
height at 100hPa
(From Li et Chen, 2012)
The torrential rain-storm occurred on 21
Jul. 2012 in Beijing
24 h accumulated precipitation
from 00UTC 21 Jul to 00UTC 22
Jul
“The torrential rain-storm occurred on 21 Jul. 2012 in Beijing
area: the worst the city has seen in more than 60 years,
dumped an average of 215 millimeters of rain in 16 hours.
Hebeizhen, a town in the suburban district of Fangshan
(South-West), saw 460 millimeters for the same period. ”。
(From Chen et al., 2012)
Heavy rainfall event on Jul.21/2012 Beijing
00z21Jul2012-00z22Jul2012
Obs.
Beijing
Mean=190.3mm/24hr
Max=460mm/24hr
Initial: global analysis
BC: global forecast
Grid size:3km
Physics:
- microphysics: WSM6
- radiation:
RRTM S&L
- pbl :
MRF
- land surface :NOAH
24-hour accumulated rainfall
GRAPES_Meso-3km
ECMWF
Fcst.
Max=341mm/24hr
(From Huanget al., 2012)
Comparison of precipitation every 6-hour forecasts against Obs.
Fcst.0-6hr
Obs.0-6hr
Fcst.6-12hr
Obs.6-12hr
Fcst.12-18hr
Obs.12-18hr
Fcst.18-24hr
Obs.18-24hr
(GRAPES_Meso-3km)
(From Huanget al., 2012)
5.2 other Research activities at CMA
– GRAPES Yin-yang dynamic core
– SV-based GRAPES ensemble forecast system
– New algorithms of dynamic core
Progress of GRAPES Yin-Yang grid
The Helmholtz equation of GRAPES in the Yin-Yang overset grid are solved.
The transplant of the whole GRAPES dynamical core is finished. However,
some bugs exist and it need to be debuged in the next step.
Helmholtz equation:
2   2  H
(From Peng et al., 2012)
3D advection results
q(day  12)  q(day  0)
alpha=0.
Instant image on the Yang grid
5
9
1
4
6
8
2
10
day12
3
alpha=45.
7
z  200m, nlev  60,     1.0d
The tracer follow the wave motion and undergo
Three oscillations in the vertical direction.
After one revolution(12 days), the tracer is back
to the initial state.
(From Peng et al., 2012)
alpha=90.
High order Multi-moment Constrained finite Volume (MCV) method
We define the moments within single cell, i.e. the cell-averaged value, the point-wise
value and the derivatives of the field variable
Solution points
Constraint points
Constraint conditons:
Approximate Riemann solvers
The unknowns (solution points) are updated in a fourth order mcv scheme, for example,
The same in multi-dimension, for example, y direction
(From Li et al., 2012)
A nonhydrostatic atmospheric governing equation sets in the
Cartesion system
Height-based terrain-following vertical coordinate (Gal-chen & Somerville
1975) is used.
is transformation Jacobian.
MCV4 results
(From Li et al., 2012)
Linear nonhydrostatic
mountain case
Analytic solution:
red dash line
Discontinuous Galerkin results (Giraldo & Restelli, JCP, 2008)
zero contours
Fourth order MCV results
(From Li et al., 2012)
6 Future Plan
3DVAR, 24-60h
Δx=3-10km, L45
GRAPES_Meso
+ RAFS
3DVAR/Bogus,
72h, Δx=10km, L45
GRAPES_TYM
SV+Sto.Phy, 10d
Δx=50km, L60
GRAPES_EPS
GRAPES_GFS
Strategic Plan
(~2015)
GRAPES_GFS
3DVAR, 10 d
Δx=25km, L60
Global GRAPES_VAR Research & Operation Plan
GRAPES
P3DVAR
a new fixed version
(res:1 deg—>0.5deg)
FY-3A MWTS、
FY-2E IR AMV
2012
•GRAPES
P3DVARM3DVAR;
•Conventional data QC
re-check
•Data Preprocessing
system re-design
GRAPES-
GRAPES-M3DVAR
oper. run
M3DVAR
FY3-B MWTS、
NPP satellite data used
FY3-A/B MWHS、 More data from
FY-2D/F IR AMV
FY2/FY3
2013
•Satellite vertical
sounding high
level channel used
GRAPES4DVAR realtime running
2015
2020
•GRAPES-4DVAR
develop;
•New version data
preprocessing used
•GRAPES4DVAR real-time
trial
• More FY
satellite data
2014
(From Gong et al., 2012)
Regional GRAPES_VAR research and operation plan
GRAPES
3DVAR
parallel version
operation
2012
•3DVAR system
improvement;
•B matrix re-estimate
and tuning
•Conventional data QC
recheck.
Radar VAD、
GPS/PW oper used.
AWS humidity
data operational
used
Cloud analysis
2013
oper used
GRAPES_RAFS quasioper. running
2015
2014
•Pressure-wind
balance re-tunning;
•Eliminate
boundary noise in
B matrix;
•GPS/PW QC;
•Cloud analysis
improvement;
•Radar precipitation
heating profile;
•Radar reflectivity
QC;
•Wind profile QC
•Continue
GRAPES_4DVAR?
Regional GRAPES_VAR: 4DVAR or
3DVAR+EnKF?
2020
•GRAPES-4DVAR
real-time test;
•Radar data QC
•Unified global
/ regional 3DVAR
•Satellite data
used in regional
model
(From Gong et al., 2012)
Future HR GRAPES -DA and Prediction System
3DVAR
EnKF
analysis 1
Innovation
member 2
forecast
Ensemble
covariance
EnKF
analysis k
H.R. Model
GRAPES
forecast
……
member k
forecast
3DVAR-ECV
data assimilation
GRAPES
analysis
member 1
forecast
GRAPES-DFL
member 2
analysis
member 2
forecast
GRAPES-DFL
member k
analysis
0 20m 40m
0 20m 40m
……
EnKF
analysis 2
member 1
analysis
……
EnKF
Re-center EnKF analysis ensemble
to control analysis
observations
……
Multi model EPS
member 1
forecast
member k
forecast
control
forecast
GRAPES-DFL
0 20m 40m
GRAPES-DFL
0 20m 40m
First guess
forecast
DAS: EnKF/3DVAR Hybrid DA System; Multi Models: WRF, GRAPES_Meso
(modified from talk of Xue et al., 2012)
GRAPES_3DVAR(or 4DVAR)-Hybrid: Method
Extended control variable method (Lorenc 2003; Wang 2010):
 
J x1' , α  1 J1   2 J e  J o

 
T
1 T
1
1
 1 x1' B 1x1'   2 α T C 1α  y o'  Hx' R 1 y o'  Hx'
2
2
2
K

x  x   α k  x ek
'
'
1
k 1


Extra term associated with
extended control variable
Extra increment associated
with ensemble
B 3DVAR static covariance; R observation error covariance; K ensemble size;
C correlation matrix for ensemble covariance localization; xek kth ensemble perturbation;
x1' 3DVAR increment; x ' total (hybrid) increment; y o ' innovation vector;
H linearized observation operator; 1 weighting coefficient for static covariance;
 2 weighting coefficient for ensemble covariance; α extended control variable.
82
(modified from talk of Xue et al., 2012)
Blending Nowcast with GRAPES_HR
• Implement a very-short term forecast system with 3km
resolution based on multi-model ensemble including
GRAPES_Meso, WRF and ARPS (collaborate with Nanjing
University)
• Data assimilation: hybrid DA (3DVAR+EnKF) (collaborate with
Ming Xue, Oklahoma Univ.)
(from Chen et al., 2012)
THANK YOU