1. Short Range NWP Strategy of JMA and Research Activities at MRI
IAMAS2005, 11 August 2005, Beijing
Short Range NWP Strategy of JMA and Research Activities at MRI
Kazuo SAITO
Meteorological Research Institute,
1. Operational mesoscale NWP at JMA
2. Recent developments for operation
3. Near future plans
4. Research activities in MRI

2. Essential factors in the mesoscale NWP
Model (Domain, Resolution, Dynamics, Physical processes)
Initial condition (Analysis method, Data)
Boundary condition

3. extra-tropical cyclone
Scale of atmospheric phenomena
year
Synoptic forcing
planetary wave
month
extra-tropical cyclone
week
mesoscale
typhoon
day
front
micro scale
heavy rain
Macro scale
thunder storm
hour
cumulus
conventional aerological observation -300 km, 2/day
local wind
turbulence
minute
conventional NWP model
6Dx = km, 2-4/day
second 1m
10m
100m
1km
10km
100km
1000km
10000km

4. Mesoscale NWP at JMA (March 2001-)
MSM
10 km L40, 3600 km x 2880 km, 18 hours forecast, 4 times a day
Hydrostatic spectral model (March 2001-August 2004)
Nonhydrostatic (September 2004-) nested with RSM
RSM
20 km L40, 6480 km x 5120 km, 51 hours , 2 times a day
Hydrostatic spectral model, nested with GSM (60 km L40)
MSM
RSM

5. Performance of JMA Mesoscale Model
Threat scores 40km 10mm/6hr
Threat scores 10km 10mm/3hr
Performance of MSM has been improving for both weak and moderate rains

6. 2. Recent developments for operational meso NWP
Start of Mesoscale NWP (Mar. 2001)
Wind profiler data (Jun. 2001)
4D-Var in MSM (Mar. 2002)
Domestic ACARS data (Aug. 2002)
4D-Var in RSM (Jun. 2003)
SSM/I precipitable amount (Oct. 2003)
QuikSCAT Seawinds (Jul. 2004)
Nonhydrostatic model (Sep. 2004)
Doppler radar radial winds ( Mar. 2005)

7. Wind Profiler Network of JMA
JMA deployed 25 wind profilers in 2001, and their data have been assimilated since June 2001.
Wind profilers measure the low level winds up to 5 km with a vertical resolution of 300m .
Currently, 31 wind profilers measure wind successively in addition to the 18 aerological sondes.

8. Initial Assimilation System for MSM (March 2001-March 2002)
03 UTC
04 UTC
05 UTC
06 UTC
3-h Forecast with RSM (20km L40) from 00UTC
1-h Forecast with MSM
1-h Forecast with MSM
1-h Forecast with MSM
18-h Forecast with MSM
Physical Initialization
OI Analysis + Physical Initialization
OI Analysis + Physical Initialization
OI Analysis + Physical Initialization
Precipitation Data
Conventional Data
Precipitation Data
Conventional Data
Precipitation Data
Conventional Data
Precipitation Data
(For Analysis at 06 UTC)

9. The Meso 4D-Var System (March 2002-)
2 x 3 hour assimilation windows.
Incremental approach using a 20-km version of MSM for inner loop.
Inner forward : nonlinear full-physics model
Inner backward : reduced-physics adjoint model (grid-scale condensation, moist convective adjustment, vertical diffusion, simplified radiation)
Precipitation analysis by radar and AMeDAS observation are assimilated.
Boundary condition in assimilation window is controlled.

10. Concept of ４D Var Jo Jo Jb Jo Cost function :
Gradient of cost function :
Adjoint model
Model
Penalty term
Observation parameter
observation Jo
initial time Jo
First guess
observation Jo Jb
Time integration of NWP model
observation
analysis Jo
observation
21UTC
00UTC
time
Assimilation window
３hrs

11. Radar-AMeDAS Precipitation Analysis
Hourly precipitation amount data with 2.5km resolution.
Radar-observed precipitation intensity is accumulated, calibrated with 1,300 AMeDAS rain-gauges.
More than 3,000 rain-gauges (not from JMA) added in 2003.
・：4-elements
・：Rain gauge

12. 4D-Var in MSM RUC with OI 4D-Var Observation FT=15-18
3 hour accumulated rain for FT=18 hr
Initial 12 UTC 9 September 2001
Ishikawa and Koizumi (2002)

13. Threat scores (40km verification grid)
1mm/3h
10mm/3h
June 2001
(h)
(h)
Sep. 2001
Red: 4D-Var
Blue: routine
(h)
(h)

14. Domestic ACARS Data (August 2002-)
Domestic ACARS data from the Japan Air Line have been assimilated in addition to the conventional AIREP and AMDAR data.
The ANA data have been added since September 2003.
More than 10,000 reports per day.

15. Impact of ACARS Data Observation (AMEDAS)
WITHOUT
ACARS DATA
Observation (AMEDAS)
Shear line
WITH
ACARS
Location of the observed local shear line near Tokyo is corrected with ACARS data.

16. Assimilation of precipitation and TPW data retrieved from TMI and SSM/I (October 2003-)
Defense Meteorological Satellite Program
Special Sensor Microwave / Imager
TRMM Microwave Imager

17. OSE for 00UTC, 25 Aug 2003 Water vapor field was improved Sato (2003)
Without SSM/I and TMI
3 hour rain at FT=18
TPW by SSM/I and TMI
Observation
With SSM/I and TMI
Sato (2003)

18. Performance of MSM with TMI and SSM/I
Period 2003 June 3～16 （2weeks 56 forecasts）10 km verification grid
Threat score
0.40
CNTL
0.38
TEST
0.36
1mm/3hr
0.34
0.32
0.30
0.28 3 6 9 12 15 18 FT
0.22
CNTL
0.20
TEST
0.18
10mm/3hr
0.16
0.14
0.12
0.10 3 6 9 12 15 18 FT

19. Assimilation of QuikSCAT SeaWinds July 2004 -
NASA
Observation
30ﾟN
T0207
（HALONG）

20. Precipitation FT=8-9. Initial: 12 UTC 18 July 2003
Threat scores 10km 30mm/3h, 3-19 June 2003
SeaWinds 10UTC 18 July 2003
Ohashi (2004)

21. Non-hydrostatic MSM (JMA-NHM) September 2004-
Developed by joint work between MRI and NPD/JMA
HE-VI, stable computation with LF scheme Dt=40 sec
Fully compressible, flux form 4th order advection with FCT
Direct evaluation of buoyancy from density perturbation
3-class bulk microphysics (water vapor, cloud water, rain, cloud ice, snow, graupel)
Modified Kain-Fritsch convective parameterization scheme
Targeted Moisture Diffusion
Box-Lagrangian scheme for rain and graupel
Full paper submitted to M.W.R. (Saito et al., 2005)

22. Modification of the Kain-Fritsch convective parameterization
Original K-F scheme. FT=12.
Observed 3 hour accumulated
precipitation (mm) at 21 UTC.
Several points (updraft property,
trigger function, closure assumption) in the K-F scheme have been modified to prevent unnatural orographic rainfall and excessive stabilization .
Submitted to MWR.
Modified K-F scheme. FT=12.

23. Case Study of Non-hydrostatic MSM
Heavy rainfall event (18 July 2003, FT=15h)
Snowfall (13 January 2004, FT=18h)
We are replacing the present hydrostatic meso-scale model by a non-hydrostatic mesoscale model with explicit representation of cloud physics in this September. This figure demonstrates advantages of the non-hydrostatic model even though the spatial resolution is the same, 10 km and 40 levels.Those positive impacts are mainly due to the sophisticated precipitation processes in NHM, especially in snowfall prediction in winter. In MSM, condensed water is instantaneously removed from the atmosphere as precipitation.
Radar-AMeDAS observation
Hydrostatic MSM
Non-hydrostatic MSM

24. Performance of Non-hydrostatic MSM
NH-MSM
Five-month total scores over forecast time 03, 06, 09, 12, 15, 18h against 3hourly rain analysis at 20 km grid
These panels are statistical verifications. Red lines are NHM and black lines are MSM. The bias score is significantly reduced in winter, although it is still much larger than unity. The threat score is almost the same, but if one-grid difference in position is tolerated, the advantage of NHM is evident. This means that NHM predicts heavy precipitation areas closer to observed positions.
Five-month total scores at FT=18h against analysis of height

25. Performance of JMA Mesoscale Model
Bias scores 10km 10mm/3hr
NHM
High bias scores in winter were removed by NHM

26. Assimilation of Doppler radar radial winds
March 2005-
Without DPR wind FT=15
With DPR winds FT=15
Observation
Koizumi and Ishikawa (2005)

27. Performance of MSM has been improved
Threat scores 10 km, 10mm/3hr for FT=6-9
0.23
0.17
0.11
NHM
4D-Var

28. Major Operational Changes in GSM
Boundary conditions for MSM
Major Operational Changes in GSM
Enhancement of vertical resolution from L36 to L40 (Mar. 2001)
3D-Var (Sep. 2001)
QuikSCAT Seawinds, ATOVS radiances (May 2003)
Modification of the cumulus parameterization (May 2003, Jul. 2004)
MODIS Arctic wind data (May 2004, Sep. 2004)
4D-Var (Feb. 2005)
Semi-Lagrangian scheme (TL319; Feb. 2005)
Major Operational Changes in RSM
Enhancement of vertical resolution from L36 to L40 (Mar. 2001)
4D-Var (Jun. 2003)
Target moisture diffusion (Apr. 2004)

29. Improvement of GSM performance
500 hPa Height
500 hPa Temperature
Cumulus, ATOVS,etc.
4D-Var
3D-Var
4D-Var
Significant improvement by major changes (cumulus, ATOVS, etc.) in May 2003.
Significant improvement by 3D-Var in September 2002.

30. RMSE of 500 hPa Height
11 years
3 years
Improvement in the recent 3 years ( ) exceeds that
in 10 years before 2002.

31. Performance of GSM in RMSE region
2 Day
1 Day
Contributes to RSM forecast through the lateral B.C.

32. 4D-Var in RSM (June 2003-) 3D-OI 4D-Var Observation
6 hour accumulated precipitation for FT=6 (upper) and
FT=12 (bottom) with RSM. Initial time 00UTC 17 June 2002.

33. Threat Scores of RSM (Verified with 40km resolution, 1 month for June 2002)

34. Performance of RSM improved
4D-Var
Time series of RMSE for 500 hPa field
Contribute to MSM forecast through the lateral B.C.

35. 3. Near Future Plans for 2006-2008 Model
High resolution MSM (5 km L50) (Mar )
- execute 8 times / day
Boundary condition
High resolution GSM (TL959=20km L60) (2007-)
- execute 4 times / day
Initial condition
Non-hydrostatic 4D-Var (JNoVA) (2008-)
- 3 hour assimilation window execute 8 times / day, inner 10 km

36. 5 km Nonhydrostatic MSM (2006-)
- 10kmL40 → 5km L50 (Mar. 2006)
- 4 times a day → 8 times a day (Mar. 2006)
- 33-hr forecast (Mar. 2007)
Radar-AMeDAS obs.
5km Nonhydro. MSM
10km MSM
(18 July UTC, FT=6-9)

37. 20km (TL959) Global Model (2007-)
- 60kmL40 → 20kmL60 (Mar. 2007)
- Twice a day → 4 times a day (Mar. 2007)
- Supply latest B.C. to MSM directly
(19 Jun UTC, FT=12)
For the 20km global model, a semi-Lagrangian method will be adopted for high speed computation. The operation frequency will be increased from twice a day to four times a day. The assimilation method will be upgraded from 3D-Var to 4D-Var. In a preliminary result shown here, the 20km resolution global model reproduces a detailed structure of rainfall around Japan, compared to the current 60 km resolution model.
60km GSM km GSM Radar-AMeDAS 12-h rain

38. Nonhydｒostatic 4D-Var (2008-)
5 km L50, 3 hour assimilation windows
Incremental approach using a 10-km version of nonhydrostatic MSM for inner loop
UL: Radar-AMeDAS 3-h rain
UR: 12 hr forecast Meso 4DVar
LL: Nonhydrostatic 4D-Var
Initial time 12 UTC 17, July 2004
Honda et al. (2005)

39. 4. Research activities at MRI
Model
- Cloud resolving NWP model
Initial condition
- GPS data, Direct assimilation of satellite data
- Cloud resolving 4D-Var
Boundary condition
- Global nonhydrostatic model
Meso-ensemble

40. Assimilation of GPS TPW data
JMA AWS AMeDas ・：4-elements
・：Rain gauge
AMeDAS (JMA)
GPS Earth Observation Network
(Geographical Survey Institute)

41. Assimilation of GPS TPW data
Heavy rain event 30 June 2004
Analysis of TPW
wsfc (with GPS) - wsfc (w/o GPS)
w/o GPS
with GPS

42. Impact of GPS TPW data Shoji et al. (2005)
w/o GPS
with GPS
Observed heavy rain is predicted by assimilation of GPS TPW data.
Shoji et al. (2005)

43. Assimilation of GPS occultation data
CHAMP/ISDC (GFZ) :
Challenging Mini-Satellite Payload for Geoscientific Research and Application Information System and Data Center
grey：１ｓｔ guess
black；observation
Height (ｋm)
Reflection ×106
occultation observation
GPS
Assimilation period 　00-06 UTC 16 July 2004
CHAMP

44. Impact of CHAMP FT=6 Initial 06UTC 16 July 2004
CNTL
CNTL+CHAMP
Radar AMeDAS 09-12UTC
FT=6
Initial 06UTC 16 July 2004
The CHAMP occultation data moisten the lower atmosphere and yield observed precipitation in MSM.
Seko et al. (2005)

45. Further activities MRI/JMA
Asian THORPEX
WWRP Beijing Olympic 2008 Forecast Demonstration Program /Research and Development Program
- participate in MEP component

46. Meso ensemble experiment for Niigata heavy rain in July 2004
Observation 00UTC 13 July 2004
03UTC
06UTC
FT=12
FT=15
FT=18
Routine hydrostatic MSM prediction from 12UTC 12 July 2004

47. Downscale experiment of weekly ensemble prediction
Initial 12 UTC 12 July 2004　T106 Global EPS
CONTROL
Member M03p
５図　６図と同じ。メンバー'M03p'

48. Precipitation in a rectangle over northern Japan 400×250km by Global EPS
FT=12-18
M03p
M03p
Mean precipitation
FT=00-06
extreme value
Only very weak rain in GSM

49. 10 km MSM downscale experiment of EPS
10kmNHM Control
FT=06
FT=18
Member 'M03p'
５図　６図と同じ。メンバー'M03p'

50. Precipitation in a rectangle over northern Japan 400×250km by 10 km MSM downscale experiment of EPS
FT=12-18
M03p
M03p
M07m
M07m
Mean precipitation
FT=00-06
extreme value

51. Downscaling experiment of the global EPS with MSM
Observation 00UTC 13 July 2004
03UTC
06UTC
FT=15
FT=18
FT=12
Location of precipitation is adjusted to south and line-shaped intense rain is reproduced

52. Summary
JMA Mesoscale NWP started Several factors (model, initial and lateral boundary conditions) have been modified, and the performance has improved.
Data assimilation of mesoscale data using variational method is the key factor.
Significant improvement of GSM and RSM also contributed to MSM through the LBC.
Further updates are scheduled in the operational system by 2008.
Research and developments are underway to realize dynamical prediction of heavy rain.
Mesoscale NWP is now entering a new stage.