Outline of the Operational Numerical Weather Prediction at the Japan Meteorological Agency (JMA)

Masaomi Nakamura

Numerical Prediction Division
Japan Meteorological Agency


1. Computer System

The computer system at JMA consists of a Hitachi S-3800/480 supercomputer, general purpose servers, workstations, X terminals, personal computers and numbers of peripheral equipment.

The supercomputer (Hitachi S-3800/480) is used primarily for operational numerical weather analysis and prediction, and development of NWP (Numerical Weather Prediction) system. The S-3800/480 has 4 high performance vector processors and 4 scalar processors which can simultaneously operate and have access to the same main memory. The peak vector performance of each vector processor is 8 GFLOPS, and the total peak performance of the S-3800/480 is 32 GFLOPS. The supercomputer has a 2-Gbyte high performance main storage memory and 12-Gbyte extended storage memory (ES) for temporary files and permanent files in order to shorten input/output access time. The operating system of the Hitachi S-3800/480 is the Hitachi VOS3/AS/HAP, which is compatible with that of the Hitachi general purpose server except vector operations.

An automated Cassette Magnetic Tape (CMT) Library (CMTL) unit is used to store NWP results. The library unit contains about 5000 volumes of CMT (compatible with IBM3490 type, 36 tracks) and CMTL has a capacity of 2.1Tbytes as each volume of CMT has a capacity of 400 Mbytes.

2. Objective Analysis System

Four kinds of objective analyses for meteorological fields are performed: global analysis for the Global Spectral Model (hereafter GSM9603 or GSM), regional analysis for the Regional Spectral Model (RSM9603 or RSM), typhoon analysis for the Typhoon Model (TYM9603 or TYM), and upper stratospheric analysis for the one-month forecast model. Specifications of the JMA analysis system are summarized in Table 1. The global and regional analyses are carried out by using a multi-variate three-dimensional optimum interpolation method on the grid of the corresponding forecast models. The typhoon analysis is basically the same as the global analysis except that the typhoon analysis is performed only in the western North Pacific and adjacent areas. The upper stratospheric analysis is carried out by using a function fitting method on pressure surfaces above 10 hPa.


Table 1. Specifications of the JMA objective analysis system
Analysis model Global analysis Regional analysis Typhoon analysis Upper stratospheric analysis
Analysis time 00, 06, 12, 18 UTC 00, 12 UTC 06, 18 UTC 00, 06, 12, 18 UTC
Data cut-off time
Early analysis:
2.5h (00 UTC),
3.0h (12 UTC)
3.0 h (00 UTC)
3.5 h (12 UTC)
1.5 h
Early analysis:
3.0h (12 UTC)
Cycle analysis:
13.0h (00 UTC),
7.5h (06 UTC),
12.0h (12 UTC),
6.5h (18 UTC)
Cycle analysis:
13.0h (00 UTC),
7.5h (06 UTC),
12.0h (12 UTC),
6.5h (18 UTC)
Grid system Gaussian grid Lambert projection Same as global analysis Latitude-longitude
Horizontal resolution 0.5625degree 20 km at 60degreeN
and 30degreeN
0.5625degree 2.5degree
Number of grid points 640 x 320 257 x 217 321 x 217 144 x 73
Vertical levels surface + 30 levels
up to 10 hPa
surface + 36 levels
up to 10 hPa
Same as global analysis 7, 5, 3, 2, 1, 0.4 hPa
Analysis variables Ps, Z, u, v, T, RH Ps, Z, u, v, T, RH Ps, Z, u, v, T, RH Z, T
First guess 6 hour forecast by global model 12 hour forecast by regional model 6 hour forecast by global model Analysis 6 hours before
Analysis method 3-dimensional optimum interpolation on model levels 3-dimensional optimum interpolation on model levels Same as global analysis Function fitting on pressure surfaces

#Global sea surface analysis on 1degree x 1degree grids is carried out everyday.




Global Sea Surface Temperature (SST) of 1degree latitude/longitude resolution is analyzed every day. Optimum interpolation (OI) is employed as its main scheme with the use of monthly mean of NCEP SST analysis for twelve years as for the first guess.

3. Forecast Models

Specifications of the main operational NWP models at JMA, the Global Spectral Model (GSM), the Regional Spectral Model (RSM), the Typhoon Model (TYM) and the One-month Forecast Model are listed in Table 2.


Table 2. Outline of the numerical weather prediction models at JMA
Global Spectral Model Regional Spectral Model Typhoon Model One-month Forecast Model
Objectives Medium-range forecast,
Typhoon track forecast
Short-range forecast Typhoon track forecast One-month forecast
Forecast Domain Global East Asia & adjacent seas Relocatable in the western North Pacific Global
Grid Size & Number of Grids 0.5625degree / 640x320 20 km / 257x217 40 km / 163x163 1.875degree / 192x96
Map Projection . Lambert Mercator (Lambert) for typhoon south (north) of 20 KN at T+0 h .
Vertical Levels 30 levels up to 10 hPa 36 levels up to 10 hPa 15 levels up to 30 hPa 30 levels up to 1 hPa
Major Physical Processes . . . .
Radiation (Solar) Every hour Every hour Every hour Every hour
(Infrared) 3 hourly Every hour Every hour 3 hourly
Cumulus Convection Arakawa-Schubert (1974) Arakawa-Schubert (1974) Arakawa-Schubert (1974) Arakawa-Schubert (1974)
Mid-level Convection Mass-flux type Moist convective adjustment Moist convective adjustment Mass-flux type
Grid Scale Cond. & Evap. Included Included Included Included
Land Surface SiB Constant beta Constant beta SiB
Forecast Hours 84 h (0000 UTC),
192 h (1200 UTC)
51 h (0000, 1200 UTC) 78 h (0600, 1800 UTC) 30 days (5 runs twice a week)
Initial Fields Global analysis Regional analysis Typhoon Interpolated from global analysis
Lateral Boundary Conditions . 0-51 h global forecast from T+0 h 6-84 h global forecast from T-6 h .
Forecast Phenomena Large- and synoptic-scale Synoptic- and meso-scale Typhoon Long-term large-scale



The high resolution version (T213L30) of GSM produces daily 8-day forecast to support operational one-week forecasts and 3-day forecasts of typhoon track. The model provides the time-dependent lateral boundary conditions to RSM and TYM. A low-resolution version (T63L30) of GSM is used to make an ensemble one-month forecast with ten members every week.

RSM is the main model for short-range weather forecast and supports objective forecast guidances processed with Model Output Statistics (MOS), Kalman Filter and Neural Network.

TYM has an identical dynamical framework and physical processes to those of RSM. Typhoon prediction is performed four times a day whenever a typhoon exists in the western North Pacific: GSM makes 84- and 192-hour predictions from 0000 and 1200 UTC, respectively, and TYM does 78-hour ones from 0600 and 1800 UTC. Both models support the activities of the RSMC (Regional Specialized Meteorological Center) Tokyo Typhoon Center of WMO (World Meteorological Organization).

4. Objective Forecast Guidances

a. Kalman Filter

The Kalman Filtering technique is used to prepare short-range forecast (up tp 51 hours) guidance products, such as probability of precipitation, 3 hour precipitation amount, surface air temperature and surface wind speed. The guidance products of precipitation and surface air temperature are calculated at every grid point with 20km intervals. The guidance products of surface wind speed is prepared for local forecast at aerodromes.

b. Neural Network

Three layered neural networks are also used to produce some of the short range-forecast guidances. Wmap (categorized weather), PoHP (probability of heavy precipitation), PoT (Probability of thunderstorm) and RHmin (daily minimum relative humidity) are prepared by neural network using RSM predictions. Wmap is calculated on every square grid of 20km x 20km area including Japan. PoHP and PoT are calculated for every local meteorological observatory's responsibility area. Daily minimum relative humidity (RHmin) is calculated at 71 meteorological observatories.

c. One-week forecast guidance

One-week forecast guidances are produced for PoP (probability of pricipitation (1mm/day)), MRR (daily area mean precipitation amount), RoCA (ratio of cloud amount), Tmax (daily maximum surface temperature), Tmin (daily minimum surface temperature) and Tmean (daily mean surface temperature). The Model Output Statistics (MOS) technique is used to make PoP, MRR and RoCA and the Kalman Filtering technique is used to make the temperature guidances. The development is under way to change the method of producing PoP, MRR and RoCA to Kalman filtering technique.

5. Tracer Transprot Model

JMA has established a RSMC for environmental emergency response (EER). When radioactive accidents occur, tracer transport model will be run to provide trajectories, low-level concentrations and depositions for the RA-II members.

6. Support to foreign countries and meteorological business

The member countries of Typhoon Committee of WMO can have an access to RSMC data server of JMA through the internet and international ISDN line and utilize NWP products of JMA for their operational forecasting.

NWP products including GPV data are provided to weather companies through MBSC (Meteorological Business Support Center) and utilized on online and offline basis.

7. Future plan

(1) It is planned to start, from the spring of 1998, the operation of very-short-range forecast system with the use of NWP model with 10km horizontal resolution. The main object of the system is to support forecasters by providing the information of severe mesoscale phenomena such as heavy rainfall and strong wind up to 12-h from the initial time (06 and 18UTC). The forecast model of the system is the same as that of RSM except for horizontal resolution and it adopts a unique data assimilation method. Before forecast 3-h prerun is done starting from the field obtained by doing the interpolation of 6-h forecast of RSM and physical initialization (PI). During this period, 1-h forecast and physical initialization (PI) with the use of observed precipitation data are repeated at 1-h interval. After the 3-h prerun the objective analysis based on the optimal interpolation method and final PI is done and the forecast is started. The lateral boundary data are provided by RSM. The development to upgrade the system is ongoing: nonhydrostatic model, prognostic cloud scheme and 4-D VAR.

(2) Ocean-atmosphere coupled model will be operated from 1998 to predict El Nino event.

(3) Ensemble medium-range forecats will be put into operation in 1999 to supprot one-week forecast.

(4) The 3-D VAR for the global analyasis will be implemented in 1999.

References

NPD/JMA, 1997: Outline of the Operational Numerical Weather Prediction at the Japan Meteorological Agency, Appendix to Progress Report on Numerical Weather Prediction, 126pp.


Editor's Note: More information on the operational numerical models of JMA can be found here.

Dr. Masaomi Nakamura
Numerical Prediction Division
Japan Meteorological Agency
1-3-4 Otemachi Chiyodaku
Tokyo, 100, Japan
Fax: 81-3-3211-8407
E-mail: nkmnaka@npd.kishou.go.jp