Japan Meteorological Agency

Transcription

1 JOINT WMO TECHNICAL PROGRESS REPORT ON THE GLOBAL DATA PROCESSING AND FORECASTING SYSTEM AND NUMERICAL WEATHER PREDICTION RESEARCH ACTIVITIES FOR 2007 Japan Meteorological Agency 1. Summary of highlights (1) Assimilation of ATOVS radiance data from AP RARS (Asia Pacific Regional ATOVS Retransmission Service) and that from EARS (EUMETSAT Advanced Retransmission Service) in the global analysis started on 22 February and 2 August 2007, respectively. (see (5)) (2) ATOVS radiance data from NOAA18 were introduced to the global analysis on 18 April 2007 and those from MetOp on 21 November 2007 in addition to those from NOAA15 17 and Aqua ATOVS radiances. (see (5)) (3) The resolution of inner loop model used in the global 4D VAR analysis was increased from T106L40 to T159L60 on 21 November (see ) (4) The resolution of GSM was increased from TL319L40 to TL959L60 with a topmost level raised from 0.4 hpa to 0.1 hpa, along with a model upgrade on 21 November (see ) (5) The resolution of One week EPS model was increased from TL159L40 to TL319L60 with a topmost level raised from 0.4 hpa to 0.1 hpa, and the SVs method was introduced to create perturbations for the One week EPS, in place of the Breeding of Growing Mode (BGM) method, on 21 November (see ) (6) Assimilation of radial velocities of Sendai, Niigata and Nagoya Doppler radars in the mesoscale analysis started on 2 August (see ) (7) The forecast time of MSM was extended from 15 hours to 33 hours at 03, 09, 15 and 21UTC initial time and the MSM0603 was upgraded into model (MSM0705) on 16 May (see ) (8) The analysis scheme of Hourly Analysis was upgraded from three dimensional optimum interpolation method to three dimensional variational method on 22 March (see (2)) (9) A preliminary operation of Typhoon EPS started on 31 May (see ) (10) In accordance with the extension of forecast time range and the upgrade of MSM, the guidances for disaster prevention and for the aviation forecast that uses the MSM output were updated. Now elements were added, and the specification of some elements were changed. The guidance for the aviation forecast that supports both TAF S (Aerodrome Forecast Short) and TAF L (Aerodrome Forecast Long) are now derived from only the MSM output. (see (1)) (11) Resolution of GSM was upgraded and RSM was retired on November 21, Therefore, all guidances for the daily weather forecast to 3 days are therefore derived from the new GSM output. (see (1)) 2. Equipment in use at the GDPFS in JMA The Computers for numerical analysis and prediction of JMA were upgraded on 1 March, The computers are located at the Headquarters in central Tokyo and Office of Computer Systems Operations in Kiyose City, which is about 30 km west from the Headquarters. The two sites are connected via a wide area network. Major features of the computers are listed in Table 2 1. Table 2 1 Major features of computers Supercomputers (Kiyose) HITACHI SR11000/K1 Number of nodes 160 (80 nodes x 2 subsystems) Processors 2560 POWER5+ processors (16 per node) 1

2 Performance 10.75TFlops per subsystem (134.4GFlops per node) Main memory 5.0 TB per subsystem (64 GB per node) Attached storage* HITACHI SANRISE 9585V (6.8 TB per subsystem) Data transfer rate 8.0 GB/s (one way), 16.0 GB/s (bi directional) (between any two nodes) Operating System IBM AIX 5L Version 5.2 UNIX servers (Kiyose) HITACHI EP8000/570 Number of nodes 3 Performance 85 SPECint rate 2000 per node Main memory 16 GB per node Attached storage* HITACHI SANRISE 9533V (1.4TB) Operating System IBM AIX 5L Version 5.2 * the dedicated storage for the supercomputers / servers. Workstations (Kiyose) HITACHI HA8000/130W Number of nodes 18 Performance 18.2 SPECint rate 2000 per node Main memory 4.0GB per node Operating System Red Hat Enterprise Linux ES release 3 Storage Area Network (Kiyose) HITACHI SANRISE 9585V Total storage capacity 22.9 TB Automated Tape Library (Kiyose) StorageTek PowderHorn 9310 Total storage capacity 0.9 PB Tape drives StorageTek 9940 B (6 drives) Workstations (HQ) HITACHI HA8000/130W Number of nodes 11 Performance 10.7 SPECint rate 2000 per node Main memory 1.0 GB per node Operating System Red Hat Enterprise Linux ES release 3 Network Attached Storage Total storage capacity 3.0 TB (HQ) TB (Kiyose) Wide Area Network (between HQ and Kiyose) Network bandwidth 200 Mbps (two independent 100 Mbps WAN) 3. Data and Products from GTS in use 3.1 Observations The summary of data received through the GTS and other sources and processed at JMA is given in Table 3 1. Table 3 1 Number of observation reports in use SYNOP/SHIP 61000/day BUOY 30000/day TEMP A/PILOT A 1700/day TEMP B/PILOT B 1100/day TEMP C/PILOT C 1200/day TEMP D/PILOT D 1000/day AIREP/AMDAR /day PROFILER 1700/day 2

3 SATOB (WIND) ATOVS SSMI QSCAT AQUA/AMSR E AIRS/AMSU Metop/AMSU A Metop/MHS NOAA/AMSU A NOAA/AMSU B NOAA/MHS TRMM/TMI /day /day /day /day /day /day /day /day /day /day /day /day 3.2 Forecast products GPV products of global prediction model of ECMWF, NCEP, UKMO are used for internal reference and monitoring. 4. Forecasting system 4.1 System run schedule and forecast ranges Table summarizes the system run schedule and forecast range. Table Schedule of the analysis and forecast system Initial Run schedule Model Time (UTC) (UTC) Global Analysis/Forecast Mesoscale Analysis/Forecast Typhoon Ensemble Forecast Ocean Wave Forecast Storm Surge Forecast , , Forecast Range (hours)

4 Medium range Ensemble Forecast Medium range forecasting system (4 10 days) Data assimilation, objective analysis and initialization In operation (1) Global analysis and initialization for the GSM A four dimensional variational (4D VAR) data assimilation method is employed for the analysis of the atmospheric state for the JMA Global Spectral Model (GSM). The control variables are relative vorticity, unbalanced divergence, unbalanced temperature, unbalanced surface pressure and the natural logarithm of specific humidity. In order to improve the computational efficiency, an incremental method is adopted, in which the analysis increment is evaluated first at a lower horizontal resolution (T159) and then it is interpolated and added to the first guess field at the original resolution (TL959). Global analyses are performed at 00, 06, 12 and 18 UTC. An early analysis with short cut off time is performed to prepare initial conditions for operational forecast, and a cycle analysis with long cut off time is performed to keep the quality of global data assimilation system. The specifications of the atmospheric analysis schemes are listed in Table The global land surface analysis system has been operated since March 2000 to provide initial conditions of land surface parameters for the GSM used in the medium range forecasts. The system includes the daily global snow depth analysis to obtain an appropriate initial condition of snow coverage and depth. Daily global snow depth analysis system is described in Table For initialization of atmospheric states of the GSM, the incremental non linear normal mode initialization (NNMI) and the vertical mode initialization (Murakami and Matsumura 2004) were introduced in February 2005, while the GSM used in the ensemble prediction system employs the conventional NNMI. The non linear normal mode initialization with full physical processes is applied to the first five vertical modes. A spatial resolution of the global analysis was upgraded from TL319L40 to TL959L60 in November The cut off time of cycle analysis was shortened by 20 minutes to save computational time for data assimilation. Table Specifications of global analysis Cut off time 2.33 hours for early run analyses at 00, 06, 12 and 18 UTC hours for cycle run analyses at 00 and 12 UTC 5.25 hours for cycle run analyses at 06 and 18 UTC Initial Guess 6 hour forecast by GSM Grid form, resolution and number of grids Gaussian grid, degree, 1920 x 640 Levels 60 forecast model levels up to 0.1 hpa + surface Analysis variables Wind, surface pressure, specific humidity and temperature Data Used 4

5 SYNOP, SHIP, BUOY, TEMP, PILOT, Wind Profiler, AIREP, NOAA/ATOVS radiances, MetOp/ATOVS radiances, Aqua/AMSU A radiances, atmospheric motion verctors (AMVs), from MTSAT 1R, GOES, METEOSAT, MODIS polar AMVs, SeaWinds, Mirowave imager radiometer radiances (AMSR E, TMI, SSM/I) and Australian PAOB Table Specifications of Snow Depth analysis Methodology Two dimensional Optimal Interpolation scheme Domain and Grids Global, 1 x 1 degree equal latitude longitude grids First guess USAF/ETAC Global Snow Depth climatology (Foster and Davy, 1988) Data used SYNOP snow depth data Frequency Daily (2) Typhoon Bogussing of the global analysis For typhoon forecasts over the western North Pacific, typhoon bogus data are generated to represent typhoon structure accurately in the initial field of forecast models. They are made up of artificial sea surface pressure, temperature and wind data around a typhoon. The structure is axiasymmetric. At first, symmetric bogus data are automatically generated based on the central pressure and 30 kt wind speed radius of the typhoon. The axi asymmetric bogus data are then generated by retrieving asymmetric components from the first guess field. Finally, those bogus profiles serve as pseudo observation data in the global analysis Research performed in this field (1)Development of 4D VAR with Semi Lagrangian scheme Development of revised 4D VAR data assimilation system has been just started to incorporate a two time level Semi Lagrangian advection scheme that has been used in operational GSM and a reduced Gaussian grid which is planned to be used in it. As a first step, 3D VAR system which employs the reduced Gaussian grid was developed. It contains all of observation operators and variational bias correction technique used in operational 4D VAR system. Some single observation experiments and short range cycle experiments have been conducted. (T. Kadowaki) (2) Improvement of data selection scheme in quality control The JMA quality control (QC) system uses relatively sparse vertical selection of atmospheric sounding observations by radiosondes, wind profilers and airplanes, because these observation data is mainly selected at around the mandatory levels: 1000, 925, 850, 700, 500, etc. Since Global Spectral Model (GSM) and its assimilation scheme adopted a higher vertical resolution (60 levels), the vertical thinning methods for above observations should be improved to make available more data. Assimilation experiments were done to examine usefulness of vertically denser observation data. When adding seven more levels, around which the data are thinned, in the troposphere, the results (in surface pressure, 500 hpa geopotential height and so on) are improved for the forecast time (FT) 48 to 216 hours but not for FT 0 to 24. The denser usage of observation data would allow the forecasts to fit more closely to observations. (M. Sakamoto) (3) Development of VarQC method for global analysis system Variational Quality Control (VarQC) method was developed and tested for JMA global analysis system. In this test, VarQC method was applied only to the conventional observation data. The data selection strategies in the pre process were modified so that almost the same amount of the data were used in the test as in the operational analysis. It is confirmed that the use of VarQC has 5

6 generally positive impact on the global forecast, although mean error of typhoon position forecasts increased. Further research will be required to reduce the typhoon track forecast error. (T. Iriguchi) (4) Development of LETKF Further progress has been made in the development of the local ensemble transform Kalman filter (LETKF) with the JMA global spectral model (GSM) at TL159L40. Several major upgrades have been made to the GSM LETKF system, which includes 1) removing local patches as in Miyoshi et al. (2007), 2) applying an additive covariance inflation method as in Whitaker et al. (2007), 3) improving parallel efficiency, and 4) applying an adaptive bias correction for satellite radiances. Recent inter comparisons with the operational 4D VAR system indicated comparable performance, although some problems have been found in assimilating satellite radiances within LETKF. If satellite radiances are not assimilated, LETKF outperformed 4D VAR overall. We will continue the development, especially focusing on improving treatments of satellite radiances. We also plan to adapt the experimental system to JMA's next generation GSM at TL319L60 with the reduced Gaussian grid. (T. Miyoshi) (5) Addition of new satellite data ATOVS sounding radiance data from NOAA18 were introduced on 18 April, 2007 and those from MetOp satellites were introduced in 21 November in addition to those from NOAA15 17 and Aqua ATOVS radiances on 18 April and on 21 November 2007, respectively. These new satellite data showed the positive impacts on forecast skills in terms of the 500hPa geopotential height. JMA also started using AP RARS/EARS ATOVS data that are directly received at 21 ground stations in Asia, Australia, Europe and North America, and retransmitted in short time to NWP centers. AP RARS and EARS increased the ATOVS radiance data used in the early analysis and improved the analysis quality. AP RARS: Asia Pacific Regional ATOVS Retransmission Service EARS: EUMETSAT Advanced Retransmission Service Model In operation The specifications of the operational Global Spectral Model (GSM0711; TL959L60) are summarized in Table In November 2007, the resolution of the GSM was increased from TL319L40 to TL959L60 with a topmost level raised from 0.4 hpa to 0.1 hpa. The numerical integration scheme was renewed from leap flog scheme to two time level scheme. A new highresolution analysis of sea surface temperature and sea ice concentration started to be used as ocean surface boundary conditions. A convective triggering scheme was introduced into the cumulus convection parameterization. A new 2 dimensional aerosol climatology derived from satellite observations started to be used for the radiation calculation (Iwamura and Kitagawa 2008; Nakagawa 2008). JMA runs the GSM four times a day (00, 06, 18UTC with forecast time of 84 hours and 12UTC with that of 216 hours). Table Specifications of Global Spectral Model for nine day forecasts Basic equation Primitive equations Independent variables Latitude, longitude, sigma pressure hybrid coordinates and time Dependent variables Surface pressure, winds (zonal, meridional), temperature, specific humidity and cloud water content 6

7 Numerical technique Integration domain Horizontal resolution Vertical resolution Time step Orography Gravity wave drag Horizontal diffusion Vertical diffusion Planetary boundary layer Treatment of sea surface Land surface and soil Radiation Convection Cloud Spectral (spherical harmonics basis functions) in horizontal, finite differences in vertical Two time level, semi Lagrangian, semi implicit time integration scheme Hydrostatic approximation Global in horizontal, surface to 0.1 hpa in vertical Spectral triangular 959 (TL959), roughly equivalent to x degrees lat lon 60 unevenly spaced hybrid levels 10 minutes GTOPO30 dataset, spectrally truncated and smoothed Longwave scheme (wavelengths > 100 km) mainly for stratosphere Shortwave scheme (wavelengths approximately 10 km) only for troposphere Linear, fourth order Stability (Richardson number) dependent, local formulation Mellor and Yamada level 2 turbulence closure scheme Similarity theory in bulk formulae for surface layer Climatological sea surface temperature with daily analyzed anomaly Climatological sea ice concentration with daily analyzed anomaly Simple Biosphere (SiB) model Two stream with delta Eddington approximation for shortwave (hourly) Table look up and k distribution methods for longwave (every three hours) Prognostic Arakawa Schubert cumulus parameterization Prognostic cloud water, cloud cover diagnosed from moisture and cloud water Research performed in this field (1) Development of the Cumulus Parameterization Scheme of the Global Spectral Model In November 2007, the convection triggering mechanism proposed by Xie and Zhang (2000)and dynamic CAPE generation rate (DCAPE) was introduced to the cumulus parameterization scheme of the operational GSM to improve the rainfall forecast (Nakagawa 2005). The introduction of DCAPE reduced the tendency of GSM to overestimate weak precipitation areas especially from local noon to late afternoon. The calculation procedure of DCAPE for the GSM was revised to consider the effect of wind crossing the isobar at the surface more precisely, which is not taken into account sufficiently in the previous version (Nakagawa 2008). An excessive limitation on cumulus upward mass flux from redundant vertical CFL condition was also removed. The equitable threat score for precipitation forecasts with the revised GSM against raingauge observations over Japan was superior to that with the previous one. The revision of the cumulus parameterization scheme also reduced the typhoon positional error substantially. JMA plans to implement this revision into operational GSM in January (M. Nakagawa) (2) Improvement of Aerosol Climate Data Aerosol climate data used in a radiation scheme of the GSM were changed in November The new climate data are monthly mean total aerosol optical depths based on the actual distribution derived from satellite observations measured by MODIS and TOMS. The new data represent more realistic seasonal changes of global aerosol distribution compared with the simple 7

8 two types (over land and sea) of constant data previously used. It is confirmed that overestimates of incoming shortwave radiation flux at the surface are decreased with the new data especially over Asia and the northern Africa where dense aerosols are observed. (S. Murai) Operationally available NWP Products The following model output products from GSM are disseminated through the JMA radio facsimile broadcast (JMH), GTS, RSMC Tokyo Data Serving System (RSMS DSS) and the WMO Distributed Data Bases project server (DDB). Table List of facsimile charts transmitted through GTS and JMH Symbols for contents: Z: geopotential height, ζ: vorticity, T: temperature, D: dewpoint depression, ω: vertical velocity, W: wind speed by isotach, A: wind arrows, P: sea level pressure, R: rainfall. Model Area Contents and Level Forecast Initial Availa Hours Time bility GSM Far East 500 hpa (Z, ζ) Analysis 00/12UTC GTS 24, 36 00/12UTC GTS/JMH 500 hpa (T), 700 hpa (D) 24, 36 00/12UTC GTS/JMH 700 hpa (ω), 850 hpa (T, A) Analysis 00/12UTC GTS 24, 36 00/12UTC GTS/JMH Surface (P, R, A) 24, 36 00/12UTC GTS/JMH East Asia 300 hpa (Z, T, W, A) Analysis 00UTC GTS 500 hpa (Z, T, A) Analysis 00/12UTC GTS/JMH 500 hpa (Z, ζ) 48, 72 00/12UTC GTS 700 hpa (Z, T, D, A) Analysis 00/12UTC GTS 700 hpa (ω), 850 hpa (T, A) 48, 72 12UTC GTS 850 hpa (Z, T, D, A) Analysis 00/12UTC GTS/JMH Surface (P, R) 24, 48, 72 00/12UTC GTS/JMH 96, UTC JMH Asia 500 hpa (Z, ζ) 96, 120, 144, 12UTC GTS 850 hpa (T), Surface (P) 168, 192 Asia Pacific 200 hpa (Z, T, W), Analysis 00/12UTC GTS Tropopause (Z) 250 hpa (Z, T, W) Analysis, 24 00/12UTC 500 hpa (Z, T, W) 00/12UTC Northern 500 hpa (Z, T) Analysis 12UTC GTS Hemisphere North 200 hpa (streamline) Analysis, 24, 00/12UTC GTS West Pacific 850 hpa (streamline) 48 00/12UTC 500 hpa (Z, Z anomaly to climatology) 8

9 Table List of GPV products (GRIB) transmitted through GTS, RSMC DSS and DDB Symbols for contents: Z: geopotential height, U: eastward wind, V: northward wind, T: temperature, D: dewpoint depression, H: relative humidity, ω: vertical velocity, ζ: vorticity, ψ: stream function, χ: velocity potential, P: sea level pressure, R: rainfall. Prefixes µ and σ stand for average and standard deviation of ensemble prediction results, respectively. Symbols * indicate limitations on forecast hours or initial time as shown in the table below. Model GSM GSM GSM Destination RSMC GTS, RSMC, DDB GTS, RSMC, DDB Area and Whole globe, S 60 N, 60 E 160 W Whole globe, resolution Levels and elements Forecast hours 10 hpa: Z, U, V, T 20 hpa: Z, U, V, T 30 hpa: Z, U, V, T 50 hpa: Z, U, V, T 70 hpa: Z, U, V, T 100 hpa: Z, U, V, T 150 hpa: Z, U, V, T 200 hpa: Z, U, V, T, ψ, χ 250 hpa: Z, U, V, T 300 hpa: Z, U, V, T, H, ω 400 hpa: Z, U, V, T, H, ω 500 hpa: Z, U, V, T, H, ω, ζ 600 hpa: Z, U, V, T, H, ω 700 hpa: Z, U, V, T, H, ω 850 hpa: Z, U, V, T, H, ω, ψ, χ 925 hpa: Z, U, V, T, H, ω 1000 hpa: Z, U, V, T, H, ω Surface: P, U, V, T, H, R 0 84 every 6 hours and every 12 hours Except analysis Initial times 00UTC, 06UTC, 18UTC and 12UTC 10 hpa: Z, U, V, T 20 hpa: Z, U, V, T 30 hpa: Z, U, V, T 50 hpa: Z, U, V, T 70 hpa: Z, U, V, T 100 hpa: Z, U, V, T 150 hpa: Z, U, V, T 200 hpa: Z, U, V, T, ψ, χ 250 hpa: Z, U, V, T 300 hpa: Z, U, V, T, D 400 hpa: Z, U, V, T, D 500 hpa: Z, U, V, T, D, ζ 700 hpa: Z, U, V, T, D, ω 850 hpa: Z, U, V, T, D, ω, ψ, χ 925 hpa: Z, U, V, T, D, ω 1000 hpa: Z, U, V, T, D Surface: P, U, V, T, D, R 0 84 every 6 hours additional every 24 hours for 12UTC every 6 hours 00UTC, 06UTC, 18UTC and 12UTC 10 hpa: Z*, U*, V*, T* 20 hpa: Z*, U*, V*, T* 30 hpa: Z, U, V, T 50 hpa: Z, U, V, T 70 hpa: Z, U, V, T 100 hpa: Z, U, V, T 150 hpa: Z*, U*, V*, T* 200 hpa: Z, U, V, T 250 hpa: Z, U, V, T 300 hpa: Z, U, V, T, D* 400 hpa: Z*, U*, V*, T*, D* 500 hpa: Z, U, V, T, D* 700 hpa: Z, U, V, T, D 850 hpa: Z, U, V, T, D 1000 hpa: Z, U*, V*, T*, D* Surface: P, U, V, T, D, R 0 72 every 24 hours and every 24 hours for 12UTC for 12UTC Except analysis * Analysis only 00UTC and 12UTC 00UTC only Model Destination Area and resolution Levels and elements GSM RSMC 20 S 60 N, 80 S 160 W hpa: Z, U, V, T 150 hpa: Z, U, V, T 200 hpa: Z, U, V, T 250 hpa: Z, U, V, T 300 hpa: Z, U, V, T 500 hpa: Z, U, V, T, D, ζ 700 hpa: Z, U, V, T, D, ω 850 hpa: Z, U, V, T, D, ω Surface: P, U, V, T, D, R Forecast 0 36 every 6 hours, hours 48, 60, and 72 Initial times 00UTC and 12UTC 9

10 4.2.4 Operational techniques for application of NWP products (1) Guidance for forecast The application techniques to both the medium and short range forecasting system are described in (1) Ensemble Prediction System (EPS) In operation JMA operates One week EPS once a day at 12 UTC up to nine days ahead. The specifications of One week EPS are shown in Table One week EPS is composed of one control forecast and 50 perturbed forecasts. Initial perturbations are generated by Singular Vectors (SVs) method (Buizza and Palmer, 1995). A tangent linear and adjoint model used for the SVs calculation are the same as those used in the four dimensional variational data assimilation system for the operational GSM. The moist total energy norm (Ehrendorfer, 1999) is employed for the metrics of perturbation growth. A forecast model used in One week EPS is a low resolution version of GSM (see ). In November 2007, One week EPS was upgraded. In initial perturbation method, the SVs method was introduced into operational system in place of the Breeding of Growing Mode (BGM) method (Toth and Kalnay, 1993). In the forecast model, horizontal and vertical resolution is enhanced from T L 159 to T L 319 and from 40 levels to 60 levels, respectively. Additionally, the improvement of deep convection scheme was implemented. Table Specifications of One week EPS Spectral triangular truncation 319 with linear grid Horizontal Resolution, Grid Size (T L 319), x degree in latitude and longitude Vertical Resolution (model top) 60 levels (0.1 hpa) Forecast Range (Initial Time) 216 hours (12 UTC) Time Step 1200 seconds Initial Field Truncated analysis field of T L 959 into T L 319 Ensemble Size 51 members (50 perturbed forecast and 1 control forecast) Perturbation Generation Method Singular Vectors method Inner Model Resolution Spectral triangular truncation 63 (T63), 40 levels Norm Moist Total Energy Targeted Area Northern Hemisphere Tropics (30N 90N) (20S 30N) Perturbation Physical Process *Simplified physics **full physics Optimization Time 48 hours 24 hours Evolved SV Used Used Number of Perturbations 25 members 25 members *Simplified physics: Initialization, horizontal diffusion, surface turbulent diffusion and vertical turbulent diffusion. **Full physics: In addition to the simplified physics processes, gravity wave drag, long wave radiation, clouds and large convection and cumulus convection Research performed in this field (1) Introduction of stochastic physics 10

11 One of the issues in the current JMA one week EPS is that the ensemble spread in summer is notably smaller than the root mean square error of the ensemble mean forecast for the later half of the forecast period. Therefore, the JMA planned to introduce stochastic physics to treat the effect of model uncertainties (R. Buizza 1999). In this technique, a random perturbation is added to physical tendencies at every time step. (H. Yonehara) 4.3 Short range forecasting system (0 72 hrs) Data assimilation, objective analysis and initialization In operation (1) Mesoscale analysis The specifications of the mesoscale analysis schemes are listed in Table A four dimensional variational (4D VAR) data assimilation method has been employed since 19 March 2002 for the analysis of the atmospheric state for the JMA Meso Scale Model (MSM) with a six hour assimilation window. Radar Raingauge Analyzed Precipitation data in addition to conventional data are used for assimilation. The control variables are surface pressure, temperature, unbalanced wind and specific humidity. In order to improve the computational efficiency, an incremental method is adopted in which the analysis increment is evaluated at a lower horizontal resolution (20 km) and then it is interpolated and added to the first guess field at the original resolution (10 km). JMA started assimilating radial velocities of Sendai, Niigata and Nagoya Doppler radars in August Table Specifications of mesoscale analysis Cut off time (mesoscale) 50 minutes for analyses at 00, 03, 06, 09, 12, 15, 18 and 21 UTC Initial Guess (mesoscale) Six hour forecast by MSM Grid form, resolution and number of grids (mesoscale) Lambert projection, 10 km at 60N and 30N, 361 x 289, grid point (1, 1) is at north west corner and (245, 205) is at (140E, 30N) Levels (mesoscale) 40 forecast model levels up to 10 hpa + surface Analysis variables Wind, surface pressure, specific humidity and temperature Data Used SYNOP, SHIP, BUOY, TEMP, PILOT, Wind Profiler, Doppler radar (radial velocity), AIREP, ATOVS SATEM, ATOVS BUFR, AMVs from MTSAT 1R, SeaWinds, Mircrowave imager radiometer retrievals (AMSR E, TMI and SSM/I) and radarraingauge analyzed precipitation (2) Typhoon Bogussing of the mesoscale analysis For typhoon forecasts over the western North Pacific, typhoon bogus data is generated to represent typhoon structure accurately in the initial field of forecast models. They are made up of artificial sea surface pressure, temperature and wind data around a typhoon. The structure is axiasymmetric. At first, symmetric bogus data is automatically generated based on the central 11

12 pressure and 30 kt wind speed radius of the typhoon. The axi asymmetric bogus data is then generated by retrieving asymmetric components from the first guess field. Finally, those bogus profiles serve as pseudo observation data in the mesoscale analyses Research performed in this field (1) Development of JMA NHM based variational data assimilation system We have improved the JMA NHM based variational data assimilation system (JNoVA, Honda et al.(2005)) to guarantee the numerical stability. Besides, the control variable transform considering the geostrophic/cyclostrophic balance was introduced. A forecast/analysis system using the JNoVA was also upgraded to the near operational system. To compare the performance of the JNoVA with that of the operational mesoscale 4D VAR (Meso 4D VAR), the one month experimentation has been conducted under almost the same conditions both in summer and in winter. In general, the JNoVA shows the similar performance. However, the degradation of the precipitation forecasts, especially for weak rain in summer, is a crucial issue. After resolving this issue, the JMA will replace the Meso 4D VAR with the JNoVA in 2008 (Honda and Sawada, 2008). (Y. Honda and K. Sawada) (2) Usage of radial velocity data of Doppler radar The JMA started using doppler radar s radial velocity data (Vr) of eight airports and four observatories with the operational mesoscale 4D VAR analysis. JMA plans the use of additional seven observatories and will conduct an investigation to improve the precision of the wind and the rainfall forecast by the increase of observation data. (Y. Ishikawa) (3) Usage of ground based GPS precipitable water vapor Precipitable water vapor (PWV) data, which was obtained from the nationwide GPS (Global Positioning System) network over Japan, operated by the Geographical Survey Institute (GSI), were assimilated into mesoscale 4D VAR on trial. Several forecasts starting from the analyses with GPS PWV made by the mesoscale model were compared with those starting from the analyses without GPS PWV. A remarkable improvement in rainfall forecasts was seen in several cases. A statistical score also showed the positive impact of GPS PWV on rainfall forecast. (Y. Ishikawa) (4) Development of high resolution analysis A three dimensional variational (3D VAR) data assimilation system, based on JNoVA (JMA Nonhydrostatic Model based Variational Data Assimilation System, Honda et al. 2005), and an observation quality control system were developed to generate initial condition for a high resolution simulation of JMA NHM (JMA nonhydrostatic model). An analysis is performed on a wider domain covering eastern Japan at five km resolution, assimilating observations through hourly assimilation cycles. After performing the assimilation cycle for three hours, the analysis field was used as an initial condition of the high resolution local forecast model (LFM) at two km resolution over Kanto area. Result of the forecast is shown in (4). (T. Fujita and H. Kurahashi) (5) Development of a hybrid LETKF 3D VAR anaysis scheme A hybrid local ensemble transform Kalman filter (LETKF, Hunt et al. 2007) and three dimensional variational (3D VAR) analysis scheme (Hamill and Snyder 2000) have been tested as a data assimilation scheme to generate initial perturbations for an ensemble forecasting with five members using JMA NHM, which was planned to run on a domain covering Japan and its surrounding areas at a resolution of about 10 km. In the analysis scheme, data assimilation systems using LETKF (Miyoshi and Aranami 2006) and 3D VAR (based on JNoVA, Honda et al. 2005) are run in parallel, communicating information with each other. The 3D VAR system uses information of flow dependent background error covariance from the ensemble forecasts of the 12

13 LETKF system. On the other hand, in the LETKF system, the ensemble mean of the updated ensemble is replaced with the analysis field from the 3D VAR system. (T. Fujita) Model In operation (1) Regional Spectral Model (RSM) JMA stopped operating the RSM when the horizontal grid spacing of the GSM became 20 km in November The GSM now supports the short range forecasting as well. The specifications of the GSM are listed in Table (2) Meso Scale Model (MSM) In May 2007, the forecast time of the MSM was extended from 15 hours to 33 hours at 03, 09, 15 and 21UTC initial time and the MSM0603 is replaced with the upgraded model (MSM0705). With the retirement of the RSM in November 2007, JMA started nesting of the MSM into the GSM. The specifications of MSM0705 are listed in Table Table Specifications of Meso Scale Model (MSM0705) Basic equations Fully compressible non hydrostatic equations Independent variables Latitude, longitude, terrain following height coordinates and time Dependent variables Momentum components in three dimensions, potential temperature, pressure, mixing ratios of water vapor, cloud water, cloud ice, rain, snow and graupel Numerical technique Finite discretization on the Arakawa C type staggered coordinates, horizontally explicit and vertically implicit time integration scheme, fourth order horizontal finite differencing in flux form with modified advection treatment for monotonicity Projection and grid size Lambert projection, five km at 60N and 30N Integration domain Japan, 721 x 577 grid points Vertical levels 50 (surface to 21.8 km) Forecast time 15 hours from 00, 06, 12, 18UTC 33 hours from 03, 09, 15, 21UTC Forecast phenomena Severe weather Initial fields 4D VAR analysis with mixing ratios of cloud water, cloud ice, rain, snow and graupel derived from preceding forecasts considering consistency with the analysis field of relative humidity MSM runs three hour before the initial time for spin up. Lateral boundary hour forecast by GSM initialized at 00, 06, 12 and 18 UTC for (03, 06), (09, 12), (15, 18) and (21, 00) UTC forecast Orography Mean orography smoothed to eliminate the shortest wave components Horizontal diffusion Linear, fourth order Laplacian + nonlinear damper Targeted moisture diffusion applied to the grid points where excessive updrafts appear Moist processes Three ice bulk cloud microphysics + Kain Fritsch convection scheme Lagrangian treatment for the fall of rain and graupel Radiation (short wave) Two stream with delta Eddington approximation (every 15 minutes) Radiation (long wave) Table look up and k distribution methods (every 15 minutes) 13

14 Cloudiness Gravity wave drag PBL Land surface Surface state Cloud water and cloud cover diagnosed using the partial condensation scheme No parameterization scheme included Improved Mellor Yamada Level 3 Scheme Similarity theory adopted for the surface boundary layer Ground temperature predicted using a four layer ground model Evaporability depending on location and season. Observed SST (fixed during time integration) and sea ice distribution Climatological values of evaporability, roughness length and albedo Snow cover over Japan analyzed every day Research performed in this field (1) Replacement of lateral boundary conditions in the operational meso scale model The operational meso scale model of JMA (MSM) had employed Regional Spectral Model (RSM) as lateral boundary conditions. With the commencement of operating the high resolution 20 km Global Spectral Model (GSM) as the model for short term weather forecast instead of RSM, the lateral boundary condition for MSM was switched to the forecasts provided by 20 km GSM from those by RSM in Nov (Hara et al., 2008). From the view of synoptic scale, some improvements, such as the reduction of errors on geopotential height and temperature, are obtained by employing 20 km GSM forecasts as the boundary conditions for MSM, but undesirable characteristics of 20 km GSM also appears in the forecasts with MSM through the boundary condition, especially in the moist processes. Sometimes dry bias in the middle layer and wet bias in the lower layer, which 20 km GSM has, make the convective parameterization (modified Kain Fritsch scheme) work so actively that overestimated precipitation brought by the convective parameterization can be observed. (T. Hara) (2) Inclusion of a temperature perturbation based on the relative humidity to the Kain Fritsch convective parameterization scheme To represent the effects of subgrid scale convection, the Kain Fritsch (KF) convective parameterization scheme (Kain and Fritsch 1990) is adopted to MSM in addition to cloud microphysics. To identify source layers for convective clouds, the KF scheme utilizes a trigger function based on the temperature at the lifting condensation level and the grid scale vertical velocity (Kain 2004). The KF scheme, applied to the humid climate area of Japan and its surroundings, sometimes fails to initiate parameterized convection when the lowest atmosphere is wet and dynamical forcing is weak. To eliminate this weakness, a temperature perturbation based on the relative humidity, which was originally developed for the High Resolution Limited Area Model by Undén et al. (2002), has been added to the trigger function. The inclusion of the temperature perturbation depending on the relative humidity also improved the forecast of diurnal convective rain. (M. Narita) (3) Replacement of the land surface model into the Simple Biosphere Model In the current land surface model of JMA NHM, the temperature of the soil is forecasted with the heat conductivity of four levels and the soil moisture is predicted with force restore method. The role of vegetation and the tide of snow are not considered in detail. Therefore, the land surface model based on Simple Biosphere Model (Sellers 1986), which has a sophisticated snow model, was introduced to JMANHM. In order to run this land surface model operationally, such adjustment as estimating initial values and the parameter of land surface was done. As a result of introducing SiB, for example, the prediction of the temperature was improved because the prediction of the snow depth became better. (D. Miura) 14

15 (4) Development of a high resolution local forecast model Since 1 June 2006, an experimental run of a high resolution local forecast model (LFM) has been executed as shown in the last report. The LFM can forecast local strong rainfalls by the assimilation of detailed surface observations and can forecast the intensity adequately with high resolution since horizontal grid spacing is 2 km. However, excessive rainfalls in grid point scale had occasionally happened in summer. Therefore, we are currently investigating the cause and the measures, for example, improvement of convective parameterization. Another problem is that the influence of each assimilated observation sometimes extends too widely, which suggests that the statistically derived background error covariance is inappropriate. The background error covariance needs to be revised, for example, to include spatial inhomogeneities and/or anisotropies. (K. Takenouchi) (5) Development of new dynamical core for mesoscale model A new dynamical core for mesoscale model is now developed for next generation numerical weather prediction system. Fully compressible equation is applied as governing equation of this dynamical core. Finite volume method with generalized coordinate system and Runge Kutta time integration scheme are employed to discrete the equation. Idealized test has been successfully conducted. Improvement of dynamical core and implement of physical processes will be performed. (C. Muroi and Y. Aikawa) (6) Development of a mesoscale ensemble prediction system To provide probabilistic forecasts and reliability information of MSM, JMA has started to develop a mesoscale ensemble prediction system (MEPS). A time lagged ensemble prediction system has been constructed using deterministic forecasts of MSM for the purpose of researching its usefulness. The mean, maximum, minimum and spread of predicted variables are generated from the system. From the medium term perspective, we need to construct a new mesoscale ensemble prediction system considering the uncertainty of initial conditions, lateral boundary conditions and forecast model. Currently, we are focusing on the initial conditions and testing two methods. One is a singular vector method, using the tangent linear model and the adjoint model of the JMA nonhydrostatic model (Honda et al., 2005). The other is a downscaled perturbation method using the weekly ensemble forecast. (H. Tsuguchi and K. Ono) Operationally available NWP products See short and medium range forecast products are described in Operational techniques for application of NWP products In operation (1) Guidance for forecast Three types of operational technique are routinely used to derive guidances from NWP model output. The first type is Kalman filter, the second is artificial neural network, and the third is logistic regression. These techniques are applied to grid point values from the GSM (0 84 hour forecast) and the MSM (0 33 or 0 15 hour forecast) output in order to reduce systematic forecast errors or to extract some useful information including that on probabilistic or categorical values in an adaptive manner. 15

16 The Kalman filter technique is applied to derive probability of precipitation, the average precipitation amount in each 20 km square grid, the maximum/minimum temperature and hourly time series temperatures, the maximum and time indicated wind speed and the associated direction at JMA s each surface weather station, from the GSM and the MSM output. This technique is also applied to derive the maximum/minimum temperature and hourly time series temperatures, the maximum and hourly time series wind speed and the associated direction, the average and minimum visibility, and provability of the minimum visibility less than 5,000 meters and 1,600 meters for the aviation weather forecast (TAF) guidance at each airport, from the MSM output. The artificial neural network technique is applied to derive weather category, probability of heavy precipitation, probability of thunderstorm in each 20 km square grid, and the minimum humidity at each of the JMA s SYNOP reporting station, from the GSM output. This technique also constitutes an essential basis for forecasting of the maximum precipitation amount and the snowfall depth. The maximum precipitation forecast is derived by multiplying the average precipitation amount in each forecast area by an optimum ratio, derived from an artificial neural network, of the observed maximum precipitation to the average precipitation amount from model output. The snowfall depth forecast, operationally produced since March 2004, is estimated by multiplying model derived precipitation amount and an artificial neural network derived optimal snow to liquid equivalent ratio, determined by the empirical relation between observed snowfall depth and precipitation. This technique is also applied to derive cloud amount and cloud base height of three layers at the minimum ceiling event for the TAF guidance at each airport, from the MSM output. The above two types of technique produce various guidances up to 84 hours (4 times a day) from the GSM output, and up to 33 hours at most (8 times a day) from the MSM output, in 1 hour, 3 hour, or 6 hour intervals. Guidances for the daily maximum/minimum temperatures and the daily minimum humidity target on forecast of today, tomorrow, the day after tomorrow, and 3 days after. The logistic regression technique is applied to derive provability of thunderstorm for the TAF guidance at each airport, from the MSM output. The TAF guidance that supports TAF S and TAF L produces visibility, cloud amount and cloud base height, wind speed and direction, weather, temperature, and provability of thunderstorm 33 hours ahead at most (8 times a day) in 1 hour or 3 hour intervals, from the MSM output. (2) Hourly Analysis JMA is providing products of hourly analysis with grid spacing of 5 km for real time monitoring of weather condition. It is based on the MSM forecast and observed data, and the latest MSM forecast output is used as a first guess. The product is made every hour within 30 minutes from hourly observation time and is provided to operational forecasters and aviation users. The specifications of the hourly analysis schemes are listed in Table Analysis scheme of the hourly analysis was upgraded from 3D OI to 3D VAR and radial velocity data from Doppler radars is used instead of VVP wind data from March Table Specifications of hourly analysis Cut off time 0.25 hours for analyses Initial Guess 2, 3 or 4 hour forecast by MSM Grid form, resolution and number of grids Lambert projection, 5 km at 60N and 30N, 721 x 577, grid point (1, 1) is at north west corner and (489, 409) is at (140E, 30N) Levels 40 forecast model levels + surface Analysis variables 16

17 Wind, temperature, surface wind and surface temperature Data Used AMeDAS, Wind Profiler (WINDAS), Doppler Radar (velocity data) of JMA, AMDAR (aircraft) and Satellite wind Research performed in this field We are developing new guidances for dew point temperature and gust speed/direction from the MSM output. We are also developing the probabilistic guidance for short range forecasting, such as guidance for probability of precipitation that exceeds the threshold amount from the typhoon ensemble prediction system and the one week ensemble prediction system. 4.4 Nowcasting and Very Short range Forecasting Systems (0 6 hrs) Since 1988, JMA has been routinely operating a fully automated system of precipitation analysis and a very short range forecasting to monitor and forecast local severe weather. In addition to these, JMA has been operating the Precipitation Nowcast since June The system has three products as below: (1) The Radar raingauge Analyzed precipitation (hereafter R/A)*, which is one hour accumulated precipitation based on observation of the radars calibrated half hourly by the raingauge measurements of the Automated Meteorological Data Acquisition System (hereafter AMeDAS) operated by JMA and other available data, such as raingauges of local governments. (2) The Very Short Range Forecast of precipitation (hereafter VSRF), which is a forecast of onehour accumulated precipitation based on extrapolation and prediction of the Meso Scale Model (MSM, See 4.3). The forecast time of VSRF is from one to six hours. (3) Precipitation Nowcast, which is a forecast of 10 minute accumulated precipitation based on extrapolation. The forecast time of Precipitation Nowcast is from 10 to 60 minutes. * Before 15 November 2006, it was called as Radar AMeDAS precipitation Nowcasting system (0 1 hr) In operation The Precipitation Nowcast predicts 10 minute accumulated precipitation by linear extrapolation up to one hour. Initial rainfall intensity distribution is derived from radar data obtained at 10 minute interval, which is calibrated by raingauge observation. Using the movement vector of VSRF, it predicts precipitation distribution by extrapolation within three minutes after the radar observation to support the local weather offices for issuing warnings of heavy precipitation. Table Precipitation nowcasting model Forecast process Linear extrapolation Physical process Simplified orographic dissipation Movement vector Taken from very short range forecasting system Time step 1 minute Grid form Cylindrical equidistant projection Resolution About 1 km Number of grids 2560 x 3360 Initial Calibrated radar echo intensities Forecast time Up to 60 minutes from each initial time (every 10 minutes = 144 times/day) 17

18 Research performed in this field The Precipitation Nowcast uses movement vectors of VSRF which are derived half hourly and focused on a time scale more than one or two hours. For the improvement of the Precipitation Nowcast, it is desirable to derive movement vectors every 10 minutes using a method suitable for a forecast in a time scale less than one hour. We have been developing some simplified methods for computing movement vectors with less time, such as an optical flow method Models for Very Short range Forecasting Systems (1 6 hrs) In operation (1) Radar Raingauge Analyzed precipitation (R/A) Radar Raingauge Analyzed precipitation (R/A) is a precipitation distribution analysis with 1 km resolution and is derived half hourly. Radar data and raingauge precipitation data are used to make R/A. The radar data is intensity data of 35 weather radars of JMA and of the Ministry of Land, Infrastructure, Transport and Tourism (MLIT). The raingauge precipitation data is collected from more than 9300 rain gauges operated by JMA, MLIT and local governments. After collecting these data, each radar intensity data is accumulated to make the one hour accumulated radar precipitation data. Each accumulated radar precipitation data is calibrated with the one hour accumulated raingauge precipitation data. The R/A is the composite of all calibrated and accumulated radar precipitation data. An initial field for extrapolation forecast is the composite of calibrated radar intensity data. (2) Very Short Range Forecast of precipitation (VSRF) The extrapolation forecast and the precipitation forecast from the Meso Scale Model (MSM; see (2)) are merged into the very short range precipitation forecast (VSRF). Merging weight of MSM forecast is nearly zero at one hour forecast and gradually increased with forecast time to a value determined from the relative skill of the MSM forecasts. Table Specifications of extrapolation model Forecast process Extrapolation Physical process Orographic enhancement and dissipation Movement vector Movement of a precipitation system is evaluated by the cross correlation method Time step 2 5 minutes Grid form Oblique conformal secant conical projection Resolution 1 km Number of grids 1600 x 3600 Initial Calibrated radar echo intensities Forecast time Up to six hours from each initial time (every 30 minutes = 48 times/day) The VSRF products are provided at about 20 minutes after the radar observation to support the local weather offices that issue weather warnings for heavy precipitation, and used for forecast calculation of applied products such as Soil Water Index and R/A Runoff Index Research performed in this field 18