WDS/DPFS & NWP Report14, Annex II, P. 1

WDS/DPFS & NWP_Report14, Annex II, p. 1

W O R L D M E T E O R O L O G I C A L O R G A N I Z A T I O N

======

ANNUAL JOINT WMO TECHNICAL PROGRESS REPORT ON THE GLOBAL DATA-PROCESSING AND FORECASTING SYSTEM (GDPFS) INCLUDING NUMERICAL WEATHER PREDICTION (NWP) RESEARCH ACTIVITIES FOR 2014

CHINA, JULY 2015

1.Summary of highlights

1.1 Developments of operational NWP

The operational GRAPES-MESO was upgraded from V3.3 to V4.0 with 10km horizontal resolution and 50 vertical levels in June 2014. A new operational Global Ensemble Prediction System(T639-GEPS)has been operationally implementedsince 7thAug. 2014 in CMA with 15 - day forecasts and 15 ensemble members replacing the old version (T213-based GEPS).A new operational GRAPES-MESO-based Regional Ensemble Prediction System(GRAPES-MEPS) has been operationally implemented since 7 Aug. 2014 with 15km horizontal resolution,72h forecast and 15 members.

1.2 Developments of GRAPES

A one year trial GRAPES-GFS (GRAPES Global Forecast System) was conducted with the horizontal resolution at0.5degree, 60 levelsin vertical and the model top atabout 3hPa. The performance of the trial is comparable to the current operational global spectral model ofCMA.

Some research activities on GRAPES-GFS include: 1) The optimization of the code of GRAPES global 4D-Var to match the new version global model, has improved much the precision and efficiency of GRAPES-4DVAR, and its performance is very similar to GRAPES 3D-Var when only GTS data are assimilated. 2) The development of GRAPES Yin-Yang grid dynamics builtin the frame of the lat-lon grid GRAPES, is almost finished. Several idealized tests are conducted with GRAEPS Yin-Yang grid dynamics and the reasonable results are shown.

An ‘on-demand’ GRAPES-3km (the horizontal resolution is 3km) was established and supported the severe weather forecast in the flood season in 2014 in China. Some results showed that this high resolution GRAPES-Meso could capture the severe rainstorm events in easternChina, and its performance was better than the operational one at 10km-resolution.

2.Equipment in use

There is no change in 2014. The total peak performance of IBM Flex System P460 is 1759 TFlops and the total storage capacity is about 6925TB. Three sets of subsystems of this HPC wereinstalled in Beijing in 2013, in which the peak performance was more than 1PFlops. More details are showed in Table 2.1.

Table 2.1Details ofsub-systems of CMA IBM Flex System and/or P460 HPC Systems

Subsystem / SS1 / SS2 / SS3 / SS4 / SS5 / SS6 / SS7
Site / Beijing / Guangzhou / Shenyang / Shanghai / Wuhan / Chengdu
Peak Performance (TFlops) / 527.10 / 527.10 / 391.69 / 77.24 / 51.80 / 77.24 / 26.35
Storage (TB) / 2109.38 / 2109.38 / 949.22 / 210.94 / 140.63 / 210.94 / 70.31
CPU Cores
（Include I/O nodes） / 18560 / 18560 / 13792 / 2720 / 1824 / 2720 / 928
Memory (GB) / 81792 / 81792 / 57856 / 10752 / 7168 / 10752 / 3584

3.Data and Products from GTS in use

Data from GTS in use are showed in table 3.1 according to one day data used by GRAPES-GFSin a batch experiment in December 2014.

Table 3.1 Number of observation reports in use

Data type / Mean / Data type / Mean / Data type / Mean
SYNOP / 17967 / AIREP/AMDAR / 123782 / NOAA15_AMSUA / 82918
SHIP/BUOY / 5166 / SATOB (WIND) / 86089 / NOAA18_AMSUA / 88896
TEMP / 1335 / AIRS / 83200 / METOPA+B-AMSUA / 32400
METOPA+B-MHS / 2943000
GNSS(including COSMIC) / 2447 / NOAA19-AMSUA / 80086 / METOPA+B-IASI / 64800
METOP-A-ASCAT / 589896

4.Forecasting system

4.1System run schedule and forecast ranges

In IBMFlex Power P460, the operational scheduleis shown in table 4.1.

Table 4.1 Operational Schedule of NWP system in CMA

Systems / Cut-off time (UTC) / Run time (UTC) / Computer used
Global Forecasting System (operational) (T639L60_GSI) / 01:40 (18Z_ASSIM+9HR_FCST) / 01:40～02:40 / IBM Flex P460
03:29 (00Z_ASSIM+240HR_FCST) / 03:29～05:50 / IBM Flex P460
10:00 (00Z_ASSIM+9HR_FCST) / 10:00～11:00 / IBM Flex P460
11:15(06Z_ASSIM+84HR_FCST) / 11:15～13:20 / IBM Flex P460
13:40 (06Z_ASSIM+9HR_FCST) / 13:40～14:40 / IBM Flex P460
15:29(12Z_ASSIM+240HR_FCST) / 15:29～17:50 / IBM Flex P460
22:00 (12Z_ASSIM+9HR_FCST) / 22:00～23:00 / IBM Flex P460
23:45 (18Z_ASSIM+84HR_FCST) / 23:45～02:00 / IBM Flex P460
09:10 (00Z_ASSIM+240HR_FCST) / 09:10～10:25 / IBM Flex P460
Global Forecasting System / 12:00 (00Z_ASSIM. +6HR_.FCST) / 12:00～12:25 / IBM Flex P460
(GRAPES－GFS1.4) / 17:30(06Z_ ASSIM +6HR_.FCST) / 17:30～17:50 / IBM Flex P460
21:10(12Z_ASSIM.+240HR_FCST) / 21:10～22:30 / IBM Flex P460
23:30(12Z_ASSIM.+ 6HR_.FCST) / 23:30～24:00 / IBM Flex P460
3:30(18Z_ASSIM.+ 6HR_.FCST) / 03:30～03:50 / IBM Flex P460
Regional Forecasting System
(GRAPES_MESO4.0) / 03:40 (00Z_ ASSIM +60HR_.FCST) / 03:40～06:40 / IBM Flex P460
16:40 (12Z_ ASSIM +60HR_.FCST) / 16:40～19:40 / IBM Flex P460
Ensemble Forecasts
With 15 members
15 members (T639L60) / 04:40 (00Z_ASSIM+240HR_FCST) / 04:30～07:10 / IBM Flex P460
12:30 (06Z_ASSIM+6HR_FCST) / 12:30～13:15 / IBM Flex P460
16:30 (12Z_ASSIM+240HR_FCST) / 16:30～19:10 / IBM Flex P460
00:30 (18Z_ASSIM+6HR_FCST) / 00:30～01:10 / IBM Flex P460
Ensemble Forecasts
With 15 members
15 members (GRAPES_MESO) / 07:30 (00Z_ASSIM+240HR_FCST) / 07:30～09:30 / IBM Flex P460
19:30 (12Z_ASSIM+240HR_FCST) / 19:30～20:30 / IBM Flex P460
Regional Typhoon Forecasting System（GRAPES） / 05:00 (00Z_ASSIM+6HR_FCST) / 05:00～07:30 / IBM Flex P460
11:00 (06Z_ASSIM+6HR_FCST) / 11:00～13:30 / IBM Flex P460
17:00 (12Z_ASSIM+6HR_FCST) / 17:00～19:30 / IBM Flex P460
01:00 (18Z_ASSIM+6HR_FCST)) / 01:00～03:30 / IBM Flex P460
Sand/dust Forecasting system
(T639) / 05:30 (00Z_72HR_FCST) / 05:30～07:00 / IBM Flex P460
18:30 (12Z_72HR_FCST) / 18:30～20:00 / IBM Flex P460
Sea Wave Forecasting System
(WW3) / 07:00 (00Z_120HR_FCST) / 07:00～08:00 / IBM Flex P460
19:00 (12Z_120HR_FCST) / 19:00～20:00 / IBM Flex P460
HAZE Forecast System
(T639) / 00:10（00Z_84HR_FCST） / 00:10~04:10 / IBM Flex P460
12:00（12Z_84HR_FCST） / 12:00~16:00 / IBM Flex P460
GRAPES Rapid Analysis and Forecast System(GRAPES_RAFS) / 01:30 (00Z_ASSIM+24HR_FCST) / 01:30～02:55 / IBM Flex P460
04:30 (03Z_ASSIM+24HR_FCST) / 04:30～05:55 / IBM Flex P460
07:30 (06Z_ASSIM+24HR_FCST) / 07:30～08:55 / IBM Flex P460
10:30 (09Z_ASSIM+24HR_FCST) / 10:30～11:55 / IBM Flex P460
13:30 (12Z_ASSIM+24HR_FCST) / 13:30～14:55 / IBM Flex P460
16:30 (15Z_ASSIM+24HR_FCST) / 16:30～17:55 / IBM Flex P460
19:30 (18Z_ASSIM+24HR_FCST) / 19:30～20:55 / IBM Flex P460
22:30 (21Z_ASSIM+24HR_FCST) / 22:30～23:55 / IBM Flex P460

4.2Medium range forecasting system (4-10 days)

4.2.1Data assimilation, objective analysis and initialization

4.2.1.1In operation

The data assimilation system used in operation is Grid-point Statistical Interpolation (GSI) which was introduced from NCEP. The observational data include those from GTS while ATOVS 1b data from NOAA-15/18 channels are assimilated in the global data assimilation system. The background fields used for analysis are 6-hour forecasts from T639 model.

4.2.1.2 Research performed in this field

• The development of GRAPES-Var

The original GRAPES variational data assimilation system(referred to GRAPES-Var hereafter)does not match the GRAPES forecast model both in the definition of grids and atmospheric state variables (Xue and Chen, 2008). These differences may lead to quite many errors, especially when GRAPES-Var extends from the three to four dimension variational assimilation systems. Recently we have updated the GRAPES-Var, which completely employs the same grids and state variables as those in the GRAPES model. The new features of GRAPES-Var(Xue et al., 2012)include: 1) the physical characteristics and location of analysed variables are consistent with those in the forecast model; 2) the balanced constraint relationship of mass and wind fields is developed on the height-based terrain-following coordinate;3) the preconditioned transformation allows different vertical covariance for each horizontal spectral mode, giving them more control over the variations in horizontal scale with height. 4) the observation operators are redesigned in order to match the new coordinate with the grids. The current GRAPES-Var is capable to assimilate radio sonde, synop, ship, aircraft, cloud drift wind, the global navigation satellite radio occultation observations(Liu and Xue, 2014), satellite radiances, scatterometer sea surface winds, radar observations, and so on.The satellite radiance observations used in GRAPES are Advanced Microwave Sounding Unit-A (AMSU-A) from the present polar satellites, and the hyper spectral infrared observations.

A LBE-regression hybrid balanced constraint has been developed in GRAPES-3DVAR (Wang et al., 2014). GRAPES 3D-Var resorted to the linear balance equation (LBE) to model the dynamically balanced constraint between mass and wind fields. However, the constraint implied by LBE is fallacious in the tropics where the coupling between mass and wind is overestimated. In the new hybrid scheme, after the calculation of LBE on each level, a vertical regression whose coefficients could vary at different latitudes and model levels is introduced.

The tangent linear model (TLM) and adjoint models (ADM) are rebuilt based on the new GRAPES Global model and the parallel efficiency was improved by reducing parallel halo partition and ADM’s base state using push/pop to save memory in 2014. The new TLM and ADM greatly improved the parallel efficiency of GRAEPS Global 4D-Var. GRAPES 4D-Var experimental version in 2014 needs 30 minutes using 512 CPU cores for a 6-hour time window with 0.5°*0.5°model resolution and 36 vertical levels .

Constrained Bias Correction (CBC) for satellite radiance assimilation in GRAPES-Var

Radiance bias correction is crucial to the successful assimilation of satellite radiance observations which are typically affected by biases that arise from uncertainties in the absolute calibration, the radiative transfer modelling, or other aspects. These biases have to be removed for the successful assimilation of the data in NWP systems. Two key problems have been identified in bias correction: Firstly, bias corrections can drift towards unrealistic values in regions where there is strong model error (especially for developing models with not well tuned physics) and relatively few “anchor” observations, i.e. observations that have few systematic errors and therefore allow the separation between model and observation bias. Examples where this has been particularly problematic are channels sensitive to stratospheric temperature. Secondly, there is undesired interaction between the quality control and bias correction for observations where bias-corrected observation departures are used for quality control and where these departures show skewed distributions (e.g. in case of cloud detection). Constrained Bias Correction (CBC) scheme is proposed using priori knowledge of radiometric uncertainty information in GRAPES in order to avoid the drift of observation bias correction to the biased model background (Han, 2014). It is a kind of Tickhonov regularization techniques in inverse problems using minimum norm solution with priori information.

The assimilation of MWTS onboard FY-3 satellites in GRAPES-Var

Fengyun-3 (FY-3) satellites are Chinese new generation polar-orbiting meteorological satellites.FY-3Awas successfully launched on 27 May 2008. The Microwave Temperature Sounder (MWTS) onboard FY-3A is the first microwave temperature sounding unit in China. In 2010 and 2013, the second and third in the FY-3 series, FY-3B and FY-3Cwere launched with more advanced MWTS on board.To assimilation these MWTS radiances into NWP system, a cloud detection scheme is needed. However, the current algorithms developed for the microwave satellite measurements cannot be directly applied to the MWTS observations. A new cloud detection algorithm is proposed for the MWTS (Li and Zou, 2013). The method is based on the cloud fraction product provided by the Visible and Infrared Radiometer (VIRR) on board FY-3satellites. A MWTS field-of-view (FOV) with a cloud fraction greater than a thresholdfVIRR will be identified as a cloudy radiance. The threshold fVIRR is determined by the AMSU-A cloud liquid water path products, obtained from the Microwave Surface and Precipitation Products System (MSPPS).Analysis of the test resultsindicates that most clouds are identifiable by applying a VIRR cloud fraction threshold of 76%.Other QC steps for FY-3A/B/C MWTS include the following: (i) two (for FY-3A/B) or eight (for FY-3C) outmost FOVs; (ii) measurements from low level channels over sea ice and land; (iii) coastal FOVs; and (v) outliers with large differences between model simulations and observations.

The impact of MWTS radiances on the prediction of Chinese NWP system-GRAPES (Global and Regional Assimilation and Prediction System) was researched. The typhoon case study shows that the assimilation of the Microwave Temperature Sounder (MWTS) data can improve the typhoon track prediction by changing the model-predicted steering flow. The impact cycle experiments conducted for 30-day periods show that the QC scheme removed the outliers efficiently. Verifications indicate that forecast skill is improved after assimilating MWTS data.

4.2.2Model

4.2.2.1 In operation

There is no change for the operational model TL639L60 model.

4.2.2.2Research performed in this field

In GRAPES global model, a new prognostic cloud scheme is implemented to improve forecast of cloud cover, cloud water, precipitation and cloud-radiation interaction. The new scheme integrates a double-moment bulk microphysics developed by Chinese scientists, a prognostic cloud fraction scheme, which formulates the impacts of the convective detriment to gridscale cloud, and parameterizes low level cloud using shallow convective and PBL process. The vertical layers of GRAPES global model have been increased from 36 to 60, and non-interpolation of temperature in semi-Lagrangian advection has been realized to improve mass conservation.

4.2.3Operationally available Numerical weather prediction products.

There is no change for available Numerical weather prediction products.The T639 model products generated from operational runs are 0-240h forecasts for 00UTC and 12UTC initial time and 0-72h forecasts for 06UTC and 18UTC initial one. A list of NWP Products is given in table 4.2.1.

Table 4.2.1 The List of CMA NWP Products

Variables / Unit / Layer / Level (hPa) / Forecast hours / Area
Geopotential height / Gpm (geopotential meters) / 26 / 10, 20, 30, 50, 70, 100, 150, 200, 250, 300, 350, 400, 450, 500, 550, 600, 650, 700, 750, 800, 850, 900, 925, 950, 975, 1000 / 000，003，006，009，012，015，018，021，024，027，030，033，036，039，042，045，048，051，054，057，060，063，066，069，072，075，078，081，084，087，090，093，096，099，102，105，108，111，114，117，120，126，132，138，144，150，156，162，168，180，192，204，216，228，240 / North-east hemisphere
(0.28125*0.28125)
0°N－180°N，90°E－0°
Temperature / K / 26
U-wind / m/s / 26
V-wind / m/s / 26
Vertical velocity / Pa/s / 26
vorticity / s-1 / 26
divergence / s-1 / 26
Specific humidity / Kg/kg / 26
Relative humidity / % / 26
10m U-wind / m/s / 1 / 10 m above ground
10m V-wind / m/s / 1 / 10 m above ground
2m Temperature / K / 1 / 2 m above ground
Surface temperature / K / 1 / surface
Sea surface pressure / Pa / 1 / mean sea level
Surface Pressure / Pa / 1 / surface
2m RH / % / 1 / 2 m above ground
The first layer of soil temperature / K / 1 / 0-0.07 m below ground
Second layer soil temperature / K / 1 / 0.07-0.28 m below ground
Third layer soil temperature / K / 1 / 0.28-1 m below ground
Fourth layer soil temperature / K / 1 / 1-2.55 m below ground
The first layer of soil moisture / m3/ m3 / 1 / 0-0.07 m below ground
Second layer soil moisture / m3/ m3 / 1 / 0.07-0.28 m below ground
Third layer soil moisture / m3/ m3 / 1 / 0.28-1 m below ground
Fourth layer soil moisture / m3/ m3 / 1 / 1-2.55 m below ground
Convective precipitation / mm / 1
Large scale precipitation / mm / 1
Total precipitation / mm / 1
Low-level cloud cover / % / 1 / cloud base
Middle-level cloud cover / % / 1 / cloud base
High-level cloud cover / % / 1 / cloud base
Total cloud cover / % / 1 / cloud base
Maximum 2m Temperature / K / 1 / 2 m above ground
Minimum2m Temperature / K / 1 / 2 m above ground
Surface sensible heat flux / W m**-2 s / 1 / surface
Surface latent heat flux / W m**-2 s / 1 / surface
Surface solar radiation / W m**-2 s / 1
Surface heat radiation / W m**-2 s / 1
Snow / M (water equivalent) / 1 / snow
Water content of Surface / m (water-e) / 1
Evaporation / m (water-e) / 1
Run-off / M / 1
Snow depth / m (water-e) / 1
Geopotential height / Gpm / 1 / surface
Sea-land marks / N/A / 1 / surface
Dew point temperature / K / 19 / 200,250,300,350,400,450,500,550,600,650,700,750,800,850,900,925,950,975,1000
Wet potential vorticity vertical component / 10-6 m-2 s-1 k kg-1 / 19
Wet potential vorticity horizontal component / 10-6 m-2 s-1 k kg-1 / 19
Temperature Advection / 10-6 K/s / 6 / 200,500,700,850,925,1000
Vorticity Advection / 10-11 /s2 / 6
Dew point temperature difference / 10-1 C / 4 / 500,700,850,925
Water vapour flux / 10-1 g/cm·hPa·s / 4
Divergence of vapour flux / 10-7 g/cm2·hPa·s / 4
Pseudo-equivalent temperature / K / 4
K index / °C / 1 / mean sea level

4.2.4Operational techniques for application of NWP products.

This item should include only a brief description of automated (formalized) procedures in use for interpretation of NWP output (MOS, PPM, Kalman filter, Expert System, etc.) for example, “the MOS from ECMWF NWP is used to derive extreme temperatures and daily precipitation”.

4.2.4.1In operation

Station daily maximum and minimum temperatures were extended to 10-day. Gridded QPF, temperature and wind from 4-day to 7-day and maximum and minimum temperaturesfrom4-day to 10-day were put in quasi-operation.

4.2.4.2Research performed in this field

Ensemble model output was used to improve forecast effect of meteorological elements from 4-day to 10-day.

4.2.5Ensemble Prediction System (EPS) (Number of members, initial state, perturbation method, model(s) and number of models used, number of levels, main physics used, perturbation of physics, post-processing: calculation of indices, clustering)

4.2.5.1In operation

The global operational ensemble prediction system (GEPS) based on T639 model (T639-GEPS) has been implemented. Data assimilation system for control forecast is GSI assimilation. The system configuration is as follows:

• Number of members: 15 members; 14 perturbed members (adding/subtracting perturbations from seven independent breeding cycles) plus one control run
• Initial state perturbation method: Breeding Growth Method(BGM)
• Number of models used: one model, T639L60 (about 30 km)
• Perturbation of physical process: Stochastic Physical Processes Tendency (SPPT) method
• Running cycle: 00UTC and 12UTC, running twice per day
• Integration time
• T639L60 control run with an integration period of 15 days
• 14 perturbed members being T639L60 with an integration period of 15 days
• Perturbations being from 7 independent breeding cycles
• Research performed in this field

The GEPS based on GRAPES_GFS model (GRAPES-GEPS) using a singular vector method as initial perturbation has been developed at CMA.The main development of GRAPES-GEPS in 2014 included: 1) the main components of SVs calculation, tangent linear and adjoint model of GRAPES_GFS, were updated, and the computation cost of SVs has been greatly reduced; 2) the forecast model in GRAPES-GEPS was updated by the latest version; 3) the experiments of increasing horizontal resolution to 50 Km and increasing ensemble size to 41 were conducted.

4.2.5.3Operationally available EPS Products

The T639-basedglobal ensemble prediction model products generated in operational are 0-360h forecasts for 00UTC and 12UTC initial time and 0-6h forecasts for 06UTC and 18UTC initial time. Ensemble size is 15 including 14 perturbed forecast and control run. The output interval is 6 hours. A list of NWP GEPS Products is given in table 4.2.2.

Table 4.2.2The list of global EPS products

Abbreviations / Elements / EPS products / Probability threshold
HGT / 500hPa Geopotential Height / Spaghetti
Ensemble Mean & Spread
RH / 700/850hPa Relative humidity / Ensemble Mean & Spread
T / 850 hPa Temperature / Ensemble Mean & Spread
RAIN_24H / 24-HR Accum. Precip. / Ensemble Mean
Mode & Maximum
PRBT / 1, 10, 25, 50 ,100mm
SLP / Sea level pressure / Ensemble Mean & Spread
T2M / Temperature at 2m / Ensemble Mean & Spread
UV10M / Wind speed at 10m / Ensemble Mean & Spread
PRBT / 10.8, 17.2m/s
EFIR / 24-HR Accum. Precip. / Extreme forecast index
EFIT / Temperature at 2m
EPS METEOGRAM / Total cloud cover
6-HR Accum. Precip.
Wind speed at 10m
Temperature at 2m / BOX & WHISKERS

4.3Short-range forecasting system (0-72 hrs)

4.3.1Data assimilation, objective analysis and initialization

4.3.1.1In operation

The GRAPES regional 3DVAR system is an incremental grid-point data analysis system with 10km horizontal resolution and 50 vertical levels the same as the GRAPES_Meso model. The data assimilated include the conventional GTS data, GPS/PWand FY_2E. The analysed variables include zonal and meridonal winds, non-dimensional pressure and specific humidity.The first guess is from the operational 6-hour prediction of T639 global model with the digital filter for initialization.

4.3.1.2Research performed in this field

Data assimilation improvements of GRAPES-MESO model included:1) computing background errors based on T639 6h forecast fields; 2) finishing assimilation tests of intensive precipitation observationdata using nudging method; 3) optimizing the thermodynamic scheme in Cloud Analysis Scheme system; 4) evaluating the effect of assimilating intensive radio-sonde data at 06UTC; 5)studyingwind-profile radar data assimilation;6)testing on partial cycle in GRAPES-Meso system.