The operational ALADIN models

Please consult also the summary of (pre-)operational configurations.

1. Introduction

Changes in the operational version of ARPEGE along the first half of 2001

• On 24th of January, some tunings of the CYCORA package became operational. A turbulent version of gusts was also implemented. This is also valid for ALADIN-France.
• On 19th of February, NOAA15 radiances are assimilated and SATOB utilisation is slightly changed.
• On 31st of May, a bug on angle of measure for NOAA15 is corrected.

First changes in the operational version of ARPEGE along the second half of 2001

"New cycles", operational from September 11th: it involves 7 changes, the impact of which is expected to be nearly neutral :

• 1. Cycles 24T1 (ARPEGE) and 15 (ALADIN) :

  
  Two weeks of parallel suite in August showed a neutral behaviour for ARPEGE in data assimilation mode and one quasi-identical ALADIN forecast in dynamical adaptation mode. However some tasks are not yet ready and optimizations still required.

• 2. Improved use of land-sea mask in CANARI :

  
  This modification avoids using land points in the computation of obs-guess increments over sea and vice-versa, with consequences on coastal areas for SST, T2m and H2m analyses (hence on initial surface temperature and moisture).

• 3. Increase in the number of potentially selected input observed data for CANARI :

  
  This leads to smoother increments for surface fields, especially SST.

• 4. Correction of erroneous values for horizontal diffusion within 4d-var :

  
  Less horizontal diffusion at low resolution, i.e. T42/T63/T95, especially on divergence, should lead to less "incrementality" of the minimisation with respect to the T199 model, hence to a slightly better analysis.

• 5. Introduction of the mesospheric drag computation in the TL/AD physics :

  
  It is just to improve consistency between the inner and outer loops of 4d-var and prepare the jump to 41 levels.

• 6. "Dry" TL/AD :

  
  The problems of Saharan fictitious precipitations (which increased from 3d-var to 4d-var) and of some erratic behaviour of the relative humidity field from one run to the next one (even on data richer areas) were partly explained and corrected. Since geopotential depends on humidity via the vertical integral of [Rd+(Rv-Rd).q].T.d(lnp) (virtual temperature effect), there exist situations where the 4d-var minimization finds it easier to fit height (or wind) data through a change of q than through a change of T, even if the q change has to be substantial in magnitude to achieve this goal. A temporary but efficient solution is to reset Rv=Rd when entering the TL/AD computations, and only there. The impact of such a change is illustrated by the figures on the cumulated (6h) precipitations in the test and the operational ARPEGE forecasts (forecasts starting on 27/08/2001 at 00 UTC, valid at 18 UTC).

• 7. Avoiding snowfall on warm surfaces : The distribution of convective/subgrid precipitations between rain and snow is now height-dependent, so as to avoid snowfall when the atmosphere is very warm and the soil simply between 0°C and +3°C. This had been also implemented in ALADIN-France.

2. Operational version at Austrian Meteorological Service

The main changes in the operational environment are described in the report on deported work.

3. Operational implementation of ALADIN-Belgium

Preparations have been made during the first six months of 2001 for the change of operational computer at RMI : the old CRAY will be shortly replaced with a more powerful SGI Origin 3400 computer.

4. Operational version at Bulgarian Meteorological Service

Last news in Newsletter16.

5. Pre-Operational version at Croatian Meteorological Service

HW configuration

•  Local network of workstations :

  
  - 4 x PC KAYAK600 2 proc. PENTIUM III-Coppermine (Intel) 800 MHz, 256 Mb RAM
  - 3 x PC KAYAK600 2 proc. PENTIUM III-Coppermine (Intel) 667 MHz, 256 Mb RAM

•  Network :

  
  - HP ProCurve Switch 2424M - 100 Mb/sec switch
  - Ethernet 100 Mb cables

• Software :

  
  - Red Hat Linux 6.2 (Kernel 2.2.16)
  - Fujitcu F90 compiler
  - MPICH package - message passing library 1.2.0

ALADIN / CROATIA application

Main news in domain size: increasing of domain and decreasing of resolution.

•  Kept performance :

  
  - version of ALADIN : cycle AL12T1
  - coupled with ALADIN/LACE (12km resolution)
  - integration for 48 hours / once per day (00 UTC)
  - number of vertical levels : 31
  - time needed for Lancelot (ee927) is ~2 minutes at one 2 proc. PC 667 MHz

•  Summer 2000 :

  
  - number of points : 72x72 (80x80) - Slovenian domain
  - resolution : 11.16 km
  - time step : 400 sec (432 steps for a 48 hour integration)
  - domain :
  SW corner (lat,lon) : 42.53 ° N, 8.68 ° E
  NE corner (lat,lon) : 49.44 ° N, 18.89 ° E
  - time needed per run : 30 min

• Winter 2000/2001 :

  
  - number of points : 144x120 (127x109)
  - resolution : 8 km
  - time step 327.273 sec (528 steps for 48 hour integration)
  - domain :
  SW corner (lat,lon) : 41.79 ° N, 8.93 ° E
  NE corner (lat,lon) : 49.53 ° N, 21.98 ° E
  - time needed per run : 90 min

Applications

• Visualization:

  
  - Vis5D : meteorological fields map of surface parameters (2m- temperature, surface pressure, 2m- relative humidity, 10m- wind, precipitation and cloudiness) and upperair parameters in low atmosphere (horizontal wind, vertical velocity, temperature and geopotential on 10 pressure levels 1000 - 500 hPa) of ALADIN/CROATIA
  - Grads : HRID's in 45 points for ALADIN/LACE and ALADIN/CROATIA models
  - visualization of ALADIN/LACE GRIB products

• Dynamical adaptation of wind field for 8 domains :

  
  - number of grid points : 80x80 (72x72)
  - resolution : 2 km
  - 15 vertical levels
  - integration time : 3 min per 1 time window
  - every day run for see-side domains : Dubrovnik, Split, Maslenica, Senj
  - run on demand for other domains : Karlovac, Varazdin, Pakrac, Osijek

• Dispersion model:

  
  - preparation of input files for HYSPLIT_4 transport and dispersion model on demand

•  Intranet products:

  
  - production of html pages with ALADIN/LACE and ALADIN/CROATIA maps

6. Operational ALADIN-FRANCE in Météo-France

On 24th of January, the ALADIN mesh-size diminished from 9.9 km to 9.5 km ; orography and land-sea mask were thus slightly improved. Part of the problems due to the finer resolution of the post-processing grid (unrealistic plots of 2m fields on lakes, estuaries, peninsulas or islands) disappeared. Both operational and test suites are used to detect any anomalies.

7. Workstation version at French Meteorological Service

The porting of the export version AL13_01 on SUN was completed. That of AL15 is under progress. A specific workstation problem was identified in AL13_01, due to new "tiling" surface parameterization in the IFS physics. It was mentioned to the alabobo list.

A high resolution ALADIN-Gourdon model, coupled to ALADIN-France, was running daily along June 2001, thanks to Jean-Daniel Gril, and the forecasts were sent to the ALATNET seminar. The mesh-size was 5 km, with 37 vertical levels.

8. Operational version at Hungarian Meteorological Service

The main changes in the operational environment are described in the report on deported work.

9. Operational ALADIN-LACE in CHMI

The ALADIN/LACE application switched to the new physical package denoted as CYCORA-BIS" on:

• 14/02/2001 for 12 UTC network time: CYCORA BIS

  
  The CYCORA-BIS package is based on the ALADIN library AL12_op6 /CY22_op6. This library also enabled to compute new instantaneous physical fluxes: CAPE, MOCON, WIND GUSTS and PBL HEIGHT. These new products were added to the distribution list of gribbed outputs of ALADIN/LACE. A short scientific description on the definition and computation of these fields was prepared by Project Scientific Officer.
  After having prepared the blending scripts and namelists (also for the library level of CYCORA-BIS) and solved the synchronization of the blending assimilation cycle and the production runs, the blending was put under operations on:

• 03/05/2001 for 12 UTC network time: BLENDING

  
  However, after a few days when comparing the scores of the blending suite to the dynamical adaptation suite (which still run in parallel for a control) a problem was noticed at the surface (see the parallel suite abf"). Therefore the suite was immediately switched back to the dynamical adaptation on:

• 16/05/2001 for 12 UTC network time: BACK TO DYNAMICAL ADAPTATION

  
  The problem of surface fields was found rapidly: it was a mishandling of the monthly constants (such like vegetation index, albedo, emissivity, etc.), when the blending algorithm retained always the monthly constants of the guess. Thus the same monthly constants were carried through since a cold start. Since the last cold start to prepare the blending suite for the operational application took place in January, the winter-type constants were retained. In the beginning of May, when the weather changed from winter to spring regime, a chock in the daily scores of 2m temperature was very nicely marked. In some way it was a good luck to obtain such a signal leading directly to the source of the problem. The technical mistake in the scripts was fixed rapidly and the blending suite was retested (test abg"). The final switch to blending took place on

• 06/06/2001 for 12 UTC network time: BLENDING O.K.

  
  Thus ALADIN/LACE is the first application where this type of downscaling algorithm was successfully implemented in the operational practice. Blending may be seen like a first step toward the data assimilation applied on a limited area, or in other words, like the first outer-loop in a multi-incremental data assimilation algorithm fixing the long wave part of the spectra.
  The technique of blending was entirely developed, comprehensively tested and implemented by RC LACE Prague Team. The blending method is described in a technical note written by Dominique Giard (available on ALADIN web server). The technical details on the concrete implementation of blending in ALADIN/LACE may be asked to Filip Vana, Maria Siroka, Dijana Klaric and Radmila Brozkova. On demand, the scripts and source codes are available (contact Filip Vana).
  The mistake in handling the monthly constants is pedagogically important. It says that when running even the simplest possible assimilation cycle, all its aspects must be very carefully checked. It has always been said within the ALADIN community that it is dangerous to run the model like a black box. This is twice more true for any data assimilation application.

Parallel Suites & Code Maintenance

The Prague Team launched the following parallel tests to assess the impact of different modifications:

• CYCORA-BIS suite abd" started at the end of the last year and was run nearly till the end of January this year. Hence, it was tested in winter conditions, like expected. The results were positive except the scores of the 2m temperature. This is caused by a missing piece in CYCORA-BIS package regarding the foreseen modifications in the radiation scheme. These modifications had to be discarded since they introduced instabilities (as tested in ARPEGE). The suite abd" became operational on 14 February 2001.
• Mixing length suite abe". Still in winter we tested a newly proposed tuning of the mixing length. This tuning was a result of the tests made on Christmas storm cases (in Toulouse and Prague) and was one of the ingredients of the CYCORA-TER package which is under preparation. However, the scores showed worsening already after a short time. Therefore the problem of mixing length went back to developers of CYCORA-TER.
• Blending suite abf". This is the suite containing the bug in the monthly climate constants.
• Blending suite abg" shows already the benefice of blending once the bug was fixed. The slight improvements are noticeable in scores of surface pressure and 2m temperature. In the altitude the scores are neutral, as expected (the radio-sounding network is too coarse to verify mesoscale structures). The suite abg" became operational on 6 June 2001.
• The results of parallel tests may be consulted on www.chmi.cz/meteo/ov/lace/aladin_lace/partests/ pages.

10. Operational ALADIN-MAROC in MAROC-Météo

ALADIN/Morocco still being run on Cray J90, using AL12 either in analysis and forecast modes. AL13 has been successfully ported in both SM and DM modes. DM mode is based on MPICH2 adapted for Cray. The performances of this mode, however, are very bad as MPICH2 is not optimized for Cray J90. The SM mode with LMLTSK=.TRUE. and NTASKS greater than 1 is not running. Theses last two reasons made it impossible to use AL13 operationally. AL13 still however be used for development aims. The configurations tested with AL13 on Cray are E001, E701, EE927, Full-Pos, Screening (screening is not working if the observation file contains SATEM messages).

To be able to go far with the new versions of ALADIN, DMN/Morocco is changing J90 machine by another new one IBM SP 2 POWER 3. This last new machine is already installed and all the above mentioned configurations are running without any difficulties and with great performances.

The machine contains three parallel nodes : two computation nodes equipped by 16 processors of 1.5 GFlops each (in scalar mode computation) and a third node equipped by 4 processors dedicated for file server tasks. The machine is 19 GB of memory and 1 TB hard disk managed by GPFS (Global Parallel File System). The communications between the nodes is ensured thanks to a high speed switch called "Colony" (1GB/s).

Briefly, the machine is 54 GFlops power : CANARI analysis followed by 60 hours forecast in parallel with 13 EE927 is performed in 18 minutes on just one node. This time become less than 12 minutes when using 2 nodes. (for remember, ALADIN/Morocco is 180/180/L31).

All the efforts are now focusing on putting this machine into operations ; the ultimate date for this purpose is the end of September. The operational suite will be based on a powerful task sequencer called LoadLeveler.

NB : The archiving system which was stopped for some technical reasons, will continue to work on a new hardware, also an IBM based one, called IBM LTO 3584, which can handle backup, archive, HSM, and disaster recovery data storage needs with ease. This will handle our growing storage requirements.

For more information concerning our new computing configuration, another more detailed article will appear in the next newsletter.

11. Operational version at Polish Meteorological Service

The main changes in the operational environment are described in the proceedings of the 10th ALADIN workshop : "Last developments of ALADIN operational environment at IMWM", by Marek Jerczynski.

12. Operational version at the Portuguese Meteorological Service

The main changes in the operational environment are described in the report on deported work.

13. Operational version at the Romanian Meteorological Service)

No change since the report in Newsletter18.

14. Operationa version at Slovak Meteorological Service

Last news in Newsletter 19.

15. Operational version at Slovenian Meteorological Service

Last news in Newsletter16 .

16. Operational version in Tunisia              

This is the first report of our new Partner who entered the ALADIN project officially on May 31, 2001 with the signature of the MoU.

• The Mediterranean games opportunity (Tunis, 1-15 sept. 2001)

  
  The "Institut National de la Météorologie" respectfully requested Météo-France to let it get starting with ALADIN by experimenting a suitable configuration of ALADIN remotely at Toulouse considering the Mediterranean games opportunity.
  To meet the project objectives for the "Institut National de la Météorologie" (INM), the new Tunis team, in cooperation with the GMAP team together with DSI staff and some other researchers working for Météo-France, have performed the work regarding this experience.

• The running configuration

  
  The suitable configuration for ALADIN/Tunisie chain which is now running on VPP5000 in Toulouse has been conceived in order to produce the needed predicted meteorological parameters in time.
  The e001 configuration as well as Full-Pos are running on the VPP50000 in Toulouse.
  The run time is 10 min. The selected products are available on Météo-France ftp server at 04h00 GMT.
  These data are available on the INM server at 05h00 GMT.
  The transfer of the data takes 30 min. using Internet as well as the local leased line connecting the INM to its Internet provider.
  Additionally, the visualization of the transferred products from Toulouse to Tunis uses the software "Chagal" which is running on the local HP755 workstation. J.D. Gril has performed the required work (modifications) and made the necessary tests in collaboration with H. Fehri from INM to adapt "Chagal" to HP workstations so that it may be used properly to display all the fields defined hereafter.

• Study and selection of potential meteorological parameters

  
  The following parameters have been selected together with the COMPAS team leader (B. Lacroix) :
  1. Surface parameters :
      - Humidity and temperature (2m), wind (10m).
      - Three hours cumulated rainfall.
  2. Potential vorticity and vertical velocity at levels 925, 850, 700 and 500 hPa.
  3. Cloudiness.
  4. Temperature and geopotentiel at level 850 hPa.
  5. Temperature and geopotentiel at level 500 hPa.
  6. Sea pressure.

• Control task

  
  The Tunisian team is now conducting the analysis of the current received products. In fact, within the last few days of obtaining the so called elaborated products of ALADIN/Tunisie from Toulouse, a daily qualitative checking of those products is made by two members of Aladin team together with some forecasters (mainly the surface parameters: nebulosity, rainfall, wind components and temperature).
  Following the data gathered (the elaborated ALADIN/Tunisie received products), we will return to conduct the in-depth analysis and control that is required.

• Conclusion

  
  The proposed approach has proven to work successfully so far now.
  Every phase of the project will be further evaluated and controlled by a panel of local experienced forecasters, led by ALADIN project manager.
  The combination of the core team and additional objective technical expertise from other ALADINers, once the whole INM's numerical weather prediction system is completely well implemented and locally operational, is expected to produce high quality study results and ensure a good economic impact. We look forward to working with all ALADINers and GMAP staff, in what we will strive to make a very successful relationship.
  In the meantime, this experience will certainly allows our local forecasters to get familiarized with ALADIN products.

• Example of ALADIN/Tunisie chain output (3 hours cumulated rainfall)

  
  Domain : 117*151 (128*162) points
  Resolution : 12.5 km
  Centre : Latitude = 9.36 East , Longitude = 36.06 North

Home