The operational ALADIN models

1. Introduction

(more details Samuel Westrelin)

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

From first of August, ATOVS radiances from NOAA16 were assimilated and SST analysis re-activated.

On eleventh of september, cycle 24T1 (new observations files format) of ARPEGE and 15 of ALADIN became operational. At the same time, a mesospheric drag, the suppression of snow falls when cold atmosphere covered warm surfaces and post-processing of CAPE were implemented.

On twelfth of november, CYCORA-ter physics package took place.

2. Operational version in Austria

(more details Thomas Haiden)

Nothing new since the last Newsletter.

3. Operational version in Belgium

(more details Olivier Latinne or Luc Gerard)

Since 7 February, "AL12 / CYCORA_bis and 41 levels" has gone operational on SGI Origin 3000 Computer, 16 processors

AL15 / CYCORA_ter is now running in double suite on the Origin 3000. This implied long work for porting, and various adaptations of post-processing tools.

We envisage passing to a linear grid.

AL12 and AL15 have been ported to Linux PCs.

A new archiving system has arrived, and we also intend to review the selection of data to archive when switching the operational chain to AL15.

4. Operational version in Bulgaria

(more details Andrey Bogatchev)

The work on validation and verification of statistical adaptation scheme continued. Some bugs were fixed and now it runs in pre-operational mode.

The scripts for retrieving the initial and LBC-s were completely rewritten and were implemented operationally in the second half of December. Since that time there were no failures and the requests to the 'pickup' ftp-servers were sensitively decreased.

5. Pre-operational version at Croatian Meteorological Service

(more details Alica Bajic)

The main change in the ALADIN environment is its installation on a new computer:

Computer type and performance:

• SGI ORIGIN 3400
• 12 CPU MIPS R12000 on IP35 400 MHz
• Main Memory size 3072 Mbytes
• OS IRIX 6.5

Test results for the LACE domain (240x216 points, 31 levels) are shown on the picture.

Figure: Integration time for 1 time-step with different numbers of processors.

AL12 version of ALADIN is installed. Scripts are modified to be used on the new computer. ALADIN tools (aft_time, aft_info, aft_dele, aft_merg, edf and decdate) are adapted and installed. Preparation of transfer of the operational suite on the SGI is almost finished. Next step will be to install a new ALADIN version, AL15.

6. Operational version at Météo-France

(more details Samuel Westrelin)

Similar changes as in ARPEGE along these 6 months.

7. Workstation version of ALADIN at Météo-France

(more details Jean-marc Audoin)

On the Sun workstation, using compiler F90 Fujitsu for Sun, the version Al15 has been implemented with the new libraries : tal, tfl. It runs for configurations e001, ee927. But some compiler problems appeared with the module procedures, as in MPL_BROADCAST_MOD with this diagnostic : "do not ensure that generic procedure MPL_BROADCAST is unambiguous".The compiler Fujitsu for Sun doesn't recognize the type of a variable that can be REAL*8, REAL*4 or INTEGER in a generic procedure. To solve this problem, one must split the module procedure for each type of variable, as in the oldest version of module ECSORT in library xrd version XR22.

8. Operational version in Hungary

(more details Andras Horanyi)

As far as the operational ALADIN/HU model suite is concerned there were not any changes. The operational model suite is still based on version AL12.

9. Operational LACE application

(more details Filip Vana)

1. Evolution of the ALADIN/LACE application.

The ALADIN/LACE suite abandoned the data packing in the blending step:

06/08/2001 for 12 UTC network time: the data packing in the work-files is abandoned in the inner part of the blending procedure.

This is a purely technical modification, which purpose is to get the highest possible precision when using any incremental approach (like in the incremental data assimilation algorithm and/or in blending).

Impact on the forecast: The impact, as demonstrated by a short parallel suite, was hardly noticeable on the scores, however this modification is a precaution in order not to accumulate very small errors within the blending cycle.

Technical impact: more volume required in order to store the ALADIN first guess file (unpacked). The increased storage demand only impacts Prague archiving device.

The ALADIN/LACE application switched to the incremental digital filter on:

10/10/2001 for 12 UTC network time: the incremental digital filter is applied

It is a well known fact that any "initialisation" procedure (such as nonlinear normal modes, digital filter, etc.) applied on the initial conditions of the model in order to restore the mass-wind balance is at the same time removing some part of the useful signal owing to various imperfections in the nonlinear calculations. Hence the classical way of initialisation (called mostly as the "external filter" due to its external application with respect to the analysis) is today in advanced NWP models replaced by a fully internal initialisation (applied within the variational data assimilation scheme) and/or by the "incremental" initialisation, which tries to balance only the analysis increments but not the first guess. Back to 1997, the incremental digital filter initialisation (IDFI) was developed in ALADIN as part of the 3d-var R & D action. This IDFI technique, however, was nowhere really applied since it has had no meaning in the dynamical adaptation mode. After completing the blending suite, the Prague Team cleaned-up the IDFI code and scripts and applied it within the blending. Though the blending itself provides already well balanced fields (it was found that no initialisation is needed in the blending cycle), a weak external digital filter was still needed in the production forecasts. This weak external filter was replaced by IDFI, still using only a weak filter.

Impact on the forecast: i) Keeping even more the well-developed mesoscale structures in the initial state of the forecast compared to the case when an external DFI was used; ii) improved score of the geopotential bias in the altitude (probably due to a better treatment of tidal waves coming in from the LBCs).

Technical impact: relevant changes in the scripts and namelists

Perspectives: the IDFI technique is fully compatible with 3d/4d-var, as proved by 5 years of ARPEGE experience.

2. Parallel Suites & Code Maintenance

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

- Suite ABH : "disallowing all packing in the assimilation procedure". This suite was quite short (one week only) in order to get sure about the technical change. The purpose of this change is explained above. The suite became operational on 06/08/2001.

- Suite ABI : "incremental DFI & blending" . The external initialization was completely removed from the ALADIN/LACE suite. The scores were neutral except the improvement of mass field bias in the upper atmosphere. Suite ABI became operational on 10/10/2001.

- Suite ABJ : "IDFI & blending on linear grid" . After a preliminary tuning of the blending on the linear grid, a parallel test was made. Concerning the scores, these are very similar to the ones of ABI. Further effort will be needed in order to assess potential improvement of the linear grid. Since the usage of linear grid increases the size of data files (for telecommunication and storage), its operational use is not envisaged till some benefices become clear.

- Suite ABK : "CYCORA-ter package" . New physical parameterization was tested, over the second half of October 2001. The scores were, however, negative nearly for all parameters. That is why the suite was stopped. The negative results were consulted with GMAP. A plausible reason to obtain these bad scores was a particular choice of the testing period (linked to a certain type of weather situation). Since CYCORA-ter was tested in ARPEGE off-line for other seasons with mostly positive impact, it was decided to repeat the test later.

- Suite ABL : "repeat of CYCORA-ter package test" . Unfortunately, the results were as bad as for Suite ABK, this time the test was run for a full winter situation (end of November to Christmas). This time it was concluded that very probably this behaviour is due to the continental character of LACE domain. The Prague Team decided NOT to apply operationally the CYCORA-ter package in contrary to the decision made for ARPEGE suite unless some further work on the physics would be done. Unfortunately, in general LACE works too little on the physical parameterizations and the team did not have a force to investigate deeper on its own.

The results of parallel tests may be consulted on / pages.

10. Operational version in Morocco

(more details Radi Ajjaji)

Aladin Morocco: actual situation and perspectives.

After several validation tests comprising all components of Albachir operational suite beginning from observation files creation until the production of charts, it was decided to put AL13 into operations on our new IBM SPII platform.

For the time being, Albachir runs in both assimilation and integration modes (assimilation is based on Canari of course). But this situation is considered as a transitional one: Moroccan Meteorology Direction decided to move to another Albachir configuration massively nested. Aladin will run first on a great area including all North Africa with 25-km resolution coupled with Arpege (320x180x41) and for 72H range. As a beginning assimilation will be based also on Canari as a first step towards BlendVar on this domain. A second finer mesh Aladin model (9 km) will be embedded and coupled with the later but without assimilation. This second model will run to 60H range with 3 hours of coupling frequency. At the end of this second model execution, four local regional models will be launched to 36H range and on a 5-km grid meshes. The schedule of the different tasks is joint to this report.

The 64 kb/s link between Toulouse and Casa is being upgraded to 128 kb/s. This will enable the easy transfer of heavy Aladin North Africa coupling files (about 10 MO per file).

On the side of observations, our BDM is enlarged to receive all observation fluxes on SMT over the world. And another study is beginning to find a solution that will allow the presence of radiances on the BDM. The aim of the huge efforts made on observations is first to be able to move easily to a variational analysis in Casablanca, and second to be able to verify our projected Albachir configuration. The later is already in an advanced stage for operational Albachir, the first verification results will be send to ``verifala'' very soon.

Intensive studies are being done to validate Albachir new configuration (Asynchronous coupling, Assimilation without observation files coming from Toulouse, Impact of DFI on Canari increments, Investigations on 3D-Var, ...etc).

Since the last December, it is possible to run 3D-Var at Casablanca thanks to Claude Fischer precious help. Lagged NMC statistics had been produced. Now a team comprising three engineers are dedicated to assimilation with the aim to put 3D-Var into operation as soon as possible.

The following table describes some technical aspects of our new projected massively nested configuration.

11. Operational version in Poland

(more details Marek Jerczynski)

See the report on deported work.

12. Operational version in Portugal

(more details Vanda Costa)

See the report on deported work.

13. Operational version in Romania

(more details Doina Banciu)

See the report on deported work.

14. Operational version in Slovakia

(more details Olda Spaniel)

Nothing new since the last Newsletter.

15. Operational version in Slovenia

(more details Neva Pristov)

A short overview of present state is given in first part since the reports of operational version have not been prepared regularly during last two years.

Characteristics of operational model configuration have not been changed. To recall:

• 11.17 km horizontal and 31 vertical resolution (80*80 points),
• 450 s time-step,
• 48 hours integration twice per day,
• coupling by ALADIN/LACE every 6 hours.

Since 23th November 1999 the operational suite is running on cluster of 5 workstations (description can be found in Newsletter 14) with Linux as operating system. Four processors are used for the integration. The fifth is used for pre-processing (ee927) and visualization. After the completed integration following products are additionally computed:

• dynamical adaptation of surface wind,
• dynamical adaptation of precipitation (since January 2000),
• meteograms,
• time cross-sections (HRID on pseudotemps since August 2000),
• space cross-sections,
• pseudo-satelite movie,
• precipitation for hydrological models.

Correction of the temperature at 2m using Kalman filter is applied in meteograms. Daily verification between direct model output and observations (SYNOP,TEMP) is performed.

The operational model version is still AL11, however AL12 was installed and is used for the research and development work (701, non-hydrostatic version). Different tools are available: programs dealing with FA files (conversion to Vis5d, GrADS format), visualization with GROM and CHAGAL, computing pseudotemps, HRID, Harpe, computing vertical cross-section (ASCS), source code browser, MANDALAY and code for rewriting SYNOP and TEMP observations from local base.

Meteogram on request as web application was ready in the end of year 2001. Users inside met-service can select any point by clicking on image with model domain, inserting geographical coordinates or selecting a station from WMO catalogue. Beside ALADIN/SI model also ALADIN/LACE or ECMWF can be chosen.

Operational performance of the ALADIN/SI model is also important information. For the most of the days the model started and finished in time. The number of unsuccessfully finished model integrations was reduced by factor 2 in year 2001 comparing to year 2000.

Still in reasonable time products were available for additional 13 days in year 2000 and 8 days in 2001, when model was re-run by intervention. Two main reasons for missing runs were hardware problems and non-arrival files with lateral and boundary conditions due various causes. The software and hardware environment is stabilized, however few times integration stopped because of floating point exception error, which did not always occur when model was re-run. A view of the execution times for the 00UTC model run in year 2001 is given in Figure 1. Delays occurred mostly because coupling files were delayed.

Figure 1  Model integration times for the model ALADIN/SI 00UTC runs in year 2001, operational (red) and re-runs (green).

16. Operational version in Tunisia

(more details Abdelwaheb Nmiri)

The skill of the youngest ALADIN application is described in a paper.

Home