Very few new news about our pre-operational ALADIN versions but important
ones : Bulgaria and Portugal went in pre-operational mode and operational
ALADIN-Belgium is running until 48 hours.
An AWOC meeting took place during the ALADIN workshop in Bucarest. Details in Newsletter 14.
The maximum forecast range of the operational ALADIN-Belgium has been extended to 48 hours. Until 36 hours, coupling files are classically obtained from ALADIN-FRANCE files. Between 36 and 48 hours, coupling files are directly obtained from ARPEGE files. Validation tests have proved the interest of this hybrid forecast : no major inhomogeneity appear between forecasts before and after 36 hours.
Since September 1998 the new workstation SUN ultra sparc 60 was installed with the following characteristics: Processor clock rate 350 MHz, 256 MB memory and 10GB disk storage. At this month the preparation of Aladin08t3 for running on this station was started. Unfortunately, native SUN f90 compiler does not support one very important for down sizing option - to change the default real size from 4 to 8 Bytes. Since the end of October we obtained monthly trial license for Fujitsu F90 and C compilers. So we have properly running Aladin08t3 on new work station. The 24 h forecast is computed for one hour with DFI. The code of Aladin09.03 export package was successfully compiled and configuration ee927 was validated (shared memory version). The problems with the new auxiliary library were almost the same as for DEC - some specific timing should be added for SUN.
Since December 1998 the new Solaris 7 - 64 bit operational system was installed.
Now we perform pre operational runs of Aladin09. The are three integration domains created for scientific tasks.
On domain with 9 km resolution integration of 48 h forecast takes two hours.
The main steps of the operational suite are:
The impact of the number of vertical levels in ALADIN-France has been tested . The number of levels was changed from 31 to 43 (in ARPEGE the number of levels kept unchanged : 31 levels)
The next table shows a 35% cpu increase on computation time for
ALADIN-France forecast with 43 levels (so there is currently no operational
possibility for such a modification)
|
|
31 levels |
43 levels |
Over cost |
|
Time real |
33'59" |
42'35" |
25% |
|
Time cpu (second) |
6431 |
8623 |
35% |
The impact on scores (especially 2m temperature and cloudiness) is good. For each forecast range, rmse (root mean squared error) of temperature over France is about 0.2K lower with test version, and bias is better by 0.5K at maximum.
The configurations 701 (CANARI analysis) and 923 (definition of surface fields) have been implemented on the SUN workstation.
The most important events around the workstation version of ALADIN during the last three months are as follows:
At the end of October it was decided to increase the horizontal (from around 14 to 12 km) and the vertical (from 27 to 31 model levels) resolution of the ALADIN/LACE application. It was already agreed during the summer that the envelope subdomain provided for the
LACE workstation versions will be slightly decreased still keeping a reasonable data transfer time to the Member States. The modification of the envelope subdomain influenced the domain of the ALADIN/HU model, which had to be also modified (also decreased a bit in order to be inside the envelope one). Parallelly with this change we also decided to increase slightly the horizontal resolution of our model (in order to keep a reasonable resolution difference between the coupling and coupled models). Finally with the decrease of the domain size and the increase of the horizontal resolution the same number of points were kept as earlier (therefore the computer cost was increased only due to the increase of vertical resolution).
So at the end of October a relatively complex set of modifications had to be tested and implemented in a relatively short time. The switch took place successfully on the 26th of October. Recently the horizontal resolution of the ALADIN/HU model is about 10 km and the vertical one is 31 model levels.
Work had been started to run analysis configurations with ALADIN and successful tests were done running Mandalay and Canari. Later on the software producing TEMP and SYNOP data in ASCII format was developed and tested. These developments result that the informatic environment is getting ready for regular execution of CANARI-related analysis based on SYNOP and TEMP messages. According to the plans in the future we would like to have hourly analysis provided by the CANARI system.
The implementation of AL09 was performed and the basic configurations were tested comparing the norms of the old and new cycles for a 6 hours integration (see the note of Gabor Radnoti circulated through the alabobo mailing list). Some new experiences were gained while trying to use the distributed memory version of ALADIN on two DEC workstations. After the installation of the MPI software with the efficient help of Jure Jerman the distributed memory version was successfully tested. Unfortunately the speed-up results were not very encouraging due to the slow speed of the network. Nevertheless it is believed that the tests were important for the future running of the distributed memory version of ALADIN at HMS.
NDLR: The tests on AL09 have been only performed until a 6 hours range : this explain why they did not spot the big bug !..
As it was anticipated in previous reports some case studies are being performed using the non-hydrostatic version of ALADIN. The forecasters have been choosing some interesting cases during the last convective season. The studies are focusing on a domain defined over the Transdanubian (Western part of Hungary) region and for every case three experiments were compared. The first one is the operational ALADIN/HU model, the second one is the hydrostatic version of ALADIN with 5 km horizontal resolution and the last one is the non-hydrostatic version with 5 km horizontal resolution. The results are going to be examined at the beginning of 1999.
The activities in October 1998 have already been described in the previous ALADIN Newsletter Number 12. So this short report comprises only the activities from the beginning of November 1998 till Christmas. In this period a considerable work was devoted to the port of graphical software needed for RC LACE activity in MAP (work of Jasna Vehovar and Metod Kozelj, SI).
The last change in the operational suite within 1998 year was the change of physiographical data (improved version of the so-called climate files), which was also a concerted action and was scheduled on 19/11/1998 for 12 UTC network time (improved climate data)
In the 1999 year, the first change is going to be devoted to the change of the operational library.
The Prague Team has succeeded to port to the NEC/SX4 platform and to validate the following libraries/software packages:
1. NCAR Graphics (release 4.01, used to prepare the RETIM charts)
2. XR19 auxiliary library (replacing the older cycle XRD04)
3. COH18T3 library (observation file treatment - part of the verification package ''verif.pack”)
4. MANDALAY (releases MA01 and MA05T1 - part of verif.pack)
5. AL09/CY19T1 Arpège/Aladin library (validation done for the operational configurations, replacing the older cycle AL08T3/CY18T3)
6. COH19T1, OS19 libraries (replacing the older cycle COH18T3)
7. Other graphical tools (CHAGAL, ASCS, HRID, GROM, ImageMagick, Ghostscript)
A source code management software has been put at disposal for the developers. Also cross-compilers are available on SUN servers.
A first version of the verification package ''verif.pack” was completed and validated in September 1998 (the used libraries are AL08T3/CY18T3/XRD04/COH18T3/MA01 for the part computing the scores, the graphical interpretation part is based for the time being on xmgr). Currently an upgrade of the libraries is going on (AL09/CY19T1/XR19/COH19T1/MA05T1), together with a development of a general verification procedure based on verif.pack. At the moment, the verifications are made against the observation file pre-treated in Météo-France, however this situation is far from being ideal.
The Prague Team installed in November 1998 a special post-processing for the type of products to be used within the MAP (Meso-scale Alpine Project) Field Experiment Phase (autumn of 1999), together with their visualization. A small part of this post-processing is running now regularly. However, it turned out that the graphical conversion programs do not run efficiently enough on the NEC/SX4 platform since they cannot be vectorized. Therefore it was decided to use a more appropriate Sun SPARC server at CHMI. This machine is already available and its monitoring will reach operational level by the end of March 1999.
A case study regarding one ''missed” forecast of the storm (July 1998) is under way. However, the results pointing to a problem of the current convection scheme have not yet been fully confirmed.
A working environment for the Prague Team has been under the construction. Thanks to the reorganization of the offices inside CHMI the Prague Team may use up-to 4 working posts for the visitors, hosted at the NWP Division.
(delays of the +48 h GRIBs production longer than 1 hour beyond the normal schedule):
1. 01/11/1998 12 UTC run: problem of data server in Toulouse. Delay about 3 hours
2. 05/11/1998 00 UTC run: problem in Toulouse - bug in procedures. Delay about 6 hours
3. 24/11/1998 12 UTC run: failure of the program generating GRIBs. Delay about 2 hours and a half
4. 30/11/1998 00 UTC run: failure of the main telecommunication device at CHMI. Delay about 3 hours
5. 03/12/1998 00 UTC run: problem in Toulouse. Delay about 4 hours
6. 04/12/1998 00 UTC run: failure in automatic launching of the operational suite. Delay about 3 hours and a half.
Summary: Within 2 months there were 6 cases with more significant delay (more than 1 hour behind the normal schedule), which is a bit worse rate than the average (2 cases per month). 3 cases were due to the problems in Toulouse, 3 cases were due to the problems in Prague.
(more details mario.almeida++at++meteo.pt)
On November 17th, ALADIN-Portugal went in pre-operational mode. Version AL08T3 of this limited area model is running on a dedicated workstation. This model is running up to H+48, with 31 vertical levels and a horizontal mesh of 100*90 points, corresponding to a resolution of 12.4 km and with a time step of 50s.
On a DecAlpha 500, it takes almost 2 hours, elapsed time, to complete the forecast, the post-processing and the conversion of all required fields into GRIB (projection used lat/lon) for further handling and visualization.
During this quarter, initial and boundary fields from the ARPEGE global model were received through the new upgraded communication link (64 Kbps, sharing TCP/IP and GTS/WMO traffic). Post processing is done every hour.
From November 27 until 29, Eric Bazile and Jean-Marcel Piriou (Météo-France/CNRM/GMAP) were in Lisbon, defining areas of development and work for our team. They delivered talks for a small group of meteorologists and forecasters.
Visualization of some surface fields from pre-operational runs are presented next, using GRIBEUSE utility after post-processing.
Work continues on version AL09 of ALADIN.
 |
Home |