The Second Medium-term (1999-2001) research plan for ALADIN has been reformulated by the 3-headed team (R. Bubnova, D. Giard & A. Mokssit) created during the 3rd Assembly of ALADIN Partners to initiate and coordinate the exchanges, collect proposals and prepare the discussions for the next Assembly. A working plan for 1999 has been prepared accordingly.
These documents are available both on our public ftp (newplan2001.ps : Second medium-term (1999-2001) research plan for ALADIN; plan1999.ps : the 1999 working plan in English; plan1999_fr.ps : the 1999 working plan in French) or (here ) on this ALADIN server.
Please consult the complete documents for details.
The 3 headed team (R. Bubnova, D. Giard & A. Mokssit) met in Toulouse in February 22nd-26th. They reformulated the second medium-term research plan (see 1.) and made the following proposals to enhance coordination in the ALADIN project.
1. Improving the diffusion of information on source code evolution
A Utilisation of the web site for ALADIN
including an access to :
- all the mails sent by the Toulouse "Gourous" (alias GCO team, or Eric Sevault and Stéphane Martinez) concerning source code modifications, namelists changes, reported problems, ...
- sets of namelists for the successive "export" versions (not yet done)
B Enhance used of the "alabobo++at++meteo.fr" mailing list
Every problem noticed in source code MUST be reported through alabobo, including obvious bugs, non-standard coding, compilation problems, ...
C Consistency between operational and research libraries
Differences must be reduced to the strict minimum, to make checkings easier and avoid the multiplication of namelists sets. This was experimented in the last export version, AL09.07.
D New management of export versions
The access to the corresponding code is now two-fold :
- the usual expanded source code package, to be transfered by ftp. Both the whole source code and a reduced set corresponding the the modified elements will be available.
- a view on clearcase, which allows to point easily at individual subroutines. Extraction scripts to recover modifications between successive versions will be designed. Such an access to AL09.07 is already available through the command "cc_getview -r AL09 -u export".
For more details contact : Eric.Sevault++at++meteo.fr.
2. Preparing new cycles
A Update of the auxiliary library
The Gourous promised to update it more regularly and thoroughly. Moreover accounts on J90 (Cray computer) and a SUN workstation will be available for Sami Saarinen so that he can improve the portability of ODB, the new management tool for observations under development at ECMWF.
B Preliminary tests
A set of simple tests is to be designed. These experiments are to be run on various platforms before any significant change in the auxiliary library or in coding norms (as the change from CY20/AL10 to CY21/AL11), to point out any portability problem. The GCO team is in charge of tests on SUN workstations and Cray platforms (J90). This should be ready for the export version of AL11.
C. New strategy for phasing
The next phasing operations will be scheduled in two successive steps, focusing first on "basic" configurations (ARPEGE, then ALADIN), second on data assimilation aspects (ARPEGE, then ALADIN). This was in fact the case for the last cycles, and must be taken into account in the management of stays. Moreover it has been decided to perform a more thorough validation of the code before the delivery of an export version for any new cycle : on VPP of course, but also on workstations (in Météo-France for basic configurations, in Romania for CANARI, any other volunteers ?) and on Cray (once in Belgium, once in Morocco).
3. Verification
A simpler procedure will be designed for the diffusion of objective scores, by observation point rather than by domain. The main goal for 1999 is to concentrate informations from all ALADIN partners in Toulouse. Methods for computing the deviation from observations may differ, since tools or strategies may change. However the output format must be the same, so that data can be easily handled and merged together. A common database will be implemented next year in Toulouse.
The last 6 months were probably not a good example of code and phasing management in the strict sense. Supposed to be available beginning of november, Arpege pre-cycle 20 arrived in Toulouse in december. Unfortunately, the Aladin phasers were present at GMAP only until christmas. The somewhat uncontrolled sliding of the deadlines gave the opportunity for some nice discussions at GMAP on how to do best, short, and after a few days nobody really remembered who was first to shift the dates. For Aladin, a race against the clock started, with a completed source phasing but absolutely no chance to have any binary for validation end of 1998. The "lost" time was filled with some developments, so that many of the phasers would probably not appreciate too much if the general feeling of my paper would be that they had holidays.
They had not. Luc Gerard coded the new formulation for the convection scheme, in link with Doina Banciu and Jean-Marcel Piriou. This rather heavy coding entered AL10, and it is now the start for new calibrations for the future operational settings (work of Vladimir Pastircak in Toulouse and Martin Bellus in Prague). Meriem Zitouni dived into distributed memory and message passing issues and developped the DM version of the diagnoctic "ECHKEVO" routine. Though this type of applications does not for the time being concern many centers outside Toulouse, the DM features will have to be more and more emphasized in the coming phasings, for example via code cleaning. Ilian Gospodinov worked for AL10 on the semi-lagrangian code both for the inclusion of some of his PhD developments and for phasing. David Dvorak and Martin Charron secured the non-hydrostatic code for the new cycle (conf 001). Maria Siroka (3d-var), Cornel Soci (Canari), Plamen Neytchev (setup), Dominique Giard and Patrick LeMoigne (923, 927) and myself (message passing, sensitivity) prepared the many other facilities for AL10.
However, as I already mentioned, nothing was really ready end of 1998. For the real validation, an extra effort was done in february-march 1999, with the help of a restricted commando of validators: Plamen Neytchev ran about 100 jobs to validate thoroughly configuration 1 in many different settings. Wafaa Sadiki ran Canari on AL09, and worked also on DM-aspects. Francoise Taillefer has now included Wafaa's developments for AL10/Canari. Meriem Zitouni performed a somewhat unusual task: she validated the non-hydrostatic model. Some additional work was also necessary to introduce the modifications required by the new parametrization of soil freezing and make 927 provide a clean extension zone, whatever the clim file. Thus, AL10 is probably the first cycle where this non-operational version is ready before any "stagiere" starts to sweat on it. In the same spirit, the variational types of configurations also are ready (test of TL, test of AD, gradient test, 3d-Var gradient test).
Was this the end for this semester ? Of course not ! You forget about the bad intentions and ill-minded inspirations of our gourous, who decided there should be a major cleaning of the code, soon after CY20/AL10. So we wait for the delivery of AL11, the free-format, implicit none, generic-type constant version of the Aladin model. Of course, those who will do "cc_diff" between the former and the cleaned code will have difficulties to recognize anything but a common line between the two versions (see Eric Sevault's comments in the previous Newsletter).
So the present text is typed on friday, april 30th, 15 UTC, by a sunny, mild and slightly windy Toulouse afternoon. The end of the overall phasing, that was supposed to lead to CY21/AL11 beginning of february 1999 is still expected to happen some day in may.
Oh, I forgot to mention some more scientific work implemented in AL10 as well:
Martin Charron's TL and AD of the non-hydrostatic version entered partially
Lateral Boundary treatment in TL and AD
some more work on RUBC
Also, for information, we have presently a proper version of AL10 under clearcase view "AL10_cleaning", in phase with the Arpege branch "CY20_al10". In future, the cleaned branch "CY21_al11" will be merged with another CY21 branch prepared by the data assimilation team. By the end of may, we should then have our AL11(T1) phased with a CY21T1.
Maybe I was slightly pessimistic in my presentation of the AL10 phasing. Please contradict me and be a lot to accept our next invitation for phasing, for november-december 1999 !
Another lesson of the recent phasing efforts is that the data assimilation facilities are validated after the other configurations. Therefore, from year 2000 onwards, it has been decided to phase in two steps : hte data assimilation parts lagging behind the model items.
NDLR (who didn't participate to this huge work): AL10 is one of the only ALADIN cycles for which absolutely ALL configurations - be they operational or in development - have been validated. We had to wait more than expected to obtain this result - granted, but wasn't it worth ?
All correspondants for operational problems should have received the following message on mid-April.
As you all know, the problem of soil temperatures exasperately keeping too close to 0DG in transition phases has been casting a shadow on the otherwise good performances of ISBA, and this for the whole of last winter. The causes of the main deficiencies in soil freezing parameterization (since part of the forecast failures were due to the snow cover description) have been identified and a well tested correction set is now ready.
In fact it has been available for some weeks but tests in ARPEGE with data assimilation have shown that, when the soil is partially frozen, there is a quite long "spin-up" time (some weeks) before finding a new equilibrium and that the situation might sometimes be worse in-between, even if results are clearly positive in the end. Since ALADIN partners are entirely dependent on ARPEGE for this issue, it was decided (after an unsuccessful attempt to "catch-up" a research mode suite) to wait for the season when basically all soil-ice would have disappeared from the ALADIN-territories and to then "piggy-back" this "(temporarily) transparent" modification on another ARPEGE parallel suite.
The occasion seems now to be ripe with an ARPEGE parallel suite testing a change of statistics for the Jb term of 3D-Var (hence a completely independent issue) that started on 8/4/99. Assuming that:
(1) the results of this parallel suite will look promising by 21/4,
(2) everything will be ready on both ARPEGE and ALADIN sides at that time,
we intend to incorporate on 22/4 in this parallel suite the freezing/melting CODE AND FILE modifications (with some corrections in soil water analysis and in Full-Pos too) with the aim of an operational switch on 3/5 (NDLR: delayed to 24/5).
As previously mentioned, the necessity to incorporate a new soil prognostic variable (superficial ice content) imposes a COORDINATED CHANGE to all ALADIN users : either moving to the new package (compulsory for ALADIN-France and ALADIN-LACE) or using a file conversion procedure (for those not yet ready), like it was already the case on 16/3/98 for the switch to ISBA. The corresponding procedure is ready and available on dino. Anyway no major damage is to be expected in case of a missed conversion.
Concerning the file-testing procedures, ALADIN-France and ALADIN-LACE will be in full parallel suite mode. One set of testing files from the parallel suites in Toulouse will be prepared for Al-Bachir, Belgique, Poland, Portugal and SELAM on a date to be negociated. The LACE Prague Team will prepare a similar set for Austria, Hungary, Slovakia and Slovenia.
In principle, owing to the above-mentioned choice of a warm season for the switch, the results should be quasi-identical, except above and around high mountains. However, since the "test of the pudding" will come at a yet unknown time in the autumn, it would be wise, once you master the new situation technically speaking, to make a scientific test on a more realistic season before that event. Coupling files for the freezing/melting episode (over France at least) of 18 to 27 of November 1998 will be made available on request in research mode to those of you interested in such a test.
We hope that these explanations will help you to prepare for the foreseen change (that can of course be postponed in case of well justified necessity (*). Since we probably forgot to cover some more detailed aspects of the problem, please do not hesitate to ask for more information. An explanation of the scientific modifications will be put soon on the ALADIN www server and a detailed description of the corresponding export version follows. In the end we are sure that everything will go as well as one year ago and that next winter will show the benefits of the action we are going to do now together.
(*) On the one hand there is no urgency before next fall but on the other hand we might not find again the opportunity of such a thematically independent ARPEGE parallel suite as the current one.
A brief scientific description can be found on our usual ftp (files : extabs.ps, pp1.ps, pp2.ps).
|
Home |