Very few contributions in this holiday period ... correspondents are strongly requested to send their contributions for the next Newsletter and anybody who needs some informations should contact them directly on the below addresses.
Last news in Newsletter 14.
Last news in Newsletter 12.
Last news in Newsletter 13.
Last news in Newsletter 13.
Last news in Newsletter 13.
Last news in Newsletter 13.
The main events during the second quarter of 1999 were related to the migration of the ALADIN/HU operational application from the DEC 600 Au Personal Workstation to the SGI ORIGIN 2000 computer (with 8 processors).
This work was carried out in two consecutive steps:
The porting of the additional softwares related to the model (e.g. preparation of meteograms in table form, preparation of netcdf output files, verification, etc.) is also started but not yet finished, therefore for the time being some auxiliary tasks are still carried out on the DEC machine.
There was also some work performed in the framework of a bilateral collaboration between Slovenia and Hungary as far as the CANARI optimal interpolation scheme is concerned (during the visit of Neva Pristov in Budapest):
Finally it is noted that we are still using AL09 for operations together
with the convertor needed for removing the field related to soil freezing from
the initial and boundary conditions.
The only scientific change of the application concerned the
modification of the soil scheme to describe better the freezing/melting
processes of soil water. This change is supposed to improve namely the
forecast of the surface soil temperature, also influencing seriously the
forecast of 2m temperature (c.f. the cases of bad 2m temperature forecasts
last winter). So, when winter will ask us what we have been doing in summer,
we are supposed to be ready. The change is known under the namelist variable
"LFGELS".
The "LFGELS" modification has been put into the operations on :
26/05/1999 for 12 UTC network time (LFGELS bugfix on the cycle
AL09/CY19T1/XR19) with expected neutral results when tested before in a
parallel suite. Since the change of soil scheme meant a change also in the
coupling files (new reservoir of soil ice at the soil surface), the
operational introduction was concerted and took place simultaneously in ARPEGE
and in many of the ALADIN applications. The change of the scheme has already
been described elsewhere and it shall not be repeated in this report.
The Prague Team launched the following parallel tests to assess the impact of different modifications:
1. Test of the LFGELS code, with neutral scores, as expected.
2. Test of the L2TLFF: a more precise computation of the Coriolis advection
term (under key LADVF) in the two-time-level semi-Lagrangian scheme. This
suite was very short, since the scores got quite worse, indicating the
presence of a bug in the L2TLFF formulatioas n coded recently in ALADIN.
3. Start of the family of parallel suites to validate the new cycle
AL11/CY21T1/XR21 released at the end of May.
The results of parallel suites may be consulted on the RC LACE web pages: www.chmi.cz/meteo/ov/lace/aladin_lace
The blending has been the most important research and
development action, which has started at the end of June. The idea of blending
has been known since the first trials done by the Toulouse Team in 1997: the
goal is to create an initial state by blending the ARPEGE analysis (supposed
to provide correct large scales) and the ALADIN forecast (used like an
analysis guess, supposed to preserve some small scales features resolved by
ALADIN but not present in ARPEGE analysis). Two years ago the results made
with blending showed rather neutral, even slightly negative results. One of
the reasons of such results was very probably the long cycling interval of 12
hours. Meanwhile, the strategy how to make the blending was thought over and a
new scheme was proposed for the current study. First, the cycling is made
exactly in the same way like for the assimilation in ARPEGE. It means that
there is now a 6-hour blending cycle relying on the ARPEGE assimilation cycle
files. To be precise, the blending "assimilation-like" cycle is
based on the long cut-off ARPEGE analysis and it is independent from the
so-called production forecasts (the usual operational runs going to +48h). The
initial state to start the production forecast is thus based on the short
cut-off ARPEGE analysis. It means, that for our two main network times at 00
and 12 UTC there are two blending events: i) the real time one based
necessarily on the short cut-off to provide the initial conditions for the
production forecast, ii) the one done as late as possible before the next
production event, based on the long cut-off and providing the initial
conditions to run the 6h assimilation forecast (the long cut-off analysis of
ARPEGE has got the maximum observations available). Further, the blending
event itself, which may be looked at like an analysis event without
observations, is now conceived in the spirit of the incremental method. The "increments"
are computed at much lower resolution, corresponding to the ones of ARPEGE
analysis. Both ALADIN guess and ARPEGE analysis are projected to this low
resolution, however, before making the sub;raction, the fields are filtered by
DFI in order to extract from them a noise-free signal. The increments are
computed afterwards and projected back to the high resolution grid. The final
sum of the ALADIN guess and the increments is made in the way to keep the long
waves from ARPEGE analysis, which we suppose to be correct:
Blended ALADIN state = ALADIN_6h_guess + (DFI'(ARPEGE_ analysis) -
DFI'(ALADIN_6h_guess))_inc
Where the prime at DFI means that it is applied at lower resolution. At the
beginning of the integration, the obtained blended state is still subjected to
the classical initialization by DFI contrary to the "incremental
DFI" technique used in ARPEGE, since here the sum of fields may be still
a bit too noisy. The first results of this very interesting study shall be
reported in the next newsletter. More details can be asked to Dijana Klaric
and the rest of the blending team (Martin Janousek, Gabor Radnoti, Radmila
Bubnova and Jean-Francois Geleyn).
Within April, a part of the tuning of new convection scheme started in Prague. It was the work of Martin Bellus, devoted to the new computation of the saturated adiabatic state and its tuning. For Martin Bellus it was also a useful training, since he was a newcomer. Apart from this development, some little studies concerning the convection were attempted but so far without concrete results yet, mostly due to the fact that stagiaires came for an extremely short time to Prague Team.
The bug in L2TLFF formulation in ALADIN was found (it was reported on alabobo mailing list and the correct code is available since the cycle AL11/CY21T1/XR21). A parallel suite is planned in July. Further, Filip Vana has ported successfully the environment of his PhD work to Prague and continue to do experiments. His tool has been taken over by David Dvorak, who is going to use it for tests of the operational choices in the semi-Lagrangian scheme.
A new tool for computation of some stability indices has been developed (Harald Seidel). This tool works on the fullpos files and the results are added to the input file. It was conceived for a possible future use in diag.pack (analysis + fullpos (making already some diagnostics) + other diagnostics computed from the results of fullpos).
The functions verif.pack program were enlarged by the possibility to compute moving averages (Klaus Stadlbacher). The verifications of the whole period from the start of the operations in Prague is under computation. It is hoped that in the next newsletter we may publish long time scores including their evolution.
A considerable effort was put to the creation of local CMAFOC files, in order to process locally the SYNOP and TEMP type of data (Zuzana Huthova). Tests with respect to Meteo-France database will start at the end of summer. Major improvement was done also on the source code management tool MaK, regarding generation of a correct list of dependencies on modified modules (Filip Vana). Other useful R & D tools were ported to Prague or developed (fortran2html, ectoplasme, EDF, GVALAG, GVAGP, see RC LACE web pages for more details).
The documentation to the first version of verif.pack is ready. Another piece of useful documentation on the Prague Team R & D environment is under construction: Prague Team Member Handbook. All these documents are available on the RC LACE web pages (see the address above).
Last news in Newsletter 12.
Last news in Newsletter 14.
(more details mario.almeida++at++meteo.pt )
Last news in Newsletter 13.
Last news in Newsletter 12.
Since March 1999, workstation version of ALADIN/SLOVAKIA is running on upgraded machine: DEC station XP1000 with EV6 processor, 640 MB of memory and 12 GB disk space.Two runs per day (00 and 12 UTC) are operational from June 1999. There is one new product provided for Central Forecasting Office and aviation forecasters based on WS version ALADIN/SLOVAKIA: meteogram for thirteen stations over Slovakia.
We also plan to prepare some vertical cross-sections from ALADIN/SLOVAKIA model outputs, using ASCS program with improved graphics, and to increase our integration domain in the near future.
ALADIN/SI operational report, April -
June 1999
OpTiM group, Hydrometeorological
Institute of Slovenia
During the abovementioned period the model has not undergone any changes.
Routinely two daily runs using AL09 have been produced with standard
disponibility.
Porting to the cluster of workstation under Linux starts to be in its final
phase due to the new, only recently available Compaq Fortran-90 b compiler for
Linux. The code compiles relatively fine but the execution suffers from some
yet unrevealed bugs, either inside the code or in the compiler itself.
Home |