Note : Only non-ALATNET work is reported here. And part of the effort of the Toulouse team was dedicated to purely ARPEGE problems.
Cycle 26T1 was completed during the spring, with the help of Andrey Bogatchev, Adam Dziedzic, Cornel Soci, Oldrich Spaniel and Piet Termonia, and a distant contribution from Gabor Radnoti. Its content is described in a dedicated paper by Claude Fischer. The preliminary investigation of consequences of the in-depth changes scheduled by ECMWF for cycles 27 (new data flow) and 28 (automatic cleaning of interfaces) started.
The portability of the procedure gmkpack was improved : see the corresponding report by Ryad El Khatib. Besides, the cleaning of the xrd library undertaken by Jean-Daniel Gril, together with building the portable library PALADIN, progresses regularly.
The cost of configuration 927 was divided by 2 after the reorganisation of I/Os in Full-Pos (Ryad El Khatib). The search for new minimization softwares to be used in configuration 131 (variational assimilation), considering that the corresponding cost-function is quadratic, has started (Karim Yessad and Gérald Desroziers).
The 1d version of ALADIN used for the validation of physics has been modified in order to allow forcing by vertical velocity, not only divergence, and use the BOMEX and TOGA datasets. For more details about the single-column model and the corresponding extraction of vertical profiles from the full 3d model, contact Eric Bazile, Yves Bouteloup or Jean-Marcel Piriou.
An experiment of intercomparison of ARPEGE and IFS forecasting systems over a 15-days winter period, starting from IFS analyses, led to an improvement of configuration 901. Whenever possible, ARPEGE (ALADIN) forecasts must use initial soil and surface prognostic fields from the ARPEGE analyses, since the differences between the two surface schemes induces a significant spin-up, spoiling forecasts especially in cold situations. Besides, this test showed that, when halving the time-step and increasing the number of vertical levels as in IFS, forecasting skills are comparable for the troposphere, ARPEGE being still worse in the stratosphere.
Martin Bellus undertook an in-depth cleaning of ARPEGE physics (APLPAR.F90 and below), especially concerning surface (only ISBA, LSOLV=.T., allowed) and cloudiness (split of PNEBH ), removing all unused parts of the source code (even a little too much, as appeared afterwards). Then he focused on the externalization of the ISBA scheme (for further use in AROME), resuming the work started by the climate team and merging it with the above-described cleaning. This work was pursued by László Kullmann.
The impact of the parameterization of deep convection was examined on an "Aladinade" (i.e. the forecast of a too deep cyclogenesis) over the Mediterranean sea in May 2003. The problem disappears when changing the closure condition or using the KFB scheme, but the underlying "why" is still a mystery. Besides, along the parallel suite testing the first change of computer, an unexpected divergence was noted, due to the propagation of very small differences in SST (single versus double precision computations). Contact points : Eric Bazile and Jean-Marcel Piriou respectively.
Rashyd Zaaboul prepared parts of the new operational suites in Morocco : computing climatological files for ALADIN-NORAF and for ALADIN-Maroc at different test resolutions, implementing locally versions 25T1 of ODB and CANARI, updating the extraction of observations... He also modified the "generic" assimilation script prepared at GMAP for 3d-var or blending, to introduce the "Blendvar" and "incremental digital filter initialization" options.
Olivier Latinne prepared the move to a far larger ALADIN-Belgium domain.
Adam Dziedzic and François Bouyssel designed a cheap assimilation suite to allow testing modifications related to the surface, thus impacting only the lowest levels, during long periods at a reasonable cost in ARPEGE. Note it cannot be used to test changes in orography, likely to have a large-scale impact both horizontally and vertically. The suite is based on a blending between the modified surface and low-level fields and the large-scale upperair analysis increments from the operational 4d-var assimilation suite, as follows :
assimilation steps |
surface variables |
upperair variables |
|
6 h forecast |
Pi-1(0) Þ Pi-1(+6) |
Pi-1(0) Þ Pi-1(+6) , assumed close to P i-1op(+6) |
|
soil/surface assimilation |
Pi-1(+6) Þ Ai |
unchanged |
|
upperair correction |
unchanged |
Pi-1(+6) , Ai
op Þ |
|
initial state |
Pi(0) = Ai |
Pi(0) = Ai |
Aiop is the corresponding operational analysis, including 4d-var upperair followed by soil/surface assimilation. The spatial filter is based on Full-Pos, with an "up-and-down" 927 to the resolution of 4d-var increments (here T298c3.5 « T161c1.0).
The cost of such a simplified assimilation suite is about 8 times less than that of a full 4d-var re-assimilation. A one-week test without modification allowed to check than the resulting upperair fields didn't differ that much from the reference run (controlling large-scale balances and spin-up).
This method was used to evaluate a set of modifications proposed along the last years by Stjepan Ivatek-Sahdan, Agnesz Mika and François Bouyssel :
As expected, the resulting soil moisture looks nicer, while the scores against SYNOP observations are slightly worse than in a similar simplified suite without changes (test along June 2003).
The computation of orography (Part 1) was significantly improved :
In the meantime, the behaviour of the two basic cost-functions ("Bouteloup" and "Jerczynski"), as the tuning parameters or the starting point of minimization change, was examined. Attempts to go further than cycles CY24T1/AL15 failed up to now. Contact points : Jadwiga Woyciechowska, Dominique Giard, and the GCO team for up-to-date scripts.
An option for "aqua-planet" models is now available. Contact point : Dominique Giard
Patricia Pottier and Jean-Daniel Gril corrected the ALADIN database, for 2002 and early 2003 contributions. The next update action will address the ALADIN web site, before a migration of both under Linux.