ALADIN developments in Toulouse during the last six months of 2000

1. Main events

The GMAP team welcomed 32 ALADIN / ALATNET visitors along these two quarters. They worked in Toulouse for 2 weeks to 6 months, often splitted into several stays, or topics, or fundings. The developments related to the ALATNET research program are presented in the 2nd ALATNET Newsletter.

Once again a large part of the effort was devoted to code maintenance : creation of cycles AL14 and CY23T1, additional code cleaning, analysis of the problems brought by the suppression of the shared-memory configuration (imposed by the IFS team) ... and renewal of the ALADIN web site. Many thanks to Radmila Brozkova, Ilian Gospodinov, Andras Horanyi, Stjepan Ivatek-Sahdan, Metod Kozelj, Richard Mladek, Gabor Radnoti, Roger Randriamampianina, Oldrich Spaniel, Malgorzata Szczech and Yong Wang for their help !

From January 1st, 2001, GMAP is reorganizing, moving from 3 teams :

• EXT : ALADIN
• MOD : ARPEGE, modelling aspects
• AAD : ARPEGE, assimilation

to 4 thematic teams :

• ALGO (rithms) : Pierre Bénard, Ryad El Khatib, Patrick Saez, Eric Sevault, Karim Yessad
• COOPE (rations) : Jean-Marc Audoin, Philippe Caille, Eric Escalière, Dominique Giard, Jean-Daniel Gril, Claude Fischer, Patricia Pottier, and Florence Nicoleau (half-time)
• PROC (essus) : Eric Bazile, Loïk Berre, François Bouyssel, Jean-Marcel Piriou, Françoise Taillefer, François Vinit
• OBS (ervations) : Pierre Alary, Olivier Clary, Elisabeth Gérard, Patrick Moll, Aline Peuch, Dominique Puech, Florence Rabier

Each team is now concerned by both ARPEGE and ALADIN. New permanent scientists are expected for 2001. Most ALADIN contact points don't change, with as usual exceptions : Jean-Marc Audoin is now responsible for the topics "workstation" and "coupling".

Beside the two new ALATNET students, Gianpaolo Balsamo and Cornel Soci, four visitors are currently at GMAP for long stays : Judith Padilla-Barbosa (ALGO), Olivier Pédoussaut (COOPE), Nadia Fourrié and Zhiqan Liu (OBS).

2. Dynamics

Most of the work in this domain was devoted to ALATNET topics (non-hydrostatism, relaxation of the thin-layer hypothesis) and the completion of the PhD thesis of Ilian Gospodinov.

The instabilities which appeared in the autumn parallel suites also required an improvement of the spectral brake, which now acts at the critical layer and above (Ryad El Khatib).

3. Physics

The design and refinements of the CYCORA-bis package was the main non-ALATNET task of the last months. This modset was tested in parallel suites for 5 months from September, with changing vertical and horizontal resolutions in ARPEGE and ALADIN. Thanks to the efforts of Martin Bellus, Jean-François Geleyn and Jean-Marcel Piriou in Toulouse and the distant help from Doina Banciu, the following modifications were introduced in operations on January 24th, 2001 :

• New computation of the Richardson critical number (USURID, USURIC, USURICE, GCCSV) : this approach reduces the top of winter PBL inversion smoothing problem, while keeping the same quality of the "Christmas storms" forecasts.
• Cloud core buoyancy as a fraction of an undilute plume (GCVNU, GCVALFA, ECMNP) : the top of the tropical deep convection clouds was a bit too low, as diagnosed in the GATE or TOGA-COARE 1d runs. While now relaxing the in-cloud profile towards the one of an undilute plume, cloud tops become higher, and tropical q_v & T biases are reduced.
• Stop deep convection if the condensation level is not reached (SCO) : this cures partially the problem of some unrealistic precipitation events near desert areas.
• Modification (introducing some smoothing) of the computation of the J_q term inside the deep convection scheme, to reduce a positive feedback between data assimilation, deep convection and dry turbulence, which sometimes led to numerical problems (LCVLIS, GCVPSI).
• Mesospheric drag above 8 hPa (LRRMES, XDRMTK, XDRMUK, XDRMTP, XDRMUP) : this friction has been introduced to prevent numerical explosions for strong winds for the first two levels of the model.
• The overall impact of these changes was quite positive, though a small bias on 2m-temperature was introduced. Finally the ARPEGE grid-type and resolutions didn't change, and the horizontal resolution of ALADIN/FRANCE increased slightly (9.5km/T99 instead of 9.9km/T95) keeping the same grid-type and domain extension.

4. Diag-Pack / Optimal Interpolation analysis

Gabor Radnoti thoroughly analysed the feasibility of an analysis of visibility, starting from the system used at UKMO, and wrote a detailed working plan.

Lora Gaytandjieva, Elisabeth Gérard and Françoise Taillefer significantly improved the analysis of the snow water content, from synoptic observations of snow height. The statistical model was improved, adding a vertical component to the correlation function :

m (Dr , Dln(p))= exp ( - Dr² / 2D² ) ´ exp ( - Dln(p)² / 2P² )

and tuning the horizontal characteristic length D (to take into account the sparse observations) and the vertical characteristic parameter P (to improve the consistency between snow cover and height). The geographical selection of observations was also corrected, and the observation operator improved. The first tests with a full set of observations showed a real improvement. This scheme was first to be implemented in the ARPEGE assimilation suite, to improve the localisation of snow-covered areas at initial time. It might as well be used for diagnostic purpose. Anyway, this is still an open research field.

The feasibility study for an operational implementation of Diag-Pack at Météo-France started on July. The first steps were the definition of the observation set (between 1700 and 2000 stations over France) and of the reference forecast, the choice of visualisation tools, and the tuning of surface analysis (T_2m, Hu_2m, V_10m, p_s). The tests situations were selected, and are now controlled, by forecasters. Upperair analysis is not considered up to now, taking into account its higher cost, the necessary retuning of statistics to ensure continuity near the surface, and the very irregular distribution of the available (aircraft) observations, both in space and time. [alias : RETIC-J98A]

Olivier Latinne spent a few weeks in July validating an assimilation suite for ALADIN / Belgium (based on CANARI).

5. Full-Pos / Changes of geometry

Vanda Sousa Da Costa and Ryad El Khatib tested the latest developments in Full-Pos.

a. 4-points versus 12-points interpolation

A thorough comparison between quadratic, monotonic-quadratic (new ! preventing overshooting) and bilinear interpolations was performed, considering only surface and boundary-layer fields. The experiments showed a deterioration along coasts with bilinear interpolations.

b. New post-processing of boundary layer fields

This new method, where boundary-layer fields are recomputed on the target grid using interpolated surface and upperair variables and the target surface constants, was compared with the previous one using 2 months of operational ARPEGE and ALADIN / FRANCE forecasts (August and November 2000). Results are quite encouraging, especially for 2m-temperature.

c. Changes for surface fields

Up to now the interception water content (water on the leaves, prognostic variable) and the resistance to evapo-transpiration (diagnostic variable) were set to constant values (0. and 1. respectively, as when starting climatic runs) in Full-Pos. This choice was motivated by the high spatial variability of these fields and the lack of controls against overshooting in Full-Pos when ISBA was implemented. This variables may now be interpolated as the other surface fields, with minima, maxima on land consistent with the physical parameterizations (0.-1., 0.- a veg LAI resp.). The impact on superficial temperature and soil moisture along short-range forecast may be significant (depending on precipitations of course).

The roughness lengths over sea are now interpolated (rather than set to their constant climatological value).

d. New diagnostics

The structure of the code was first changed to make the introduction of new fields easier, with a first application to :

* water-vapour column

* improved wind-gusts formulation

* various formulations of MOCON

* duration of precipitations

* min./max. of relative humidity at 2m

* heights of iso T=0 K and iso q^'_w=0 K

* ...

And the autumn parallel suites enabled to diagnose and correct a bug in E927, active when a pole of the ARPEGE grid is in the target ALADIN domain.

for more details contact : Ryad.ElKhatib++at++meteo.fr

6. Configuration 923 / Surface description

a. New databases

Igor Kos, Rashyd Zaaboul and Eric Bazile, with some help from François Bouyssel and Dominique Giard, contributed to a complete renewal of the climatological values of surface variables over land. The new global databases have a resolution of 1° (instead of 1.5°) and various origins :

• results of GSWP-type experiments for soil moisture,
• average of ARPEGE analysed fields over two years (Dec. 1998, Nov. 2000) for temperature and snow; the corresponding database was also built and evaluated for soil moisture.

The initial data were carefully controlled and compared, and transformed to a suitable format (regular latitude ´ longitude grid, handling of missing values, conversion of soil moisture fields). The new fields of moisture and snow look more realistic, when considering the annual cycle and the geographical distribution.

Two 3-weeks long assimilation experiments were launched for a winter and a summer period, to asses the impact of the new soil moisture : operational assimilation versus fixed to the new climatological value. They gave useful informtions on :

• the relative parts of the analysis and relaxation corrections in the soil moisture assimilation,
• the impact of soil moisture on the undue precipitations over desertic areas (none),
• the importance of assimilation (scores of 2m-temperature and relative humidity were globally worse when imposing the new climatological values, though the soil moisture fields looked more realistic).

Longer assimilation experiments were performed afterwads on January-February 2000, showing a slight improvement in the description of soil moisture and snow cover when using the new climatology for all surface fields.

These studies led to introduce some more limits on climatological surface fields :

• limitation of climatological soil moisture to field capacity,
• correction of the climatological snow cover according to soil moisture whenever the latter is changed (assuming that the soil cannot be completely dry under snow)
• correction of the climatological snow cover according to the temperature whenever the latter is changed. This is also done in Full-Pos with a fixed threshold (triple point), while in 923 the limit is fixed by namelist. Some more coordination is required !

An analysis of the correlations between the snow water content, the surface temperature and its mean value, using 2 months of ARPEGE data, was also undertaken. It aimed at improving the interpolation of snow fields, in configurations 701, 927 or 923, defining a correction with altitude. But results were quite puzzling, showing the persistence of snow up to 10°C.

The huge work related to the new land-cover map is described in the ALATNET section.

b. Evolution of the code

Apart from the above-mentioned modifications for climatological surface variables, several developments were undertaken (Dominique Giard) :

• distributed-memory version (the optimization of spectral orography is still not working)
• reduced memory requirements, finer definition of masks
• consistency between land fraction, urbanisation and vegetation
• new computation of roughness length
• case of pure land domains
• new fields : albedo of vegetation (introduced in all configurations), aerosols (computed either from large-scale data or from the surface characteristics following the formulas of Richard Mladek)
• ideal case of an "aqua-planet" domain

7. Support

• Jean-Marcel Piriou built new tools to make the development and validation of physical parameterizations easier :
• an external library of thermodynamical functions,
• a routine "sounding" to plot vertical stability, wind profiles, etc..., for instance from the 1d model, as shown in the figure below,
• an interface to new observational datasets for the 1d model
• a routine " model to satellite" to compute brightness temperatures from the 3d model,

An updated documentation for the 1d version of ARPEGE /ALADIN (or Single Column Model) will soon be available.

for more details contact : Jean-Marcel.Piriou++at++meteo.fr

• Jean-Daniel Gril add another option to CHAGAL : observations may now be superimposed to model fields. It proved quite useful for the validation of changes in snow analysis. An example is shown below.

• A new cycle, AL14 / CY23T1, was created and partly validated. Its contents are described in the first pages of the Newsletter. The introduction of ozone as a spectral thermodynamical variable was completed by Malgorzata Szczech, with still some problems in E001. The analysis of the potential problems for the next cycles began, especially on the following topics : external libraries for spectral transforms and physics (Gabor Radnoti, Andras Horanyi, Claude Fischer, Jean-Marcel Piriou, François Bouyssel), suppression of the shared-memory and multi-tasking options (Oldrich Spaniel, Jean-Marc Audoin, Ryad El Khatib : the memory requirements increase by about 50% for E001 and 75% for E927 when moving from shared do distributed memory). The effort of unifying ARPEGE and ALADIN subroutines was pursued (Stjepan Ivatek-Sahdan, Ryad El Khatib), while moving to a distributed-memory framework introduces new differences between ARPEGE and ALADIN in other parts of the code...
• Richard Mladek developed programs to use ALADIN for "climate" runs.
• The linear-grid option was successfully tested on some situations in ALADIN / France last August, after some modifications to allow the use of an orography with a lower spectral resolution (2/3). But the parallel tests failed. New attempts in Spring ?
• To end with , good news ! Thanks to the renewal of the ALADIN web site by Metod Kozelj, the search for informations will be easier.

Home