Prof. Joze RAKOVEC (University of Ljubljana) was the external co-ordinator for the RESEAU FORMATION RECHERCHE which started in February 1996 with a duration of 3 years. He visited Toulouse in the period 6-13 February and presented a final report on the progress of the research at CNRM/GMAP in the scope of the RFR for five PhD students from countries from Central and Eastern Europe.
His report is available here . It contains a "short report and results of the research of the five candidates" with also "detailed reports on the research of the four researchers" (one of the five candidates, Marta Janiskova, has successfully defended her thesis last November in Bratislava).
His "general remarks and conclusion" are favourable :
The action in scope of the Reseau formation-recherche is a quite successful one: one PhD has already been completed and four are on the horizon for this or next year (or two). The main reason for such successful progress are the personal qualities of the students, and the fruitful collaboration with their supervisors.
As in my previous report I must also stress the importance of the pleasant research environment being enabled at Météo-France in Toulouse. Not only the logistics and the efficient equipment: each candidate has his own X-terminal with all communicational capabilities, access to all libraries of the model and to the initial and boundary conditions, access to the Météo-France's library with main journals and textbooks. Also the friendly atmosphere is one of the important supporting factors for a fruitful research work. What is for some of the students a constraining factor is the sometimes limited possibility to run the model quickly, at each desirable time immediately. Namely bigger jobs which demand great computing power are normally waiting in a queue to be executed (which is a rather common situation in every shared and busy computing environment).
Next important factor, contributing to the efficient work of the candidates is certainly the fact that model ALADIN (or the system Arpège/ALADIN), with which most of the research is connected, was developed by the international team in which most of the students also participated. The model has become operational for most of the students' countries and so they feel the model to be ``their'' model. Such personal involvement is certainly a very positive and supporting factor for the success.
The side effect of the research work in Toulouse, not being connected with the students' research, is also the improving ability of using foreign languages. The researchers and their supervisors communicate in English and in French. And outside offices the language is of course French. This does not mean that the candidates will learn the ``high'' French with all grammatical finesses - but already now are all of them more or less capable to communicate with their environment in Toulouse.
The last experiments have revealed an insufficient vertical development of the convective clouds in ARPEGE/ALADIN model over tropical region (as shown by Jean Marcel Piriou by comparison with the ECMWF and Meso-NH models). In the case of such deep clouds, it seems to be an important buoyancy loss in the lower part of the atmosphere, that could be related to a too strong entrainment rate. On the other hand, the simulations of a strong squall line over middle latitude region (CLEOPATRA case: 1992, July 21) have been improved by increasing the entrainment rate at the cloud basis.
As a result, a new formulation for convective entrainment, suggested by Jean-Francois Geleyn, has been developed. The entrainment rate is supposed to be dependent not only on the altitude but also on the cloud depth.
The new formulation was firstly tested in 1D ARPEGE/ALADIN model version for the 22 February 1993 TOGA COARE squall line (thanks to the efforts of Jean Marcel Piriou and Jozef Vivoda for updating the model version and preparing the model forcing data). The results showed the new formulation could be a possible solution to obtain deeper clouds. The comparison with the results of a 3D cloud resolving model (Redelsperger and Sommeria), for the 22 February 1993 TOGA COARE squall line, proved a better agreement for the convective mass flux. The results of CLEOPATRA case 3D simulations were less encouraging. Therefore for computing the entrainment rate instead of the cloud depth, the integral of cloud excess in moist enthalpy was used. This lead to an improvement of the simulations, but further experiments and free parameters tuning are needed.
There are two separate problems within any two-time-level semi-implicit semi-Lagrangian (2TLSISL) scheme : the determination of the semi-Lagrangian trajectories and the treatment of the non-linear residuals. By using our simplified testing tool which is a 1D "Shallow Water" model, we have shown that the accuracy of the trajectory scheme is essential for the accuracy and the stability of any 2TLSISL scheme. By expanding in Taylor series and applying the advection aperator to both sides of a 2TLSISL discretization of the one dimensional forced advection aquation we have proven 2 theorems. The first of them introduces a class of second order accurate in time 2TLSISL schemes under the condition that we know the exact trajectories. The second one determines the accuracy of any 2TLSISL scheme under the conditions of our first theorem. Thus it has been shown that the treatment of the non-linear residual is more adaptable than the trajectory scheme.
Another part of our work other the last 6 months was the implementation of the second order accurate in time trajectory scheme in the hydrostatic version of the ARPEGE/ALADIN NWP model. Equivalently to what has been developed with our 1D model the new trajectory scheme on the horisontal is a second order accurate in time approximation of the horisontal advection equation where the acceleration appears exactly as it can be found at the right-hand-side of the 2D momentum equation including all forces - the Coriolis force, the pressure force and the diabatic forcing (physical tendencies). In other words uniformly accelerating motion along the trajectory is assumed on the horisontal. In any hydrostatic atmospheric model the vertical motion governing equation is diagnostic or, in other words, there is no forcing and it has an exact solution representing uniform motion. Thus in our new trajectory scheme we assume uniform motion on the vertical.
We have examined our new scheme together with the Classical 2TLSL trajectory scheme employing linear extrapolation in time and the SETTLS scheme, developed last year in the ECMWF, employing extrapolation of the acceleration from the previous time step along the same trajectory. The famous "Baltic Jet" case from February 25, 1997 has been chosen as a testing case. What is specific for this case is that a numerical instability patern in the core of a mid troposphere jet over a flat region has been related to the extrapolation in time in the classical 2TLSL trajectory scheme which was operational at that time. By performing 2 groups of experiments - with and without diffusion - with the three examined schemes and comparing the results we arrived to the conclusion that only our new scheme cures entirely the instability.
The research will continue with more detailed study of the non-linear residual treatment regarding the stability, accuracy and conservation ability of the 2TLSISL scheme.
Impact of different semi-Lagrangian interpolators used was studied on adiabatic frontogenesis process. For the typical semi-Lagrangian time step (around CFL criterium) the effect of different interpolators used was negligible for this case so the timestep was extended to the limit of of semi-Lagrangian stability to allow the scheme to be more sensitive to used interpolators. Then some conditions were defined and tested, to be able to choose the best interpolator for any given point keeping as much accuracy and stability (in necessary situation applying more diffusive interpolator) as possible. This solution allows either increase of accuracy of the scheme keeping the same timestep or increase of timestep keeping the same accuracy.
To conclude his least stay (before his PhD defense in Ljubljana), on Monday 26th of April, Mark presented the main results of his work during a CNRM seminar in Toulouse. English and French summaries of his presentation follow :
Prediction of Small-Scale Events Using Dynamic Adaptation
Even though the progress in numerical modelling of the atmosphere is continuous and considerable, only the increase of the model's spatial resolution can bring more details to the meteorological prediction in the regions with complex orography. In order to be able to obtain a useful forecast of small-scale events, such as local wind features or local, orographically induced rainfall extremes, the described dynamic adaptation approach can be used. This way the computing time is reduced significantly comparing to the full model integration.
The principle is the following: first, the fields from a large scale model output (which presents the large-scale prediction for a chosen range) are interpolated onto a dense grid of the adaptation model. This is followed by a short, typically 30 minutes integration of the high resolution, adaptation model. Physical parametrization in this model are deprived of all thermodynamics and the vertical resolution is also reduced except in the low levels. Resulting surface wind is in a great majority of cases very close to the one, obtained by a full model's, full range integration. Theoretical considerations and the application of ALADIN model for the above-mentioned purpose will be discussed.
The field of vertical velocity is a dynamic consequence of horizontal wind modification due to more detailed orography in the hydrostatic adaptation model. As the rainfall intensity is in undoubting relation to the vertical motion, better vertical velocity field can hopefully improve the rainfall intensity and distribution prediction. One statistic and one deterministic method will be presented.
Prévision des phénomènes en petite-échelle par adaptation dynamique
Malgré le progrès important et permanent dans la domaine de la modélisation numérique d'atmosphère, la prévision détaillée et précise des phénomènes, dûs au relief montagneux, peut toujours profiter d'augmentation de résolution spatiale du modèle. Les phénomènes, tels que les petits traits distinctifs du vent ou les extrêmes locaux de la pluie, peuvent être obtenus par la méthode présentée. Le temps de calcul est sérieusement diminué par rapport à celui d'une intégration complète du modèle en haute-résolution.
Le principe en est le suivant: d'abord il faut faire une interpolation des valeurs des champs atmosphériques du modèle grande-échelle (prévus par celui-ci pour une échéance choisie) sur la grille fine du modèle adaptant. Ensuite ce dernier tourne pendant une période de 30 minutes. Les paramétrisations physiques dans le modèle adaptant ne contiennent que la partie sèche, et les niveaux verticaux sont moins denses en haut altitude, ce qui fait le modèle encore moins cher. Le vent en surface ainsi obtenu est assez proche de celui du modèle dit "plein", ayant tourné pendant toute l'échéance. Les considération théoriques et l'application du modèle ALADIN pour les objectifs décrits seront présentées pendant le discours.
On suppose que le champ de vitesse verticale est une conséquence dynamique directe de changement du vent par orographie dans un modèle adaptant hydrostatique. Le lien entre le mouvement vertical et l'intensité de pluie est bien évident. Par conséquent une meilleure connaissance du champ de vitesse verticale donne une base pour une prévision plus précise de distribution et quantité de pluie. Deux façons d'y pourvenir vont être montrées: une méthode statistique et une approche déterministe.
|
Home |