All the LACE partners were represented in Prague during this quarter (14 people from 6 countries) and worked on various subjects.
1. Data assimilation related developments.
Blending of surface prognostic variable fields completes the one of spectral variables. The following strategy has been adopted, starting from these principles:
1. We need to use the information from observations. This we may do only via ARPEGE analysis in case of blending.
2. We want to use the information in the ALADIN guess.
3. We do not have much information on structures of the surface fields (they have different evolution equations than the atmospheric part and different nature). Their structure is mostly dictated by the physiography of the model surface boundary (orography, land/sea mask, vegetation). Therefore it is the LANCELOT procedure that is selected as the base of blending filter, since this procedure takes into account the bottom boundary features in a most complex way.
We may summarize the base of the surface blending algorithm:
Step 1: take ARPEGE analysis and project it by LANCELOT to ALADIN grid (AR2AL).
Step 2: take ARPEGE guess and project it by LANCELOT to ALADIN grid (AR2AL).
Step 3: obtain ARPEGE analysis increment : ARPE_anal_ar2al minus ARPE_guess_ar2al. The increments are available on ALADIN grid and they are filtered by LANCELOT (AR2AL).
Step 4: add the increments to ALADIN guess.
Blended surface variables = ARPE_anal_ar2al minus ARPE_guess_ar2al plus ALAD_guess.
BUT:
We have to still care for the following problems:
1. We have to re-check the physical limits and consistency of the computed blended state. For example, we cannot have snow and surface soil temperature above zero at the same time and place. We cannot exceed the saturation limits of soil reservoirs, and so on.
2. We have to be careful that ARPEGE and ALADIN cycles do not diverge. Therefore we apply a sort of weak relaxation to ARPEGE analysis.
The equation then becomes:
Blended surface variables = Check_limits ( (ARPE_anal_ar2al minus ARPE_guess_ar2al plus ALAD_guess) * 0,95 + ARPE_anal_ar2al * 0,05)
The other surface fields (not variables) should be taken from ALADIN guess.
It should be further noted, that the ''analysis” of surface variables in ARPEGE rely mostly on the analysis of temperature at 2 meters. The question here is, whether an analysis of T_2m and consequently of the surface variables should not be done directly in ALADIN. This hypothesis may be verified in parallel with the tests of surface fields blending.
The relevant software for surface blending has been developed and tested on two forecast cases in order to find out possible bugs or remaining problems.
The aim of this tuning has been to find out the spectral resolution ratio between the resolution of the operational model and the resolution related to ARPEGE long waves. Further, the parameters of internal DFIs should be refined as well (see two previous Newsletters for more detailed description). Concerning the spectral resolution ratio, a table of experiments has been done, for the values 3.3 (initially tested), 2.9, 2.5, 2.1 and 1.7. The ratio of 2.5 has been selected, i.e. NSMAX=28, NMSMAX=31 for the lower resolution (NSMAX=71, NMSMAX=79 are the operational values of ALADIN/LACE). Tuning of internal DFI (the one of blending increments) still continues. Regarding the external DFI, it was figured out that: the initialization is not needed in the short runs producing the guess; however some initialization is still needed in the production runs but probably a weaker one than it is currently.
More details can be asked to: Dijana Klaric, Stjepan Ivatek-Sahdan, Martin Janousek, Gabor Radnoti, Radmila Bubnova and Jean-François Geleyn.
A set of ''standard” NMC statistics was computed using the raw ALADIN/LACE forecasts. As it has been expected, the single observation experiment has shown the same syndrome like in case of ALADIN/FRANCE domain: the contribution of the observation goes through the extension zone and appears on the opposite side of the integration domain C+I. This effect is due to bi-periodic nature of spectral ALADIN fields (the minimization of the penalty function J is done in the spectral space because of the reduced dimension of the problem) and due to the typical correlation lengths of several hundreds of km, exceeding the usual length of ALADIN extension zone (since the purpose of ALADIN is to run on a very fine mesh, hence with a small Delta x, the physical length of the E-zone is typically about 150 km or shorter). It should be noted here that a sufficient increase of the E-zone length is impossible: either it would mean to increase substantially the number of grid-points in the E-zone in which case we would have more points in the E-zone than in the meteorological domain itself, or to increase Delta x in which case there is no point to run ALADIN. Together with the fact that ALADIN domains are relatively small (the domains LACE and FRANCE are only about 2500 km wide to allow using the fine mesh at still reasonable computing cost) and thus the sampling for doing well the analysis of long waves is rather doubtful, it leads to the idea of a decremental approach: ALADIN has to rely on ARPEGE for doing the analysis of long waves while the ALADIN fine mesh should be used to analyze smaller scales as a complement to ARPEGE. In the spirit of this strategy some other types of background errors were examined still using the NMC method, the diagnostics being now tuned on eliminating the large scale component present in the error covariances. For details, there is a first draft of a special ALADIN Note on this issue.
More details can be asked to: Maria Siroka, Claude Fischer, Radmila Bubnova.
2. Developments in the physics.
There has been no development in the physics within the last quarter of 1999. Regarding the convection issues, a newcomer to the ALADIN team has started the training.
3. Developments in the dynamics.
A progress has been made in the NH model, using the vertical plane version to simulate the non-linear non-hydrostatic idealized mountain flow. A diagnostic tool to compute and visualize the momentum flux and surface drag was finalized, enabling to measure the precision of model simulations. Further, the experiments has shown that:
Finally, a remaining problem in the set-up of sponge layer has been noticed and will need a correction.
More details can be asked to: Jozef Vivoda, Tamas Szabo, Radmila Bubnova, Martin Janousek
4. Developments in the diagnostics.
There has been no development in the last quarter of 1999.
5. Developments in the verification.
The verif.pack tool should be extended to provide the scores of precipitation. The work has been started but it turned out that the length of measured precipitation period is not properly coded in the local SYNOP database in order to be correctly converted to the CMA format. A necessary technical development has began.
6. Technical developments.
An effort has been devoted to the porting of SMS (Scheduler-Monitor-Supervisor) software to the NEC SX4 platform and to the design of the operational suite under SMS. Its practical implementation is planned for year 2000.
More details can be asked to: Metod Kozelj, Roman Zehnal, Martin Janousek.
7. Work on documentation.
A number of reports of the work carried on is available from the Prague Team. In particular, a practically useful paper on Idealized studies with 2D version of ALADIN and associated tools was written by Jozef Vivoda.
|
Home |