Stratus prediction
In a cooperation between ZAMG and the University of Vienna (S. Lichtenauer), the forecast errors of the ALADIN model with regard to fog and low cloudiness are investigated. The research is part of COST Action 722 "Short-range forecasting methods of fog, visibility and low clouds". It was found that the frequent under-prediction of low stratus in the Danube area is associated with incorrect prediction of temperature inversion characteristics. In the model, inversions are generally too much smeared out, and too close to the ground. Above the inversion, the model is typically 1-2 degrees too cold compared to sounding observations, whereas it is up to several degrees warmer than observed just below the inversion. The lack of sufficiently cold air just below the inversion leads to an underestimation of cloudiness. With less than 100% cloud cover, the PBL begins to warm during morning hours. This further reduces cloud cover, and a reinforcing feedback loop sets in. The problem of too smooth temperature structures is present from the initial time (+00h) because the data assimilation in ARPEGE rejects cold temperatures just below the inversion if they differ too much from the model state. This was shown for a stratus case in January 2002 (E. Bazile, personal communication). A description of numerical experiments that were made on the stratus / inversion problem, with both the SCM and the full 3d model, can be found in this Newsletter. [Further details: alexander.kann++at++zamg.ac.at ]
Prognostic convection
The prognostic convection scheme of Luc Gerard was tested on a case of diurnally forced mountain convection. This is a type of convection that is typically predicted too early in the day. The prognostic scheme, which takes into account the "inertia" of deep convection in the form of prognostic draft strength and active mesh fraction both for convective up- and downward motion, did not show a significant delay in the onset of precipitation. However, the spatial patterns were more selective compared to the standard scheme, and at least for the case investigated, closer to observations (radar data). Additional tests during the convective season in 2002 need to be carried out to confirm this potentially beneficial behaviour. [Further details: stefan.greilberger++at++zamg.ac.at]
Operational verification
The first in a series of bi-annual verification reports was issued in February 2002. It can be obtained in pdf format from the ZAMG homepage http://www.zamg.ac.at by clicking on the British flag, following the "Numerical Weather Prediction" link and choosing "Verification" from the pull-down menu. The report contains point verification statistics and shows ALADIN / ECMWF model intercomparisons for near-surface variables over a period of several years. The next issue will appear in September 2002. As a special topic it will contain a preliminary analysis of the performance of ALADIN during the catastrophic rainfall events in August 2002. [Further details: thomas.haiden++at++zamg.ac.at]
More information about research activities of the NWP group at ZAMG can be obtained from the ZAMG homepage http://www.zamg.ac.at by clicking on the British flag and following the "Numerical Weather Prediction" link.
Most efforts focussed on ALATNET related research, and consequently are described in the ALATNET Newsletter.
Most of the effort was devoted to the implementation of a new operational suite.
Great deal of time in the previous period was devoted to the organization of the 12th ALADIN workshop on "Scientific Development and Operational Exploitation of ALADIN model" and preparing the presentations for the workshop.
Following work has been presented :
Martina Tudor : MAP IOP 15 Case Study
Marija Mokoric : ALADIN products in "Taming the Dragon-Dalmatia 21 to 24 May 2002"
Zoran Vakula : Adriatic Storm Cases
Stjepan Ivatek-Sahdan : Analysis of surface variables
Note: all ALATNET related R & D, representing the majority of the effort, is reported in ALATNET Newsletter. Here we sum-up topics which are not referred as ALATNET ones for Prague centre.
Developments in physics : Problems with unrealistic "large-scale" precipitations
Last spring we noticed a few cases when the model produced unrealistic spots of precipitation reaching sometimes 60 mm /6 h (Figure 1). Although the meteorological situation was favourable for strong rain, such an amount was surely exaggerated. Several tests were made, using both past and newest versions of the physics (pre-CYCORA, CYCORA, CYCORA-bis, CYCORA-ter, very last changes and corrections in CYCORA-ter) but there was almost no sensitivity to this. We tried to diminish the length of the time-step (we used half the nominal one for LACE) in order to exclude any possible numerical instability within the physics. However, the result was even worse since the amount of precipitation was almost doubled. Since the heavy rain was produced by the large-scale or better to say by the explicitly resolved precipitation computation, we could think that the problem was linked to the spatial resolution (in connection to the convection scheme which seemed almost inactive), perhaps to the recent increase in vertical resolution (last summer nobody reported similar cases). After discussing with the physics team in Toulouse (J.-F. Geleyn, F. Bouyssel) we tried to introduce the parameterization of "shear-linked convection" proposed by J.-F. Geleyn.
The impact on forecast was quite significant. For the studied case the amount of precipitation was diminished to more realistic values (Figure 2), more comparable with the rain gauge observations (reaching a maximum of 21 mm / 6 h in the concerned area, see Figure 3). The resulting precipitation field moderately changed everywhere. We have tried the physics package including the "shear-linked convection" scheme also for other cases reported this spring. For most of them we obtained more realistic precipitation fields, however we cannot say that the forecast was systematically better.
In addition, we tried to recompute the cases of unrealistic cyclogenesis in the Adriatic (there were three such cases reported last summer). Once more we obtained a nice improvement: the cyclone over the sea was weaker and the pressure field remained almost the same elsewhere.
However, since the impact on forecast is quite important, more extensive testing and perhaps retuning, including parallel suites, must be done before switching to the use of shear-linked convection in the operational application.
More details can be asked to : Jean-François Geleyn, François Bouyssel, Radmila Brozkova, Filip Vana.
Figure 1 : Operational ALADIN/LACE forecast starting on 05/06/2002 at 00 h UTC : cumulated precipitations between +24 h and +30 h. Convective (unresolved) precipitations are on the left; the ones resulting from the explicit computation are on the right. The maximum coming from the explicit computation reaches more than 52 mm / 6 h in the Jesenik mountains.
Figure 2 : Same as in Figure 1, but for the parallel test ABR, using the new parameterization with the operational initial and lateral boundary files. The maximum of explicitly resolved precipitation is by 20 mm / 6 h lower in the Jesenik mountains when compared to the operational forecast. On the other hand there is a stronger maximum in Germany. The amount of convective rain is even smaller than in the operational case.
Figure 3: Observed sum of precipitation over 6 h interval between 00 h and 06 h UTC on 06/06/2002 at synoptic stations.
Developments in data assimilation : CANARI surface analysis combined with upperair blending
Currently the blending algorithm used in ALADIN/LACE comprises two parts: blending of the spectral dynamical prognostic variables (BLEND) and blending of the surface variables (BLENDSUR). The latter is simply based on a sophisticated interpolation of the ARPEGE analysis increments onto the ALADIN grid and summing them with the ALADIN guess. The aim of this work was to replace this blending of surface variables by a surface analysis / assimilation, using the optimal interpolation scheme CANARI. At the first stage of this work, beside the simple technical change, the aim was to study various possibilities to implement this surface analysis step. Then, another particular problems were looked at. Concerning the above mentioned possibilities of the combination of the surface analysis and upperair blending, three approaches were tried:
Experiment 1: both schemes are independent in the sense that the result of one does not influence the other one, that is both parts work with the original guess. This scheme corresponds to the current arrangement of the BLENDSUR and BLEND.
Experiment 2: The dynamical part BLEND is done first and the guess for CANARI gets refreshed upperair fields (the comparison to observations in CANARI is done at the level of SYNOP stations where the model value is influenced by both the upperair and surface variables). Resulting surface fields are thus dependent on BLEND via the first guess.
Experiment 3: The surface analysis is done first, modifying the state of surface variables. The digital filtering of upperair fields which is the base of the BLEND step is performed with the updated surface fields. Thus the result of BLEND is influenced by surface corrections.
All three experimental setups were tried over a short period last February, characterized by a very windy but relatively warm weather. The results were compared regarding the impact on the forecast (subjective evaluation), on the scores (VERAL) and on the noise level for spectral fields (ECHKEVO). There were not many differences noticeable on the charts except perhaps for precipitation patterns. Concerning the noise, Experiment 3 was the most balanced one, as foreseeable. All three experiments provided slightly better scores against SYNOP observations than pure blending, although Experiment 1 and Experiment 2 were a bit more superior to Experiment 3.
Within the surface analysis, the observations of screen-level temperature and humidity were used to initialize the soil temperature and moisture. A particular point is the sea surface temperature. Within the ALADIN/LACE domain there are small pieces of sea basins where it is not sure to get any ship or buoy observations. It would be rather risky to rely on the sea surface temperature analysis within CANARI-ALADIN. Hence, the sea temperature shall be retained from the ARPEGE analysis.
More details can be asked to : Adam Dziedzic.
Developments in the verification
The long-term verification of ALADIN/LACE was done. Please, refer to the special contribution of Dijana Klaric in this Newsletter.
Note : Only non-ALATNET work is reported here. And a large part of the work of the Toulouse team was dedicated to purely ARPEGE problems (new horizontal diffusion, new observations, 4d-var problems, ...) during the present period. See also the PhD report from Jean-Marcel Piriou.
There were only few ALADIN visitors in Toulouse along the last months. Embassy fundings for Central and Eastern Europe are still blocked for 2001 and 2002. Morocco is now in a similar situation for 2002. Only Tunisia escaped from such problems !
The main achievement was the creation of libraries 25T1, with the help of Bodo Ahrens, Andrey Bogatchev, Gergö Bölöni, Marek Jerczynski, Zahra Sahlaoui and Siham Sbii. The corresponding changes are described in a separate paper by Claude Fischer.
Significant effort was also devoted to the strengthening or even to the setup of networking in various domains. Several mini-workshops were organized in Toulouse (non-hydrostatic dynamics and predictor-corrector approach, orographic forcing, convection, turbulence), Kranjska-Gora or Medulin. This was highly necessary for physics, where real networking is just starting, with difficulty and frictions. This also allowed to open new issues. Most reports from the working groups are now available on the ALADIN web site, and in the present Newsletter.
Stjepan Ivatek-Sahdan refined his previous work on the statistical model for optimal interpolation analyses of 2m-temperature and relative humidity, using more classes (28 vs 14) and more domains (11 vs 4) in the computation of forecast errors. Its newly proposed formulations for the coefficient of correlation and the dependency on resolution proved again to better fit data. After some retunings, mainly smaller characteristic lengths, analysis experiments were performed for two situations (15/01/2001 and 15/08/2001, 00 h and 12 h UTC). Changes were globally positive, when compared to the operational scheme : smaller corrections in areas with only sparse observations, analysed fields closer to observations. A full data assimilation suite is required to draw more conclusions. A second step was the introduction of an horizontal smoothing of the mean soil moisture just after the correction of soil fields in CANARI. This aims at damping the large heterogeneity present in operational fields, and reducing the associated forecast errors on 2m fields. A "Laplacian" filter, already used to smooth roughness lengths for instance ((e)lislap.F90 for specialists), is applied on the soil wetness index, taking into account the land / sea mask. The radius of impact and the number of applications of the filter were tuned. Results were promising, but an evaluation within along assimilation cycle is also required here. This will be done next September, with an operational application in ARPEGE beyond and an expected subsequent improvement in the initialization of soil moisture and temperature in ALADIN. A direct application of smoothing to soil / surface analyses in ALADIN is underway (see the ALATNET report from Gianpaolo Balsamo).
A lot a work was dedicated to the cleaning of physics, correcting sleeping or active bugs, ensuring reproducibility and improving safety locks. The situation looks safer now, even if the correspondence between correction actions and model versions (operational and export ones) is not fully clear yet. A simplified version of a document written by Jean-François Geleyn, describing the latest changes in the physics package, is given in Table 1. The full version was sent to aladin++at++meteo.fr and is available on the ALADIN web site, as an appendix to the report of Medulin meeting on operations : http://www.cnrm.meteo.fr/aladin/scientific/2002/WG_oper_0602.html . In the framework of these studies, Jean-Marcel Piriou designed a useful tool to identify gridpoints where fields or tendencies reach unrealistic values. The corresponding modset should enter the next libraries (CY25T2, autumn 2002).
Table 1 : 18 months of problems/evolutions in the ARPEGE/ALADIN physics (status on 12-08-2002)
n° |
Step |
Mod code |
Action |
Mod. namelist |
Previous |
Comments |
0 |
CYCORA-bis (whole package) |
CYCORA |
Real clean starting point ; no back-phasing beyond it for all what follows; reference for changes below | |||
1 |
Code |
acdrov aplpar |
Only a marginal impact. | |||
2 |
Modification of the melting |
accvimp |
Fraction of snowfall |
GCVMLT=1.6E-04 (added parameter) |
º GCVMLT=0. |
In case of surface inversion one could previously have snow reaching the soil from convective precipitation for a slightly cold surface temperature after crossing very warm layers. The previous situation is still maintained via a local key but the switch to the new scheme is recommended. The two tunings are equivalent for "standard" vertical temperature profiles |
3 |
CYCORA-ter |
achmt accoefk aplpar |
Several consistency |
To allow further PBL developments in a better environment. Bugged twice, see [12] | ||
4 |
Idem |
New mixing length |
ALMAV=300. BEDIFV=0.05 UHDIFV=8.E-04 |
ALMAV=75. BEDIFV=0.4 UHDIFV=4.E-04 |
Better results at global scale and over oceans, but not appropriate for all conditions (cold continent) . Requires local decisions/tuning | |
5 |
Idem |
New dependency of Kuo -closure on resolution |
GCOMOD=1. REFLKUO=10000. + LSRCON=.T. |
GCOMOD=0. |
To restore the correct dependency of both LS and CV precipitations when Dx -> 0 while keeping the "local" advantages of LSRCON; it may suppress some instabilities, as in ALADIN-France | |
6 |
Idem |
GCCSV=0. |
GCCSV=1. |
Better simulation of top of PBL inversions | ||
7 |
Idem |
USURID=0.035 |
USURID=0.016 |
Tuning for consequences of [3-4-5-6] | ||
8 |
Reproductibility |
acnebn |
Bug correction |
Dependency on NPROC |
Allows clean test + bug fixing; the results are changed even if 1-proc mode | |
9 |
Blow-up of model fixed + consistency |
aplpar |
LSRCONT=.T. (added switch) (+ LSRCON=.T.) |
º LSRCONT=.F. |
Makes the convective closure more consistent and symmetric for q & s Bugged, see [19] | |
10 |
Blow-up of model fixed |
accvimpd |
Modified Kuo-2 test |
Prevents the activation of a "sleeping" design bug in downdrafts computations. | ||
11 |
Preventive action |
accvimp |
Suppression of false virga |
Risk spotted while working on [10]. | ||
12 |
Bug corrections |
accoefk |
Two of them fixed; |
USURID=0.042 |
USURID=0.035 (if [4]) |
Bug correction compulsory with [3]; interaction with the retuning of [7] |
13 |
Anti-fibrillation adaptation to 41/37 levels |
XMULAF=-1.85 |
XMULAF=-1.75, |
Change linked to the new D t and Dz near the surface; this tuning should in principle be redone for each resolution change. | ||
14 |
Vertical slicing via the anti-fibrillation scheme cured |
acdifus |
b cannot decrease when going downwards |
LMULAF=.T. (added switch) |
º LMULAF=.F. |
The antifibrillation scheme ceases to create stationary spurious PBL wind patterns. Necessary if one wants to run with long time steps or XMULAF further away from -2. XMULAF to be retuned |
15 |
Avoiding 0 / 0 risk |
accvimp accvimpd |
Changing thresholds |
Better security; hardly modifies results | ||
16 |
Model blow-up cured |
accvimpd accvimp |
Better consistency |
One avoids trying to reach the ground with a downdraft across a well mixed and very dry PBL. For the time being the precipitations spuriously reach the ground in such a case, with this fix. Further treatment required soon (see 20-21-22). | ||
17 |
Clarification of the downdraft's role |
accvimpd |
Protection against a risk of sub-cloud precip.-generating downdraft computations (esp. if 15-16 are active) |
This weakness existed from the start of the downdraft parameterisation but its role was probably minor up to now. | ||
18 |
Safety action |
accvimpd |
Equivalent of [11] for downdrafts |
May be superfluous because never used. | ||
19 |
Bug correction |
aplpar |
Avoid ing a double substraction of LS precipitations in the CV closure |
Bug correction compulsory with [9] | ||
20 |
Additional |
accvimpd accvimp |
Avoiding a "zero" total effect when the limit is acting |
Physically necessary since the strong increase of the protection values via [15]; small impact. | ||
21 |
Idem |
accvimp |
Modulated |
RCVEVAP=0.25 (added parameter) |
LCVEVAP .F. <-> 0. .T. <-> 1. |
With [16] one may have convective precipitations reaching the surface despite crossing very dry levels. The idea is to have at the start the same value for RCVEVAP than for GDDEVA, in order to get the best possible compensation between the two effects of [16] and [21] |
22 |
Idem |
accvimpd |
Updated protection against negative |
With non-zero values of RCVEVAP a yet unprotected consequence of [17] becomes far more likely to lead to negative fluxes. Hardly modifies the results even if clearly necessary. |
Some more modifications will be introduced soon, related to new developments ("shear-linked" convection, temporal smoothing in the shallow convection scheme, "moist" gustiness). Items 1 to 18 are already operational in ARPEGE, and the order is close to the chronological one. The export version ald_export_AL15_03 corresponds to items 1 to 11. For teams at least at the level of AL12_op6_CYCORA_bis, the Toulouse team can prepare source codes for the relevant routines, given a consistent list of items.
The previous work of Martin Bellus on the shallow-convection scheme was pursued, with a careful coding and more tests. This is detailed in a separate paper, and also in the above-mentioned mail, sent to aladin++at++meteo.fr.
Jean-Marcel Piriou coded a new diagnostic of PBL height, based on a proposal of Ayotte (1996) and further improved to get smoother fields, both in space and time. Thanks to Thibault Montmerle, the RTTOV-7 radiation code is now available within the model-to-satellite tool developed by Siham Sbii and Jean-Marcel Piriou.
New physiographic databases are now easily available on VPP and archive machine, under directory ~mrpe603, as for the operational ones : the GTOPT030 dataset for orography and a new global database for soil and vegetation (the old one without ice on New Caledonia).
Several ODB tools were prepared for Diagpack (the porting of which is now finished), in cooperation between Françoise Taillefer and Dominique Puech (e.g. to visualize the deviation to observations). Some documentation on ODB for ALADIN is now available on the ALADIN web site. A more extensive documentation on ODB, but in French, is also available by Dominique Puech or via the GMAP web site. Meanwhile Philippe Caille designed a procedure to introduce pseudo-profiles of humidity in CMA files, for the needs of 3d-var experiments.
Jean-Daniel Gril undertook to rewrite the auxiliary ( xrd) library, to improve the portability of ALADIN and the associated tools. A first package, PALADIN 1.0.1, is ready. It includes the following applications : ecto, edf, frodo, pseudo, and basic libraries. It has been tested successfully on 5 mainframes with 4 operating systems and 7 compilers, with the help of Neva Pristov :
|
O.S. version |
Mainframe |
Compiler |
|
Linux 2.4.18 # |
alpha (DEC) |
egcs-2.91.66 (Compacq Fortran compiler) |
|
Linux 2.4.17 #1 |
i686 |
Lahey / Fujitsu Fortran 95 Express Release L6.00a S/N: LX072865 |
|
Linux 2.4.3-20mdk #1 |
i686 |
pgf90 3.2-4 Copyright 1989-2001, The Portland Group, Inc. |
|
Linux 2.4.18-6mdk #1 |
i686 |
pgf90 4.0-2 Copyright 1989-2000, The Portland Group, Inc. Copyright 2000-2002, STMicroelectronics, Inc. |
|
HP-UX B.11.00 A |
9000 / 782 |
HP F90 v2.4 |
|
HP-UX B.11.00 U |
9000 / 800 |
HP F90 v2.5.3 |
|
SunOS 5.6 |
SUN W, Ultra-4 |
Fujitsu Fortran Compiler Driver Version 4.0.2.1 |
|
SunOS 5.6 |
SUN W, Ultra-4 |
Sun WorkShop 6 2000/04/07 FORTRAN 95 6.0 |
|
IRIX64 irix 6.5.10 |
IP30 mips |
MIPSpro Compilers : Version 7.30 |
This work is to be pursued, with support from the Budapest workshop on maintenance. The portability of the ALADIN-ALATNET database was also improved, with a move to Linux. To end with, a documentation on geometry was written and will be very soon available on the ALADIN web site.
Most of the ALADIN related activities at the Hungarian Meteorological Service were concentrated on ALATNET topics, therefore important additional details can be found in the ALATNET report of HMS.
The main topics of progress can be identified as follows:
- As mentioned at the operational aspects lot of efforts were carried out migrate all the model versions and applications to our new Regatta server. The most important part of this work had been completed until the end of June.
- The ODB software was fully installed in the Regatta machine and certainly all the corresponding softwares as cma2odb, ODB-viewer, lamflag-ODB, etc.
- The verification procedures were upgraded and completed making possible a regular verification of the available models with special emphasis on the ALADIN/HU, ALADIN/LACE and ECMWF models. Plots and quarterly report were already produced and planned to be produced on a regular basis.
- The transform package was phased in Budapest and the new version was given to the Toulouse team.
- During the last RC-LACE Council meeting, Hungarian Meteorological Service was volunteering to organise an ALADIN "maintenance and phasing" workshop. The dates of this training workshop was fixed for the end of November 2002, and the first announcement was circulated. More details under the http://omsz.met.hu/ismeretterjesztes/rendezvenyek/aladin_ws/paper.html homepage.
Nothing new.
A first release of an objective verification package is now ready. A comparison between different data assimilation systems : O.I., 3d-var, blending and blendvar, was also performed using the ALBACHIR application. (see the joint papers)
Nothing really new since the last Newsletter (routine problems).
An objective verification was carried out for the period from March 2001 to February 2002 using the two-meters temperature, two-meters dew-point temperature and ten-meters wind speed as surface predictands and the temperature, dew-point temperature, specific humidity, geopotential and wind speed for upper level checks. Besides, total precipitation was also verified. A warm surface bias in the coastal region of Portugal and the underestimation of dew point at two meters are some of the interesting conclusions that will be published soon.
Concerning the analysis subject, not so much technical advances have taken place, however a pre-operational verification of the CANARI / DIAGPACK package is being prepared.
On the diagnostics topic, some more case studies of extreme precipitation events were performed using the home-developed stability indexes (see the joint paper).
Post-processing of minimum and maximum temperature for verification (Simona Stefanescu)
The minimum and maximum temperatures for 10 observation stations are extracted from the 3-hour post-processed files. The extreme temperatures are computed for intervals corresponding to the observations and they are corrected taking into account the difference between model and station altitudes. They are used in the verification procedure.
Implementation of Al12_04 version of the ALADIN model (Simona Stefanescu)
The AL12_04 version has been implemented on the ALPHA DEC platform in order to use the last modifications and corrections. Comparisons with the AL12_02 version were carried out; no significant differences were found.
Spatial variability of the forecast error covariances (Simona Stefanescu under the supervision of Loïk Berre)
Continuing the experiments made on the spatial variability of temperature forecast error covariances, the 1d spatial variations of 2d spectral covariances were investigated. The latitudinal and longitudinal variability of standard deviation and length-scale of the correlation function were calculated and plotted for different levels. The results for the latitudinal variations were successfully validated by comparison with the results of Mohamed Raouindi based on 1d spectral coefficients. The study will be deepened with diagnostics related to the 2d spatial variations of 2d spectral covariances.
Implementation of the 1D Shallow water spectral model (Raluca Radu)
The last version (3.0) of the one-dimensional "Shallow water" spectral model developed by Ilian Gospodinov was implemented on the DEC platform. This simplified model can be used for testing and improving the coupling method in ALADIN (see details in the ALATNET report).
Implementation of version AL15_03 of the ALADIN model (Doina Banciu)
In order to transfer the operational suite from the 1 processor ALPHA DEC platform to the 8 processors E4500 SUN platform, the version AL15_03 has been implemented. For the present operational domain the duration of a 48-hour model integration (including digital filter initialization and in-line post-processing) is between 22-27 minutes, depending on the situation. The results of the AL12_04 (using CYCORA-bis package) and of the AL15_03 (using CYCORA-ter) versions have shown differences especially for precipitation fields. Further tuning of the free parameters will be considered together with a systematic verification against the real data.
See the joint paper by Rastislav Kvaltyn.
The preparation work for the verification-project prototype application started in early June. The main lines are described on the ALADIN web site:
Kay Suselj continued his work on high-resolution verification of precipitations.