The paper report was sent to the EC on April 27, 2001 with the corresponding costs statements and a hard copy of the database report updated electronically on the RTN Internet site. A copy was sent to the correspondant of each ALATNET centers.
Despite the fact that the amount of work has been slightly below the originally set target, the scientific progress linked to the ALATNET selected actions has been satisfying and sometimes even spectacular.
In the opinion of the ALATNET management group, the progress is such as could reasonably be expected, with a few delays compensated by some spectacular breakthroughs. The main cause of satisfaction is the good integration of the work of the Young Researchers in the network research aspects and its link with some of the above-mentioned areas of welcome advances.
These publications involve the Toulouse and Prague research centres. The first one also involves the Budapest research centre.
The programme of work is split in 12 main topics. Each partner is responsible for 1 to 4 topics, even if the basic work is shared between all teams. Let us recall the numbering of the involved research centres :
1 for Toulouse (Fr) ; 2 for Bruxelles (Be) ; 3 for Prague (Cz) ; 4 for Budapest (Hu) ; 5 for Ljubljana (Si)The thematic reports correspond to the initial work plan.
This work has been pursued along two main lines: i) examining the stability of the time-schemes used in the non-hydrostatic (NH) version of the model, ii) examining the discretisation used for terms specific to the NH equations. The first study was based on the linear stability analysis and brought quite positive results. It became clearer how the linear stability depends on the choice of the so-called "background" temperature and surface pressure used in the semi-implicit formulation of the time-scheme. Further, it was shown that a proper choice of the prognostic variables may lead to an enhanced stability of the scheme and to the possible use of considerably longer time-steps. The linear analysis results were confirmed by a family of academic experiments. Besides, the "predictor-corrector" (PC) scheme was proved to be necessary to preserve the stability of the model in presence of a steep orography. The full-iteration PC scheme made it possible to integrate the two-time-level semi-Lagrangian NH scheme. The effort on the second study brought also significant improvements of the NH version of ALADIN. The most important one was the redefinition of the computation of a divergence term in the compressible continuity equation, where a volume conserving scheme was changed to a mass conserving scheme. Other improvements were done in the discretisation of the pressure gradient force, where the formulation is now closer to the hydrostatic primitive equation case. The new 2d (vertical plane) version of ALADIN proved very useful for such studies, providing a simplified, and cheap, framework for academic tests.
involved partners so far : P1, P3, P4First of all, the decision was taken to concentrate especially on the IOP (Intensive Observation Periods) cases from MAP (Mesoscale Alpine Project) in the lead centre (Ljubljana). For such situations, additional observations are available, either conventional ones with significantly higher spatial and time resolution, or new measurements such as wind profiler or aircraft data, useful in this framework. The main goal of the research in Ljubljana is to systematically evaluate the behaviour of high resolution models. As a consequence a thorough search for the cases where the reference / coupling model (here ALADIN / LACE) is already close enough to "reality" (observations) was performed. The main question to answer is "which additional quality can higher resolution bring" and not "why was the coupling model wrong for this situation".
Due to high computer usage demands of non-hydrostatic model experiments, there was quite a lot of work done in order to make the preparation of "climatological" files in high resolution easier. These files provide the basic framework of experiments : domain extension, resolution, orography, surface characteristics, ... This rather technical work is expected to make the creation of initial files with arbitrary horizontal and vertical resolution much faster. The main advantage will be a freedom in moving the integration domain to the area of interest, where additional data are available, so that the computational domain could be much smaller, only covering the region of interest for the particular situation.
Various experiments were performed to obtain informations about the model behaviour, comparing the hydrostatic and non-hydrostatic dynamics each time. Some knowledge about the typical difference between the two versions was gathered, as well as some insight into the behaviour of physics at higher resolutions, focusing on precipitations first. Some additional experiments were performed to isolate reasons for problems (and find solutions) and new sets of parameterisations were tested. Such resolutions of about 5 km are in the twilight region of numerical modelling, where the frontier between parameterized and resolved phenomena is rather hazy. Further investigations of the impact of non-hydrostatic dynamics will wait for the definition of a stable version.
involved partners so far : P1, P3, P5The first attempts to alleviate the problem of orographic resonance within semi-Lagrangian advection schemes were based on the so-called "double spatio-temporal averaging of orographic terms". When combined to an improved formulation of the continuity equation, this approach looked quite promising in simplified academic experiments. Unfortunately, the results using the full 3d model were negative. The topic is suspended for the time being, waiting for further investigations of the potential of the PC method (cf. Topic 1).
The study of the damping properties of a semi-implicit semi-Lagrangian scheme is being summarized in order to publish the obtained results. The underlying idea is to implement a new horizontal diffusion scheme in the model using the damping ability of semi-Lagrangian interpolators. The investigation of the potential of decentered semi-implicit schemes was also suspended.
involved partners so far : P1, P3The work on this topic started earlier than scheduled. The modifications to the model equations, required to relax the thin-layer hypothesis and simultaneously reintroduce some neglected Coriolis terms, were carefully analysed and documented. The hydrostatic and non-hydrostatic dynamics were considered, as well as the interface with physics. This was done in the ARPEGE framework first, i.e. not investigating the additional problems related to plane geometry and coupling. The code was modified consistently in the simple case of Eulerian advection, and 3d experiments performed to check that conservative properties and forecast skill are not affected at the current scales.
involved partners so far : P1The influence of the ratio in horizontal resolution between the coupling and the coupled models is under study. First results show that this is quite sensitive topic (there seems to be more noise in the meteorological fields near the border of integration domain when this ratio is bigger). Some experiments were also performed focussing on the part of vertical resolution.
Meanwhile, a new idea to combine the traditional grid-point values and spectral coefficients coupling was proposed in order to improve the way how the information from the coupling to the coupled model is transformed. The combined method works very well in the 2d horizontal barotropic model and it is clearly superior to the classical method. However there are unresolved problems yet with this scheme in the 3d baroclinic model.
involved partners so far : P1, P2, P3, P4, P5The basic blending technique was refined for both spectral and grid-point fields, and is now ready for an operational implementation in the ALADIN / LACE suite. Retunings and pre-operational tests also started for ALADIN / France. The specific case of nested models is now under study.
Blending was also combined for the first time with 3d-var (three-dimensional variational) assimilation in ALADIN. Its role is to take care of the large scales already analysed by the coupling model while the meso-scale analysis should possibly act only on the smaller ones. Various combinations were tested and compared to a classical 3d-var approach and to the dynamical adaptation mode. The results point in favour of the blended initial state to be used as a first guess for 3d-var analysis.
Another development concerns the coupling of the surface-pressure tendency instead of its absolute values. The preliminary results obtained with an academic mountain flow show an improvement when the new technique is used. However there is an additional difficulty to be solved for combining this new method of coupling with the semi-implicit time-scheme.
involved partners so far : P1, P2, P3, P4The new prognostic convection scheme (cf. Topic 8) was adapted to non-hydrostatic dynamics (use of additional variables, representation of new effects).
The robustness of the present physics-dynamics interface to variations of the time-step was controlled again, as a first step.
involved partners so far : P1, P2A wide re-analysis of the different features of the present convection scheme as well as its insertion among the other parameterisations was performed, in order to address the different limitations and hypotheses which loose their validity when going to high resolution meshes. This lead to different developments, applied presently for the hydrostatic runs of the model : representation of the pressure gradient between the cloud and the environment, downdraughts, ensemble effects. Currently under study is a prognostic approach for the draughts velocities and mesh fractions, with promising preliminary tests. The other enhancements will imply the parameterization of suspended condensed phases, and the use of the large-scale non-hydrostatic variables, especially the non-hydrostatic pressure departure, but probably also the pseudo-vertical divergence.
The relative parts of changes in the physics and increases in horizontal resolution were analysed on the case of the July 1997's flash floods over Central Europe. The impact of the last changes in the parameterisation of convection was examined within three nested models: ARPEGE, ALADIN/LACE and ALADIN/CZ.
The behaviour of several convection schemes were compared, in cooperation with the Spanish HIRLAM team : "Bougeault" (reference) and a first version of "Kain-Fristsch" in ALADIN, another version of "Kain-Fristsch" and a combination "Kuo-Sundquist" (reference) in HIRLAM. The tests were performed on two severe situations with the same domain extension and resolution for the two models.
Significant effort was devoted to the improvement of the parameterisation of vertical diffusion. A new formulation of the modified Richardson number was proposed, with a dependency on height. It enables to prevent the too frequent erosions of the inversion layer encountered last winter, while preserving the ability of the model to forecast cyclogenesis (such as the 1999's Christmas storms). The anti-fibrillation scheme was modified consistently. A detailed study of the sensitivity of the new exchange coefficients (dependency on height and on Richardson number) to the tunable parameters of vertical diffusion enables to check that sensible values were chosen for the first parallel tests.
The problem of low-level cloudiness was also addressed, focusing on the too weak development of stratocumulus over sea in anticyclonic areas. Tuning the current formulation of stratiform cloudiness or introducing other parameterisations (two new formulations were tested) didn't yield to an improvement. A diagnostic study showed that there is enough liquid water in the corresponding areas to produce a sensible cloudiness provided the formulation is slightly modified. Besides a new tuning of the present formulation of convective cloudiness was proposed, which could lead to a better spatial distribution of clouds.
Recent work has been done on the representation of the resolved and sub-grid scale orography features and their influence on the atmospheric circulation. In this scope, the parameterisations of the "form drag" and "lift" forces as currently implemented in ARPEGE / ALADIN have been evaluated and their effects compared to those of the "envelope orography" still in use in the operational versions. Following the negative results of this comparison, another way of approach to this problem is being investigated through the refinement of the concept of orographic roughness length, determining partially the amount of vertical diffusion of momentum in the lower layers of the atmosphere. This new approach includes a modification of the definition of the orographic roughness length by introducing in it some characteristics of the unresolved orography and the possibility to create a directional roughness length taking into account some slope or anisotropy characteristics.
involved partners so far : P1, P2, P3, P5A "functional boxes" approach for the parameterisation of liquid water and ice, defined jointly with the European HIRLAM NWP teams, is now working in the single-column version of the model. This will enable the comparison of various (existing or new) schemes, and even the evaluation of individual elements of parameterisations.
New procedures were designed to improve the initialization of the temperature of lakes in the model (i.e. along changes of geometry). A detailed impact study was performed using a high resolution version of the model centred around Lake Balaton (Hungary), to check the impact of such changes (up to 15 K). However results were somewhat deceiving, and showed that correcting the water temperature was often not enough to significantly improve forecasts. Atmospheric fields in the lowest layers have to be corrected accordingly. To this end a first parameterization of the evolution of the temperature of lakes was introduced in the model.
Two modifications of the parameterisation of evaporation over
sea were tested in one-month long assimilation experiments :
- reduction of the residual roughness length over sea,
- the same but adding a new parameterisation, which aims at increasing
evaporation by introducing a dependency on convective activity.
The global hydrological cycle and the trade winds are better described, though an improvement of scores is noticed only for buoy observations in the Tropics. A comparison to some field experiments was also performed.
A new high-resolution (~ 1km) global land-cover map was prepared (aggregation to a 5' resolution, derivation of useful fields) for use in ARPEGE and ALADIN and tested along data assimilation experiments in the global as well as in the limited-area cases. New model fields describing soil and vegetation were added simultaneously. Local (national) high-resolution data for soil characteristics were also tested in Budapest.
The parameterisation of the snow cover was deeply modified in
order to take into account :
- the time-evolution of the albedo of snow,
- the impact of vegetation (density, type, foliage) on the snow coverage,
- the impact of vegetation on the mean albedo.
The first tests are quite encouraging (improvement of the surface temperature, reduction of the correlations between forecast errors and vegetation cover).
New global databases for orography and urbanisation, with a better resolution, were compared to the previous ones, at the global and local scales. They were shown to lead to a real improvement in the description of surface. A method to improve the representation of coasts in the model was also designed (within the optimization procedure for spectral orography).
Some refinements to the thermodynamics were tested. To obtain a more exact computation of the saturating vapour pressure, the hypothesis of constant specific heat for liquid water and ice was relaxed. In the new version they depend on temperature as a second- and a first- order polynomials respectively. All the rest of thermodynamics and parts of the physics were modified accordingly. However the impact on forecasts was "small" considering the effort devoted to this improvement.
The radiation scheme was also considered, with an attempt to improve the mutual exchange between the model layers using more precise computations of the optical depths. Preliminary tests, performed with the single column model, showed a real improvement, especially at the surface. More validation (and eventually retuning) on real 3d cases is required anyway.
involved partners so far : P1, P2, P4Along this first year, the effort focused on unused SYNOP observations, as scheduled. Studies were performed in the framework of optimal interpolation analysis, waiting for progress in variational assimilation. Observations of snow height and visibility were considered. Hourly "classical" surface observations (mean-sea-level pressure, temperature and relative humidity at 2m, wind at 10m) are also used quasi-operationally now in Budapest and Toulouse.
Some work was also devoted to the use of new satellite data, but mainly in the framework of global large-scale assimilation. However the pre-processing tools designed for this purpose will benefit also to the meso-scale model.
involved partners so far : P1, P4The activities related to the three-dimensional variational (3d-var) analysis in the ALADIN model started in the framework of the ALATNET project in three different centres: Budapest, Prague and Toulouse. The working plan and schedule is fully respected and there are already significant advances in several subtopics, which gives a very good basis for the completion of the overall objectives within this scientific topic. The main achievements are described hereafter.
The background error statistics were already computed by several scientists on the basis of the classical NMC method for different domains of interest. The results of these investigations showed that in the covariances the large scale information still dominated, instead of the smaller scale information coming from the high resolution model. This fact might cause a problem in the process of 3d-var resulting in the re-analysis of the large scales (which are already analysed by the forcing model) instead of providing new information in the small scale part of the spectra. Realising this, new alternatives were searched, replacing the classical NMC method with derived algorithms likely to allow a higher dominance for the small scales.
Hereafter we shortly describe all the methods used until now for computing the background error covariances :
¨ STANDARD NMC statistics : Computation of error covariances based on the differences between two forecasts valid for the same time (typically 36h forecast and 12h forecast from the next day).
¨ LBC statistics : The same as above just taking into account the lateral boundary information from the global model. This algorithm gives an idea of the contribution of the large scale effects in the classical statistics.
¨ DIFF statistics : Computation of error covariances based on the differences of the inputs of the first two methods. This method was the first alternative algorithm, which tried to remove the large scale effects from the background error statistics.
¨ LAGGED constant coupling statistics : As in the classical case with the difference that the 12h forecasts are recomputed using the same lateral boundary information as it is for the 36h forecast. This method is a more sophisticated way to decrease the importance of the large scale forcing in the statistics.
The final evaluation of all the different methods showed that the lagged statistics algorithm has the best potential to decrease the large-scale energy in the covariances spectra, therefore it can be successfully used in the process of 3d-var. It was furthermore found that it is worthwhile to have an intensive sensitivity experimentation of the lagged method with respect to the forecast lengths and the time shift between the beginning of the forecasts. This study is going to create the following difference fields ("diff" refers to the forecast differences, "P" to the operational model run and "Q" to the new run with the same boundary conditions, both letters being associated to the range of the forecast):
diff=06 : P12-Q06, P18-Q12, P24-Q18, P30-Q24, P36-Q30,
P42-Q36, P48-Q42
diff=12 : P18-Q06, P24-Q12, P30-Q18, P36-Q24, P42-Q30, P48-Q36
diff=18 : P24-Q6, P30-Q12, P36-Q18, P42-Q24, P48-Q30
diff=24 : P30-Q6, P36-Q12, P42-Q18, P48-Q24
diff=30 : P36-Q6, P42-Q12, P48-Q18
diff=36 : P42-Q6, P48-Q12
diff=42 : P48-Q6
The background error statistics will be computed on these differences, which will give an idea, what is the best combination from the point of view of scale selection of the background errors.
The 3d-var scheme for the ALADIN model was first debugged and
informatically validated in Toulouse and then ported to Prague and Budapest
successively. The related work is now ongoing in these three centres. The
preliminary validations were concentrating on the following issues :
¨ Informatic validation, to ensure
that all the technical parts of the 3d-var procedure (from the data treatment
through the screening until the minimisation of the cost-function) are working
well and providing reasonable results for the different components of the
cost-function.
¨ Single-observation experiments,
to assess the basic characteristics of the structure functions.
¨ Full observation experiments, to
give an idea of the robustness of the scheme in case of taking into account
all the available information.
Based on these tests it can be said that the suites installed at the above
mentioned centres are behaving reasonably well and the scientific validation
can be started for the scientific assessment of the performance of the 3d-var scheme.
The scientific validation concerns in fact two main aspects :
i) the structure of analysis increments, and ii) the impact of the initial
conditions provided by 3d-var on the consecutive model integration. For the
time being the following main activities were carried out and results were
obtained :
¨ The initialisation technique is
of crucial importance on the analyses increments. The applied classical
digital filter technique might destroy some relevant small scale features,
therefore there is a strong need to find an alternative and relevant
initialisation procedure (for instance incremental digital filter
initialisation).
¨ The initialisation problem might
be treated together with the proper choice of the first guess used for the analyses.
¨ Investigations were carried out
to find the optimal ratio between the "observation" and
"backgoround" components of the cost-function with a-posteriori
evaluation of the scheme.
¨ Data assimilation experiments
taking into account different first-guess fields depending on the choice of
the background error statistics were launched.
¨ Experiments using a blended first
guess in combination with the lagged statistics were performed (see the report
on Topic 6 for more details).
In these above mentioned activities there is significant convergence towards a
scientifically sound 3d-var scheme.
During the first year of the ALATNET project there were lot of activities on the field of three-dimensional variational data assimilation. The results start to form a consistent basis for the full scientific validation of the scheme. The tasks to be tackled in the near future are as follows: further investigations for including more observations into the data assimilation process, final conclusions on the background error statistics to be used, final solution for the initialisation problem, final tuning of the relative importance of Jo and Jb terms, finding the optimal coupling strategy for the scheme. It is expected that for the next report most of these problems will be solved giving a first chance for the operational application of the 3d-var scheme.
Besides the investigation of the potential use of variational tools for sensitivity studies progressed in Toulouse with the work of an ALATNET Pre-Doc student and the work on singular vectors started.
involved partners so far : P1, P3, P4The main effort on this topic is likely to be delayed by about one year, since it appears safer not to split manpower between too many facets of assimilation as long as 3d-var assimilation is not fully validated and the work on observations not yet fully on a good way. However the operational implementation of 4d-var assimilation in ARPEGE in summer 2000 gave the opportunity to first face unexpected problems linked to the interaction between physics and the analysis of humidity at "small" scales. A new digital filtering scheme, fully embedded within 4d-var assimilation, was designed to improve the balance between fields at the full model resolution (and using the full set of physical parameterisations). Besides, the parts of the incremental approach (i.e. computing analysis increments at lower resolution and / or using a simplified physics) and the physics were investigated. This experience will benefit to the further work in ALADIN.
involved partners so far : P1Some more details may be found in the ALATNET Newsletters, available on the ALATNET Web-site : http : // www.cnrm.meteo.fr / alatnet / .
There is no change to mention here. (see Research Method in the initial programme of work)
The completion of the work plan was discussed and the initial program adjusted during the third meeting of the ALATNET steering committee (Bruxelles, 19 March 2001).
Breakdown of tasks
The detailed work plan was modified in order to :
¨ take into account the main observed time-shifts (in both directions),
¨ include new sub-items, relevant to the main objectives, but not identified or left out when the first version was designed (nearly two years ago),
¨ improve consistency, by moving some items from one main topic to another.
The new version may be found hereafter, with changes in bold characters.
1. Theoretical aspects of non-hydrostatism (NH) P3
1a: Development of the vertical plane version of ALADIN, as a powerful and cheap tool for further studies in dynamics [0.;1.5] P1
1b: Refinement and test of a radiative upper boundary condition for hydrostatic and non-hydrostatic dynamics [1.;3.] P1
1c: Improvement of the lower boundary condition [0.;3.] P1 & P4
1d: Re-examination of the semi-implicit approach [0.,2.] P1
1e: Solving residual instability problems in the two-time-level semi-Lagrangian advection scheme (so as to enable larger time-steps and a correspondingly lower computing cost) [1.;3.] P1 & P5
1f: Problems not yet identified [2.75;4.] P1, P2, P4 & P5
2. Case studies aspects P5
2a: Definition of the framework of experiments, choice of a set of reference situations [0.;0.75]
2b: Validation of the current physics and non-hydrostatic dynamics : comparison to hydrostatic dynamics, to observations, identifying problems [0.25;2.]
2c: Validation of the new developments in dynamics or coupling [1.75;4.] P3 & P4
2d: Validation of the refinements in physics, identifying feed-backs and residual problems [0.75;4.] P1 & P2
3. Noise control in high resolution dynamics P3
3a: Further damping of orographic resonance in semi-Lagrangian advection [0.;2.] P1
3b: Improved use of the damping properties of a decentered semi-implicit semi-Lagrangian advection scheme [1.;2.] P1
3c: Investigation of the potential of the predictor-corrector approach [0.;3.] P1
4. Removal of the thin layer hypothesis P1
4a: Analysis of the required modifications and coding [0.;2.]
4b: Impact studies in the vertical plane model then on real cases [1.;4.] P3
5. Coupling and high resolution modes P5
5a: Bi-directional coupling in spectral mode [1.5;4.] P1, P3 & P4
5b: Well posed lateral coupling in NH mode [1.25;4.] P1
5c: Problems of jump in resolution / domain sizes in modelling and data assimilation modes [0.;4.] P3, P4, P2
5d: Design of a more precise method for spectral coupling [.5,2.5] P4
6. Specific coupling problems P3
6a: Blending of fields in data assimilation for preserving high resolution forecast details [0.;2.] P1
6b: Tendency coupling for surface pressure and other technical variations around Davies' technique of field coupling in a buffer zone [0.;4.] P4 & P5
7. Reformulation of the physics-dynamics interface P2
7a: Study of the interactions between non-hydrostatic features and physical parameterisations [1.;3.25] P1 & P3
7b: Analysis of the problems related to a 1-dimensional physics, impact of an exact introduction of diabatic forcing [3.;4.] P1 & P5
7c: Sensitivity of the physics/dynamics interface to vertical resolution [1.;2.] P3, P1
8. Adaptation of physics to higher resolution P2
8a: Parameterisation of the small-scale features of convection [0.;1.25] P1
8b: Test, retuning and improvement of the various physical parameterisations in the framework of a very high resolution [0.75;3.75] P5
8c: Improved representation of boundary layer [0;3.25] P1
8d: Improved representation of orographic effects [0.;4.] P5, P1
9. Design of new physical parameterisations P1
9a: Implementation of a new parameterisation of turbulence [1.;3.] P5, P2
9b: Use of liquid water and ice as prognostic variables, implementation of a new microphysics parameterisation [0.;2.] P2
9c: New parameterisation of exchanges at sea and lake surface [0.;1.25] P4, P2
9d: Improved representation of land surface, including the impact of vegetation and snow [0;4.] P2, P4 & P5
9e: Refinements in the parameterisations of radiation and cloudiness [0., 4.] P4, P2
10. Use of new observations P1
10a: Yet unused SYNOP observations [0.;1.25] P2, P3, P4 & P5
10b: GPS and/or MSG observations [1.;3.]
10c: Doppler radar observations [2.25;4.] P3 & P4
10d: METOP (IASI) observations [3.;4.] P4
11. 3d-var analysis and variational applications P4
11a: Definition and calculation of new background error statistics, impact of domain resolution and extension, identification of horizontal relevant scales [0.;2.] P1 & P3
11b: Scientific investigation of the problem of the extension and coupling zone, analysis of the impact of initialization. [0.;1.25] P1
11c: Management of observations in 3d-var : from academic single-observation experiments to the use of any available data [0.,2.] P1 & P3
11d: Coupling problems in variational data assimilation, interaction with blending [0.;2.] P1, P2, P3, P5
11e: Intensive scientific validation and improvement of 3d-var [1.;3.] P1 & P3
11f: Development of variational type applications (adjoint methods for sensitivity studies, singular vectors), as a project itself and to provide more insight into the coupling problem for 4d-var [1.25;4.] P1
12. 4d-var assimilation P1
12a: Basic validation and tests [2;4.] P3 & P4
12b: Definition of a coupling strategy [2.;4.] P4 & P5
12c: Scientific validation [3;4.] P2, P3, P4 & P5
12d: Improvement of the treatment of humidity in data assimilation [0.;4.] P2, P3, P4 & P5
Schedule and Milestones
Changes in the schedule are detailed in the thematic reports and hereabove. The milestones were modified accordingly, but also (and mainly) since they were not consistent with the initial work plan, especially concerning the use of new observations.
The new milestones are defined as :
· After one year and a quarter :
- non-hydrostatic «reference version» of ALADIN at high resolution (with a «state of the art» parameterisation set) ready and tested on a few cases;
- blending of initial fields and 3d-var data assimilation ready for first parallel testing;
· After two years (Mid-Term Review) :
- completion of an intensive validation of the «reference version» with an appropriate coupling strategy and an improved physics;
- development of new observation operators started.
· After three years :
- operational implementation(s) of an update of the «reference version»;
- consistent data assimilation ensemble ready for operational implementation;
- well defined strategy for the assimilation of humidity at fine scales available.
· After four years (Final Review) : completion of the whole programme.
Research effort of the participants
The total research effort of the participants is very likely to differ from the initial estimation. It will increase with the necessary conversion of two Post-Doc into Pre-Doc grants (in Budapest and Ljubljana, participants 4 and 5 respectively), for lack of candidates. This is discussed in section B5. The old and new values are mentioned hereafter, together with the current stage.
Professional research effort on the network project | |||||||||
Participant |
Young researchers to be financed by the contract (person-months) |
Researchers to be financed from other sources (person-months) |
Researchers likely to contribute to the project (number of individuals) | ||||||
initial |
final |
so far |
initial |
final |
so far |
initial |
final |
so far | |
1 2 3 4 5 |
78 38 46 24 38 |
78 38 46 28 41 |
8.5 0 11 0 6 |
283 136 173 106 133 |
283 136 173 106 133 |
98 19 59 19 5 |
17 9 12 7 7 |
17 9 12 7 7 |
19 6 8 7 7 |
Totals |
224 |
231 |
25.5 |
831 |
831 |
200 |
52 |
52 |
47 |
The modified target efforts are in bold characters.
The research effort financed from other sources involves three contributions :
¨ the research (and organisation) effort from the permanent staff of each centre,
¨ the research effort from visitors invited on other fundings,
¨ the participation to the ALATNET training course,
the respective parts of which are :
Partition of the research and training efforts along the first year | |||||
Permanent staff |
Invited researchers |
Training courses | |||
Partner |
effort (person-months) |
researchers (number) |
effort (person-months) |
researchers (number) |
effort (person-months) |
1 2 3 4 5 |
54 19 19 18 4.5 |
17 6 7 7 7 |
43 0 10.5 0 0.5 |
24 0 12 0 1 |
0 0 25 0 0 |
Totals |
114.5 |
42 |
54 |
34 |
25 |
The total research and training effort on ALATNET fundings (young researchers excluded) lies around 7 person-months and, since small, is embedded the total effort.
One thus sees that the total scientific and teaching effort is roughly on target (both in volume and number of individuals touched), except for the Young Researchers part, this being linked to the difficulty in filling the Post-Doc positions, despite an equal publicising effort for the latter than for the PhD ones.
An ALATNET Web-site was created and is updated regularly : http://www.cnrm.meteo.fr/alatnet/. It provides informations on ALATNET events, young researchers, coordinators, research plan, training courses, ... and is linked to the Web-sites of the European Commission and of the ALADIN project (hence to the web pages of ALATNET partners).
A specific and open ALATNET e-mail list was created for exchanges between students, mentors, and coordinators : alatnet@meteo.fr. At the end of March 2001, it gathers 18 addresses. Students can also use the other ALADIN e-mail lists.
An ALATNET Newsletter is published every 6 months, jointly with the ALADIN Newsletter (but with a clear distinction between contributions). It mentions ALATNET events and describes the progress in scientific work. The two first Newsletters are a hundred pages long. There are available on the ALATNET Web-site and were sent to all ALADIN partners and to each of the five SRNWP coordinators (Short Range Numerical Weather Prediction, a European network involving National (Hydro-) Meteorological Services, hereafter NMSs).
Given the attendance of the Radostovice Seminar by scientists from non-ALADIN SRNWP groups and the recruitment of two non-ALADIN Young Researchers among six (see Item B.5.5), and considering the effort towards ALADIN non-ALATNET countries in these two occasions, one can probably say that the communication strategy of the network is adequate but needs to be further maintained. The Web-site as well as the Newsletter are now well established communication facilities and proved very useful for the circulation of information among the participants (as well as for gathering some of the information for the annual report !).
The visibility of the Young Researchers work is perhaps yet insufficient, but the programme of participation to workshops that has been prepared for them (see Item B.4.5) should hopefully improve the situation there. Furthermore, they will have to contribute to the next ALATNET / ALADIN Newsletters, as do the other ALADIN PhD students.
§ First meeting of the ALATNET steering committee :
- Toulouse (Fr), 17 April 2000
- rules for candidates selection and employment, effort registration, use of fundings, ...
- presentation of ALATNET to ALADIN people
§ Kick-off meeting for Research Training Networks :
- Bruxelles (EC), 9 June 2000
§ Second meeting of the ALATNET steering committee :
- Toulouse (Fr), 12 October 2000
- problems encountered so far, sending students to workshops, ...
§ Third meeting of the ALATNET steering committee :
- Bruxelles, 19 March 2001
- next call for candidacies, revision of the work plan, ...
The corresponding reports, as well as an historical account of the project life, are available on the ALATNET Web-site.
§ Short visits between research centres along the first year (on ALATNET fundings or not, excluding the participation to the first ALATNET training course) :
¨ Toulouse Þ Bruxelles
J.F. Geleyn , |
09/11/2000 |
: coordination |
´ Toulouse Þ Prague
¨ Toulouse Þ Budapest
C. Fischer , |
28/05/2000 - 09/06/2000 |
: 3d-var |
´ Toulouse Þ Ljubljana
¨ Bruxelles Þ Toulouse
A. Quinet , |
17/04/2000 |
: coordination |
O. Latinne , |
04/09/2000 - 02/11/2000 |
: improved representation of vegetation |
P. Nomérange, |
04/09/2000 - 18/10/2000 |
: orographic effects at high resolution |
P. Termonia , |
10/12/2000 - 17/12/2000 |
: coupling problems |
A. Deckmyn , |
11/02/2001 - 23/02/2001 |
: blending |
´ Bruxelles Þ Prague
´ Bruxelles Þ Budapest
´ Bruxelles Þ Ljubljana
¨ Prague Þ Toulouse
R. Brozkova , |
06/03/2000 - 01/05/2000 |
: 3d-var |
R. Mladek , |
24/08/2000 - 08/09/2000 |
: management of surface fields |
D. Merkova , |
29/11/2000 - 24/12/2000 |
: physics and high resolution runs |
R. Brozkova , |
01/01/2001 - 28/02/2001 |
: non-hydrostatic dynamics, 3d-var |
P. Smolikova , |
01/02/2001 - 28/02/2001 |
: non-hydrostatic dynamics |
´ Prague Þ Bruxelles
¨ Prague Þ Budapest
M. Siroka , |
28/05/2000 - 09/06/2000 |
: 3d-var |
´ Prague Þ Ljubljana
¨ Budapest Þ Toulouse
T. Szabo , |
01/03/2000 - 14/04/2000 |
: non-hydrostatic dynamics |
S. Kertesz , |
01/03/2000 - 14/04/2000 |
: parameterisation of lakes |
A. Horanyi , |
17/04/2000 |
: coordination |
R. Randriamampianina |
01/04/2000 - 25/06/2000 |
: parameterisation of radiation |
A. Horanyi , |
02/10/2000 - 13/10/2000 |
: 3d-var |
G. Radnoti , |
23/10/2000 - 04/11/2000 |
: visibility observations coupling problems |
G. Radnoti , |
20/11/2000 - 15/12/2000 |
: coupling problems |
G. Radnoti , |
22/01/2001 - 18/02/2001 |
: coupling problems |
´ Budapest Þ Bruxelles
¨ Budapest Þ Prague
T. Szabo , |
02/10/2000 - 29/10/2000 |
: non-hydrostatic dynamics |
G. Boloni , |
20/11/2000 - 17/12/2000 |
: 3d-var |
T. Szabo , |
27/11/2000 - 23/12/2000 |
: coupling of surface pressure tendency |
¨ Budapest Þ Ljubljana
G. Radnoti |
20/11/2000 - 15/12/2000 |
: coupling problems |
¨ Ljubljana Þ Toulouse
M. Zagar , |
17/04/2000 |
: coordination |
N. Pristov , |
09/10/2000 - 13/10/2000 |
: coordination |
´ Ljubljana Þ Bruxelles
¨ Ljubljana Þ Prague
D. Cemas , |
14/08/2000 - 03/09/2000 |
: high resolution experiments |
U. Strajnar , |
30/10/2000 - 12/11/2000 |
: high resolution experiments |
N. Pristov , |
06/11/2000 - 19/11/2000 |
: high resolution experiments |
M. Zagar , |
01/03/2000 - 23/06/2000 |
: horizontal diffusion of humidity |
¨ Ljubljana Þ Budapest
N. Pristov , |
02/07/2000 - 07/07/2000 |
high resolution dynamical adaptation of wind |
§ Invitations on ALATNET fundings concerning European non-ALATNET countries
¨ in Bruxelles (Be, P2) : Elena Cordoneanu (Romania), Andrey Bogatchev (Bulgaria), Martin Bellus (Slovakia), each for a few days (experts for Topic 4);
¨ in Radostovice (Cz, P3) : free accomodation provided for students from non-ALATNET and EC-eligible countries.
§ Discussions between involved scientists during European workshops :
¨ 8th ALADIN workshop in Cracow, Pl, (19-20/06/2000), with the participation of :
E. Gérard, D. Giard (Toulouse centre)
L. Gérard, P. Nomérange, P. Termonia (Bruxelles centre)
D. Merkova (Prague centre)
G. Radnoti (Budapest centre)
M. Zagar (Ljubljana centre)
¨ EWGLAM / SRNWP meeting in Toulouse, Fr, (09-13/10/2000), with the participation of :
G. Balsamo, C. Soci, I. Gospodinov (ALATNET young researchers)
E. Bazile, P. Bénard, L. Berre, P. Caille, C. Fischer, J.F. Geleyn, E. Gérard, D. Giard, J.D. Gril, P. Moll, J. Pailleux, P. Pottier (Toulouse centre)
O. Latinne, P. Nomérange, L.M. Trépant, J. Vanderborght (Bruxelles centre)
R. Brozkova, M. Janousek, D. Klaric, M. Siroka (Prague centre)
A. Horanyi (Budapest centre)
N. Pristov (Ljubljana centre)
¨ 9th ALADIN workshop in Bruxelles, Be, (06-08/11/2000), with the participation of :
J.F. Geleyn (Toulouse centre)
L. Gérard, O. Latinne, P. Nomérange, P. Termonia, J. Vanderborght (Bruxelles centre)
I. Bonta, F. Wantuch (Budapest centre)
A. Poredos, A. Hrabar (Ljubljana centre)
¨ 5th Assembly of ALADIN Partners in Vienna, Au, (24/11/2000), with the participation of :
J.F. Geleyn, D. Giard (Toulouse centre)
R. Brozkova (Prague centre)
A. Horanyi (Budapest centre)
J. Jerman (Ljubljana centre)
§ Additional distant supervision of young researchers between ALATNET centres :
J.F. Geleyn (Fr) >> L. Gérard (Be)
J.F. Geleyn (Fr) >> P. Nomérange (Be)
G. Radnoti (Hu) >> P. Termonia (Be)
P. Bénard (Fr) >> P. Smolikova (Cz)
R. Brozkova (Cz) >> T. Szabo (Hu)
§ Regular e-mail exchanges between contact points for each topic in ALATNET centres : as usual.
For each call, vacancies are published on the Web-site of the E.C. and a notice is sent to each European ALADIN partner, and to SRNWP coordinators who are responsible for broadcasting informations to other European NMSs. The publicity to universities is left to individual NMSs.
§ First call for candidacies :
- publication : 19 April 2000
- deadline : 12 May 2000
- selection : 30 May 2000
- positions :
4 Pre-Doc positions opened (2 in Toulouse, 1 in Prague, 1 in Ljubljana), and filled (10 candidates)
3 Post-Doc positions opened (1 in Bruxelles, 1 in Prague, 1 in Budapest), only 1 filled (1 candidate)
§ Second call for candidacies :
- publication : 26 December 2000
- deadline : 1 February 2000
- selection : 2 February 2001
- positions :
3 Post-Doc positions opened (2 in Bruxelles, 1 in Budapest), only 1 filled (1 candidate)
§ Problems encountered so far
The ALATNET RTN had to face a very severe lack of candidacies on the opened Post-Doc grants : 1 candidate only for 3 positions opened at each call. Oppositely there was a great demand for Pre-Doc grants (10 candidates for 4 positions).
This was discussed at the last ALATNET meeting, and lead to the following proposals :
¨ to prolong the contract of Christopher Smith in Prague from 12 to 23 months
He is doing an excellent job, he applied last year with the intention to profit as much as possible of the two years leave of absence he took from his job at UKMO and it is really doubtful we could find any other suitable candidate without this step (given the lack of candidacies and the need to also fill the position remaining open in Bruxelles). This would mean to keep exactly the original schedule but in one piece rather than in two.
¨ to transform the 16 months long Post-Doc position in Budapest in a 20 months Pre-Doc stay
Originally it was thought that due to the need of basic knowledge on data assimilation issues it would be nice to have an experienced Post-Doc student for the proposed work on three-dimensional variational data assimilation. In our understanding the main reason for the failure of both calls to candidacy is the age limit of 35 years, far more penalising at the Post-Doc level than for a Pre-Doc offer. We still believe that the topic is very challenging and that it is worthwhile to keep it active in the framework of the ALATNET project. Therefore it seems reasonable to try to transfer the Post-Doc stay to an equivalent Pre-Doc position (same amount of funds, longer stay), especially given the fact that there are now already research activities performed at HMS for 3d-var, which makes the start of such work much more easier now than it would have been be the case earlier. Furthermore, from experience of the first call, there is a chance to have a strong competition for this position, this giving a good possibility of efficient choice.
¨ to do the same in Ljubljana (moving from 13 to 16 months)
The arguments are roughly the same as for the previous case, except that the change is even more needed for logistic reasons (the further failure of a call for candidacy would create a bottleneck of hosting/supervision possibilities towards the end of the contract) and that it should take place with a redefinition of subject and a common proposal with Budapest, using the 4 remaining months of "Post-Doc visit" available there to get a more comfortable total of 20 months altogether.
Unless the European Commission disagrees with the previous proposals, the partition between Pre-Doc and Post-Doc positions will be changed as follows :
Pre-Doc positions |
Post-Doc positions | |||||||
Centres |
old (7) |
new (9) |
old (6) |
new (3) | ||||
Starting Year |
Length (months) |
Starting Year |
Length (months) |
Starting Year |
Length (months) |
Starting Year |
Length (months) | |
Toulouse |
2000 2000 2001 2001 2001 |
16 16 16 15 15 |
2000 2000 2001 2001 2001 |
16 16 16 15 15 |
||||
Bruxelles |
2000 2001 |
19 19 |
2001 2001 |
19 19 | ||||
Prague |
2000 |
23 |
2000 |
23 |
2000 2001 |
12 11 |
2000 |
23 |
Budapest |
2001 2002 |
4 (*) 4 (*) |
2001 2002 2001 |
4 (*) 4 (*) 20 |
2000 |
16 |
||
Ljubljana |
2000 |
25 |
2000 2001 |
25 16 |
2001 |
13 |
(*) visit from an ALATNET PhD student from Toulouse (2001) or Ljubljana (2002)
The positions already filled are in italics.
and the progress in recruitment compared to the initial and final objectives reads :
Participant |
Young Researchers to be financed by the contract initial version (person-months) |
Young Researchers to be financed by the contract final version (person-months) |
Young Researchers to be financed by the contract so far (28/02/2001) (person-months) | ||||||
Pre-doc |
Post-doc |
Total |
Pre-doc |
Post-doc |
Total |
Pre-doc |
Post-doc |
Total | |
1 |
78 |
0 |
78 |
78 |
0 |
78 |
10 |
0 |
10 |
2 |
0 |
38 |
38 |
0 |
38 |
38 |
0 |
0 |
0 |
3 |
23 |
23 |
46 |
23 |
23 |
46 |
5 |
6 |
11 |
4 |
8 |
16 |
24 |
28 |
0 |
28 |
0 |
0 |
0 |
5 |
25 |
13 |
38 |
41 |
0 |
41 |
6 |
0 |
6 |
TOTAL |
134 |
90 |
224 |
170 |
61 |
231 |
21 |
6 |
27 |
The modified target efforts are in bold characters.
§ Third call for candidacies :
- publication : 30 March 2001
- deadline : 31 May 2001
- selection : 7 June 2001
- positions :
5 Pre-Doc positions opened (3 in Toulouse, 1 in Budapest, 1 in Ljubljana)
1 Post-Doc position opened (in Bruxelles)
Given the standing structure of the ALADIN action, the experience already acquired in Toulouse and Prague for structured visits and the past benefits of the bilateral Sloveno-French and Hungaro-Slovenian actions, there was no particular surprise in the integration of the five Young Researchers in their new teams. Administrative difficulties from the host countries were however noticeable in Prague and are still delaying the start of the stay of Ilian Gospodinov in Bruxelles.
§ Individual training of young researchers
The basic training of young researchers :
- on ALADIN whenever required (i.e. for Gianpaolo Balsamo and Christopher Smith),
- on the computing environment,
- on the bases of the chosen research topic,
relies on individual hosting NWP teams and mentors, as well as the attendance to local seminars or training courses.
§ Participation to European workshops
Each ALATNET young researcher will be sent to some European workshops from 2001 :
- all of them shall attend the EWGLAM / SRNWP workshops
(Cracow, Poland, in 2001 ; De Bilt, Netherlands, in 2002 )
- each of them shall attend thematic SRNWP workshops, the next ones being :
¨ on Numerical Methods (Bratislava, Sk, July 2001) for Ilian Gospodinov, Christopher Smith, Klaus Stadlbacher and Jozef Vivoda (with an informal ALATNET coordination meeting on non-hydrostatic dynamics),
¨ on Surface Processes (joint with HIRLAM workshops) for Gianpaolo Balsamo,
¨ replaced by a stay within an HIRLAM team working on variational assimilation for Cornel Soci (next meeting in 2002),
¨ to be defined for the next ones;
¨ some of them (Gianpaolo Balsamo, Ilian Gospodinov and Cornel Soci) already attended the last EWGLAM / SRNWP workshop (Toulouse, Fr, 09-13/10/2000).
They may also attend ALADIN workshops (2 per year, next one in Toulouse, Fr, 07-08/06/2001).
§ ALATNET training courses
The ALATNET program of work includes the organization of 3 training courses covering the main aspects of Numerical Weather Prediction (NWP) :
¨ on High Resolution Modelling (Radostovice, Cz, 15-26/05/2000) ,
¨ on Data Assimilation (Gourdon, Fr, 11-22/06/2001) ,
¨ on Numerical Technics (Si, 2002) .
These seminars are open to European students and to non-European ALADIN students. All European NMSs are informed in due time, as for ALATNET vacancies. The ALATNET young researchers have to attend this advanced training (once employed of course).
The first training course was a real success, with 57 participants from 15 countries, along two weeks. Some details may be found on the ALATNET Web-site. Its organization was shared by Prague (practical problems) and Toulouse (definition of lectures, choice of teachers).
For the first call, the rules defined by the ALATNET steering committee during its first meeting were as follows :
«
* the rank he gives to each of the group of 2 subjects : from the 5 answers, a priority order will be given to these groups (in case of equality, casting vote of Météo-France) ;
* the ordered list of candidates for each group of subjects.
¨ Inside each group of subjects, candidates will be ranked according to the weighted average (2 points for the country who will receive the candidate, 1 point for the other 4; in case of dead heat, the opinion of the receiving country will be followed)
¨ The first candidate to the first group of subjects (depending of the priority order of the subjects) will be accepted; in case he / she applies for other vacancies, these candidacies will be cancelled.
¨ Then, the first remaining candidate for the second group of subjects is accepted and his others candidatures are cancelled; idem for the 3rd and 4th groups of Pre-Doc subjects.
¨ One subject will be chosen within each group : choice of the retained candidate (if he applied for only one subject) ; collective decision otherwise (not necessary following the order of priority of the retained candidate) ;
¨ Waiting list :
* elimination of the candidatures on subjects not retained;
* elimination of the retained candidates of the first list;
* selection of the remaining candidates with the same procedure as for the first list (people will be in one single waiting list on their best position).
¨ If necessary, a similar procedure will be defined for the Post-Doc subjects / candidates.
¨ Phone conference between ALATNET correspondents on March 30th, 11.30 to validate the results or correct any stupid result.
»
For the second call, no rules were necessary : only one candidate, with the required experience.
This lead to the following partition of selected candidates :
Young researchers / Candidates | ||||
Pre-Doc |
Post-Doc | |||
Country |
Women |
Men |
Women |
Men |
ALADIN ^{(*)} |
0 / 2 |
3 / 5 |
0 / 0 |
1 / 1 |
non ALADIN |
0 / 1 |
1 / 2 |
0 / 0 |
1 / 1 |
^{(*) }For the first two calls, candidates from ALATNET countries were not allowed, to better spread over Europe the benefits of the network.
For the third (and last) call, the steering committee will keep in mind the need to have a better partition between women and men, and possibly involve also the HIRLAM community (HIgh Resolution Limited Area Modelling, a NWP partnership involving Denmark, Finland, Ireland, Iceland, Netherlands, Norway, Spain, Sweden and cooperating with Baltic States).
The parity women / men was better preserved in the participation to the first ALATNET training course. A few students came from non-ALADIN countries (Italy, Germany, Sweden). There was no selection, of course.
Involved Countries |
Parity |
First Training Course | |
Students |
Teachers | ||
ALATNET (with young researchers) |
Women |
7 |
2 |
Men |
12 |
6 | |
European , ALADIN , non-ALATNET |
Women |
12 |
0 |
Men |
11 |
1 | |
European , non-ALADIN |
Women |
2 |
0 |
Men |
3 |
1 | |
non-European , ALADIN |
Women |
0 |
0 |
Men |
0 |
0 | |
total contribution |
Women |
21 |
2 |
Men |
26 |
8 |
The recruitment procedure for the first ALATNET Young Researchers, despite being pushed as early as possible once the contract was signed came too late for enforcing a systematic attendance of the selected persons to the Radostovice Seminar. It simply happened that Klaus Stadlbacher and Jozef Vivoda were among the trainees. The situation will of course be better controlled for the forthcoming ALATNET training events (see Item B.4.5).
This part of the ALATNET programme is indistinguishable from the global effort to make the ALADIN project a real link between training, research and operational aspects of Numerical Weather Prediction, a multi-disciplinary theme in itself (fluid dynamics, atmospheric physics, signal processing, numerical analysis, optimal control, behaviour of stochastic systems, interface with computing techniques, ...). A few numbers are probably sufficient to explain how structured is this part of the environment of the ALATNET effort: more than 30 equivalent-persons, between 1 and 2 PhD per year, 35% of the effort realised through research-training stays outside the 15 home institutions, this corresponding to 13% of the total cost of the project (salaries and computing-costs included).
NWP activities are a rather internal business since they encompass their own "industry" (i.e. the production of daily operational weather forecasts, mainly for public service aims and for regulated aeronautical forecast procedures) and one more striking figure of the ALADIN project is the existence of 13 (pre-)operational versions. This mixed research/operational environment thus offers a good opportunity for witnessing the transfer process from research to daily application (and its specific problems), even if the Young Researchers are not directly participating in this part of the NWP effort. The link with computer manufacturers is also important in such a "simulation bound" scientific activity. For example the Radostovice Seminar was indirectly sponsored by COMPAQ who offered an "alpha" workstation for remote connections and local computing of demonstration examples.
Apart from the expected level of basic problems when starting such a large structured effort like ALATNET, two major obstacles have been encountered :
* administrative difficulties :
(i) the Czech Embassy in London was unable to correctly state all required documents for the entry of Christopher Smith into the scheme in Prague and later even forced him to make a return trip to London to regularise things (for matters that could well have been treated on the spot in Prague itself, but for a "surprising" legislation);
(ii) the Belgian Government is enforcing an employment (rather than grant) system for the Young Researchers, requiring therefore very long administrative delays for non-EC students, which made the recruitment of Ilian Gospodinov by IRM really late and difficult (more than 4 months delay for obtaining a visa).
In both cases there are signs of contradiction between a willingness to benefit from the EC programmes and an exaggeratingly suspicious application of national rules; the ALATNET management group believes that it is up to its contact points in the TMR / IHP EC programme to see how to best deal with this type of annoying information.
* the difficulty to find suitable candidates for Post-Doc positions; the common age limit with Pre-Doc candidates may play some part in this problem but it also appears that the situation is similar for other types of funding and quite different there from that a few years ago; our analysis was therefore that a redefinition of the structure of the training programme was necessary in order to face such a strong conjunctural obstacle (see Item B.5.2).
Position : Pre-Doc, Ljubljana
Research topic : Systematic qualitative evaluation of high-resolution non-hydrostatic model
The work in the framework of the ALATNET-project started in Ljubljana in September 2000 and this actual first part is lasting for twelve months, while a second thirteen-month stay will start in September 2002.
The main purpose of the work is to systematically evaluate the non-hydrostatic version of the ALADIN model together with the results for very high resolutions (down to 1 km), respectively.
The first part of this stay was mainly used to set up a proper working environment, to learn all the 'wheres', 'whens', 'whos' and 'hows', that are needed for starting model experiments and displaying results and all the additional stuff, that is necessary to establish perfect working conditions.
To reach the above mentioned goal, it is first necessary to find a database of cases, that allows to investigate the performance of the different model settings. The MAP database provides a proper source for finding such cases, cause for the IOP's large amounts of data (model and observation ones) are available.
The choice of the cases was influenced by the fact, that the main differences between hydrostatic and non-hydrostatic dynamics can be expected in situations, where vertical velocity plays a major role for the prevailing weather situation, but besides this also weather situations for strong wind and 'quiet' situations are going to be examined.
Such first mentioned cases occur during the MAP-SOP in 1999 for example on 20.9.,4.10. and the 23.10., where in all of this situations remarkable amounts of precipitation were forecast for the area, for which the test-runs were done (north-east part of Italy and western half of Slovenia).
In the following pictures one example of comparison between non-hydrostatic and hydrostatic dynamics and the differences between the choice of the lateral boundary conditions is shown as well. The latter means, whether the boundary conditions for the 2.5 km resolution are taken from the 'operational' (11 km resolution) or from the 'intermediate' (5 km resolution) model results.
The spatial resolution is about 5 km or 2.5 km respectively. The 3 time-level semi-lagrangian scheme was used for the model integration and the length of the time-step was 60 s or 30 s respectively.
The relevant synoptic feature for the model domain is a cold front, which approaches from the west and causes partly large amounts of precipitation in the domain of interest, although the real front was already a bit weaker than expected by the model.
All following pictures show the forecasted precipitation amount for a duration of two hours. The forecast time is 12 hours and the model run starts on 20.9.1999 at 12:00 UTC.
The precipitation patterns for non-hydrostatic and hydrostatic run look quite similar, although there is a difference, when focus is put on the areas of maximum precipitation (Croatian coast and border area Italy-Slovenia). The maximum value forecasted by the hydrostatic run exceeds the non-hydrostatic in both cases (the 'direct' coupling and the 'nested 'coupling) and both amounts are higher, than the observed ones.
Nevertheless the forecasted patterns are similar to the observed precipitation. So it seems that the non-hydrostatism damps the maximum values towards more realistic ones.
Comparison of the direct coupling to the nested one shows clearly, that precipitation patterns in the further case are sharper than in the latter one and that this difference is more remarkable compared to the difference in non-hydrostatic versus hydrostatic dynamics.
When the differences between different resolutions should be focused on, it can be stated, that for the 5 km resolution the maxima are far lower than for the 2.5 km resolution (just 50% of the value) and again that the differences between H and NH dynamics are quite small (It's the same as for the higher resolution). Additionally it can be seen, that the pattern for the lower resolution are still very smooth, while in the 2.5 km resolution there is much more (unrealistic?) structure, that have some 'wave-look'.
For the next future an intensive study of this and several other cases is planned with focusing on forecasts also for other domains (for example to check the influence of orography in detail), higher resolution and higher frequency of coupling (the presented experiments were done with 6-hour coupling frequency).
I would also like to use this first official report for saying a big "THANK YOU" to everybody in HMIS (and especially in the OpTiM group), who was endeavoured to create excellent working conditions for me here. This includes providing new computer devices and supporting me in any imaginable way.
Precipitation forecast [mm/2h] 1999092012+12h
hydrostatic dynamics versus non-hydrostatic dynamics
resolution of the model : 2.5 km - resolution of the coupling model : 11 km
resolution of the model : 2.5 km - resolution of the coupling model : 5 km
resolution of the model : 5 km - resolution of the coupling model : 11 km
Position : Pre-Doc, Toulouse
Research topic : Sensitivity studies using a limited-area model and its adjoint for the meso-scale range
The purpose of this study is to establish the role played by the initial conditions (IC) and lateral boundary conditions (LBC) in a short range mesoscale forecast. The adjoint method has been used to compute the gradients of the forecast errors with respect to the ICs and LBCs. These were introduced as small perturbations of the type dx=-a*Ñx in the initial data, x being the model state fields and a a scaling factor. The sensitivity tests were carried out for the different forecast range and integration area. We have tried to distinguish between the errors coming from a lack of information in LBCs and inaccuracies in the initial state.
The domain used in our experiments has a grid size of 30 km, elliptic truncation E49 and 150x144x19 points. The integration time chosen in this study was mainly 36 hours starting from 23 December 1997, 00 UTC. The results have demonstrated that the ICs play a leading role for a good forecast. On the contrary, the forecast was less sensitive to the modifications in the LBCs. In Fig. 1a the gradients with respect to the IC, ¶J/¶T, are shown. Large gradients values are concentrated in the south-western part of the domain. This is the region where small perturbations introduced in ICs have a positive impact on the forecast. For example, 2.5 hPa have been added in the sea level pressure field taken the difference between operational ICs and the modified one. Starting from these modified initial data, the model reach a solution which gives birth to a cyclonic core over the British Islands as it is illustrated in Fig. 1b. The pressure of the cyclonic centre is 983 hPa, close to the verifying analysis available 24 December 1997, 12UTC whose core is 988 hPa but is far from the reality since at Valentia meteorological station, the mean sea level pressure was 975 hPa. Even if the forecast cyclone is not as deep as in reality, we can say that the errors coming from ICs are responsible for the failure of the operational forecast. These experiments are in harmony with those undertaken for the same case but with the global model ARPEGE (G.Hello et al., 2000).
For this case, the domain used was the operational ALADIN / France (E95, 288x288x31 points, dx=9.9 km). For this high-resolution integration area a first problem regarding the gradients computation has been revealed. Strong punctual gradients over the Alps have been found (Fig. 2a). These structures are present in the adiabatic model and they do not vanish when the Eulerian time-step is reduced from 40 to 10 s. It was found that even for 10 s time-step there are high values over the orography for certain levels. For diminishing these unrealistic values, the boundary layer height (h) and the wind velocity scale factor (u*) in the adjoint linear simplified physics (Buizza, 1993) have been modified. Thus, we increased h from 1 to 1.5km and the value of u* for land and sea to 1 and 0.4 respectively. Another way to get rid of these structures was to change the trajectory truncation in the adjoint model, for example from E95 to E35.
It is very likely that using simplified physical parameterisation (Janiskova, 1999) in ALADIN, the unrealistic structure of the gradients will disappear. A first step forward during this study was to verify that this package is functional in the model and that it works properly.
Again, like for the Irish storm, it was found that the influence of the perturbations in LBCs is in positive sense but is rather week. However, an improved forecast is obtained if the ICs are modified. In the sensitivity forecast the model was able to move the cyclone centre closer to the verifying analysis but with a very high price. For example, more than 5 hPa have been added into a region where some useful gradient was present over the Bay of Biscay. It is very likely that an important part of the gradient has already gone out from the domain crossing the western boundaries and it should be found over the Atlantic Ocean.
During this first ALATNET stay, the study has been concentrated on the effect of the perturbations in both lateral boundary data and initial conditions in the prediction of a rapid evolving cyclogenesis. In every experiment in which only lateral boundary perturbations was injected, the impact in the forecast was very marginal. On the contrary, the results have shown that the errors in ICs are more responsible for the forecast failure. This is not because the information coming from the coupling area is perfect. The reason could be that the utilised technique for deriving LBCs gradients is not proper for ALADIN. We also think that this failure is due to the coupling method (Davies) used and we could say that for a future 4d-var assimilation in ALADIN a new coupling method should be imagined which, eventually, to be able to distinguish between inflow and outflow region of the domain.
In the future, other case studies will be carried out mainly to study the impact of simplified physical parameterisation in ALADIN adjoint model. We will concentrate then on the sensibility of the mesoscale phenomenon inside the forecast area.
Position : Pre-Doc, Toulouse
Research topic : Coupling a variational assimilation of gridpoint surface fields with a 4d variational assimilation of upperair spectral fields
The initialization of the temperature and the humidity in the soil in Numerical Weather Prediction models is important for short- and medium- range forecasting. The model biases (precipitation, radiation, etc.) may have a cumulative effect on surface fields and lead to large errors which will degrade the forecast of surface fluxes (sensible and latent heat). The aim of this work is to study a variational surface analysis of temperature and humidity in the ground over continental areas in a limited area model. The analysis of sea surface temperature, snow or frozen water in the soil is not investigated. The surface analysis should assimilate indirect observations like precipitation, 2m- observations, satellite observations, etc., since the observations of temperature and humidity in the soil are very sparse and not taken routinely. The operational surface analyses at Météo-France, ECMWF, DWD are assimilating 2m- observations considering the importance of this observational network and their links with the soil state. The variability of these links on meteorological conditions is important and has to be taken into account in the surface analysis. First we have evaluated 2m forecast errors for temperature and relative humidity since they are used as input of the surface analysis. For that, 48 h-forecasts are performed with the ALADIN / France model and compared hourly to 2m-observations in order to study the diurnal cycle for 2m-temperature and relative humidity errors over the ALADIN / France domain.
A catalogue of the critical biases and root mean square errors (rms) is built over a selection of significant cases likely to be highly sensitive to soil initialization. The synoptic situations, subjectively chosen, are characterized by anticyclonic circulation interesting the western European areas over a several days period. In such cases, a weaker dependency from the upper level variables allows to state a clearer correlation between screen level variables and soil conditions. Cyclonic patterns are also considered for completeness. The domain chosen for testing is centred over France where the ground-based network gathering hourly observations is dense. A quality control based on comparison to forecast fields is applied to remove invalid data. The stations located at elevation higher than 800 m are not considered, to avoid to weight the biases due to the 2m errors over orographic context, not a primary focus of this study. Three anticyclonic (in January, June and September 2000) and three cyclonic periods of about 10 days were chosen as case-studies. The mean bias and rms for temperature (T) and relative humidity (RH) at 2m, although of small entity (1-2 °C for T2m, 10-20 % for RH2m), show a cyclic diurnal pattern that suggest the possible improvement of physical parameterization. Assuming that the three selected periods are representative of seasonal anticyclonic circulations, the presence of a cyclic bias over the year is also mentioned. A less regular behaviour is present in cyclonic cases, as expected, due to the greater number of processes involved in the 2m forecast errors, like misrepresentation of cloudiness or horizontal advection.
The June case presents the largest values for rms and bias of 2m temperature and relative humidity with peaks scattered all over the domain. This case is interesting for weighting the soil wetness impact on 2m forecast errors since the solar radiative forcing is strong. The initial soil wetness index (SWI) is considered for purpose of comparison, as it provides a fractional value of the total water content of the soil between the wilting point and the field capacity. A clear relation between this index and 2m temperature and relative humidity errors is found during well developed anticyclonic days. The 16th of June forecast is chosen to compare the initial SWI and the 42 h temperature forecast errors at 2m (for 17th June 18 UTC). The comparison is done at this time (18 UTC) to weight the action of the whole day forcing of solar radiation. Two main considerations are extracted. Firstly, it is possible to observe that large areas have extreme SWI values, higher than 1 (very wet) and below 0 (very dry). By comparing these patterns to the large errors of 2m temperature it is possible to observe that the overestimated temperatures areas are well matched by dry soil patterns and similarly for underestimated temperature areas and wet soil patterns. This correspondence of peaks both in magnitude and position highlights the influence of the soil moisture initialization on 2m errors. More over, this shows that the surface analysis is not well tuned over France for this period. Secondly, the heterogeneity observed for SWI is rather unrealistic, and probably comes from the strong readjustments provided by the analysis. The magnitude of these readjustments is considered by computing the variation of the SWI during the whole period. The mean and the maximum absolute value of the SWI variations show large corrections applied to the field during the relatively short period. As the analysed water content used in ALADIN comes from the ARPEGE analysis interpolated to the finer mesh, a check of consistency is provided taking the initial SWI for the 16 of June 2000 directly from ARPEGE analysis. It is observed that the structure is preserved and all the gradients are reproduced, excluding a problem in the interpolation between ARPEGE and ALADIN grids. These results suggest that the use of a finer tuned analysis on the ALADIN mesoscale domain could benefit from the use of the complete information from soil/screen level relation so it is considered for development. The application of climatological constraints or the use of a spatial filtering in the analysis will be also considered in order to reach a more realistic initialization of the soil moisture.
As a further step then, the use of a sequential assimilation technique for soil variable in the mesoscale ALADIN France model is studied. The background error covariances are estimated in order to obtain a new set of optimal interpolation coefficients. The variability of these statistics on the cumulated solar radiation (in the past 3 hours) is preferred because of its closer links to evaporation. The computation of OI coefficients is performed separately for soil temperature and soil wetness analysis according to the different nature of the dependencies and constraints involved. For soil temperature the only constraints applied is limiting to 8 °C the 2m temperature errors and to 60% the relative humidity errors, as it is assumed that stronger variations would not related to a single surface-screen level interaction. For soil moisture statistics, in addition, are not considered grid points characterized by glaciers, snow covered points, precipitation in the past 3 hours, frozen soil, for which the analysis increments should be small, and could introduce spurious contributions. The assumption for observational errors are of 1 °C for 2m-temperature and 10 % for 2m-relative humidity. The statistics are obtained applying meteorological constraints on the grid-point using a NMC method. The background error covariances are extracted from inner integration area, as full physical space, by comparison of subsequent forecasts valid for the same time (i.e. 6 and 30 hours forecasts). Vegetation cover sensitivity is also provided in the study, as relevant in evapotranspiration process. The optimal coefficients are evaluated for winter and summer season. Two independent statistics are built from 30 days in winter and 30 days in summer at the 4 synoptic hours (06, 12, 18, 24 UTC). The optimal coefficients obtained show similar behaviours to general results coming from ARPEGE statistics. A remarkable and general difference between winter and summer statistics consists in the larger variances in the latter that leads, for any radiation forcing (and vegetation cover, where considered), to a larger value of the OI coefficients. The 2m- temperature correlation with surface temperature and mean soil temperature is similar in winter and summer statistics. The 2m variables correlation with soil moisture is produced considering different vegetation covers, although no significant differences are observed for the winter statistics. A remarkable feature is observed instead in the summer for mean soil moisture, where the vegetation cover leads the order of correlation with 2m-relative humidity and anti-correlation with 2m- temperature, as a confirmation of the important relation involved.
Further tuning of the OI coefficients is necessary and a test on real cases will be considered in later work before stepping into a variational assimilation. This technique is especially considered for future studies as it will keep count also of the non-linearities between observations and surface variables and can use different assimilation windows according to the evolution time scale of the variable retrieved.
Position : Pre-Doc, Prague
Research topic : Application of the predictor-corrector method to non-hydrostatic dynamics
The non-hydrostatic version of ALADIN is based on the Euler system of equations with the shallow atmosphere approximation. It could potentially predict meso-scale phenomena unpredictable by operational models based on Hydrostatic Primitive system of equations. The future operational exploitation of the non-hydrostatic version of ALADIN is crucial for the ALADIN community. It will be possible only if a sufficiently robust and efficient scheme is developed. It was proven in the past that the two-time-level semi-implicit scheme with semi-Lagrangian treatment of advection is both very efficient and robust. This remains true, so long as the semi-implicit scheme controls effectively all the gravity and acoustic oscillations. Since ALADIN is a spectral model, the traditional semi-implicit scheme had to be implemented such that it retains stability by treating a linear constant coefficient Helmholtz equation in model spectral space. This appears, however, to be insufficient to control acoustic oscillations in the non-hydrostatic model. The traditional semi-implicit scheme is therefore unstable for both three and two-time-level schemes. It was detected experimentally that semi-implicit treatment must be applied also on the nonlinear residual of three-dimensional divergence. Due to nonlinear character of three-dimensional divergence, the additional semi-implicit treatment was performed by simplified predictor-corrector algorithm and iteration of Helmholtz solver. This stabilized the three-time-level schemes but the two-time-level semi-implicit semi-Lagrangian scheme remains unstable. The purpose of my work, within the ALATNET Training Network, is to examine sources of instability in the existing implementation of the two-time-level semi-implicit semi-Lagrangian scheme, subsequently to design a method which will control such instabilities and to implement these solutions in the ALADIN model. It is believed that additional semi-implicit treatment of nonlinear terms, currently explicitly treated, will stabilize the two-time-level scheme. This can be achieved through generalization of a predictor-corrector algorithm.
During the reported period, October 2000 - February 2001, I concentrated my work on the following activities.
I presented my preliminary results at an internal seminar in Prague (20th February 2001).
We used a traditional semi-implicit time step as a predictor. The corrector was designed as an iteration of the Helmholtz solver, where iterated nonlinear terms on the right-hand side of the equation were re-evaluated at each iteration using the most recent values (values from predictor when first iteration is performed). We tested the various predictor-corrector formulations when different nonlinear terms were included in the corrector step. The main result of linear analysis of stability is that the model is unconditionally unstable independently of the predictor-corrector scheme formulation. However, a properly designed predictor-algorithm can ensure a sufficiently small growth rate of the solution for practical purposes. The most robust formulation was found one with all tangent-linear model terms included in the corrector step. One corrector step appears to be generally sufficient with that formulation. If not, then it is enough to increase the number of iterations, because the algorithm is convergent. Therefore the speed of the convergence of the corrector-predictor scheme should be investigated rather than stability itself. It is known from the theory that the pressure SHB instability should not exist when s vertical coordinate is used. This was found not to be the case with the current model formulation. Therefore a new prognostic pressure variable was proposed that solved this problem. This new variable was implemented and successfully tested in ALADIN. The thermal SHB instability in the non-hydrostatic case has different behaviour than in the Hydrostatic Primitive system of equations. It is not enough for stability that the reference temperature used in the semi-implicit scheme, is higher than a maximum actual temperature in the model domain. The stability and speed of convergence requires that the magnitude of the difference between those two temperatures should be bounded by some reasonable value. The reference temperature profile used in the semi-implicit scheme should be as close as possible to the actual average temperature profile. This sensitivity to the chosen reference temperature was also observed in other non-hydrostatic models. To solve the thermal problem, the semi-implicit scheme with a non-isothermal reference state was studied. It was found that the elimination of the linear system to one single equation for one variable is not possible, as is the case with an isothermal reference state. It is possible to eliminate the system until two equations for two variables. Logarithmic pressure thicknesses, needed to evaluate discrete vertical operators, can not be expressed as a simple function of reference state pressure, if a non-isothermal reference temperature is used. It was found that the logarithmic pressure thicknesses currently used in the model would be a sufficient approximation.
I implemented following changes into model ALADIN.
Using all currently implemented changes, I performed stable integration of nonlinear non-hydrostatic flow-regime case. I found that at least three corrector steps have to be made to stabilize the model sufficiently, although linear analysis of stability showed that one step should be enough. The reason for slow convergence has to be investigated in the full model and a method has to be found which will allow the number of corrector steps to be decreased.
Position : Post-Doc, Prague
Research topic : Stability analysis and precision aspects of the boundary condition formulation in the non-hydrostatic dynamics and exploration of the alternatives for discrete formulation of the vertical acceleration equation both in Eulerian and semi-Lagrangian time marching schemes
The purpose of my project, within the ALATNET Training Network, is to investigate problems associated with the upper and lower boundary conditions of the ALADIN limited area atmospheric model. Such problems are to be investigated, and solutions sought, with particular attention to the formulation of non-hydrostatic dynamics within the model.
During the period September 2000 - February 2001 my principal activities have been as follows.
¨ Acquiring understanding of the physical and mathematical theory underlying the ALADIN model. The two main areas of study required for this were the use of spectral methods in atmospheric modelling and the formulation of the equations of non-hydrostatic fluid flow, using hydrostatic pressure as the vertical coordinate.
¨ Familiarisation with the ALADIN code and the computer platform on which it runs.
¨ Setting up an idealised test case (trapped lee wave test) for use as a tool in investigating model performance. This test case provides a convenient test for the ability of a numerical model to simulate non-hydrostatic dynamics. It is a particularly "clean" test, in the sense that no artificial wave-damping mechanisms are required.
¨ Proposing changes to various algorithms within the ALADIN model. Then changing model code and making numerical experiments to evaluate these changes.
¨ Reading published literature relevant to numerical weather prediction and computer simulation of fluid flow.
¨ Discussions with colleagues to maintain awareness of overall research and development associated with ALADIN.
Also within this period I presented ongoing work at an internal seminar in Prague (20th February 2001).
The trapped lee wave test, as mentioned above, is an idealised test suitable for evaluating the performance of a non-hydrostatic atmospheric model. The test is formulated for inviscid, adiabatic dynamics in a two-dimensional vertical slice. An isolated bell-shaped hill forms the lower boundary for this model. At the inflow boundary the vertical profile of the atmosphere has linearly increasing vertical shear in the horizontal wind; the thermal stratification is such that buoyancy oscillations are supported. At steady state, the analytical solution for this problem is a trapped lee wave structure. That is, gravity waves are generated by the flow over the isolated mountain and these waves undergo internal reflection. Thus a steady wave-train is established downstream of the mountain. This effect is purely non-hydrostatic.
When this idealised test is applied to the ALADIN model, it is found that the vertical wind field strongly displays a spurious artefact directly over the mountain. Through a number of numerical experiments this was shown to be related to the discretisation employed for the diagnosed divergence term in the model equations. Analysis of this discretisation showed that it had the effect of conserving fluid volume rather than mass. An alternative discretisation was proposed, which avoids this non-physical constraint. This new discretisation removes the spurious behaviour of the model, seen in the trapped lee wave test.
The trapped lee wave test has so far only been run with relatively small amplitude orography. The reason for this is to allow comparison with the results of analytical linear analysis. Results from other idealised tests, simulating non-hydrostatic flow over large amplitude orography, indicate the presence of spurious behaviour similar to that described above. However, in this case it is found that the problem is specifically related to the semi-Lagrangian scheme. Possible sources of this problem are currently being investigated.
In June 2001 I shall attend the ALATNET Seminar on Data Assimilation in Gourdon, France. In July 2001 I shall attend a workshop on Numerical Techniques, to be held in Bratislava, Slovakia. This workshop is organised by the SRNWP (Short Range Numerical Weather Prediction) research network, which comprises several European consortia concerned with numerical weather prediction.