|
SECOND MEDIUM-TERM (1999-2001) RESEARCH PLAN FOR ALADINapproved (for its purely scientific part and accordingly edited) at the third ALADIN Assembly of Partners , Prague, 6/11/98 J.-F. Geleyn & D. Giard (Météo-France), A. Horányi (Hungarian MS)FOREWORD As implied by the above title, this is the second exercise of this type in the history of ALADIN; the distance between the two plans, i.e. four years, is about right; both dates also coincide with important events in the life of the project, the quasi-operational start of ALADIN PECO in 1994 and, for now, both the repatriation of ALADIN-LACE to Central Europe and the operational status of ALADIN-France. There are however many changes in the background:
Despite these new and difficult background conditions, let us however hope that we are right if we conclude this foreword with the same sentence as four years ago: «If all partners are convinced of its benefits and act accordingly, the scientific collaboration is profitable to everyone and it remains the best justification of all efforts associated with the project.». INTRODUCTIONI) SUPPORT TO NWP OPERATIONSII) APPLICATIONSIII) RESEARCH FOR TOMORROW'S NUMERICAL WEATHER PREDICTIONCONCLUSIONSAppendix 1 : DRAFT MEDIUM-TERM RESEARCH PLAN FOR ALADIN, presented to the second meeting of the LACE steering committee, Bratislava, 29-30/06/94Appendix 2 : LIST OF TOPICS AND POTENTIALLY ASSOCIATED ALADIN SCIENTISTS, version of 26/01/1998 technically edited for the inclusion in the Research PlanINTRODUCTIONThis plan, like the draft first one, has been constructed starting from a few basic assumptions:
Should any of these two assumptions become unrealistic, then the plan should be quickly updated with less ambitious features. Consequently, again like four years ago, we shall take into account the necessity to maintain a strong compatibility with ARPEGE/IFS, the streamlining of research activities towards achievable operational goals (as often as possible of general interest) as well as the choice of "windows of opportunity for excellence" created by the already achieved research/development steps. The structure of the current plan is completely different from that of the previous one. The reason is that it is considered important to have a good level of structuration for an easier assessment of the plan and with an easier addition of items if necessary. Three main topics will be discussed hereafter: OPERATIONS, APPLICATIONS and INVESTMENT, this sequence reflecting no priority order but only some kind of technological logic. Some of the sub-items appearing in the main topics are of course somehow interconnected and of a more general scope than simply to be used for short range forecasting at fine scale, but, like last time, we assume that these two characteristics are implicit through the work in the IFS/ARPEGE/ALADIN framework. It was attempted to find co-ordinators for every main item and also proposed to them to produce some summary reports twice a year (for instance every February and September) in the given area of interest. The scientific plan might then always be upgraded before the Assembly of Partners preserving the same basic structure as used in this version. I) SUPPORT TO NWP OPERATIONSThis chapter is of primary interest since more and more operational applications of the ALADIN model exist (currently the model is operational in Belgium, France, Hungary, Morocco, Poland, Romania, Slovenia and operated for the RC-LACE community in the Czech Republic; its status is quasi-operational in Austria, Bulgaria, Portugal and Slovakia). The operational and quasi-operational versions ensure a solid and smooth basis for the development and research work as well, but the need to maintain a reasonable increase in resolution at each coupling stage adds scientific and technical constraints to the whole process. The code maintenance aspect, as recalled in the introduction, is a key question for all operational versions, ensuring to everyone the application of the best available version of the code at all times. But it is equally (even if not more) important for the common research and development work. By now several new ALADIN configurations are available (following a strong requirement of the Partners) for research and development like tangent linear, adjoint and sensitivity configurations, variational analyses or singular vector calculations, but this makes the maintenance efforts even more complex and difficult. At the same time it is expected that more and more deported work will be included into the main library, which will be a new challenge for maintenance. It is also recognised that the maintenance should take care about documentation and code optimisation aspects, as well as specific training. I.1. DYNAMICS (Proposed co-ordinator: Gábor Radnóti, Hungary)This part of the ALADIN development activities is at a turning point: the solution (thanks to ECMWF) of some outstanding problems with the two-time-level semi-Lagrangian scheme, the recent progress concerning fibrillations and the apparent impossibility to go to linear or semi-linear grids at high resolution seem to mark the end of a cycle for the hydrostatic version of ALADIN, even if some details of the formulation remain interesting to study. Therefore there are no more major obstacles to overtake for the hydrostatic resolution of the model; however for the higher resolutions many questions are raised (see in the investment part of the plan). Regarding the boundary conditions, the question how to modify details of the initial choices for the coupling was never successfully raised, probably because those choices were quite good ones. This does not mean that the superiority of the Davies-method (and of the way it is implemented, for spectral conditions, in ALADIN) is definitive, but simply that any revision will have to be an important one. Whether the effort would be worth the price also remains an open question. I.2. PHYSICS (Proposed co-ordinator: Marta Janiskova, Slovakia)This part of the plan is probably one of the less problematic ones. The work on the associated topics has been quite successful in recent years (with a strong influence on the evolution of the physical package in both ARPEGE and ALADIN -or on the confirmation of past choices by «negative» results-) and it is easy to see that the problem of «distributed work» is far less acute here than elsewhere. Furthermore, the above-mentioned back-influence on ARPEGE should not be considered as a negligible point: progress in the quality of the coupling model is directly felt in the quality of any ALADIN application; this statement is of course very general but its most obvious field of application is «physics». In such a situation, in order to avoid dispersion and/or redundancies, it is important to list broad tasks and to try and give them some priorities, which will be attempted now:
The other important issue here is not to particularise too much the efforts on the ALADIN operational configurations: the (currently optimal) compatibility with the physics of ARPEGE is probably crucial for the treatment of LBCs and the code could not live easily with a dozen of parameterisation schemes for convection! Hence it will become more and more important to have objective diagnostic tools to judge potential improvements at all scale, from the very local one (1D-modelling and/or very high resolution versions of ALADIN) to the global one (ARPEGE and/or imbedded ALADIN in simulation mode on the monthly time scale). Recently started steps into this good direction have to be encouraged and their results (in terms of «tools») have to be widely used in the ALADIN community, at all levels (this is a nice advantage here -in contrast to most other items of the plan- that most of these tools are not specific to one given type of environment). The main problem will thus be how to organise such «use», for a continuation of progress at the current rhythm or even better. I.3. OBSERVATION HANDLING (Proposed co-ordinator: Liviu Dragulanescu, Romania)Due to the fact that the application of independent data assimilation cycles (see more at the investment chapter) is to be used at more and more Partners, this technical question is an important one. Since recently the CANARI OI scheme can be used in several operational contexts, which all require the application of special observation files. These observation files with standard GTS information might be created in Toulouse and put at disposal of the Partners. The local information sources can be added into the system at the place of the operational application. Certainly the added information has to be monitored and controlled as far as its quality is concerned, however this part of the work should be under the responsibility of each Partner. It is also mentioned here that the observation files will be soon replaced by databases; this major change will force a significant amount of modifications in their handling, therefore this will mean a significant amount of additional work to be supported by any Partner interested in local data assimilation activities. The CANARI optimal interpolation analysis can be used for four different purposes: data assimilation, diagnostic-oriented analysis, monitoring of local observations and verification. The diagnostic-oriented analysis is meant, when the system is used without data assimilation mode for nowcasting purposes, with every available information being used to produce a maximally reasonable «static» analysis for the forecasters. It was already proved that the CANARI system is capable in this mode and in some circumstances to diagnose forthcoming severe thunderstorms with the help of special post-processed fields, like moisture convergence or convective available potential energy. But the original choices about ARPEGE/ALADIN mean that, from the code point of view, all aspects are interrelated and this should always be taken into account: a specific development that would forbid the continued use of CANARI for other purposes would be unacceptable; this obviously extends to the case when O/I, 3D-Var and 4D-Var are considered together (see below). Regarding verification, despite the realisation of many individual projects, this item has not been very successful from the research and development point of view in the past four years. All actions were of limited scope and/or driven by other projects, in fact mainly by operational contingencies (see for instance the efforts for verification of pseudo-TEMP messages or precipitation); but this is not sufficient, especially in a distributed project like ours where duplication is a luxus we should not afford; the only good solution to this problem is to make such actions part of a real structured research effort, but this message seems very difficult to get through in front of other day to day priorities (surprisingly enough it is in the area of subjective verification that the project is most innovative!). Yet, the work at finer scale than other models and the need to search objective ways to verify the continuous improvement of the ALADIN software call for more ambitious goals in this area. Finally it is recalled that it is of crucial importance to have a good co-ordination on the CANARI related work. The first steps were already taken in the last ALADIN workshop (in Prague, June, 1998), but the efforts should be continued. II) APPLICATIONSThis chapter deals with the application of the model itself and its products for further use and for other special applications. II.1. MODELLING PROBLEMS (Proposed co-ordinator: Claude Fischer, France)This paragraph discusses those experiments, which can be basically performed running the model with different settings or for studying different interesting cases. This type of work can be done at any Partner where the execution of the ALADIN model is possible. Wide ranges of topics can be considered starting from «simple» case studies, through studying the coupling strategy (for instance the frequency of coupling) until more complicated sensitivity studies or predictability studies. Of course the necessary tools should be put into operation-like status, therefore it would be desirable to create a list of tools for ALADIN in order to stimulate activities in that field. Therefore this problem is thus more one of documentation (and producing publications) and co-ordination of already on-going activities than of fostering new ideas. II.2. PRODUCTION RELATED PROBLEMS (Proposed co-ordinator: Jure Jerman, Slovenia)At every Partner applying the ALADIN model, in some way some special applications are used for the improvement of the forecasts provided by the ALADIN model. The first in that row is the application of special NWP fields for giving specialised forecasts (e.g. for aviation) with simple computations of special derived quantities, which helps the forecast of a given phenomena (in the case chosen here as example these parameters are well-known like different lability indices, Richardson number, etc.). The other way of applying the NWP outputs are their further adaptation, which can be done in a statistical or dynamical (this also encompasses using the ALADIN model itself as tool for dynamical adaptation for a smaller domain with very short integration periods) way. The outputs of the ALADIN model can be used as input by other models, like snow or hydrological models or they can drive special other models as air pollution models for instance. At that point the information exchange is also of crucial interest to know what kind of applications are used at the Partners. It is proposed to compile a list of existing tools around ARPEGE/ALADIN in this field of interest, as well as a list of the existing operational applications. Projects to exchange such tools or copy such applications should be encouraged, even if they are not strictly speaking under the responsibility of the ALADIN scientific community (except for the dynamical adaptation use of ALADIN itself).
III) RESEARCH FOR TOMORROW'S NUMERICAL WEATHER PREDICTIONIII.1. VARIATIONAL ASPECTS (Proposed co-ordinator: András Horányi, Hungary)This part of the plan is the only one that has benefited in the past years of a real co-ordination effort and, given its growing importance inside the project, it was a welcome fact! Nevertheless most of the effort is still ahead and the huge complexity of the issues makes it difficult both to classify actions and to ask the right questions on «how to proceed now?». We shall however try and do it, keeping in mind that the calendar of actions in ARPEGE (after 3D-Var in 1997, 4D-Var in a «basic» version in 1999 and in an «improved» one in 2000) is determining in order to let problems be «explored» before their specific treatment in ALADIN. At the same time the recent success of the operational implementation of 4D-Var at ECMWF (albeit only with T63 increments), raises important strategic questions for any LAM-community in Europe (HIRLAM and ALADIN). If one admits that the previous choice made by the Partners (insist on 3D-Var on the path to a still potential 4D-Var) is confirmed and that recent evidence reinforces the 4D-Var promises at meso-scale, there are still several key questions raised from which the most immediate one is that of the balance in ALADIN at high resolution. The next most important issues are to get the observation error term for limited area domain working, the acquisition and evaluation of the relevant model error statistics for different ALADIN domains and the adaptation of the background error term calculation from the global statistical balance option to one appropriate for ALADIN. Putting together these basic ingredients some preliminary tests might be started in a third step. Beside that and with lower priority, work should continue on high resolution aspects of the tangent linear and adjoint versions of the diabatic and non-hydrostatic part of the code and on the study the role of digital filter initialisation in the dynamical balance. By now all these topics are more or less covered and the co-ordination seems to be working rather well including for the common work with HIRLAM that is crucial here. All what is thus required is a confirmation of the objectives and of the overall support for reaching them. The role of the three data assimilation items (O/I, 3D-Var and 4D-Var) inside the whole of the ALADIN project is already a far more delicate subject. In other words it practically means that an optimal choice is needed for using the above mentioned tools for operations, research and development. During the past few months two things have become more and more obvious:
In fact the question of the «window of opportunity» for data assimilation in ALADIN, that has been with us since the start of the project, will have to be reconsidered in the light of the two above-mentioned pieces of information. In principle this is more a matter of operational problems than of research, but the whole history of NWP shows that such aspects can never be separated without some backlash, especially for data assimilation matters. Indeed, while there are more and more hopes that 4D-Var at high resolution will represent a breakthrough for forecasting intense meso-scale events (success of the ECMWF implementation at the synoptic scale and «physically sound» first multivariate background error statistics obtained from ALADIN-France), its availability in all ALADIN version and its flexible maintenance seem goals for a very distant future. Furthermore, using data, be it in the more advanced manner, is never a guarantee to improve a given forecast process, often the opposite if one is too eager to go forward without careful preliminary studies. On the other hand, the fact to have several tools that can be used in different manners (taking into account the possibility of lagged-mode coupling as well as the different hours of availability of all data types) may prove a big advantage if one finds imaginative combinations of them and if everyone continues to work to progress on all fronts at the same time while using for itself only the most appropriate combination of the available tools. But those questions are obviously at risk to be treated with different rules than other we raise in this document. The associated decisions will therefore be more complex (a rearrangement of priorities requires a more sustained co-ordination than the replacement of one action by another one), but the future of the project is somehow at stake here. Finally, when it comes to the use of new data (e.g. new type of remote sensed data, or more flight reports, or imagery, etc.) at high resolution and in an environment strongly disturbed by orography, we enter a strange area: everyone claims to be interested but nobody volunteers to take the lead. This point, even if less urgent than the previous one, will have to be somehow addressed soon. Indeed, without enough well-controlled data there is no hope to have good results even with the most sophisticated scheme; therefore the development of schemes and making the data available should go in parallel. This data availability problem of course also encompasses the very political question of exchange of more traditional but non-GTS sources of information inside the ALADIN community. While speaking about variational data assimilation (3DVAR and 4DVAR) some words should be mentioned about other applications using the same tools as those for variational data assimilation. Typical such applications are sensitivity studies using the adjoint method (exploring the dynamically most important aspects in the initial conditions), calculation of singular vectors (which characterise the most unstable structures) or Kalman filter meant in its most general data assimilation sense. Some of these aspects are about to be explored, which might open some new perspectives in stochastic forecasting at the shorter ranges and-or finer scales, for in a few years time. The research and development for data assimilation is a key future aspect of the ALADIN project. On the one hand the algorithmic work simply needs a confirmation of its priority in all aspects (O/I -mainly for diagnostics-, 3D-Var and 4D-Var). On the other hand, this may remain only a research exercise if steps are not taken to ensure an increased maintenance, training and policy (for instance code protection) effort. Furthermore, some plans about use of unconventional data should start to be drawn and individual efforts are first encouraged here, in the spirit of course of the whole project. III.2. HIGH RESOLUTION DYNAMICS (Proposed co-ordinator: Radmila Bubnova, Czech Republic)The difficulties associated with the non-hydrostatic version (extension to the two-time-level framework, lower boundary condition, link with the radiative upper boundary condition currently developed in hydrostatic mode, need to probably revisit the «thin layer hypothesis», all problems of roughly equal priority if one wants to get soon a «prototype» of the future numerics of ALADIN) call here for either a lowering of ambitions or a structured effort. Indeed, either we can have the same relative efficiency (basically ratio of length of time step to mesh size) in the non-hydrostatic version as in the hydrostatic one and the transition to non-hydrostatism will be a soft one; or we give up this goal for lack of willingness to invest steadily into it and the march to higher and higher resolution will be stopped for about three years at some stage. Since neither choice seemed to be acceptable (!) when last discussed (second Assembly of Partners, December 1997), Météo-France took the initiative to secure some continuity for one year by adding one visiting scientist to its work force on the subject. However it is clear that this is only a way to buy time and that the longer term question should be addressed frankly: in the area of NWP where ALADIN can distinguish itself most from ARPEGE (something also very important from the psychological point of view!), one cannot at the same time require high ambitions for future beta-meso-scale versions and dedicate to the subject only a very intermittent and too often renewed workforce. On a less worrying subject, topics of interest have become more numerous around the general LBC question: coupling between hydrostatic and non-hydrostatic versions (low priority), spectral coupling as a possible tool for 4D-Var (medium priority), ratio of mesh sizes and domain extensions in «optimal coupling strategies» (high priority), coupling in data assimilation mode, especially in the forthcoming so-called continuous mode, i.e. 4D-Var, (very high priority). Given the associated time scales for future use, none of these topics seems to be on a critical path for the project, but all are surely of good scientific status. The non-hydrostatic version of ALADIN opens a new perspective for the operational application of ALADIN at the beginning of the next century, but it obviously needs a more sustained approach. CONCLUSIONSThe level of scientific activity around the ALADIN-project is good and innovative. However it lacks more and more political guidance and day-to-day co-ordination. One has now to realise that the aims to have people «running the model» in its current status and people contributing to the progress of the quality of the product are getting increasingly contradicting goals requiring more and more coordination if a structured training effort is not done (cf. the relevant discussions at the recent Prague Workshop, recalled in the Foreword). Indeed the only solid bridges between the two above-mentioned aspects, i.e. «high level maintenance of the code» and «verifications» are the areas where the project is less healthy than it used to be! Despite these weaknesses there are several promising ongoing activities for the future but they clearly lack a sustained commitment. All these worrying diagnostics call for a renewed interest in a carefully planned scientific investment in the Project, for a more systematic and more structured «home» initial training and for more collective maintenance efforts. We all have to feel that we are at the end of a cycle, that progress linked to the capitalisation of previous investment and to technological advances will get slower and slower, but for an effort similar to the one made at the time of ALADIN's birth. Despite all the above, one can attempt to close this document on the same optimistic last sentence as last time: «Finally let us hope that this debate will be as large as possible, as an indication that its (quasi-) operational character has made ALADIN a real "community" tool!?». Appendix 1 : DRAFT MEDIUM-TERM RESEARCH PLAN FOR ALADINPresented to the second meeting of the LACE steering committee, Bratislava, 29-30/06/94by J.-F. Geleyn, Météo-FranceAPPENDIX 2 : LIST OF TOPICS AND POTENTIALLY ASSOCIATED ALADIN SCIENTISTSVersion of 26/01/1998 technically edited for the inclusion in the Research Planby J.-F. Geleyn (Météo-France) & A. Horányi (Hungarian MS)CO-ORDINATORS:
Scientific topics for 1998 around ALADIN (mostly to be done in Toulouse)(not in priority order) DATA ASSIMILATION1. Adapt the Jb term from ARPEGE/IFS (statistical balance) to ALADIN Loik Berre (France) and Maria Siroka (Slovakia)2. Possible use of incremental/decremental approach together with spectral coupling in 4DVAR; if not problem of the consistency of 4DVAR strategies between the coupling model and the coupled one Vladimir Ivanovici and Liviu Dragulanescu (Romania)3. Incremental DFI in coupled data assimilation mode Vladimir Ivanovici and Liviu Dragulanescu (Romania)Waafa Sadiki (Morocco)4. Extension of DFI to grid-point variables for a potential use in spectral coupling mode and in 3.5DVAR Dominique Giard (France)5. Scientific study of sensitivities of forecast errors to initial and lateral boundary conditions ("poor man's 4DVAR") Cornel Soci and Liviu Dragulanescu (Romania)6. Study the balance requirements for variational data assimilation Samir Issara (Morocco)7. Single observation experiments using optimal interpolation (CANARI) Elena Cordoneanu and Cornel Soci (Romania)8. Calculation of observation errors using the Hollingsworth-Lonnberg technique Young Wang (Austria)9. Refinement of observation operators for ALADIN Lora Gaytandjieva (Bulgaria)10. Impact of surface analysis, with CANARI and ISBA in ALADIN Meriem Zitouni (Croatia)11. 4DVAR experiments using simplified physics Marta Janiskova (Slovakia)12. Diagnostics on the calculation of background error covariances using the NMC method Maria Monteiro (Portugal)13. ISBA in CANARI, relative humidity from TOVS data Samir Issara and Radi Ajjaji (Morocco)DYNAMICS1. Technical problems related with the TL and AD of the non-hydrostatic SL dynamics Martin Charron (Canada) and Martin Janousek (Czech Republic)2. Calculation of singular vectors using Lanczos algorithm (configuration 601) Frederic Chome (Belgium)3. SL version of the elastic version of the model Martin Janousek and Radmila Bubnova (Czech Republic)4. Increased sophistication of the "free-slip" lower boundary condition for the elastic version of the model Radmila Bubnova (Czech Republic)5. Introduction of the SL formalism for the lower boundary condition Danijel Cemas (Slovenia)6. Relaxation of the thin layer hypothesis and simultaneous introduction of the vertical part of the Coriolis force David Dvorak (Czech Republic)7. Study of the coupling between a hydrostatic coupling model and a non-hydrostatic coupled one Fernando Prates (Portugal) and Jaouad Boutahar (Morocco)8. Trial of linear (semi-linear) grids Metodi Marku (Albania)9. Experiments with increased vertical resolution Mihaela Caian (Romania) and Ervin Zsoter (Hungary) and Jaouad Boutahar (Morocco)10. Experiments using spectral coupling Mehdi El Abed (Morocco)11. 1D vertical model Mark Zagar (Slovenia)12. Conservation properties of a two-time level semi-implicit SL scheme Ilian Gospodinov (Bulgaria)13. Dynamical and physical control of kinetic energy spectra Filip Vana (Czech Republic)14. Radiative upper bundary conditon for accustic waves Jozef Vivoda (Slovakia)15. Interaction between orography and gravity wave drag Richard Mladek (Czech Republic)16. TL of post-processing Liviu Dragulanescu (Romania)PHYSICS1. Extension of the physics to the non-hydrostatic framework Luc Gérard (Belgium)2. Introduction of ozone as prognostic variable Svjezdana Klaic (Croatia)3. More sophisticated treatment of the radiative properties of the clouds Cristina Madeira (Portugal) and Jean-Marcel Piriou (France)4. Exploration of the evolution towards higher order closure representation of the turbulence Eric Bazile (France)5. Study of the problem of internal downdrafts (also as a potential replacement for a separated par. of shallow convection) Thomas Haiden (Austria)6. Cloudwater as prognostic variable Doina Banciu (Romania)7. CROCUS, snow model Valery Spiridonov (Bulgaria)8. Closure for convection Mehdi El Abed (Morocco)9. Intercomparison of physical parametrization packages Jozef Vivoda (Slovakia) |