Submission of the following Expression of Interest by the Members of the Short-Range Numerical Weather Prediction Programme
In the frame of a Network of Excellence:
Improvement of Weather Forecasting and Risk Assessment in Europe by Very High Resolution Simulations, Ensemble Forecasting and Modelling System Developments
Presented by the National Weather Services of the following Consortia:
Austria, Belgium, France, Portugal, Bulgaria, Croatia, Czech Republic, Hungary, Poland, Romania, Slovakia, Slovenia
Germany, Greece, Italy, Switzerland
Denmark, Finland, Iceland, Ireland, The Netherlands, Norway, Spain, Sweden
and by the
NWS of the United Kingdom of Great Britain
Text prepared by:
Gerard Cats (KNMI, The Netherlands)
Terrence Davies (UKMO, Great Britain)
Jean-François Geleyn (M-F, France)
Nils Gustafsson (SMHI, Sweden)
Jean Quiby (MeteoSwiss, Switzerland)
Need and Relevance
The need for better regional, even local, short-range (up to 48, better 72 hours) weather predictions is increasing in many circles of our society, not only in the agricultural, industrial and economic circles but also in relation with outdoor recreational activities.
But in the first place, risk forecasting has to de improved, at the foremost place the risk for human life and heavy damages. In our modern society, also the risk of disruption has to be better assessed: disruptions in water and power distribution, disruptions in surface and air transportation. Heavy rain, snowfalls, strong winds and fog are all weather elements causing disruptions.
The weather elements have to become better forecast geographically at the regional scale as well as temporally during the day.
The development and implementation of the three following new techniques will be necessary for a decisive short-range, regional weather forecasting improvement. They are all three equally important:
- a very large increase in the resolutions of today's NWP models
- the use of the ensemble forecasting technique
- the development of a Community comprehensive modelling system.
The challenges of the very high resolution
From its very beginning and without any interruption, the NWP technique has witnessed a steady increased in model resolution. It is realistic to envisage for the next 10 years resolutions of 1km for regional and local models.
Let us see what the scientific issues and problems are for
- the data assimilation, i.e. the determination of the initial conditions
- the numerical aspects, i.e. the techniques used to solve the basic system of differential equations
- the physical aspects, i.e. the mathematical representation of the sub-grid atmospheric processes.
Some Scientific Issues on Data Assimilation:
(1) How do we couple the high-resolution model data assimilation to the data assimilation of the larger scale model providing lateral boundary conditions? Do we use information from the larger scale model assimilation as constraints in the high-resolution model data assimilation? Should we attempt to control also lateral boundary conditions in the high-resolution model data assimilation?
(2) What is the optimal design of data assimilation for very high resolution forecasting? 4-dimensional variational techniques have been very successful for synoptic scale data assimilation, but will this also be the case for very high-resolution models? If we have to carry out 4D-Var at the same scale as that of the model, the so-called "regularised" parameterisation sets will have to evolve while facing new challenges. Indeed the latter will play a more important role in tangent linear and adjoint computations since the leading role of dynamical forcing disappears at these scales. But they may at the same time rely less and less on similarities with the physics package of the main model, whose internal increased complexity will make any stable linearisation attempt worthless. Thus we are entitled to ask ourselves whether the non-linearities of the physical processes will not limit the usefulness of tangent-linear and adjoint models.
Optional assimilation techniques, for example ensemble Kalman filters, may be able to handle these non-linearities and should be investigated.
(3) Due to the large degrees of freedom of high-resolution models as compared to the number of observations, assimilation structure function to spread the observed information spatially will certainly be needed. Flow-dependent structure functions will probably be needed to separate different mesoscale flow regimes, and a spatial dependency to better handle, for example, orography effects will
probably also be needed.
(4) The relevant dynamical-physical balances to constrain the assimilation at the mesoscale need also to be investigated. Moist processes probably need to be included in these balance conditions. As it is the case for synoptic scale assimilation, there may be possibilities to apply statistical balance constraints in the spatial dimensions, but it may also be possible that we need to be restricted to filtering constraints in the time dimension. Such filters essentially prohibit unwanted, for example high frequency, variations.
(5) The assimilation related to moist processes will become crucial, in particular since many observation systems will provide high resolution information mainly for clouds and precipitation. Furthermore, these observing systems are associated with complicated error structures, with weather radar reflectivity as one example. These observation error structures need to be modelled.
(6) Since many mesoscale and local weather systems are forced by the lower boundary condition, a proper assimilation of surface variables is needed for very high-resolution models. The surface variable assimilation needs to be able to mix remote sensing information (e.g. satellite images) with observations from the relatively sparsely distributed in-situ measurements (e.g. surface automatic stations).
Some Scientific Issues on Numerical Aspects:
There are a number of different options. The main split is between semi-Lagrangian advection with semi-implicit timestepping and Eulerian advection with either explicit or semi-implicit timestepping.
There are also differences in underlying grids, in variable staggering and in the choice of the variables themselves. It is unlikely that we will be able to say which scheme is best unequivocally or even which combination of components will perform best.
Comparison projects have not been very developed in the past, with some exceptions (the December 1999 storms). Centres tend to do their own internal assessment and are usually content to show year-on-year in-house improvements.
Examples of other open issues:
- What is the lowest resolution (largest gridlength) we need to produce useful forecasts of severe local weather?
- How fine must be the vertical resolution?
Some Scientific Issues on Physical Aspects:
The so-called «physical» part of the model equations will undergo an important transformation when going to the scales where non-hydrostatism is crucial for the dynamics, since one shall have to get rid simultaneously of two currently widely used key assumptions:
- the forcing of parameterisation calculations will cease to be treated as horizontally homogeneous, which means that the three space dimensions will play a role in the computation of turbulent and radiative fluxes
- there will be less and less need to take into account details of the flow organisation internal to the mesh-box. Typically, organised convection and details of the sub-grid scale underlying topography will cease to generate fluxes of a magnitude comparable to what the explicit computation will be producing between neighbouring grid points.
This double revolution in the perception that NWP code designers have of the physics of the atmosphere will also have a strong influence on the way the parameterisation codes are treated algorithmically:
- old sources of inaccuracies and of latent instabilities will disappear, while one can safely bet that they shall be replaced by new ones;
- the organisation of the time step itself (in which order and with which input are the differing items called) may have to be reviewed.
The use of the ensemble forecasting technique
Of the three new techniques that will be necessary in order to meet the meteorological challenges of the future, the ensemble technique for short-range, meso-scale forecasting is still in its infancy. The very first European workshop on this topic will take place next autumn in Spain.
But there is a profound reason why one must use this technique:
the atmospheric processes or, on the large scale, the weather systems are only partially deterministic phenomena. The meteorological processes are chaotic systems, in the sense given to this term in theoretical physics. But now we apprehend meteorology, particularly with the classical NWP technique, by a pure deterministic method. This cannot be the best way.
The ensemble forecasting technique is already successful on the global scale. There is no intrinsic scientific reason why this method cannot be equally successful for high resolution limited area models.
The ensemble forecasting technique allows formulating forecasts in probabilistic terms. That is the basic property on which methods for risk assessment can be developed.
The development of a common modelling system
The objective of this part of the project is to build an infrastructure to enable an effective European collaboration for the development of a very high-resolution numerical weather prediction system.
A high quality development of such a system requires a concerted effort of research into the physical properties of those systems (see Point "The challenges of the very high resolution"). The required intensive interactions between researchers can only be provided for by a state of the art infrastructure. For effective collaboration across Europe, the infrastructure should
1. support facilities for exchange of knowledge between researchers
2. facilitate exchange of researchers between participants
3. facilitate exchange of software components
4. provide access to distributed data bases and other compute facilities.
A variety of developments in the field of information technology will provide basic components to build the required infrastructure on. These developments include the results of three current EU projects
= PRISM (supportive to objective 3 above)
= DataGRID and EuroGrid (supportive to objectives 3 and 4)
and new web-based technologies (supportive to all objectives), like emerging standards for remote procedure invocation (e.g. SOAP, www.w3.org/TR/SOAP; WSDL, www.w3.org/TR/wsdl) and data exchange (XML).
For effective exchange of researchers between participants (objective 2), novel methods to design and write graphical user interfaces must be applied.
This enumeration of relevant developments is far from complete, but it is sufficient to demonstrate that there is a wealth of developments, which have to be done.
This infrastructure will be built with components developed in earlier projects, both in Europe and in other parts of the world; where components or the "glue" between components are missing they will be developed.
The infrastructure should be flexible enough not to inhibit the deployment of any meteorological component, to allow competitive developments. It should also be flexible enough for implementation in all participating institutes. This implies that it should not compromise their operational procedures, or, preferably, even be able to support those procedures. It also implies that the new infrastructure should not be based on expensive proprietary software.
The primary aim of the system to be developed is to facilitate European cooperation on research on very high resolution NWP. There is a genuine research effort contained in this part of the project, as it aims at developing novel methods to support cooperation in research.
A second research effort in this part of the project is to design the target infrastructure in such a way that it supports a wide range of implementations of NWP systems, both for research and operations.
Today's Numerical Weather Prediction (NWP) models have reached a degree of complexity that makes it difficult for a country, even for a large country, to develop and test with its own resources all the modules of a leading edge numerical model for the simulation of the atmospheric processes, inclusive the difficult determination of adequate initial conditions from the observations. As a consequence, groups of NWS have formed to share the work of the development and subsequent improvement of a NWP model.
We have today in Europe four groups, called Consortia, each one centred on a model.
These Consortia are: ALADIN, COSMO, HIRLAM and the UKMO. The list of the participants in each Consortium is given at the beginning of this document.
The work done inside each Consortium can be qualified of very good; this is proved by the fact that our 4 meso-scale Limited Area Models (LAM)
- the ALADIN model of the ALADIN Consortium
- the Local Model of the COMO Consortium
- the HIRLAM model of the HIRLAM Consortium
- the LAM UM of the UKMO
are top models, leading models when compared to similar models in other parts of the world.
This shows that there is a profound know-how and remarkable skills in our Consortia supported by top European scientists filling all the present needs for methodological and theoretical expertise.
The exchange of information and some common actions between the Consortia are co-ordinating in the EUMETNET Short-Range NWP Programme (http://srnwp.cscs.ch).
This Programme is responsible for series of workshops on the most actual and/or most difficult topics related to the short-range NWP on the meso-scale (srnwp.cscs.ch -> LC Workshops for all the past and future workshops).
Two of the 4 Consortia organise each year a Seminar of several days for young scientists, many of them PhD students. These Seminars are very successful (For more information, see srnwp.cscs.ch -> Education).
It is very important to consider that the scientific developments in the Consortia is not only due to the NWS involved, by is also due to the contributions of a significant number of Universities: many European NWS work together with one or more universities of their country.
- the German Weather Service with the University of Bonn (Department of Meteorology)
- the Swedish Meteorological and Hydrological Institute with Stockholm University (Department of Meteorology)
- Meteo-France with the Paul-Sabatier University of Toulouse
- MeteoSwiss with the Swiss Federal Institute of Technology (Institute for Atmosphere and Climate)
- the UKMO with Reading University (Institute of Meteorology)
- DNMI with the Oslo University.
The National Weather Services grouped in their 4 Consortia together with the Universities they are working with form the strongest network in atmospheric sciences in Europe, a true Network of Excellence.
And that Network has two great assets: it already exists and works well; it has the critical mass required for a project of that size.
But this Network - surely one of the strongest in the world in its field - has not presently the personal capacities and the financial strengths for an integrated, full-scale development of these 3 new technologies (very-high resolution, ensemble forecasting and comprehensive modelling system).
A strong external financial support is needed.
Such a support could allow a decisive step forward in weather forecasting, inclusive risk assessment that would respond to the legitimate expectations of the future.
Priority Thematic Area, Partnerships, Management of the Project
This Expression of Interest fits exactly in Thematic Area 22.214.171.124, sub-area "Operational forecasting and modelling" as well as, in the sub-area "Complementary research", with the topic "Development of advanced methods for risk assessment".
Next to the reinforcement of the collaboration between the NWS forming the four Consortia for NWP in Europe, it is also a strong intention of this Project not only to reinforce the collaboration between the NWS and the Universities they are already working with, but also to enlarge the collaboration of the NWS to other Universities in order to draw the maximum of the European research potential in meteorology.
We think that a project of this magnitude should be professionally managed. But the work of the Project Manager must be supervised by a Scientific Committee made of the most prominent scientists active in the Project, half coming from the NWS, half from the Universities.