WORLD METEOROLOGICAL ORGANIZATION COMMISSION FOR BASIC SYSTEMS CBS/CG-NERA/Doc. (number) (26.04.2006) OPAG on DPFS NUCLEAR EMERGENCY RESPONSE ACTIVITIES COORDINATION GROUP Original: ENGLISH VIENNA, AUSTRIA, 2-6 MAY 2006 EC-JRC ENSEMBLE APPROACH TO LONG RANGE ATMOSPHERIC DISPERSION (Submitted by Stefano Galmarini Institute for Environment and Sustainability, Joint Research Centre Ispra, Italy) Summary and purpose of document Long range transport and dispersion (LRTD) models are used operationally by meteorological offices and environmental protection agencies to forecast the consequences of the release of harmful material to the atmosphere at regional scale. The differences in model results are inevitable and can pose serious problems to their use for civil protection or countermeasures management in the case of trans-boundary dispersion. The multi-model Ensemble Dispersion Modeling (EDM) approach presented consists of the treatment and analysis of several long range transport and dispersion model forecast. It is based on the real-time consultation of the results of several LRTD models and on the use of ad-hoc ensemble statistical parameters to determine the level of reliability of the forecast. The system is periodically used for dry-runs and it has also been. The ENSEMBLE system has been existing since 6 years and it is now migrating to its version 2.0. The system has been recently adopted by the EU 25+3 members states Competent Authorities as the official system for the exchange, inter-comparison and redissemination of dispersion forecast is case of nuclear accidents at the long range. 1. Introduction
The ENSEMBLE project was set up with the intent of developing a system for the rapid exchange of long-range transport and dispersion forecasts produced by meteorological offices and radiation-protection agencies in Europe and around the world. The dispersion models are used in case of accidental releases of radioactive material in the atmosphere. ENSEMBLE not only is intended as a platform where several forecasts are exchanged, but also as a tool for the simultaneous analysis of the results of several models and for the ensemble analysis. Briefly, ENSEMBLE is an Internet-based server-side system (http://ensemble.jrc.it) that collects in real-time the long range dispersion forecast produced by 22 European, one US, and one Canadian institutes with more than 25 models in total. The forecasts are produced by operational long-range transport and dispersion models based on different concepts that use meteorological fields produced by different NWP models (namely ECMWF, various versions of HIRLAM, ARPEGE, ALADIN, RAMS, GME, UM and LM. All dispersion forecasts submitted are initially based on a completely forecasted meteorology, then on partly forecasted and partly analyzed meteorology and finally on completely analyzed meteorology. 2. ENSEMBLE products The flexibility of the web based technology allows the possibility to share in real time the results of many atmospheric dispersion predictions which can be consulted by the whole forecasting community for the sake of improving the support to decision making. In this section, the way in which such information can be used an in what sense it represents an asset for support to emergency response will be presented. 2.1 Availability of multiple model results for eye-ball evaluation This is the most obvious advantage of a system like ENSEMBLE. Providing support with a single model is off-course possible but the availability of several model results is by no means a indispensable asset. This is particularly true when dispersion forecast are provided by different atmospheric dispersion systems that use different NWP data. The consultation of several model results in real time allows: Immediate identification of error in setting up the case Identification on how your model results are placed within the distribution of the others, at a single location over time or over the domain of concern at fixed time intervals Identification of the overall level of agreement of the model results and indirect estimate of the reliability of the forecast. These are all indispensable assets which become even more valuable when, as it happens within ENSEMBLE, all model results are harmonized on the same temporal and geographical coordinates and they all produced results on exactly the same case; the results are collected in real time and can be consulted in real time;
concentration or deposition levels as well as time of the plot or domain portion to be displayed are controlled by the single remote user who can define is personal settings autonomously; every remote user is independent in the final judgement of the situation is only provided with a large and harmonized amount of information, no filtering or preelaboration of the data is performed. Additional features have been added to version 2 of the ENSEMBLE system, that further emphasize what listed above; briefly: Possibility to display the simulation domain anywhere on the globe (originally the system was working over Europe only). Possibility to define the domain size according to scale of the problem (originally the system was working on affixed domain size). Possibility to define the resolution on a case by case basis (originally the system was working with a fix 0.5x0.5 deg resolution). Possibility to acquire simulation on the dispersion of several nuclides at the same time (originally results on a single nuclide where accepted). GIS plotting facility with several thematic layers relevant at long range scale. Coupling of the system with the EUropean Radiological Data Exchange Platform (EURDEP) that collects real-time measurements of gamma dose rates from almost 4000 station in Europe and Russia. This allows real time on-the-fly consultation of real-time monitoring data and combination with model results. The availability of more than 20 model results and data allows immediate spotting of obviously erroneous results and to concentrate eventually on a reduced set of models. Possibility to submit results that relate to any substance and not necessarily to radioactive material. Within ENSEMBLE model results can also be combined into groups and compared in various ways for a better understanding of the differences and the possible origin. Even in this case the grouping operation can be performed in real-time and on-the-fly, every single user as a possibility to act independently from the others in this kind of operation. The grouping can be performed, for example, with respect to: Results of different versions of the same model Different dispersion models using NWP data from the same source Date of NWP data production 2.2 Space overlap at fixed threshold level Examples of this feature are given in Figure 18(a-c) of the Covex-3 report (see link in meeting web page, or Galmarini (2005)). The figure shows the overlapped contours of independent model results for a fixed threshold level. This type of representation can be useful for the identification of fine structure differences between model results.
Figures 17 and 19 of the same document give an example of overlapping 2 groups of models. The grouping relates to the time of the forecast (based on forecasted weather or analysed weather). This representation allows also the quantification of the overlapping region shared by the two clouds. The clouds can be constructed as maximum or average, meaning maximum or average value in every single point in space (time) of the two groups. An example of an overlap in time is given in Figure 21. 2.3 Agreement in threshold level and in percentile level These are the most advanced analysis tools provided by the ENSEMBLE system. They allows the direct identification of the areas of agreement in predicting the exceedance of a specific threshold level or the levels of concentration or deposition associated to pre-defined percentile level. Examples of application in the case of Convex-3 are given in figures 5a, 8, 9, 11, 12 for the Agreement in Threshold Level and 5b, and 10 for the agreement in Percentile level. Even in this case the user is free in the selection of: The models to include in the treatment The level of the threshold or the percentile Time of the analysis The graphics is produced in real time and on-the-fly. The one presented are three of the several tools available in ENSEMBLE. We refer the reader to the peer-reviewed publications listed in the reference section for more examples and description of the whole set of tools. 3. Scientific achievements The active participation of the large community of operational modelling groups to the ENSEMBLE activities has also led to active research in the field of ensemble dispersion modelling. In the paper Galmarini et al. (2004b) the technique was applied to the ETEX experiment and allowed to identify the median as the percentile of the model distribution that better represented the measured data, this in contrast with the usual assumption that the average should be the best result. A clear indication that the median should be preferred to the average in the analysis of model predictions where outliers might be present was given. Furthermore it was also proved that the multi-model ensemble prediction out-performs the single deterministic model results and that a large critical mass of model predictions is necessary to improve the prediction. This result was already anticipated in Galmarini et al. (2001) based on a pure model inter-comparison results. Although these results were confined to the ETEX-1 experiments they prompt clearly the necessity of further investigation in this fields and the there is a lot to gain by joining forces and competences. Further investigations are being performed at JRC in the field. 4. ENSEMBLE availability to other communities
As anticipated earlier the ENSEMBLE system has been chosen for the dissemination, and consultation of dispersion forecasts by the EU-25+3 (Switzerland, Romania and Bulgaria) countries. Almost all these countries have an in-house forecasting capacity and are contributing to the ENSEMBLE activities. Some are setting it up, others have no plans in that sense, but consider advantageous to have access to the system for consultation in case of emergency. Other countries outside the EU are also taking active part namely US, CANADA, JP. The future plans are to enlarge even more the community by offering the possibility to take part to the ENSEMBLE activities to any other group that is willing to participate. We are working on two different lines. The first relates to the possibility of using ENSEMBLE as a testbench for new model development or as permanent model inter-comparison platform. In this respect we are coupling the system with all existing long range datasets and even tracer experiment data relating to the mesoscale. This will allow any operational or research group a huge opportunity for evaluation and inter-comparison within a harmonized and user-friendly facility. On top of this R&D application we would welcome any other group that is interested to participate to the normal activities or investigate the possibility of providing dedicated sections to specific communities of models (e.g. RSMC). 5. REFERENCES Galmarini S. (2005). ENSEMBLE for ConVEX-3: Contribution of the ENSEMBLE System to the International Emergency Response Exercise EUR Report 2005 EUR Report 21744 EN. Office for Official Publications of the European Communities, Luxembourg. ISBN 92-894- 9759-9. Galmarini S. et al. (2004a) Ensemble dispersion forecasting, Part I: concept, approach and indicators, Atmospheric Environment, 38, 28, 4607-4617 Galmarini S. et al. (2004b) Ensemble dispersion forecasting, Part II: application and evaluation, Atmospheric Environment, 38, 28, 4619-4632 Galmarini S. et al. (2004c) Can the confidence in long range atmspheric transport models be increased? The pan European experience of ENSEMBLE Rad. prot. Dos., Vol. 109, Nos 1-2, pp. 19-24 Bianconi R., S. Galmarini and R. Bellasio (2004d) Web-based system for decision support in case of emergency: ensemble modelling of long-range atmospheric dispersion of radionuclides J. of Environmental Modelling and Software, 19, 401-411, 2004 Galmarini S. et al. (2001) Forecasting the consequences of accidental releases of radionuclides in the atmosphere from ensemble dispersion modelling. Journal of Environmental Radioactivity, 57, 3, 203-219.